This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug target/70721] Suboptimal code generated when using _mm_min_sd
- From: "rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Tue, 19 Apr 2016 09:40:10 +0000
- Subject: [Bug target/70721] Suboptimal code generated when using _mm_min_sd
- Auto-submitted: auto-generated
- References: <bug-70721-4 at http dot gcc dot gnu dot org/bugzilla/>
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70721
--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> ---
IMHO vec_merge should go (it doesn't scale to arbitrary large vector modes, it
already uses all bits of the CONST_INT for AVX512 V64QImode). vec_merge
would need to change to a CONST_{DOUBLE,WIDE} to accomodate larger vectors.
OTOH the other canonical form with (vec_select (vec_concat ...)) requires
larger vector modes. What we lack is sth similar to VEC_PERM_EXPR here,
unifying the vec_merge and the (vec_select (vec_concat ...)) cases.