public inbox for [email protected]  
help / color / mirror / Atom feed
[PATCH 1/2] Allow TAP test to excecise tablespace.
15+ messages / 5 participants
[nested] [flat]

* [PATCH 1/2] Allow TAP test to excecise tablespace.
@ 2019-04-22 11:10  Kyotaro Horiguchi <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Kyotaro Horiguchi @ 2019-04-22 11:10 UTC (permalink / raw)

To perform tablespace related checks, this patch lets
PostgresNode::backup have a new parameter "tablespace_mapping", and
make init_from_backup handle capable to handle a backup created using
tablespace_mapping.
---
 src/test/perl/PostgresNode.pm  | 10 ++++++++--
 src/test/perl/RecursiveCopy.pm | 33 +++++++++++++++++++++++++++++----
 2 files changed, 37 insertions(+), 6 deletions(-)

diff --git a/src/test/perl/PostgresNode.pm b/src/test/perl/PostgresNode.pm
index 76874141c5..59a939821d 100644
--- a/src/test/perl/PostgresNode.pm
+++ b/src/test/perl/PostgresNode.pm
@@ -540,13 +540,19 @@ target server since it isn't done by default.
 
 sub backup
 {
-	my ($self, $backup_name) = @_;
+	my ($self, $backup_name, %params) = @_;
 	my $backup_path = $self->backup_dir . '/' . $backup_name;
 	my $name        = $self->name;
+	my @rest = ();
+
+	if (defined $params{tablespace_mapping})
+	{
+		push(@rest, "--tablespace-mapping=$params{tablespace_mapping}");
+	}
 
 	print "# Taking pg_basebackup $backup_name from node \"$name\"\n";
 	TestLib::system_or_bail('pg_basebackup', '-D', $backup_path, '-h',
-		$self->host, '-p', $self->port, '--no-sync');
+		$self->host, '-p', $self->port, '--no-sync', @rest);
 	print "# Backup finished\n";
 	return;
 }
diff --git a/src/test/perl/RecursiveCopy.pm b/src/test/perl/RecursiveCopy.pm
index baf5d0ac63..c912ce412d 100644
--- a/src/test/perl/RecursiveCopy.pm
+++ b/src/test/perl/RecursiveCopy.pm
@@ -22,6 +22,7 @@ use warnings;
 use Carp;
 use File::Basename;
 use File::Copy;
+use TestLib;
 
 =pod
 
@@ -97,14 +98,38 @@ sub _copypath_recurse
 	# invoke the filter and skip all further operation if it returns false
 	return 1 unless &$filterfn($curr_path);
 
-	# Check for symlink -- needed only on source dir
-	# (note: this will fall through quietly if file is already gone)
-	croak "Cannot operate on symlink \"$srcpath\"" if -l $srcpath;
-
 	# Abort if destination path already exists.  Should we allow directories
 	# to exist already?
 	croak "Destination path \"$destpath\" already exists" if -e $destpath;
 
+	# Check for symlink -- needed only on source dir
+	# (note: this will fall through quietly if file is already gone)
+	if (-l $srcpath)
+	{
+		croak "Cannot operate on symlink \"$srcpath\""
+		  if ($srcpath !~ /\/(pg_tblspc\/[0-9]+)$/);
+
+		# We have mapped tablespaces. Copy them individually
+		my $linkname = $1;
+		my $tmpdir = TestLib::tempdir;
+		my $dstrealdir = TestLib::real_dir($tmpdir);
+		my $srcrealdir = readlink($srcpath);
+
+		opendir(my $dh, $srcrealdir);
+		while (readdir $dh)
+		{
+			next if (/^\.\.?$/);
+			my $spath = "$srcrealdir/$_";
+			my $dpath = "$dstrealdir/$_";
+
+			copypath($spath, $dpath);
+		}
+		closedir $dh;
+
+		symlink $dstrealdir, $destpath;
+		return 1;
+	}
+
 	# If this source path is a file, simply copy it to destination with the
 	# same name and we're done.
 	if (-f $srcpath)
-- 
2.16.3


----Next_Part(Mon_Apr_22_21_19_33_2019_510)--
Content-Type: Text/X-Patch; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="0002-Add-check-for-recovery-failure-caused-by-tablespace.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* [PATCH 2/4] Allow TAP test to excecise tablespace.
@ 2019-04-22 11:58  Kyotaro Horiguchi <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Kyotaro Horiguchi @ 2019-04-22 11:58 UTC (permalink / raw)

To perform tablespace related checks, this patch lets
PostgresNode::backup have a new parameter "tablespace_mapping", and
make init_from_backup handle capable to handle a backup created using
tablespace_mapping.
---
 src/test/perl/PostgresNode.pm  | 45 +++++++++++++++++++++++++++++++++++++-----
 src/test/perl/RecursiveCopy.pm | 38 +++++++++++++++++++++++++++++++----
 2 files changed, 74 insertions(+), 9 deletions(-)

diff --git a/src/test/perl/PostgresNode.pm b/src/test/perl/PostgresNode.pm
index 76874141c5..e951acc461 100644
--- a/src/test/perl/PostgresNode.pm
+++ b/src/test/perl/PostgresNode.pm
@@ -157,6 +157,7 @@ sub new
 		_host    => $pghost,
 		_basedir => "$TestLib::tmp_check/t_${testname}_${name}_data",
 		_name    => $name,
+		_tablespaces => [],
 		_logfile_generation => 0,
 		_logfile_base       => "$TestLib::log_path/${testname}_${name}",
 		_logfile            => "$TestLib::log_path/${testname}_${name}.log"
@@ -342,6 +343,26 @@ sub backup_dir
 
 =pod
 
+=item $node->make_tablespace_dir()
+
+make a tablespace directory 
+
+=cut
+
+sub make_tablespace_dir
+{
+	my ($self, $name) = @_;
+	my $basedir = $self->basedir;
+
+	die "tablespace name contains '/'" if ($name =~ m#/#);
+	my $reldir = "../$name";
+	mkdir $self->data_dir() . "/". $reldir;
+	push($self->{_tablespaces}, $reldir);
+	return $reldir;
+}
+
+=pod
+
 =item $node->info()
 
 Return a string containing human-readable diagnostic information (paths, etc)
@@ -540,13 +561,27 @@ target server since it isn't done by default.
 
 sub backup
 {
-	my ($self, $backup_name) = @_;
-	my $backup_path = $self->backup_dir . '/' . $backup_name;
+	my ($self, $backup_name, %params) = @_;
+	my $backup_path = $self->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $name        = $self->name;
+	my @rest = ();
+
+	if (defined $params{tablespace_mapping})
+	{
+		foreach my $p (split /,/, $params{tablespace_mapping})
+		{
+			push(@rest, "--tablespace-mapping=$p");
+		}
+	}
+
+	foreach my $p ($self->{_tablespaces})
+	{
+		mkdir "$backup_path/$p";
+	}
 
 	print "# Taking pg_basebackup $backup_name from node \"$name\"\n";
 	TestLib::system_or_bail('pg_basebackup', '-D', $backup_path, '-h',
-		$self->host, '-p', $self->port, '--no-sync');
+		$self->host, '-p', $self->port, '--no-sync', @rest);
 	print "# Backup finished\n";
 	return;
 }
@@ -592,7 +627,7 @@ sub backup_fs_cold
 sub _backup_fs
 {
 	my ($self, $backup_name, $hot) = @_;
-	my $backup_path = $self->backup_dir . '/' . $backup_name;
+	my $backup_path = $self->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $port        = $self->port;
 	my $name        = $self->name;
 
@@ -655,7 +690,7 @@ unconditionally set to enable replication connections.
 sub init_from_backup
 {
 	my ($self, $root_node, $backup_name, %params) = @_;
-	my $backup_path = $root_node->backup_dir . '/' . $backup_name;
+	my $backup_path = $root_node->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $host        = $self->host;
 	my $port        = $self->port;
 	my $node_name   = $self->name;
diff --git a/src/test/perl/RecursiveCopy.pm b/src/test/perl/RecursiveCopy.pm
index baf5d0ac63..f165db7348 100644
--- a/src/test/perl/RecursiveCopy.pm
+++ b/src/test/perl/RecursiveCopy.pm
@@ -22,6 +22,7 @@ use warnings;
 use Carp;
 use File::Basename;
 use File::Copy;
+use TestLib;
 
 =pod
 
@@ -97,14 +98,43 @@ sub _copypath_recurse
 	# invoke the filter and skip all further operation if it returns false
 	return 1 unless &$filterfn($curr_path);
 
-	# Check for symlink -- needed only on source dir
-	# (note: this will fall through quietly if file is already gone)
-	croak "Cannot operate on symlink \"$srcpath\"" if -l $srcpath;
-
 	# Abort if destination path already exists.  Should we allow directories
 	# to exist already?
 	croak "Destination path \"$destpath\" already exists" if -e $destpath;
 
+	# Check for symlink -- needed only on source dir
+	# (note: this will fall through quietly if file is already gone)
+	if (-l $srcpath)
+	{
+		croak "Cannot operate on symlink \"$srcpath\""
+		  if ($srcpath !~ /\/(pg_tblspc\/[0-9]+)$/);
+
+		# We have mapped tablespaces. Copy them individually
+		my $linkname = $1;
+		my $rpath = my $target = readlink($srcpath);
+
+		#convert to pgdata-based if relative
+		$rpath =~ s#^\.\./##; 
+
+		my $srcrealdir = "$base_src_dir/$rpath";
+		my $dstrealdir = "$base_dest_dir/$rpath";
+
+		mkdir $dstrealdir;
+		opendir(my $dh, $srcrealdir);
+		while (readdir $dh)
+		{
+			next if (/^\.\.?$/);
+			my $spath = "$srcrealdir/$_";
+			my $dpath = "$dstrealdir/$_";
+
+			copypath($spath, $dpath);
+		}
+		closedir $dh;
+
+		symlink $target, $destpath;
+		return 1;
+	}
+
 	# If this source path is a file, simply copy it to destination with the
 	# same name and we're done.
 	if (-f $srcpath)
-- 
2.16.3


----Next_Part(Wed_Apr_24_13_18_45_2019_938)--
Content-Type: Text/X-Patch; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="0003-Add-check-for-recovery-failure-caused-by-tablespace.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* [PATCH 2/5] Allow TAP test to excecise tablespace.
@ 2019-04-22 11:58  Kyotaro Horiguchi <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Kyotaro Horiguchi @ 2019-04-22 11:58 UTC (permalink / raw)

To perform tablespace related checks, this patch lets
PostgresNode::backup have a new parameter "tablespace_mapping", and
make init_from_backup handle capable to handle a backup created using
tablespace_mapping.
---
 src/test/perl/PostgresNode.pm  | 45 +++++++++++++++++++++++++++++++++++++-----
 src/test/perl/RecursiveCopy.pm | 38 +++++++++++++++++++++++++++++++----
 2 files changed, 74 insertions(+), 9 deletions(-)

diff --git a/src/test/perl/PostgresNode.pm b/src/test/perl/PostgresNode.pm
index 76874141c5..e951acc461 100644
--- a/src/test/perl/PostgresNode.pm
+++ b/src/test/perl/PostgresNode.pm
@@ -157,6 +157,7 @@ sub new
 		_host    => $pghost,
 		_basedir => "$TestLib::tmp_check/t_${testname}_${name}_data",
 		_name    => $name,
+		_tablespaces => [],
 		_logfile_generation => 0,
 		_logfile_base       => "$TestLib::log_path/${testname}_${name}",
 		_logfile            => "$TestLib::log_path/${testname}_${name}.log"
@@ -342,6 +343,26 @@ sub backup_dir
 
 =pod
 
+=item $node->make_tablespace_dir()
+
+make a tablespace directory 
+
+=cut
+
+sub make_tablespace_dir
+{
+	my ($self, $name) = @_;
+	my $basedir = $self->basedir;
+
+	die "tablespace name contains '/'" if ($name =~ m#/#);
+	my $reldir = "../$name";
+	mkdir $self->data_dir() . "/". $reldir;
+	push($self->{_tablespaces}, $reldir);
+	return $reldir;
+}
+
+=pod
+
 =item $node->info()
 
 Return a string containing human-readable diagnostic information (paths, etc)
@@ -540,13 +561,27 @@ target server since it isn't done by default.
 
 sub backup
 {
-	my ($self, $backup_name) = @_;
-	my $backup_path = $self->backup_dir . '/' . $backup_name;
+	my ($self, $backup_name, %params) = @_;
+	my $backup_path = $self->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $name        = $self->name;
+	my @rest = ();
+
+	if (defined $params{tablespace_mapping})
+	{
+		foreach my $p (split /,/, $params{tablespace_mapping})
+		{
+			push(@rest, "--tablespace-mapping=$p");
+		}
+	}
+
+	foreach my $p ($self->{_tablespaces})
+	{
+		mkdir "$backup_path/$p";
+	}
 
 	print "# Taking pg_basebackup $backup_name from node \"$name\"\n";
 	TestLib::system_or_bail('pg_basebackup', '-D', $backup_path, '-h',
-		$self->host, '-p', $self->port, '--no-sync');
+		$self->host, '-p', $self->port, '--no-sync', @rest);
 	print "# Backup finished\n";
 	return;
 }
@@ -592,7 +627,7 @@ sub backup_fs_cold
 sub _backup_fs
 {
 	my ($self, $backup_name, $hot) = @_;
-	my $backup_path = $self->backup_dir . '/' . $backup_name;
+	my $backup_path = $self->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $port        = $self->port;
 	my $name        = $self->name;
 
@@ -655,7 +690,7 @@ unconditionally set to enable replication connections.
 sub init_from_backup
 {
 	my ($self, $root_node, $backup_name, %params) = @_;
-	my $backup_path = $root_node->backup_dir . '/' . $backup_name;
+	my $backup_path = $root_node->backup_dir . '/' . $backup_name . '/' . "pgdata";
 	my $host        = $self->host;
 	my $port        = $self->port;
 	my $node_name   = $self->name;
diff --git a/src/test/perl/RecursiveCopy.pm b/src/test/perl/RecursiveCopy.pm
index baf5d0ac63..f165db7348 100644
--- a/src/test/perl/RecursiveCopy.pm
+++ b/src/test/perl/RecursiveCopy.pm
@@ -22,6 +22,7 @@ use warnings;
 use Carp;
 use File::Basename;
 use File::Copy;
+use TestLib;
 
 =pod
 
@@ -97,14 +98,43 @@ sub _copypath_recurse
 	# invoke the filter and skip all further operation if it returns false
 	return 1 unless &$filterfn($curr_path);
 
-	# Check for symlink -- needed only on source dir
-	# (note: this will fall through quietly if file is already gone)
-	croak "Cannot operate on symlink \"$srcpath\"" if -l $srcpath;
-
 	# Abort if destination path already exists.  Should we allow directories
 	# to exist already?
 	croak "Destination path \"$destpath\" already exists" if -e $destpath;
 
+	# Check for symlink -- needed only on source dir
+	# (note: this will fall through quietly if file is already gone)
+	if (-l $srcpath)
+	{
+		croak "Cannot operate on symlink \"$srcpath\""
+		  if ($srcpath !~ /\/(pg_tblspc\/[0-9]+)$/);
+
+		# We have mapped tablespaces. Copy them individually
+		my $linkname = $1;
+		my $rpath = my $target = readlink($srcpath);
+
+		#convert to pgdata-based if relative
+		$rpath =~ s#^\.\./##; 
+
+		my $srcrealdir = "$base_src_dir/$rpath";
+		my $dstrealdir = "$base_dest_dir/$rpath";
+
+		mkdir $dstrealdir;
+		opendir(my $dh, $srcrealdir);
+		while (readdir $dh)
+		{
+			next if (/^\.\.?$/);
+			my $spath = "$srcrealdir/$_";
+			my $dpath = "$dstrealdir/$_";
+
+			copypath($spath, $dpath);
+		}
+		closedir $dh;
+
+		symlink $target, $destpath;
+		return 1;
+	}
+
 	# If this source path is a file, simply copy it to destination with the
 	# same name and we're done.
 	if (-f $srcpath)
-- 
2.16.3


----Next_Part(Wed_Apr_24_17_02_28_2019_644)--
Content-Type: Text/X-Patch; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline;
 filename="v2-0003-Add-check-for-recovery-failure-caused-by-tablespace.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* Re: Remove source code display from \df+?
@ 2023-01-23 01:23  Isaac Morland <[email protected]>
  0 siblings, 2 replies; 15+ messages in thread

From: Isaac Morland @ 2023-01-23 01:23 UTC (permalink / raw)
  To: Justin Pryzby <[email protected]>; +Cc: Tom Lane <[email protected]>; Alvaro Herrera <[email protected]>; Pavel Stehule <[email protected]>; Magnus Hagander <[email protected]>; pgsql-hackers

On Sun, 22 Jan 2023 at 17:27, Justin Pryzby <[email protected]> wrote:

> On Sun, Jan 22, 2023 at 04:28:21PM -0500, Isaac Morland wrote:
> > On Sun, 22 Jan 2023 at 15:04, Tom Lane <[email protected]> wrote:


> But now I'm having a problem I don't understand: the CI are still
> failling,
> > but not in the psql test. Instead, I get this:
> >
> > [20:11:17.624] +++ tap check in src/bin/pg_upgrade +++
>
> You'll find the diff in the "artifacts", but not a separate "diff" file.
>
> https://api.cirrus-ci.com/v1/artifact/task/6146418377752576/testrun/build/testrun/pg_upgrade/002_pg_...
>
>  CREATE FUNCTION public.regress_psql_df_sql() RETURNS void
>      LANGUAGE sql
>      BEGIN ATOMIC
> - SELECT NULL::text;
> + SELECT NULL::text AS text;
>  END;
>
> It's failing because after restoring the function, the column is named
> "text" - maybe it's a bug.
>

OK, thanks. I'd say I've uncovered a bug related to pg_upgrade, unless I’m
missing something. However, I've adjusted my patch so that nothing it
creates is kept. This seems tidier even without the test failure.

Tom's earlier point was that neither the function nor its owner needs to
> be preserved (as is done to exercise pg_dump/restore/upgrade - surely
> functions are already tested).  Dropping it when you're done running \df
> will avoid any possible issue with pg_upgrade.
>
> Were you able to test with your own github account ?
>

I haven’t had a chance to try this. I must confess to being a bit confused
by the distinction between running the CI tests and doing "make check";
ideally I would like to be able to run all the tests on my own machine
without any external resources. But at the same time I don’t pretend to
understand the full situation so I will try to use this when I get some
time.


Attachments:

  [application/octet-stream] 0001-Remove-source-code-display-from-df-v6.patch (6.5K, ../../CAMsGm5f97iYDV9DF8sLjH2Lc+2cjis7OuYVEtGw+c21rGPUy2g@mail.gmail.com/3-0001-Remove-source-code-display-from-df-v6.patch)
  download | inline diff:
From 114c154fbb3a4e75a6abed9dcecf4d19b8a344ad Mon Sep 17 00:00:00 2001
From: Isaac Morland <[email protected]>
Date: Tue, 17 Jan 2023 14:17:42 -0500
Subject: [PATCH] Remove source code display from \df+

The column is renamed to "Internal name" and will still show the internal
name for C and Internal functions.
---
 doc/src/sgml/ref/psql-ref.sgml     |  3 +-
 src/bin/psql/describe.c            | 11 ++------
 src/test/regress/expected/psql.out | 45 ++++++++++++++++++++++++++++++
 src/test/regress/sql/psql.sql      | 37 ++++++++++++++++++++++++
 4 files changed, 87 insertions(+), 9 deletions(-)

diff --git a/doc/src/sgml/ref/psql-ref.sgml b/doc/src/sgml/ref/psql-ref.sgml
index dc6528dc11..afdf8668d6 100644
--- a/doc/src/sgml/ref/psql-ref.sgml
+++ b/doc/src/sgml/ref/psql-ref.sgml
@@ -1650,7 +1650,8 @@ INSERT INTO tbl1 VALUES ($1, $2) \bind 'first value' 'second value' \g
         If the form <literal>\df+</literal> is used, additional information
         about each function is shown, including volatility,
         parallel safety, owner, security classification, access privileges,
-        language, source code and description.
+        language, internal name (for C and Internal language functions only),
+        and description.  Source code can be shown using <literal>\sf</literal>.
         </para>
 
         </listitem>
diff --git a/src/bin/psql/describe.c b/src/bin/psql/describe.c
index c8a0bb7b3a..e487fcd480 100644
--- a/src/bin/psql/describe.c
+++ b/src/bin/psql/describe.c
@@ -410,14 +410,9 @@ describeFunctions(const char *functypes, const char *func_pattern,
 		appendPQExpBuffer(&buf,
 						  ",\n l.lanname as \"%s\"",
 						  gettext_noop("Language"));
-		if (pset.sversion >= 140000)
-			appendPQExpBuffer(&buf,
-							  ",\n COALESCE(pg_catalog.pg_get_function_sqlbody(p.oid), p.prosrc) as \"%s\"",
-							  gettext_noop("Source code"));
-		else
-			appendPQExpBuffer(&buf,
-							  ",\n p.prosrc as \"%s\"",
-							  gettext_noop("Source code"));
+		appendPQExpBuffer(&buf,
+						  ",\n CASE WHEN l.lanname IN ('internal', 'c') THEN p.prosrc END AS \"%s\"",
+						  gettext_noop("Internal name"));
 		appendPQExpBuffer(&buf,
 						  ",\n pg_catalog.obj_description(p.oid, 'pg_proc') as \"%s\"",
 						  gettext_noop("Description"));
diff --git a/src/test/regress/expected/psql.out b/src/test/regress/expected/psql.out
index 8fc62cebd2..fffb6baace 100644
--- a/src/test/regress/expected/psql.out
+++ b/src/test/regress/expected/psql.out
@@ -5247,6 +5247,51 @@ reset work_mem;
  pg_catalog | &&   | anyarray      | anyarray       | boolean     | overlaps
 (1 row)
 
+-- check \df+
+BEGIN;
+CREATE ROLE regress_psql_df;
+CREATE OR REPLACE FUNCTION regress_psql_df_c (integer, integer, cstring, internal, integer)
+  RETURNS VOID
+  LANGUAGE c
+  IMMUTABLE PARALLEL SAFE STRICT
+AS '$libdir/utf8_and_iso8859', 'utf8_to_iso8859';
+CREATE OR REPLACE FUNCTION regress_psql_df_internal (regclass, text)
+  RETURNS bigint
+  LANGUAGE internal
+  PARALLEL SAFE STRICT
+AS 'pg_relation_size';
+CREATE OR REPLACE FUNCTION regress_psql_df_sql ()
+  RETURNS VOID
+BEGIN ATOMIC
+  SELECT NULL;
+END;
+CREATE OR REPLACE FUNCTION regress_psql_df_plpgsql ()
+  RETURNS VOID
+  LANGUAGE plpgsql
+AS $$
+BEGIN
+  RETURN;
+END;
+$$;
+ALTER FUNCTION regress_psql_df_c (integer, integer, cstring, internal, integer)
+  OWNER TO regress_psql_df;
+ALTER FUNCTION regress_psql_df_internal (regclass, text)
+  OWNER TO regress_psql_df;
+ALTER FUNCTION regress_psql_df_sql ()
+  OWNER TO regress_psql_df;
+ALTER FUNCTION regress_psql_df_plpgsql ()
+  OWNER TO regress_psql_df;
+\df+ regress_psql_df_*
+                                                                                                        List of functions
+ Schema |           Name           | Result data type |             Argument data types              | Type | Volatility | Parallel |      Owner      | Security | Access privileges | Language |  Internal name   | Description 
+--------+--------------------------+------------------+----------------------------------------------+------+------------+----------+-----------------+----------+-------------------+----------+------------------+-------------
+ public | regress_psql_df_c        | void             | integer, integer, cstring, internal, integer | func | immutable  | safe     | regress_psql_df | invoker  |                   | c        | utf8_to_iso8859  | 
+ public | regress_psql_df_internal | bigint           | regclass, text                               | func | volatile   | safe     | regress_psql_df | invoker  |                   | internal | pg_relation_size | 
+ public | regress_psql_df_plpgsql  | void             |                                              | func | volatile   | unsafe   | regress_psql_df | invoker  |                   | plpgsql  |                  | 
+ public | regress_psql_df_sql      | void             |                                              | func | volatile   | unsafe   | regress_psql_df | invoker  |                   | sql      |                  | 
+(4 rows)
+
+ROLLBACK;
 -- check \sf
 \sf information_schema._pg_expandarray
 CREATE OR REPLACE FUNCTION information_schema._pg_expandarray(anyarray, OUT x anyelement, OUT n integer)
diff --git a/src/test/regress/sql/psql.sql b/src/test/regress/sql/psql.sql
index 2da9665a19..9fb2e1e095 100644
--- a/src/test/regress/sql/psql.sql
+++ b/src/test/regress/sql/psql.sql
@@ -1275,6 +1275,43 @@ reset work_mem;
 \do - pg_catalog.int4
 \do && anyarray *
 
+-- check \df+
+BEGIN;
+CREATE ROLE regress_psql_df;
+CREATE OR REPLACE FUNCTION regress_psql_df_c (integer, integer, cstring, internal, integer)
+  RETURNS VOID
+  LANGUAGE c
+  IMMUTABLE PARALLEL SAFE STRICT
+AS '$libdir/utf8_and_iso8859', 'utf8_to_iso8859';
+CREATE OR REPLACE FUNCTION regress_psql_df_internal (regclass, text)
+  RETURNS bigint
+  LANGUAGE internal
+  PARALLEL SAFE STRICT
+AS 'pg_relation_size';
+CREATE OR REPLACE FUNCTION regress_psql_df_sql ()
+  RETURNS VOID
+BEGIN ATOMIC
+  SELECT NULL;
+END;
+CREATE OR REPLACE FUNCTION regress_psql_df_plpgsql ()
+  RETURNS VOID
+  LANGUAGE plpgsql
+AS $$
+BEGIN
+  RETURN;
+END;
+$$;
+ALTER FUNCTION regress_psql_df_c (integer, integer, cstring, internal, integer)
+  OWNER TO regress_psql_df;
+ALTER FUNCTION regress_psql_df_internal (regclass, text)
+  OWNER TO regress_psql_df;
+ALTER FUNCTION regress_psql_df_sql ()
+  OWNER TO regress_psql_df;
+ALTER FUNCTION regress_psql_df_plpgsql ()
+  OWNER TO regress_psql_df;
+\df+ regress_psql_df_*
+ROLLBACK;
+
 -- check \sf
 \sf information_schema._pg_expandarray
 \sf+ information_schema._pg_expandarray
-- 
2.32.1 (Apple Git-133)



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* Re: Remove source code display from \df+?
@ 2023-01-23 02:37  Justin Pryzby <[email protected]>
  parent: Isaac Morland <[email protected]>
  1 sibling, 1 reply; 15+ messages in thread

From: Justin Pryzby @ 2023-01-23 02:37 UTC (permalink / raw)
  To: Isaac Morland <[email protected]>; +Cc: Tom Lane <[email protected]>; Alvaro Herrera <[email protected]>; Pavel Stehule <[email protected]>; Magnus Hagander <[email protected]>; pgsql-hackers

On Sun, Jan 22, 2023 at 08:23:25PM -0500, Isaac Morland wrote:
> > Were you able to test with your own github account ?
> 
> I haven’t had a chance to try this. I must confess to being a bit confused
> by the distinction between running the CI tests and doing "make check";
> ideally I would like to be able to run all the tests on my own machine
> without any external resources. But at the same time I don’t pretend to
> understand the full situation so I will try to use this when I get some
> time.

First: "make check" only runs the sql tests, and not the perl tests
(including pg_upgrade) or isolation tests.  check-world runs everything.

One difference from running it locally is that cirrus runs tests under
four OSes.  Another is that it has a bunch of compilation flags and
variations to help catch errors (although it's currently missing
ENFORCE_REGRESSION_TEST_NAME_RESTRICTIONS, so that wouldn't have been
caught).  And another reason is that it runs in a "clean" environment,
so (for example) it'd probably catch if you have local, uncommited
changes, or if you assumed that the username is "postgres" (earlier I
said that it didn't, but actually the mac task runs as "admin").

The old way of doing things was for cfbot to "inject" the cirrus.yml
file and then push a branch to cirrusci to run tests; it made some sense
for people to mail a patch to the list to cause cfbot to run the tests
under cirrusci.  The current/new way is that .cirrus.yml is in the
source tree, so anyone with a github account can do that.  IMO it no
longer makes sense to send patches to the list "to see" if it passes
tests.  I encouraging those who haven't to try it.

-- 
Justin






^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* Re: Remove source code display from \df+?
@ 2023-01-23 02:50  Isaac Morland <[email protected]>
  parent: Justin Pryzby <[email protected]>
  0 siblings, 1 reply; 15+ messages in thread

From: Isaac Morland @ 2023-01-23 02:50 UTC (permalink / raw)
  To: Justin Pryzby <[email protected]>; +Cc: Tom Lane <[email protected]>; Alvaro Herrera <[email protected]>; Pavel Stehule <[email protected]>; Magnus Hagander <[email protected]>; pgsql-hackers

On Sun, 22 Jan 2023 at 21:37, Justin Pryzby <[email protected]> wrote:

> On Sun, Jan 22, 2023 at 08:23:25PM -0500, Isaac Morland wrote:
> > > Were you able to test with your own github account ?
> >
> > I haven’t had a chance to try this. I must confess to being a bit
> confused
> > by the distinction between running the CI tests and doing "make check";
> > ideally I would like to be able to run all the tests on my own machine
> > without any external resources. But at the same time I don’t pretend to
> > understand the full situation so I will try to use this when I get some
> > time.
>
> First: "make check" only runs the sql tests, and not the perl tests
> (including pg_upgrade) or isolation tests.  check-world runs everything.
>

Thanks very much. I should have remembered check-world, and of course the
fact that the CI tests multiple platforms. I’ll go and do some
reading/re-reading; now that I’ve gone through some parts of the process
I’ll probably understand more.

The latest submission appears to have passed:

http://cfbot.cputube.org/isaac-morland.html

However, one of the jobs (Windows - Server 2019, MinGW64 - Meson) is paused
and appears never to have run:

https://cirrus-ci.com/task/6687014536347648

Other than that, I think this is passing the tests.


^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* Re: Remove source code display from \df+?
@ 2023-01-23 04:09  Justin Pryzby <[email protected]>
  parent: Isaac Morland <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Justin Pryzby @ 2023-01-23 04:09 UTC (permalink / raw)
  To: Isaac Morland <[email protected]>; +Cc: Tom Lane <[email protected]>; Alvaro Herrera <[email protected]>; Pavel Stehule <[email protected]>; Magnus Hagander <[email protected]>; pgsql-hackers

On Sun, Jan 22, 2023 at 09:50:29PM -0500, Isaac Morland wrote:
> However, one of the jobs (Windows - Server 2019, MinGW64 - Meson) is paused
> and appears never to have run:
> 
> https://cirrus-ci.com/task/6687014536347648

Yeah, mingw is currently set to run only when manually "triggered" by
the repository owner (because it's slow).  There's no mechanism to tell
cfbot to trigger the task, but you can do it if you run from your own
github.  Anyway, there's no reason to think this patch needs extra
platform-specific coverage.

-- 
Justin






^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* Re: Remove source code display from \df+?
@ 2023-03-02 22:20  Tom Lane <[email protected]>
  parent: Isaac Morland <[email protected]>
  1 sibling, 1 reply; 15+ messages in thread

From: Tom Lane @ 2023-03-02 22:20 UTC (permalink / raw)
  To: Isaac Morland <[email protected]>; +Cc: Justin Pryzby <[email protected]>; Alvaro Herrera <[email protected]>; Pavel Stehule <[email protected]>; Magnus Hagander <[email protected]>; pgsql-hackers

Isaac Morland <[email protected]> writes:
> [ 0001-Remove-source-code-display-from-df-v6.patch ]

Pushed after some editorialization on the test case.

One thing I noticed while testing is that if you apply \df+ to an
aggregate function, it will show "Internal name" of "aggregate_dummy".
While that's an accurate description of what's in prosrc, it seems
not especially useful and perhaps indeed confusing to novices.
So I thought about suppressing it.  However, that would require
a server-version-dependent test and I wasn't quite convinced it'd
be worth the trouble.  Any thoughts on that?

			regards, tom lane






^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* Re: Remove source code display from \df+?
@ 2023-03-03 04:49  Isaac Morland <[email protected]>
  parent: Tom Lane <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Isaac Morland @ 2023-03-03 04:49 UTC (permalink / raw)
  To: Tom Lane <[email protected]>; +Cc: Justin Pryzby <[email protected]>; Alvaro Herrera <[email protected]>; Pavel Stehule <[email protected]>; Magnus Hagander <[email protected]>; pgsql-hackers

On Thu, 2 Mar 2023 at 17:20, Tom Lane <[email protected]> wrote:

> Isaac Morland <[email protected]> writes:
> > [ 0001-Remove-source-code-display-from-df-v6.patch ]
>
> Pushed after some editorialization on the test case.
>

Thanks!

One thing I noticed while testing is that if you apply \df+ to an
> aggregate function, it will show "Internal name" of "aggregate_dummy".
> While that's an accurate description of what's in prosrc, it seems
> not especially useful and perhaps indeed confusing to novices.
> So I thought about suppressing it.  However, that would require
> a server-version-dependent test and I wasn't quite convinced it'd
> be worth the trouble.  Any thoughts on that?
>

I think it’s OK. Right now \df+ claims that the source code for an
aggregate function is “aggregate_dummy”; that’s probably more untrue than
saying that its internal name is “aggregate_dummy”. There are several
features of aggregate functions that are always defined the same way by the
creation process; who’s to say they don’t all have a shared dummy internal
name?


^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* [PATCH v3 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26  Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--gBBFr7Ir9EOA20Yy
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v3-0003-optimize-pg_lfind32-by-processing-tail-with-fewer.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* [PATCH v2 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26  Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--6c2NcOVqGQ03X4Wi
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v2-0003-test_lfind32-benchmark.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* [PATCH v3 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26  Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--gBBFr7Ir9EOA20Yy
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v3-0003-optimize-pg_lfind32-by-processing-tail-with-fewer.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* [PATCH v2 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26  Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--6c2NcOVqGQ03X4Wi
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v2-0003-test_lfind32-benchmark.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* [PATCH v3 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26  Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--gBBFr7Ir9EOA20Yy
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v3-0003-optimize-pg_lfind32-by-processing-tail-with-fewer.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread

* [PATCH v2 2/3] add avx2 support in simd.h
@ 2024-03-15 17:26  Nathan Bossart <[email protected]>
  0 siblings, 0 replies; 15+ messages in thread

From: Nathan Bossart @ 2024-03-15 17:26 UTC (permalink / raw)

---
 src/include/port/simd.h | 58 ++++++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/src/include/port/simd.h b/src/include/port/simd.h
index 597496f2fb..767127b85c 100644
--- a/src/include/port/simd.h
+++ b/src/include/port/simd.h
@@ -18,7 +18,15 @@
 #ifndef SIMD_H
 #define SIMD_H
 
-#if (defined(__x86_64__) || defined(_M_AMD64))
+#if defined(__AVX2__)
+
+#include <immintrin.h>
+#define USE_AVX2
+typedef __m256i Vector8;
+typedef __m256i Vector32;
+
+#elif (defined(__x86_64__) || defined(_M_AMD64))
+
 /*
  * SSE2 instructions are part of the spec for the 64-bit x86 ISA. We assume
  * that compilers targeting this architecture understand SSE2 intrinsics.
@@ -107,7 +115,9 @@ static inline Vector32 vector32_eq(const Vector32 v1, const Vector32 v2);
 static inline void
 vector8_load(Vector8 *v, const uint8 *s)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u8(s);
@@ -120,7 +130,9 @@ vector8_load(Vector8 *v, const uint8 *s)
 static inline void
 vector32_load(Vector32 *v, const uint32 *s)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	*v = _mm256_loadu_si256((const __m256i *) s);
+#elif defined(USE_SSE2)
 	*v = _mm_loadu_si128((const __m128i *) s);
 #elif defined(USE_NEON)
 	*v = vld1q_u32(s);
@@ -134,7 +146,9 @@ vector32_load(Vector32 *v, const uint32 *s)
 static inline Vector8
 vector8_broadcast(const uint8 c)
 {
-#if defined(USE_SSE2)
+#if defined(USE_AVX2)
+	return _mm256_set1_epi8(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi8(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u8(c);
@@ -147,7 +161,9 @@ vector8_broadcast(const uint8 c)
 static inline Vector32
 vector32_broadcast(const uint32 c)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_set1_epi32(c);
+#elif defined(USE_SSE2)
 	return _mm_set1_epi32(c);
 #elif defined(USE_NEON)
 	return vdupq_n_u32(c);
@@ -270,7 +286,9 @@ vector8_has_le(const Vector8 v, const uint8 c)
 static inline bool
 vector8_is_highbit_set(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_movemask_epi8(v) != 0;
+#elif defined(USE_SSE2)
 	return _mm_movemask_epi8(v) != 0;
 #elif defined(USE_NEON)
 	return vmaxvq_u8(v) > 0x7F;
@@ -308,7 +326,9 @@ vector32_is_highbit_set(const Vector32 v)
 static inline uint32
 vector8_highbit_mask(const Vector8 v)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return (uint32) _mm256_movemask_epi8(v);
+#elif defined(USE_SSE2)
 	return (uint32) _mm_movemask_epi8(v);
 #elif defined(USE_NEON)
 	/*
@@ -337,7 +357,9 @@ vector8_highbit_mask(const Vector8 v)
 static inline Vector8
 vector8_or(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u8(v1, v2);
@@ -350,7 +372,9 @@ vector8_or(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_or(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_or_si256(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_or_si128(v1, v2);
 #elif defined(USE_NEON)
 	return vorrq_u32(v1, v2);
@@ -368,7 +392,9 @@ vector32_or(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_ssub(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_subs_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_subs_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vqsubq_u8(v1, v2);
@@ -384,7 +410,9 @@ vector8_ssub(const Vector8 v1, const Vector8 v2)
 static inline Vector8
 vector8_eq(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi8(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u8(v1, v2);
@@ -396,7 +424,9 @@ vector8_eq(const Vector8 v1, const Vector8 v2)
 static inline Vector32
 vector32_eq(const Vector32 v1, const Vector32 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_cmpeq_epi32(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_cmpeq_epi32(v1, v2);
 #elif defined(USE_NEON)
 	return vceqq_u32(v1, v2);
@@ -411,7 +441,9 @@ vector32_eq(const Vector32 v1, const Vector32 v2)
 static inline Vector8
 vector8_min(const Vector8 v1, const Vector8 v2)
 {
-#ifdef USE_SSE2
+#if defined(USE_AVX2)
+	return _mm256_min_epu8(v1, v2);
+#elif defined(USE_SSE2)
 	return _mm_min_epu8(v1, v2);
 #elif defined(USE_NEON)
 	return vminq_u8(v1, v2);
-- 
2.25.1


--6c2NcOVqGQ03X4Wi
Content-Type: text/x-diff; charset=us-ascii
Content-Disposition: attachment;
	filename="v2-0003-test_lfind32-benchmark.patch"



^ permalink  raw  reply  [nested|flat] 15+ messages in thread


end of thread, other threads:[~2024-03-15 17:26 UTC | newest]

Thread overview: 15+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2019-04-22 11:10 [PATCH 1/2] Allow TAP test to excecise tablespace. Kyotaro Horiguchi <[email protected]>
2019-04-22 11:58 [PATCH 2/5] Allow TAP test to excecise tablespace. Kyotaro Horiguchi <[email protected]>
2019-04-22 11:58 [PATCH 2/4] Allow TAP test to excecise tablespace. Kyotaro Horiguchi <[email protected]>
2023-01-23 01:23 Re: Remove source code display from \df+? Isaac Morland <[email protected]>
2023-01-23 02:37 ` Re: Remove source code display from \df+? Justin Pryzby <[email protected]>
2023-01-23 02:50   ` Re: Remove source code display from \df+? Isaac Morland <[email protected]>
2023-01-23 04:09     ` Re: Remove source code display from \df+? Justin Pryzby <[email protected]>
2023-03-02 22:20 ` Re: Remove source code display from \df+? Tom Lane <[email protected]>
2023-03-03 04:49   ` Re: Remove source code display from \df+? Isaac Morland <[email protected]>
2024-03-15 17:26 [PATCH v3 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v3 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v2 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v3 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v2 2/3] add avx2 support in simd.h Nathan Bossart <[email protected]>
2024-03-15 17:26 [PATCH v2 2/3] add avx2 support in simd.h Nathan Bossart <[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