Sender: M DOT A DOT Bukin AT inp DOT nsk DOT su
To: djgpp AT delorie DOT com
Subject: Re: binutils 2.8.1 performance = slow
References: <Pine DOT SUN DOT 3 DOT 91 DOT 971126132246 DOT 960H-100000 AT is>
	<347C86AC DOT 7AAC AT seanet DOT com>
From: Michael Bukin <M DOT A DOT Bukin AT inp DOT nsk DOT su>
Date: 29 Nov 1997 07:59:23 +0600
In-Reply-To: "Alan M. Doerhoefer"'s message of Wed, 26 Nov 1997 20:29:32 GMT
Message-Id: <20en40qx38.fsf@Sky.inp.nsk.su>
Lines: 34
Precedence: bulk

"Alan M. Doerhoefer" <aland AT seanet DOT com> writes:

> Eli Zaretskii wrote:
> > 
> > On Tue, 25 Nov 1997, Alan M. Doerhoefer wrote:
> > 
> > > using binutils 2.8.1 instead of 2.7 caused a remarkable
> > > slowing of my program.
> > 
> > Did you consider the possibility of different alignment of the functions
> > in the version linked with 2.8.1?  What processor did you run your
> > program on?

  Different versions of binutils produce different instructions
for .align directive.  E.g. for the following function:

int test (void) {
  __asm__ __volatile__ (".align 4\n\tmovl $0, %eax\n\t.align 4");
  return 0;
}

bnu2.7:
leal 0x0(%esi),%esi
leal 0x0(%esi,1),%esi

bnu2.8.1:
leal 0x0(%esi),%esi
leal 0x0(%edi,1),%edi

This difference may have effects on instructions pipelining or something.
I think, speed may depend on instructions near .align.

Probably for other alignment quantities (or assembler directives)
assemblers from different versions will produce different results too.