From: Andrew Crabtree <andrewc AT typhoon DOT rose DOT hp DOT com>
Message-Id: <199706031343.AA100525409@typhoon.rose.hp.com>
Subject: Re: New PGCC version available
To: ao950 AT freenet DOT carleton DOT ca
Date: Tue, 03 Jun 1997 6:43:29 PDT
Cc: djgpp AT delorie DOT com
In-Reply-To: <199706030434.AAA04635@freenet2.carleton.ca.carleton.ca>; from "Paul Derbyshire" at Jun 3, 97 12:34 (midnight)
Precedence: bulk


> >it is very easy to crash.  Try compiling a floating point 
> >using program with -ffast-math -funroll-loops.  Watch it SIGFPE.
> 
> Why does it do that, a bug?

Yep.  I believe that -ffast-mast -mno-ieee is broken in the 
FSF snapshot.  Also -funroll loops.  Several of the more
obscure pentium only options don't work right now from the PCG 
stuff.  But, everything that is enabled by -O6 does work.  

> >all of the optimizations have been accepted by the FSF ( I know the 
> >schedule-insns2 is the PCGs but aside from that not sure...).
> 
> What optimizations specifically?

There are a lot, you would have to check the PCG faq.  It has details.

> 
> >> (POV uses DJGPP?! I never knew that.)
> >
> >The default DOS binary provided is compiled by Watcom.  The source 
> >distribution (for DOS) works with Watcom, Borland, DJGPP, and others.
> 
> How does that work... it uses graphics for godsakes. Conditional compilation?
> Also, which compiler makes the fastest POV, Watcom, Borland, or PGCC?
> 

POVRay is also supported on Unix systems, as well as Mac and Amiga I 
believe.  But yes, they use conditional compilation.  The DJGPP version
doesn't have a UI like Watcom, but it can still render to a VESA 2.0
driven screen.  

WRT Fastest - I'm not a normal POVRay user.  I just picked it up as 
something good to test with.  Maybe some other DJGPPers could
post benchmark results...

> >> How did Pentium-optimized Allegro perform?
> 
> >Slight improvement.
> 
> Slight? Well, of course a lot is already Pentium-optimized asm.
> What about the 3d?

Don't know haven't written anything using the allegro 3D stuff.
 
> >> (Would using PGCC affect Allegro much? Much of it, except the 3d stuff, is
> >> assembly.
> 
> >Yep - PGCCs no help here 
> 
> Optimizer won't touch files already in asm then?

Nope.  A static code analyzer would, but I don't think GCC has ever used one
of those.  All optimization is done from the .c->.s step.

> 
> >> Is the assembly in Allegro's .S files pentium-optimal?)
> 
> >Its pretty good assembly, but can be improved for pentium and pentium pro.
> 
> Can it be? Has anyone taken a whack at that?

The new GCC comes with a new profiling program called gcov (gnu coverage).
Its kind of like BFA.  I am also going to write up a tutorial 
for the Pentium MSR optimization techniques.  I was planning 
on working on a section of Allegro for it.  I should be done in a 
month or so.

Andrew