agora inbox for [email protected]
help / color / mirror / Atom feedImplications of multi-byte support in a distribution
275+ messages / 9 participants
[nested] [flat]
* Implications of multi-byte support in a distribution
@ 1999-08-30 21:19 Oliver Elphick <[email protected]>
0 siblings, 2 replies; 275+ messages in thread
From: Oliver Elphick @ 1999-08-30 21:19 UTC (permalink / raw)
To: [email protected]; +Cc: [email protected]
I have had a request to add multi-byte support to the Debian binary
packages of PostgreSQL.
Since I live in England, I have personally no need of this and therefore
have little understanding of the implications.
If I change the packages to use multi-byte support, (UNICODE (UTF-8) is
suggested as the default), will there be any detrimental effects on the
fairly large parts of the world that don't need it? Should I try to
provide two different packages, one with and one without MB support?
--
Vote against SPAM: http://www.politik-digital.de/spam/
========================================
Oliver Elphick [email protected]
Isle of Wight http://www.lfix.co.uk/oliver
PGP key from public servers; key ID 32B8FAA1
========================================
"For what shall it profit a man, if he shall gain the
whole world, and lose his own soul?" Mark 8:36
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-08-31 07:04 Thomas Lockhart <[email protected]>
parent: Oliver Elphick <[email protected]>
1 sibling, 1 reply; 275+ messages in thread
From: Thomas Lockhart @ 1999-08-31 07:04 UTC (permalink / raw)
To: Oliver Elphick <[email protected]>; +Cc: [email protected]; [email protected]
> I have had a request to add multi-byte support to the Debian binary
> packages of PostgreSQL.
> Since I live in England, I have personally no need of this and therefore
> have little understanding of the implications.
> If I change the packages to use multi-byte support, (UNICODE (UTF-8) is
> suggested as the default), will there be any detrimental effects on the
> fairly large parts of the world that don't need it? Should I try to
> provide two different packages, one with and one without MB support?
Probably. The downside to having MB support is reduced performance and
perhaps functionality. If you don't need it, don't build it...
- Thomas
--
Thomas Lockhart [email protected]
South Pasadena, California
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-08-31 08:36 Oleg Broytmann <[email protected]>
parent: Oliver Elphick <[email protected]>
1 sibling, 0 replies; 275+ messages in thread
From: Oleg Broytmann @ 1999-08-31 08:36 UTC (permalink / raw)
To: Oliver Elphick <[email protected]>; +Cc: [email protected]
On Mon, 30 Aug 1999, Oliver Elphick wrote:
> I have had a request to add multi-byte support to the Debian binary
> packages of PostgreSQL.
>
> Since I live in England, I have personally no need of this and therefore
> have little understanding of the implications.
>
> If I change the packages to use multi-byte support, (UNICODE (UTF-8) is
I consider Unicode as a compromise, and as such, it is the worst case. I
don't know anyone who need Unicode directly. Russian users need koi8 and
win1251, Chineese, Japaneese and other folks need their apropriate
encodings (BIG5 and all that).
Don't know what should be reasonable default; in any case installation
script should ask about user preference and run initdb -E with user
encoding to set default.
> suggested as the default), will there be any detrimental effects on the
> fairly large parts of the world that don't need it? Should I try to
> provide two different packages, one with and one without MB support?
But of course. Many people do not want MB support out of distributive.
Suspicious sysadmin should reject such package, if (s)he do not understand
what/where/why MB - and it is right.
Suporting two different packages is hard, but support only MB-enabled
package will led to many demands "please provide smaller/better/faster
PostgreSQL package".
Oleg.
----
Oleg Broytmann http://members.xoom.com/phd2/ [email protected]
Programmers don't die, they just GOSUB without RETURN.
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-08-31 09:29 Tatsuo Ishii <[email protected]>
parent: Thomas Lockhart <[email protected]>
0 siblings, 1 reply; 275+ messages in thread
From: Tatsuo Ishii @ 1999-08-31 09:29 UTC (permalink / raw)
To: Thomas Lockhart <[email protected]>; +Cc: Oliver Elphick <[email protected]>; [email protected]; [email protected]
>> I have had a request to add multi-byte support to the Debian binary
>> packages of PostgreSQL.
>> Since I live in England, I have personally no need of this and therefore
>> have little understanding of the implications.
>> If I change the packages to use multi-byte support, (UNICODE (UTF-8) is
>> suggested as the default), will there be any detrimental effects on the
>> fairly large parts of the world that don't need it? Should I try to
>> provide two different packages, one with and one without MB support?
>
>Probably. The downside to having MB support is reduced performance and
>perhaps functionality. If you don't need it, don't build it...
Not really. I did the regression test with/without multi-byte enabled.
with MB: 2:53:92 elapsed
w/o MB: 2:52.92 elapsed
Perhaps the worst case for MB would be regex ops. If you do a lot of
regex queries, performance degration might not be neglectable.
Load module size:
with MB: 1208542
w/o MB: 1190925
(difference is 17KB)
Talking about the functionality, I don't see any missing feature with
MB comparing w/o MB. (there are some features only MB has. for
example, SET NAMES).
--
Tatsuo Ishii
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-08-31 11:56 Oleg Bartunov <[email protected]>
parent: Tatsuo Ishii <[email protected]>
0 siblings, 1 reply; 275+ messages in thread
From: Oleg Bartunov @ 1999-08-31 11:56 UTC (permalink / raw)
To: Tatsuo Ishii <[email protected]>; +Cc: Thomas Lockhart <[email protected]>; Oliver Elphick <[email protected]>; [email protected]; [email protected]
On Tue, 31 Aug 1999, Tatsuo Ishii wrote:
> Date: Tue, 31 Aug 1999 18:29:21 +0900
> From: Tatsuo Ishii <[email protected]>
> To: Thomas Lockhart <[email protected]>
> Cc: Oliver Elphick <[email protected]>, [email protected],
> [email protected]
> Subject: Re: [HACKERS] Implications of multi-byte support in a distribution
>
> >> I have had a request to add multi-byte support to the Debian binary
> >> packages of PostgreSQL.
> >> Since I live in England, I have personally no need of this and therefore
> >> have little understanding of the implications.
> >> If I change the packages to use multi-byte support, (UNICODE (UTF-8) is
> >> suggested as the default), will there be any detrimental effects on the
> >> fairly large parts of the world that don't need it? Should I try to
> >> provide two different packages, one with and one without MB support?
> >
> >Probably. The downside to having MB support is reduced performance and
> >perhaps functionality. If you don't need it, don't build it...
>
> Not really. I did the regression test with/without multi-byte enabled.
>
> with MB: 2:53:92 elapsed
> w/o MB: 2:52.92 elapsed
>
> Perhaps the worst case for MB would be regex ops. If you do a lot of
> regex queries, performance degration might not be neglectable.
It should be. What would be nice is to have a column-specific
MB support. But I doubt if it's possible.
>
> Load module size:
>
> with MB: 1208542
> w/o MB: 1190925
>
> (difference is 17KB)
>
> Talking about the functionality, I don't see any missing feature with
> MB comparing w/o MB. (there are some features only MB has. for
> example, SET NAMES).
> --
> Tatsuo Ishii
>
> ************
>
_____________________________________________________________
Oleg Bartunov, sci.researcher, hostmaster of AstroNet,
Sternberg Astronomical Institute, Moscow University (Russia)
Internet: [email protected], http://www.sai.msu.su/~megera/
phone: +007(095)939-16-83, +007(095)939-23-83
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-01 02:30 Tatsuo Ishii <[email protected]>
parent: Oleg Bartunov <[email protected]>
0 siblings, 1 reply; 275+ messages in thread
From: Tatsuo Ishii @ 1999-09-01 02:30 UTC (permalink / raw)
To: Oleg Bartunov <[email protected]>; +Cc: Tatsuo Ishii <[email protected]>; Thomas Lockhart <[email protected]>; Oliver Elphick <[email protected]>; [email protected]; [email protected]
>> Perhaps the worst case for MB would be regex ops. If you do a lot of
>> regex queries, performance degration might not be neglectable.
>
>It should be. What would be nice is to have a column-specific
>MB support. But I doubt if it's possible.
That shouldn't be too difficult, if we have an encoding infomation
with each text column or literal. Maybe now is the time to introuce
NCHAR?
BTW, it is interesting that people does not hesitate to enable
with-locale option even if they only use ASCII. I guess the
performance degration by enabling locale is not too small.
--
Tatsuo Ishii
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-01 02:55 Thomas Lockhart <[email protected]>
parent: Tatsuo Ishii <[email protected]>
0 siblings, 3 replies; 275+ messages in thread
From: Thomas Lockhart @ 1999-09-01 02:55 UTC (permalink / raw)
To: [email protected]; +Cc: Oleg Bartunov <[email protected]>; Oliver Elphick <[email protected]>; [email protected]; [email protected]
> That shouldn't be too difficult, if we have an encoding infomation
> with each text column or literal. Maybe now is the time to introuce
> NCHAR?
I've been waiting for a go-ahead from folks who would use it. imho the
way to do it is to use Postgres' type system to implement it, rather
than, for example, encoding "type" information into each string. We
can also define a "default encoding" for each database as a new column
in pg_database...
> BTW, it is interesting that people does not hesitate to enable
> with-locale option even if they only use ASCII. I guess the
> performance degration by enabling locale is not too small.
Red Hat built their RPMs with locale enabled, and there is a
significant performance hit. Implementing NCHAR would be a better
solution, since the user can choose whether to use SQL_TEXT or the
locale-specific character set at run time...
- Thomas
--
Thomas Lockhart [email protected]
South Pasadena, California
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-01 05:48 Oleg Bartunov <[email protected]>
parent: Thomas Lockhart <[email protected]>
2 siblings, 0 replies; 275+ messages in thread
From: Oleg Bartunov @ 1999-09-01 05:48 UTC (permalink / raw)
To: Thomas Lockhart <[email protected]>; +Cc: [email protected]; Oliver Elphick <[email protected]>; [email protected]; [email protected]
On Wed, 1 Sep 1999, Thomas Lockhart wrote:
> Date: Wed, 01 Sep 1999 02:55:48 +0000
> From: Thomas Lockhart <[email protected]>
> To: [email protected]
> Cc: Oleg Bartunov <[email protected]>, Oliver Elphick <[email protected]>,
> [email protected], [email protected]
> Subject: Re: [HACKERS] Implications of multi-byte support in a distribution
>
> > That shouldn't be too difficult, if we have an encoding infomation
> > with each text column or literal. Maybe now is the time to introuce
> > NCHAR?
Yes, postgres after 6.5 and especially recent win becomes very popular
and additional performance hit would be very in time. Does implementing
of NCHAR only could solve all problem with text, varchar etc ?
>
> I've been waiting for a go-ahead from folks who would use it. imho the
> way to do it is to use Postgres' type system to implement it, rather
> than, for example, encoding "type" information into each string. We
> can also define a "default encoding" for each database as a new column
> in pg_database...
go-ahead, Tom :-) I would use it.
>
> > BTW, it is interesting that people does not hesitate to enable
> > with-locale option even if they only use ASCII. I guess the
> > performance degration by enabling locale is not too small.
>
> Red Hat built their RPMs with locale enabled, and there is a
> significant performance hit. Implementing NCHAR would be a better
> solution, since the user can choose whether to use SQL_TEXT or the
> locale-specific character set at run time...
>
> - Thomas
>
> --
> Thomas Lockhart [email protected]
> South Pasadena, California
>
> ************
>
_____________________________________________________________
Oleg Bartunov, sci.researcher, hostmaster of AstroNet,
Sternberg Astronomical Institute, Moscow University (Russia)
Internet: [email protected], http://www.sai.msu.su/~megera/
phone: +007(095)939-16-83, +007(095)939-23-83
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-01 11:12 Milan Zamazal <[email protected]>
parent: Thomas Lockhart <[email protected]>
2 siblings, 0 replies; 275+ messages in thread
From: Milan Zamazal @ 1999-09-01 11:12 UTC (permalink / raw)
To: [email protected]
>>>>> "TL" == Thomas Lockhart <[email protected]> writes:
>> That shouldn't be too difficult, if we have an encoding
>> infomation with each text column or literal. Maybe now is the
>> time to introuce NCHAR?
TL> I've been waiting for a go-ahead from folks who would use
TL> it. imho the way to do it is to use Postgres' type system to
TL> implement it, rather than, for example, encoding "type"
TL> information into each string. We can also define a "default
TL> encoding" for each database as a new column in pg_database...
What about sorting? Would it be possible to solve it in similar way?
If I'm not mistaken, there is currently no good way to use two different
kinds of sorting for one postmaster instance?
Milan Zamazal
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-01 11:17 Milan Zamazal <[email protected]>
parent: Thomas Lockhart <[email protected]>
2 siblings, 1 reply; 275+ messages in thread
From: Milan Zamazal @ 1999-09-01 11:17 UTC (permalink / raw)
To: [email protected]
>>>>> "TL" == Thomas Lockhart <[email protected]> writes:
>> That shouldn't be too difficult, if we have an encoding
>> infomation with each text column or literal. Maybe now is the
>> time to introuce NCHAR?
TL> I've been waiting for a go-ahead from folks who would use
TL> it. imho the way to do it is to use Postgres' type system to
TL> implement it, rather than, for example, encoding "type"
TL> information into each string. We can also define a "default
TL> encoding" for each database as a new column in pg_database...
What about sorting? Would it be possible to solve it in similar way?
If I'm not mistaken, there is currently no good way to use two different
kinds of sorting for one postmaster instance?
Milan Zamazal
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-02 05:25 Thomas Lockhart <[email protected]>
parent: Milan Zamazal <[email protected]>
0 siblings, 1 reply; 275+ messages in thread
From: Thomas Lockhart @ 1999-09-02 05:25 UTC (permalink / raw)
To: Milan Zamazal <[email protected]>; +Cc: [email protected]
> >> That shouldn't be too difficult, if we have an encoding
> >> infomation with each text column or literal. Maybe now is the
> >> time to introuce NCHAR?
> TL> I've been waiting for a go-ahead from folks who would use
> TL> it. imho the way to do it is to use Postgres' type system to
> TL> implement it, rather than, for example, encoding "type"
> TL> information into each string. We can also define a "default
> TL> encoding" for each database as a new column in pg_database...
> What about sorting? Would it be possible to solve it in similar way?
> If I'm not mistaken, there is currently no good way to use two different
> kinds of sorting for one postmaster instance?
Each encoding/character set can behave however you want. You can reuse
collation and sorting code from another character set, or define a
unique one.
- Thomas
--
Thomas Lockhart [email protected]
South Pasadena, California
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-02 06:52 Hannu Krosing <[email protected]>
parent: Thomas Lockhart <[email protected]>
0 siblings, 1 reply; 275+ messages in thread
From: Hannu Krosing @ 1999-09-02 06:52 UTC (permalink / raw)
To: Thomas Lockhart <[email protected]>; +Cc: Milan Zamazal <[email protected]>; [email protected]
Thomas Lockhart wrote:
>
> > >> That shouldn't be too difficult, if we have an encoding
> > >> infomation with each text column or literal. Maybe now is the
> > >> time to introuce NCHAR?
> > TL> I've been waiting for a go-ahead from folks who would use
> > TL> it. imho the way to do it is to use Postgres' type system to
> > TL> implement it, rather than, for example, encoding "type"
> > TL> information into each string. We can also define a "default
> > TL> encoding" for each database as a new column in pg_database...
> > What about sorting? Would it be possible to solve it in similar way?
> > If I'm not mistaken, there is currently no good way to use two different
> > kinds of sorting for one postmaster instance?
>
> Each encoding/character set can behave however you want. You can reuse
> collation and sorting code from another character set, or define a
> unique one.
Is it really inside one postmaster instance ?
If so, then is the character encoding defined at the create table /
create index
process (maybe even separately for each field ?) or can I specify it
when sort'ing ?
-----------------
Hannu
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-02 15:03 Thomas Lockhart <[email protected]>
parent: Hannu Krosing <[email protected]>
0 siblings, 1 reply; 275+ messages in thread
From: Thomas Lockhart @ 1999-09-02 15:03 UTC (permalink / raw)
To: Hannu Krosing <[email protected]>; +Cc: Milan Zamazal <[email protected]>; [email protected]
> > Each encoding/character set can behave however you want. You can reuse
> > collation and sorting code from another character set, or define a
> > unique one.
> Is it really inside one postmaster instance ?
> If so, then is the character encoding defined at the create table /
> create index process (maybe even separately for each field ?) or can I
> specify it when sort'ing ?
Yes, yes, and yes ;)
I would propose that we implement the explicit collation features of
SQL92 using implicit type conversion. So if you want to use a
different sorting order on a *compatible* character set, then (looking
up in Date and Darwen for the syntax...):
'test string' COLLATE CASE_INSENSITIVITY
becomes internally
case_insensitivity('test string'::text)
and
c1 < c2 COLLATE CASE_INSENSITIVITY
becomes
case_insensitivity(c1) < case_insensitivity(c2)
- Thomas
--
Thomas Lockhart [email protected]
South Pasadena, California
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-03 00:55 Tatsuo Ishii <[email protected]>
parent: Thomas Lockhart <[email protected]>
0 siblings, 1 reply; 275+ messages in thread
From: Tatsuo Ishii @ 1999-09-03 00:55 UTC (permalink / raw)
To: Thomas Lockhart <[email protected]>; +Cc: Hannu Krosing <[email protected]>; Milan Zamazal <[email protected]>; [email protected]
> > > Each encoding/character set can behave however you want. You can reuse
> > > collation and sorting code from another character set, or define a
> > > unique one.
> > Is it really inside one postmaster instance ?
> > If so, then is the character encoding defined at the create table /
> > create index process (maybe even separately for each field ?) or can I
> > specify it when sort'ing ?
>
> Yes, yes, and yes ;)
But we can't avoid calling strcoll() and some other codes surrounded
by #ifdef LOCALE? I think he actually wants is to define his own
collation *and* not to use locale if the column is ASCII only.
> I would propose that we implement the explicit collation features of
> SQL92 using implicit type conversion. So if you want to use a
> different sorting order on a *compatible* character set, then (looking
> up in Date and Darwen for the syntax...):
>
> 'test string' COLLATE CASE_INSENSITIVITY
>
> becomes internally
>
> case_insensitivity('test string'::text)
>
> and
>
> c1 < c2 COLLATE CASE_INSENSITIVITY
>
> becomes
>
> case_insensitivity(c1) < case_insensitivity(c2)
This idea seems great and elegant. Ok, what about throwing away #ifdef
LOCALE? Same thing can be obtained by defining a special callation
LOCALE_AWARE. This seems much more consistent for me. Or even better,
we could explicitly have predefined COLLATION for each language (these
can be automatically generated from existing locale data). This would
avoid some platform specific locale problems.
---
Tatsuo Ishii
^ permalink raw reply [nested|flat] 275+ messages in thread
* Re: [HACKERS] Implications of multi-byte support in a distribution
@ 1999-09-03 01:45 Thomas Lockhart <[email protected]>
parent: Tatsuo Ishii <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Thomas Lockhart @ 1999-09-03 01:45 UTC (permalink / raw)
To: [email protected]; +Cc: Hannu Krosing <[email protected]>; Milan Zamazal <[email protected]>; [email protected]
> But we can't avoid calling strcoll() and some other codes surrounded
> by #ifdef LOCALE? I think he actually wants is to define his own
> collation *and* not to use locale if the column is ASCII only.
Right. But there would be a fundamental character type which is *not*
locale-aware, and there is another type (perhaps/probably NCHAR?)
which is...
> Ok, what about throwing away #ifdef
> LOCALE? Same thing can be obtained by defining a special callation
> LOCALE_AWARE.
Or moving the locale-aware stuff to a formal NCHAR implementation.
istm (and to Date and Darwen ;) that there is a tighter relationship
between collations, character repertoires, and character sets than
might be inferred from the SQL92-defined capabilities.
> This seems much more consistent for me. Or even better,
> we could explicitly have predefined COLLATION for each language (these
> can be automatically generated from existing locale data). This would
> avoid some platform specific locale problems.
Right. We may already have some of this with the "implicit type
coersion" conventions I introduced in the v6.4 release.
- Thomas
--
Thomas Lockhart [email protected]
South Pasadena, California
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
* [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms
@ 2023-01-17 04:38 Andres Freund <[email protected]>
0 siblings, 0 replies; 275+ messages in thread
From: Andres Freund @ 2023-01-17 04:38 UTC (permalink / raw)
Until now we used struct timespec on all platforms but windows. Using struct
timespe causes a fair bit of memory (struct timeval is 16 bytes) and runtime
overhead (much more complicated additions). Instead we can convert the time to
nanoseconds in INSTR_TIME_SET_CURRENT(), making the remaining operations
cheaper.
Representing time as int64 nanoseconds provides sufficient range, ~292 years
relative to a starting point (depending on clock source, relative to the unix
epoch or the system's boot time). That'd not be sufficient for calendar time
stored on disk, but is plenty for runtime interval time measurement.
On windows instr_time already is represented as cycles. It might make sense to
represent time as cycles on other platforms as well, as using cycle
acquisition instructions like rdtsc directly can reduce the overhead of time
acquisition substantially. This could be done in a fairly localized manner as
the code stands after this commit.
Because the windows and non-windows paths are now more similar, use a common
set of macros. To make that possible, most of the use of LARGE_INTEGER had to
be removed, which looks nicer anyway.
To avoid users of the API relying on the integer representation, we wrap the
64bit integer inside struct struct instr_time.
Author: Andres Freund <[email protected]>
Author: Lukas Fittl <[email protected]>
Author: David Geier <[email protected]>
Reviewed-by: Tom Lane <[email protected]>
Discussion: https://postgr.es/m/[email protected]
---
src/include/portability/instr_time.h | 162 ++++++++++++++-------------
1 file changed, 86 insertions(+), 76 deletions(-)
diff --git a/src/include/portability/instr_time.h b/src/include/portability/instr_time.h
index 9ea1a68bd94..c0ed491395d 100644
--- a/src/include/portability/instr_time.h
+++ b/src/include/portability/instr_time.h
@@ -34,6 +34,8 @@
*
* INSTR_TIME_GET_MICROSEC(t) convert t to uint64 (in microseconds)
*
+ * INSTR_TIME_GET_NANOSEC(t) convert t to uint64 (in nanoseconds)
+ *
* Note that INSTR_TIME_SUBTRACT and INSTR_TIME_ACCUM_DIFF convert
* absolute times to intervals. The INSTR_TIME_GET_xxx operations are
* only useful on intervals.
@@ -54,8 +56,32 @@
#ifndef INSTR_TIME_H
#define INSTR_TIME_H
+
+/*
+ * We store interval times as an int64 integer on all platforms, as int64 is
+ * cheap to add/subtract, the most common operation for instr_time. The
+ * acquisition of time and converting to specific units of time is platform
+ * specific.
+ *
+ * To avoid users of the API relying on the integer representation, we wrap
+ * the 64bit integer in a struct.
+ */
+typedef struct instr_time
+{
+ int64 ticks; /* in platforms specific unit */
+} instr_time;
+
+
+/* helpers macros used in platform specific code below */
+
+#define NS_PER_S INT64CONST(1000000000)
+#define NS_PER_MS INT64CONST(1000000)
+#define NS_PER_US INT64CONST(1000)
+
+
#ifndef WIN32
+
/* Use clock_gettime() */
#include <time.h>
@@ -80,93 +106,43 @@
#define PG_INSTR_CLOCK CLOCK_REALTIME
#endif
-typedef struct timespec instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_clock_gettime_ns(void)
+{
+ instr_time now;
+ struct timespec tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).tv_nsec == 0 && (t).tv_sec == 0)
+ clock_gettime(PG_INSTR_CLOCK, &tmp);
+ now.ticks = tmp.tv_sec * NS_PER_S + tmp.tv_nsec;
-#define INSTR_TIME_SET_ZERO(t) ((t).tv_sec = 0, (t).tv_nsec = 0)
+ return now;
+}
-#define INSTR_TIME_SET_CURRENT(t) ((void) clock_gettime(PG_INSTR_CLOCK, &(t)))
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_clock_gettime_ns())
-#define INSTR_TIME_ADD(x,y) \
- do { \
- (x).tv_sec += (y).tv_sec; \
- (x).tv_nsec += (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (t).ticks)
-#define INSTR_TIME_SUBTRACT(x,y) \
- do { \
- (x).tv_sec -= (y).tv_sec; \
- (x).tv_nsec -= (y).tv_nsec; \
- /* Normalize */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- } while (0)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- do { \
- (x).tv_sec += (y).tv_sec - (z).tv_sec; \
- (x).tv_nsec += (y).tv_nsec - (z).tv_nsec; \
- /* Normalize after each add to avoid overflow/underflow of tv_nsec */ \
- while ((x).tv_nsec < 0) \
- { \
- (x).tv_nsec += 1000000000; \
- (x).tv_sec--; \
- } \
- while ((x).tv_nsec >= 1000000000) \
- { \
- (x).tv_nsec -= 1000000000; \
- (x).tv_sec++; \
- } \
- } while (0)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).tv_sec) + ((double) (t).tv_nsec) / 1000000000.0)
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).tv_sec * 1000.0) + ((double) (t).tv_nsec) / 1000000.0)
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- (((uint64) (t).tv_sec * (uint64) 1000000) + (uint64) ((t).tv_nsec / 1000))
#else /* WIN32 */
+
/* Use QueryPerformanceCounter() */
-typedef LARGE_INTEGER instr_time;
+/* helper for INSTR_TIME_SET_CURRENT */
+static inline instr_time
+pg_query_performance_counter(void)
+{
+ instr_time now;
+ LARGE_INTEGER tmp;
-#define INSTR_TIME_IS_ZERO(t) ((t).QuadPart == 0)
+ QueryPerformanceCounter(&tmp);
+ now.ticks = tmp.QuadPart;
-#define INSTR_TIME_SET_ZERO(t) ((t).QuadPart = 0)
-
-#define INSTR_TIME_SET_CURRENT(t) QueryPerformanceCounter(&(t))
-
-#define INSTR_TIME_ADD(x,y) \
- ((x).QuadPart += (y).QuadPart)
-
-#define INSTR_TIME_SUBTRACT(x,y) \
- ((x).QuadPart -= (y).QuadPart)
-
-#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
- ((x).QuadPart += (y).QuadPart - (z).QuadPart)
-
-#define INSTR_TIME_GET_DOUBLE(t) \
- (((double) (t).QuadPart) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MILLISEC(t) \
- (((double) (t).QuadPart * 1000.0) / GetTimerFrequency())
-
-#define INSTR_TIME_GET_MICROSEC(t) \
- ((uint64) (((double) (t).QuadPart * 1000000.0) / GetTimerFrequency()))
+ return now;
+}
static inline double
GetTimerFrequency(void)
@@ -177,11 +153,45 @@ GetTimerFrequency(void)
return (double) f.QuadPart;
}
+#define INSTR_TIME_SET_CURRENT(t) \
+ ((t) = pg_query_performance_counter())
+
+#define INSTR_TIME_GET_NANOSEC(t) \
+ ((int64) (((double) (t).ticks * NS_PER_S) / GetTimerFrequency()))
+
#endif /* WIN32 */
-/* same macro on all platforms */
+
+/*
+ * Common macros
+ */
+
+#define INSTR_TIME_IS_ZERO(t) ((t).ticks == 0)
+
+
+#define INSTR_TIME_SET_ZERO(t) ((t).ticks = 0)
#define INSTR_TIME_SET_CURRENT_LAZY(t) \
(INSTR_TIME_IS_ZERO(t) ? INSTR_TIME_SET_CURRENT(t), true : false)
+
+#define INSTR_TIME_ADD(x,y) \
+ ((x).ticks += (y).ticks)
+
+#define INSTR_TIME_SUBTRACT(x,y) \
+ ((x).ticks -= (y).ticks)
+
+#define INSTR_TIME_ACCUM_DIFF(x,y,z) \
+ ((x).ticks += (y).ticks - (z).ticks)
+
+
+#define INSTR_TIME_GET_DOUBLE(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_S)
+
+#define INSTR_TIME_GET_MILLISEC(t) \
+ ((double) INSTR_TIME_GET_NANOSEC(t) / NS_PER_MS)
+
+#define INSTR_TIME_GET_MICROSEC(t) \
+ (INSTR_TIME_GET_NANOSEC(t) / NS_PER_US)
+
#endif /* INSTR_TIME_H */
--
2.38.0
--vt42kbp2j7rjcapp
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
filename="v8-0003-instr_time-Add-INSTR_TIME_SET_SECONDS-INSTR_TIME_.patch"
^ permalink raw reply [nested|flat] 275+ messages in thread
end of thread, other threads:[~2023-01-17 04:38 UTC | newest]
Thread overview: 275+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
1999-08-30 21:19 Implications of multi-byte support in a distribution Oliver Elphick <[email protected]>
1999-08-31 07:04 ` Thomas Lockhart <[email protected]>
1999-08-31 09:29 ` Tatsuo Ishii <[email protected]>
1999-08-31 11:56 ` Oleg Bartunov <[email protected]>
1999-09-01 02:30 ` Tatsuo Ishii <[email protected]>
1999-09-01 02:55 ` Thomas Lockhart <[email protected]>
1999-09-01 05:48 ` Oleg Bartunov <[email protected]>
1999-09-01 11:12 ` Milan Zamazal <[email protected]>
1999-09-01 11:17 ` Milan Zamazal <[email protected]>
1999-09-02 05:25 ` Thomas Lockhart <[email protected]>
1999-09-02 06:52 ` Hannu Krosing <[email protected]>
1999-09-02 15:03 ` Thomas Lockhart <[email protected]>
1999-09-03 00:55 ` Tatsuo Ishii <[email protected]>
1999-09-03 01:45 ` Thomas Lockhart <[email protected]>
1999-08-31 08:36 ` Oleg Broytmann <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
2023-01-17 04:38 [PATCH v8 2/5] instr_time: Represent time as an int64 on all platforms Andres Freund <[email protected]>
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox