Re: revise -mfpmath=sse comparisons (povray test picture failurewith -ffast-math)

[Uros Bizjak <uros at kss-loka dot si>]

Hello Richard!

> > Tested on i686 and x86-64 linux, and with a povray 3.6 benchmark run.
> > The runtime of the povray benchmark improved 2.0% on the p4, and 7.9%
> > on the athlon64.  The later is surely helped by the amd64 abi being
> > designed for use with fpmath=sse.
> With todays mainline [gcc version 4.0.0 20050114 (experimental)] the 
> pictures rendered by povray-3.50c are wrong. They look like overexposed 
> photographs with their colors reversed. A rendered picture from 
> scenes/advanced/abyss.pov is attached to this message.
>
> These are CFLAGS from src/Makefile:
> CFLAGS = -O3 -march=pentium4 -mfpmath=sse -mno-fancy-math-387 
> -ffast-math -D__NO_MATH_INLINES -finline-functions -fomit-frame-pointer 
> -funroll-loops -fexpensive-optimizations -malign-double 
> -foptimize-sibling-calls -minline-all-stringops -Wno-multichar
> CXXFLAGS = $(NOMULTICHAR) -O3 -march=pentium4 -mfpmath=sse 
> -mno-fancy-math-387 -ffast-math -D__NO_MATH_INLINES -finline-functions 
> -fomit-frame-pointer -funroll-loops -fexpensive-optimizations 
> -malign-double -foptimize-sibling-calls -minline-all-stringops 
> -Wno-multichar
>
> If -mfpmath=sse is substituted with -mfpmath=387, rendered pictures look 
> OK to me.

Without -ffast-math, SSE version produces corrects results. These are 
CFLAGS:

CFLAGS = -O3 -march=pentium4 -mfpmath=sse -D__NO_MATH_INLINES 
-finline-functions -fomit-frame-pointer -funroll-loops 
-fexpensive-optimizations -malign-double -foptimize-sibling-calls 
-minline-all-stringops -Wno-multichar
CXXFLAGS = $(NOMULTICHAR) -O3 -march=pentium4 -mfpmath=sse 
-D__NO_MATH_INLINES -finline-functions -fomit-frame-pointer 
-funroll-loops -fexpensive-optimizations -malign-double 
-foptimize-sibling-calls -minline-all-stringops -Wno-multichar

Uros.