[PATCH] D74431: [X86] Don't disable code in combineHorizontalPredicateResult just because we have avx512
Sanjay Patel via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue Feb 11 12:42:22 PST 2020
spatel accepted this revision.
spatel added a comment.
This revision is now accepted and ready to land.
LGTM
================
Comment at: llvm/test/CodeGen/X86/vector-compare-all_of.ll:31
; AVX512-NEXT: vcmpltpd %xmm0, %xmm1, %xmm0
-; AVX512-NEXT: vpermilps {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; AVX512-NEXT: vandpd %xmm1, %xmm0, %xmm0
-; AVX512-NEXT: vmovq %xmm0, %rax
+; AVX512-NEXT: vmovmskpd %xmm0, %ecx
+; AVX512-NEXT: xorl %eax, %eax
----------------
This and the similar v2i64 case may warrant revisiting. We assumed movmsk was generally cheaper even on a 2-way reduction, but it depends how recent uarch have implemented it.
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D74431/new/
https://reviews.llvm.org/D74431
More information about the llvm-commits
mailing list