[PATCH] D35978: [AVX-512] Don't use unmasked VMOVDQU8/16 for 8-bit or 16-bit element stores even when BWI instructions are supported. Always use VMOVDQA32/VMOVDQU32.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added inline comments.


================
Comment at: test/CodeGen/X86/avx512-insert-extract.ll:5
 ; RUN: llc < %s -march=x86-64 -mtriple=x86_64-apple-darwin -mcpu=skx -mattr=avx512vbmi | FileCheck --check-prefix=SKX --check-prefix=SKX_VBMI %s
 
 define <16 x float> @test1(<16 x float> %x, float* %br, float %y) nounwind {
----------------
Maybe add a  --check-prefix=CHECK first option?


================
Comment at: test/CodeGen/X86/subvector-broadcast.ll:993
+; X32-AVX512-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
+; X32-AVX512-NEXT:    retl
 ;
----------------
Is this a missed execution domain opportunity? Same for the others below


================
Comment at: test/CodeGen/X86/x86-interleaved-access.ll:4
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2 | FileCheck %s --check-prefix=AVX --check-prefix=AVX2
 ; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx512f -mattr=+avx512bw | FileCheck %s --check-prefix=AVX --check-prefix=AVX3
 
----------------
Just noticed this is called AVX3?! Is that a good idea?


https://reviews.llvm.org/D35978