Date: Tue, 4 Oct 2005 13:02:53 -0400
From: mark@mark.mielke.cc
To: Martijn van Oosterhout <kleptog@svana.org>
Cc: Simon Riggs <simon@2ndquadrant.com>,
	Tom Lane <tgl@sss.pgh.pa.us>, Ron Peacetree <rjpeace@earthlink.net>,
	pgsql-hackers@postgresql.org
Subject: Re: [PERFORM] A Better External Sort?
Message-ID: <20051004170253.GA26380@mark.mielke.cc>
References: <20051002123240.GA30492@svana.org>
	<20051002193835.GJ30492@svana.org>
	<1128376292.8603.142.camel@localhost.localdomain>
	<20051004100441.GC17589@svana.org>
	<1128425094.8603.296.camel@localhost.localdomain>
	<20051004122443.GF17589@svana.org> <7788.1128434784@sss.pgh.pa.us>
	<20051004143041.GH17589@svana.org>
	<1128437813.8603.305.camel@localhost.localdomain>
	<20051004152336.GI17589@svana.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20051004152336.GI17589@svana.org>
User-Agent: Mutt/1.4.2.1i

On Tue, Oct 04, 2005 at 05:23:41PM +0200, Martijn van Oosterhout wrote:
> On Tue, Oct 04, 2005 at 03:56:53PM +0100, Simon Riggs wrote:
> > I've been using gcc 3.4 and saw no warning when using either "-Winline"
> > or "-O3 -Winline".
> Ok, I've just installed 3.4 and verified that. I examined the asm code
> and gcc is inlining it. I concede, at this point just throw in -Winline
> and monitor the situation.
> As an aside, the *_getattr calls end up a bit suboptimal though. It's
> producing code like:
>   cmp attlen, 4
>   je $elsewhere1
>   cmp attlen, 2
>   je $elsewhere2
>   ld byte
> here:
> --- much later ---
> elsewhere1:
>   ld integer
>   jmp $here
> elsewhere2:
>   ld short
>   jmp $here
> No idea whether we want to go down the path of hinting to gcc which
> size will be the most common.

If it will very frequently be one value, and not the other values, I
don't see why we wouldn't want to hint? #ifdef it to a expand to just
the expression if not using GCC. It's important that we know that the
value would be almost always a certain value, however, as GCC will try
to make the path for the expected value as fast as possible, at the
cost of an unexpected value being slower.

  __builtin_expect (long EXP, long C)

     You may use `__builtin_expect' to provide the compiler with branch
     prediction information.  In general, you should prefer to use
     actual profile feedback for this (`-fprofile-arcs'), as
     programmers are notoriously bad at predicting how their programs
     actually perform.  However, there are applications in which this
     data is hard to collect.

     The return value is the value of EXP, which should be an integral
     expression.  The value of C must be a compile-time constant.  The
     semantics of the built-in are that it is expected that EXP == C.
     For example:

          if (__builtin_expect (x, 0))
            foo ();

     would indicate that we do not expect to call `foo', since we
     expect `x' to be zero.  Since you are limited to integral
     expressions for EXP, you should use constructions such as

          if (__builtin_expect (ptr != NULL, 1))
            error ();

     when testing pointer or floating-point values.

Cheers,
mark

-- 
mark@mielke.cc / markm@ncf.ca / markm@nortel.com     __________________________
.  .  _  ._  . .   .__    .  . ._. .__ .   . . .__  | Neighbourhood Coder
|\/| |_| |_| |/    |_     |\/|  |  |_  |   |/  |_   | 
|  | | | | \ | \   |__ .  |  | .|. |__ |__ | \ |__  | Ottawa, Ontario, Canada

  One ring to rule them all, one ring to find them, one ring to bring them all
                       and in the darkness bind them...

                           http://mark.mielke.cc/