p4 opts

Sam Halliday fommil@yahoo.ie
Wed Jan 8 15:12:00 GMT 2003


hi there,

in the INSTALL file about P4 opts... you may also wish to add "-msse" as
well just before "-msse2"

CFLAGS="-O2 -march=pentium4 -mfpmath=sse -msse -msse2"

"-msse" utilises the sse instruction set for floating point ops and
-msse2 uses sse2 for double precision ops, so certainly in this case
sse2 is more imprortant. (this doesnt effect integer precision
operations...)

make check's are all OK under 1.3 with these CFLAGS (gcc-3.2.1). also,
the 30% i previously said may be misleading... that is against
unoptimised code. it is more like 10% against -O2 optimised code. dont
expect code to output EXACTLY the same values as before... it will be
more accurate this time ;) (the 387 chip was replaced by sse in order to
remove numerical instabilities, thats why x64 is purely "sse")

note that P3 machines can also use the above CFLAGS, iff the "-msse2" is
removed and "pentium4" is changed to "pentium3". future 64-bit intel
CPUs will use sse by default.

cheers,
Sam
-- 
Words have a longer life than deeds.
-- Pindar



More information about the Gsl-discuss mailing list