This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [BENCHMARK] -mfpmath=sse should disable x387 intrinsics


On Thu, 25 Nov 2004 15:39:47 +0100, Uros Bizjak <uros@kss-loka.si> wrote:
> Richard Guenther wrote:
> 
> >>Yes, Roger's patch is a good thing, and should stay in. Previously, you
> >>didn't have any switch available to turn i387 intrinsics off.
> >>
> >>
> >
> >Err, but if I want the previous -mfpmath=sse behavior there is now no
> >way for it.
> >-mfpmath=sse,387 generates different code than -mfpmath=sse before.
> >
> >
> Indeed. -mfpmath=sse,387 enables TARGET_MIX_SSE_I387, and this enables
> fadd/fsub/fmul/fdiv and fild instructions. These arithmetic instructions
> have higher latency than their sse equivalents. AFAIK, arithmetic i387
> insns and their sse equivalents fight for the same resources in case of
> pentium4.

Exactly.  I guess mixing SSE and fadd/fsub/etc. is not very helpful - maybe we
can just drop this feature and retain 387 flag for x87 intrinsics without an SSE
(synthesized) equivalent?  Though I'd very much like inlined SSE synthesized
instruction streams for pow and sqrt.

May I ask to revert the current patch?

Richard.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]