This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] [x86_32] Don't use SSE4_2 instructions on Intel Silvermont Micro Architecture.
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: Liubov Dmitrieva <liubov dot dmitrieva at gmail dot com>
- Cc: "Carlos O'Donell" <carlos at redhat dot com>, GNU C Library <libc-alpha at sourceware dot org>
- Date: Mon, 1 Jul 2013 08:12:39 -0700
- Subject: Re: [PATCH] [x86_32] Don't use SSE4_2 instructions on Intel Silvermont Micro Architecture.
- References: <CAHjhQ93YUtMz14Aj7JfhJ7Tq++eC+23cA6yubh1KPRpKTvvkaw at mail dot gmail dot com> <51D0967A dot 7050102 at redhat dot com> <CAHjhQ92mTi8OmY7RBC3=sCaX=-xbgGH82pCBz4fCQkBLJcNqOA at mail dot gmail dot com>
On Mon, Jul 1, 2013 at 5:46 AM, Liubov Dmitrieva
<liubov.dmitrieva@gmail.com> wrote:
> It doesn't look very good when I summarized the results because there
> are too many small lengths cases in your benchmarks.
> And small cases are more inaccurate.
> Ssse3 looks asymptotically faster. Actually it would better to use
> sse2 unaligned versions for Silvermont but we don't have it in glibc
> at the moment.
> Bot sse4_2 and ssse3 are not the best for SLM.
>
Why is SSE4.2 strcasecmp disabled only for libc.a,
not libc.so? Libc.so is used by most of applications.
--
H.J.