[PATCH] D54658: [X86] Add an AND with 255 to the v16i8 LowerMUL path with AVX2, but not AVX512

[Craig Topper via Phabricator via llvm-commits]

craig.topper added inline comments.


================
Comment at: test/CodeGen/X86/vector-reduce-mul.ll:2257
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
----------------
This an extra truncate on the last step. Maybe need some SimplifyDemandedBits/SimplifyDemandedVectorElts enhancement here?


================
Comment at: test/CodeGen/X86/vector-reduce-mul.ll:2802
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
+; AVX2-NEXT:    vpackuswb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
----------------
Another extra truncate


https://reviews.llvm.org/D54658