This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
RE: [RFC PATCH, x86_64] Use -mno-sse[,2] to fall back to x87 FP argument passing convention
- From: "Menezes, Evandro" <evandro dot menezes at amd dot com>
- To: "Roger Sayle" <roger at eyesopen dot com>, "Michael Matz" <matz at suse dot de>, "Uros Bizjak" <ubizjak at gmail dot com>, "Jan Hubicka" <hubicka at ucw dot cz>
- Cc: "GCC Patches" <gcc-patches at gcc dot gnu dot org>
- Date: Mon, 9 Oct 2006 15:25:29 -0500
- Subject: RE: [RFC PATCH, x86_64] Use -mno-sse[,2] to fall back to x87 FP argument passing convention
Roger,
>
> There was a time when I'd completely agree with you. Benchmarking our
> numeric applications on x86_64 with and without -mfpmath=387 using
> gcc-3.4 shows/showed SSE math to be a significant win.
That needs some clarifications.
SUSE and other distros ship with faster and more accurate routines for x86-64 than those in GLIBC and x87.
As I explained a while ago, due to compatibility issues, the accuracy of x87 instructions cannot be changed, even though they can be off by thousands of ulps. Moreover, it is unlikely that any manufacturer will put any effort speeding x87 up (but I don't mean to speak officially). It is possible to achieve better accuracy and speed with discrete routines using the SSE ISA extension.
Any such attempt, IMHO, misses the real cause of the performance issue that some observe.
> Has anyone recently revisited the -mfpmath=* issue with SPEC
> or acovea?
See SPEC CPU200x submissions by AMD: they were all run on OS'es sporting such x86-64-optimized routines and none uses -mfpmath=387, no matter which compiler (http://tinyurl.com/j779l & http://tinyurl.com/owzj2).
Thanks,
_______________________________________________________
Evandro Menezes GNU Tools Team
512-602-9940 Advanced Micro Devices
evandro.menezes@amd.com Austin, TX