[PATCH] D56756: [x86] lower shuffle of extracts to AVX2 vperm instructions

[Sanjay Patel via Phabricator via llvm-commits]

spatel marked 4 inline comments as done.
spatel added inline comments.


================
Comment at: test/CodeGen/X86/avx512-shuffles/partial_permute.ll:1052
 ; CHECK-NEXT:    vptestnmd %xmm1, %xmm1, %k1
-; CHECK-NEXT:    vpshufd {{.*#+}} xmm0 {%k1} {z} = xmm0[1,3,2,1]
+; CHECK-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
 ; CHECK-NEXT:    vzeroupper
----------------
RKSimon wrote:
> This looks like a missed opportunity to merge a avx512 mask select across an extract_subvector(vec,0) ? There a few more below as well, please can you raise a bug about them.
> ```
> vmovdqa {{.*#+}} ymm2 = <5,3,2,5,u,u,u,u>
> vptestnmd %xmm1, %xmm1, %k1
> vpermd %ymm0, %ymm2, %ymm0 {%k1} {z}
> ```
Filed here:
https://bugs.llvm.org/show_bug.cgi?id=40332


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D56756/new/

https://reviews.llvm.org/D56756