This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 1/4 v2] Optimized generic expf and exp2f with wrappers
On Thu, 14 Sep 2017, Szabolcs Nagy wrote:
> but overall the symbol version fixes need a lot of
> changes.. do we really need to carry asm math code
> for barely supported targets? do ppl really care
> if m68k/ia64/i386 get suboptimal math code?
The first question would be whether it's suboptimal anyway. That needs
benchmarking on hardware, which presumably architecture maintainers could
do for m68k/ia64. For i386 there are both generic and SSE2 versions of
expf to benchmark.
It's far from obvious that just because the x87 f2xm1 instruction exists
it's efficient to use (especially since it needs to produce a result to
the precision of long double, and expf only needs to produce a result to
the precision of float).
--
Joseph S. Myers
joseph@codesourcery.com