This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH] Compile AVX libm functions with -mavx -mprefer-avx128
On Thu, Oct 4, 2012 at 6:56 AM, Dmitrieva Liubov
<liubov.dmitrieva@gmail.com> wrote:
> We will check performance on unit tests for each affected function separately.
> I guess unit tests are good benchmark here.
I don't think so. The change prevents the stalls caused by switching
between 256-bit AVX and legacy SSE instructions, so just running these
functions won't tell you anything.
Maybe running these functions with a single legacy SSE instruction
added to the end will get the results you want?