[PATCH] D61189: [X86][SSE] Extract i1 elements from vXi1 bool vectors (WIP)

Fri Apr 26 08:11:30 PDT 2019

RKSimon added inline comments.

================
Comment at: test/CodeGen/X86/avx2-masked-gather.ll:55
 ; NOGATHER-NEXT:    vmovdqa %xmm1, %xmm0
 ; NOGATHER-NEXT:    retq
 entry:
----------------
Annoyingly this doesn't drop through the code, skipping zero-element gathers, instead repeating tests+branches.

================
Comment at: test/CodeGen/X86/avx2-masked-gather.ll:393
+; NOGATHER-NEXT:    vpacksswb %xmm0, %xmm2, %xmm4
+; NOGATHER-NEXT:    vpmovmskb %xmm4, %eax
+; NOGATHER-NEXT:    testb $2, %al
----------------
Repeated PACKSS+MOVMSK instructions - I assume due to it having a undef argument in the xmm0 slot.

================
Comment at: test/CodeGen/X86/avx512-insert-extract.ll:1024
+; KNL-NEXT:    shrq $63, %rax
+; KNL-NEXT:    cmpq $1, %rax
+; KNL-NEXT:    sete %al
----------------
we should be able to replace both shifts and the cmp with a single BT $32, %RAX ?

================
Comment at: test/CodeGen/X86/masked_compressstore.ll:51
+; HSW-NEXT:    vpacksswb %xmm0, %xmm1, %xmm2
+; HSW-NEXT:    vpmovmskb %xmm2, %eax
+; HSW-NEXT:    testb $2, %al
----------------
More repeated PACKSS+MOVMSK

================
Comment at: test/CodeGen/X86/movmsk-cmp.ll:4376
+; AVX-NEXT:    andb %dl, %al
+; AVX-NEXT:    andb %cl, %al
 ; AVX-NEXT:    retq
----------------
We should be able to reduce this to a TEST

================
Comment at: test/CodeGen/X86/movmsk-cmp.ll:4655
+; SSE2-NEXT:    ucomisd %xmm1, %xmm0
+; SSE2-NEXT:    cmovbel %ecx, %eax
 ; SSE2-NEXT:    retq
----------------
Repeated comparisons

================
Comment at: test/CodeGen/X86/setcc-combine.ll:10
+; CHECK-NEXT:    pxor %xmm0, %xmm0
+; CHECK-NEXT:    pcmpeqd %xmm1, %xmm0
+; CHECK-NEXT:    movmskps %xmm0, %eax
----------------
I think this is effectively a NOT that we should be able to handle somehow.

Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D61189/new/

https://reviews.llvm.org/D61189