delorie.com/archives/browse.cgi   search  
Mail Archives: djgpp/1997/05/22/07:39:46

Message-ID: <33843002.7C0@silesia.top.pl>
Date: Thu, 22 May 1997 13:37:38 +0200
From: Michal <wapex AT silesia DOT top DOT pl>
MIME-Version: 1.0
To: djgpp AT delorie DOT com
Subject: Re: PGCC
References: <199705201606 DOT AA182944385 AT typhoon DOT rose DOT hp DOT com>

Andrew Crabtree wrote:
> 
> > What you said IS TRUE, but the nops are on the end of all functions. The
> > amount of nops is so big thet it makes my code 20% bigger (don't know if
> > only this), I've heard that PGCC produces smaller code, becouse it
> > doesn't use nops. Me expirience is different, it uses nops but in
> > different way.
> 
> That's quite odd.  I haven't seen it add any no-ops really.  Doing it at
> the end of functions makes even less sense.  It should just jump over
> data padding if anything to align stuff.  I'll go back and look
> at the output of some bigger files and see what it does.
> 
> > Just simple fps counting. I'm shure there is some difference (maybe
> > small) but I didn't noticed any with this method (I've disabled all my
> > assembler code in triangle rendering functions).
> 
> What about your blit code?  Is that written in assembly?  Only suggestion
> I would have is to profile and see where things are getting stuck.  Then
> send me the old gcc assembly output and pgcc of that section.
> 
> Andrew
I think I know the anserw. PGCC and as are understanding .align x in
different ways (in PGCC x is the boudary, in as 2^x is the boudary).
That explaines the nops inside my code. Also with vary big gaps
inbetween variables the performance of cache is worse. I've compiled my
program with -S option, corrected the alignment, used as to get .0 file,
linked it and I've got a little speed up.
So, is there a version of as that can work well with PGCC?

- Raw text -


  webmaster     delorie software   privacy  
  Copyright © 2019   by DJ Delorie     Updated Jul 2019