Re: [PATCH 02/10] i386: Replace assembly versions of e_expf with generic e_expf.c

["H.J. Lu" <hjl dot tools at gmail dot com>]

On Thu, Oct 19, 2017 at 10:44 AM, Joseph Myers <joseph@codesourcery.com> wrote:
> I think there should also be --disable-multi-arch benchmark results here.
> Likewise for other i386 functions in this series (as otherwise all of the
> x86 benchmarks you give are testing the multiarch/*-sse2.c case only,
> nothing is testing the generic fallback of the C version built to use x87
> arithmetic - which I expect is still very likely to be faster than the old
> .S implementation, and you might justify removing the .S versions even if
> the .c versions built for x87 aren't faster, but benchmark results for
> that case should still be given).
>

With --disable-multi-arch, generic version is faster than x87 assembly
version:

  On IvyBridge with --disable-multi-arch, it improves performance by:

                           Before            After     Improvement
reciprocal-throughput      133.707          40.3778       231%
latency                    149.191          63.2515       135%


-- 
H.J.