This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH 7/N v2] [x86_64] Vectorized math functions
- From: Joseph Myers <joseph at codesourcery dot com>
- To: Andrew Senkevich <andrew dot n dot senkevich at gmail dot com>
- Cc: libc-alpha <libc-alpha at sourceware dot org>
- Date: Fri, 24 Apr 2015 20:30:34 +0000
- Subject: Re: [PATCH 7/N v2] [x86_64] Vectorized math functions
- Authentication-results: sourceware.org; auth=none
- References: <CAMXFM3sG=Uo6pJsNn34VtU+VDd4+VO+YyoUw3hCYpFi_K1cLsg at mail dot gmail dot com> <alpine dot DEB dot 2 dot 10 dot 1412031814130 dot 20515 at digraph dot polyomino dot org dot uk> <CAMXFM3sTwm+pFdALMg4aaBroFbDds=LK1rTH4ksouR=MK-75Wg at mail dot gmail dot com> <alpine dot DEB dot 2 dot 10 dot 1412111756090 dot 21111 at digraph dot polyomino dot org dot uk> <CAMXFM3tOpEoMb-oLoeSJNzOWYnJgte3bm_OR=oZ172n845WeZw at mail dot gmail dot com> <alpine dot DEB dot 2 dot 10 dot 1504241735021 dot 30193 at digraph dot polyomino dot org dot uk> <CAMXFM3uu7FuJCJSC=80FCXocyDmEK5ZoCH5OB0buxeP=36F3Ng at mail dot gmail dot com>
On Fri, 24 Apr 2015, Andrew Senkevich wrote:
> 2015-04-24 20:37 GMT+03:00 Joseph Myers <joseph@codesourcery.com>:
> > On Fri, 17 Apr 2015, Andrew Senkevich wrote:
> >
> >> My question is do we need to test this new SSE2 version (which is just
> >> simple wrapper to scalar one)?
> >
> > Well, we need to test each entry point. One of the existing deficiencies
> > of IFUNC testing (which would also apply to the CPU tests here) is that we
> > don't test the different variants of libm functions for different
> > processors - I don't think you need to address that in your patch. (So
> > make sure the SSE2 entry point is tested, but which function that testing
> > calls may depend on the processor.)
>
> Yes, if not build separate test for SSE2 entry point then SSE2 version
> will be tested only on according processor if called through IFUNC.
>
> Do we need to move to multiarch folder implementations with IFUNC?
In general use of IFUNCs is the preferred way of handling different
function implementations for different CPU variants, in the absence of a
clear reason to do it differently in a particular case.
--
Joseph S. Myers
joseph@codesourcery.com