Path: number1.nntp.dca.giganews.com!border1.nntp.dca.giganews.com!nntp.giganews.com!local01.nntp.dca.giganews.com!nntp.adelphia.com!news.adelphia.com.POSTED!not-for-mail
NNTP-Posting-Date: Wed, 11 Jan 2006 19:12:38 -0600
Newsgroups: fidonet.linux
From: Kenneth_Parrish@f345.n10.z1
Date: Wed, 11 Jan 2006 12:21:14 -0500
Subject: Re: gcc 4.1
Message-ID: 
References: 
Organization: BBS Networks @ www.bbsnets.com 808-839-6036
 1414
Lines: 25
NNTP-Posting-Host: 24.54.66.112
 properly
Xref: number1.nntp.dca.giganews.com fidonet.linux:460

-=> In article 11 Jan 06  18:05:42, Maurice Kinal wrote to Kenneth Parrish <=-

 MK> Hey Kenneth!

 MK> With the "-Os -march=pentium3 -D_REENTRANT" optimizations added to
 MK> Makefile.micro (perl-5.8.7) and a 'make -j3 -f Makefile.micro' I get the
 MK> following results;

 MK> -rwxr-xr-x  1 root root 611364 2006-01-11 09:55 microperl-gcc-4.0.2*
 MK> -rwxr-xr-x  1 root root 605760 2006-01-11 09:51 microperl-gcc-4.1*

depending on where the savings are, a few k can help much. for example,
reducing the pid hash size offers a bit more responsiveness here

 MK> So far so good but I haven't taken the 4.1 microperl out for a cruise
 MK> yet. I'll benchmark it against the 4.0.2 one as I already know that one
 MK> works excellent.  Seems to me that 4.1 compiles faster but I am not
 MK> positve about that.  More testing is needed but offhand it looks like 4.1
 MK> is a keeper.
you can change the -Os and procesor instruction weightings at
gcc/config/i386/i386.c.  i've fiddled some: fun

.. They're only trying to make me LOOK paranoid!
-+- MultiMail/Linux v0.47