This is the mail archive of the
glibc-bugs@sourceware.org
mailing list for the glibc project.
[Bug math/21967] When 512-bit AVX2 wrapper functions in mathvec are used?
- From: "hjl.tools at gmail dot com" <sourceware-bugzilla at sourceware dot org>
- To: glibc-bugs at sourceware dot org
- Date: Fri, 25 Aug 2017 17:14:31 +0000
- Subject: [Bug math/21967] When 512-bit AVX2 wrapper functions in mathvec are used?
- Auto-submitted: auto-generated
- References: <bug-21967-131@http.sourceware.org/bugzilla/>
https://sourceware.org/bugzilla/show_bug.cgi?id=21967
--- Comment #4 from H.J. Lu <hjl.tools at gmail dot com> ---
(In reply to Andrew Senkevich from comment #3)
> (In reply to H.J. Lu from comment #1)
> > avx2_wrapper is supposed to be used when AVX512F isn't profitable. But
> > with
> >
> > if (CPU_FEATURES_ARCH_P (cpu_features, AVX512DQ_Usable))
> > return OPTIMIZE (skx);
> >
> > if (CPU_FEATURES_ARCH_P (cpu_features, AVX512F_Usable))
> > return OPTIMIZE (knl);
> >
> > it is never used. There is a Prefer_No_AVX512 bit in cpu_features. But
> > we don't want to use it here since it will disable AVX512 mathvec on Skylake
> > server. We need a separate bit to control AVX512 mathvec.
>
> Do you think AVX512 mathvec functions on SKX should be wrappers to AVX2?
Not by default, but with a new ARCH bit.
--
You are receiving this mail because:
You are on the CC list for the bug.