X-pop3-spooler:	POP3MAIL 2.1.0 b 4 980420 -bs-
Message-Id:	<3.0.32.19980717065813.0098dc00@xs4all.nl>
X-Sender:	diep AT xs4all DOT nl
X-Mailer:	Windows Eudora Pro Version 3.0 (32)
Date:	Fri, 17 Jul 1998 06:58:16 +0100
To:	nuke AT bayside DOT net
From:	Vincent Diepeveen <diep AT xs4all DOT nl>
Subject:	Re: PGCC's optimizations (continued)
Cc:	Misha <vulcao AT netvision DOT net DOT il>, beastium-list AT Desk DOT nl
Mime-Version:	1.0
Sender:	Marc Lehmann <pcg AT goof DOT com>
Status:	RO
X-Status:	A
Lines:	93

At 07:12 AM 7/18/98 -0400, nuke AT bayside DOT net wrote:
>On Thu, 16 Jul 1998, Vincent Diepeveen wrote:
>
>> At 10:22 AM 7/16/98 +0300, Misha wrote:
>> >Hi.
>> >  First, thanks to Marc and others who have responded promptly and
>> >informatively to my questions.
>> >
>> >Recap: Sometime ago I posted that PGCC does not optimize PII(PPro)
>> >code that good. I said that GCC 2.7.2.1(libc5) produced better code for
>> >my number-crunching app.
>> >
>> >Now to the point. I did all what you have adivised me to do (compiler
>> >flags). Only after adding "-march=pentiumpro", the code gained about
>> >23% (the integer only version), and now it is slightly (1.1%) better than
>> >GCC 2.7.2.1 produced code.
>> >
>> >When I used the float version none of the flags you've sent me gained
>> >anything.
>> >
>> >I am not dissapointed with PGCC, but I think some more work has to be
>> >done in order to gain the full potential of the P6 (PII+PPro)
architecture.
>> 
>> that's clear, but also true for all other compilers.
>> 
>> >BTW, has anyone ever done a comparison of the same code under PGCC
>> >and DOS/Win95 compilers (VC++ 5.0, Watcom C/C++, Symantec, Borland,
>> >Metaware, etc...)?
>> 
>> Yes Watcom is very fast for DOS applications and optimizes best at
>> pentium. PGCC is at the same level like watcom, and gcc follows with a big
>> gap. Borland is for optimizing laughable. For programming features
>> it's at no #1 though. Nothing as easy as making interface for windoze
>> with borland products like builder. 
>> 
>> For windows NT/95/98 however there is only 1 number 1 and that's msvc.
>> Nothing comes near msvc++ speed. Too bad. Not only win98 seems
>> to get the world at its feet (even the most stupid computeruser can
>> install win98), but also you compile with a compiler from that company
>> to speeds no other compiler can manage. At least 10% speed diff with
>> MSVC and gcc. that's in fact not much, but the new msvc++ 6.0 which is
>> in beta is 20% faster than 5.0, so you guys have a lot of work to do!
>
>you must understand that msvc is for microsoft windows ONLY. since it's for
>alpha and x86 ONLY (matter of fact, i'd be willing to bet that 20%
increase is
>due to removal of alpha support - NT 5.0 will likely be x86 only) and so it
>doesn't need to do any weird hacks for every which architecture that gcc
>supports. it's perfectly ok for the msvc++ people to stick x86 assembly code
>all over the place and assume exactly what's in the system, because if your
>system is weird enough to break the compiler, windows 95/98/NT wouldn't have
>installed on it.
>
>make no assumptions. a very large part of gcc is its platform
independantness.
>if you were to make gcc say, a compiler for FreeBSD/x86 *ONLY*, cut out any
>code that platform doesn't use, do no checking for headers/libraries
>because you know exactly what it has, you'd make a very tiny and probably
>a much faster compiler.
>
>> About latest version pgcc: I tried installing the 'new'
>> PGCC quite a number of times, but so far
>> didn't manage under Linux. It can't work with 2.8.1 header files
>> for some reason. I'll try it tonight again.
>> 
>> >I had some difficulties porting my own code but I am working on it now. As
>> >soon as I have results I will post them here.
>> 
>> Porting from (p)gcc to msvc/watcom is hard because msvc/watcom are horrible
>> in casting. You must cast everything.
>
>which is another thing. those compilers don't do any intelligent variable
>checking to see if you absolutely need a cast or not. while you're making
your
>single-platform single-os gcc, go ahead and take out all the checks it
does to
>see wether a cast is needed or not. variables of different types should
ALWAYS
>have a cast. this would probably be a pretty big increase in compile speed.
> _        _  __     __             _ _                                  _
>|        / |/ /_ __/ /_____         |       Nuke Skyjumper               |
>|       /    / // /  '_/ -_)        |         "Master of the Farce"      |
>|_     /_/|_/\_,_/_/\_\\__/        _|_           nuke AT bayside DOT net       _|
>

Nuke, 1 big note: i'm not interested in HOW long it takes to compile stuff,
nor did i compare executable size. I'm only interested in how fast
my program is.

Vincent

>

- Raw text -