On Tue, 28 Dec 2004, Isaac Richards wrote: ]> This patch removes the -mfpmath=sse performance recommendation, but ]> keeps the additional archs, such as the winchip, the updated C3 info, ]> and other Intel and AMD processors. ] ]Have you profiled on each of those processors to see if setting the arch ]actually creates faster code? I've seen -march=athlon-xp create 50% slower ]code than -march=pentiumpro with a recent gcc.. ] ]Isaac
No, I only have Intel processors at my disposal. Compiling for pentium4 gives me a 30% speed boost. For the rest, I based this on the gcc manual. I've heard rumors about xp optimization being broken, but I can't verify this myself.
I made this patch to the documentation because someone was having problems compiling mmx_gcc.cpp because they were targeting the pentiumpro, which did not support MMX. The compiler, rightly, did not generate the MMX code.
Well I have an AMD XP so I could compile with the two different arch flags, question is: how do you measure the difference (for now I just basically stare at top during liveTV, but I guess there's something more scientific ;-).
Bye
Manu
pgpsKeSekn8Xv.pgp
Description: PGP signature
_______________________________________________ mythtv-dev mailing list mythtv-dev@mythtv.org http://mythtv.org/cgi-bin/mailman/listinfo/mythtv-dev
