This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Small pre-AVX512F optimization
- From: Uros Bizjak <ubizjak at gmail dot com>
- To: Jakub Jelinek <jakub at redhat dot com>
- Cc: Evgeny Stupachenko <evstupac at gmail dot com>, "H.J. Lu" <hjl dot tools at gmail dot com>, GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Fri, 3 Oct 2014 16:36:07 +0200
- Subject: Re: [PATCH] Small pre-AVX512F optimization
- Authentication-results: sourceware.org; auth=none
- References: <20141003143215 dot GS1986 at tucnak dot redhat dot com>
On Fri, Oct 3, 2014 at 4:32 PM, Jakub Jelinek <jakub@redhat.com> wrote:
> Hi!
>
> I've noticed that expand_vec_perm_1 completely uselessly builds GC garbage
> (CONST_VECTOR at least) when AVX512F isn't enabled at all.
>
> Ok to just call it for AVX512F?
OK.
> Even better would be to check the modes first too depending on target
> (AVX512F will only handle V{8D,16S}{I,F}mode, AVX512BW would handle
> also V32HImode (not yet implemented?), AVX512VL could handle
> V{2D,4D,4S,8S}{I,F}mode (not yet implemented?, though is there any
> const permutation not handled yet earlier?), and AVX512VL+AVX512BW
> could handle V{8,16}HImode (not yet implemented?) before creating a
> CONST_VECTOR.
AVX512BW and AVX512VL are currently moving targets, I propose to look
at this issue once all changes are committed to the repository.
> 2014-10-03 Jakub Jelinek <jakub@redhat.com>
>
> * config/i386/i386.c (ix86_expand_vec_perm_vpermi2): Fix up formatting.
> (ix86_expand_vec_perm): Only call ix86_expand_vec_perm_vpermi2 if
> TARGET_AVX512F.
> (expand_vec_perm_1): Likewise.
OK.
Thanks,
Uros.