[X86] combineX86ShuffleChain - ensure we only peek through bitcasts t…

…o vectors (PR51858) When searching for hidden identity shuffles (added at rG41146bfe82aecc79961c3de898cda02998172e4b), only peek through bitcasts to the source operand if it is a vector type as well. (cherry picked from commit dcba994)
llvm · Sep 20, 2021 · b96ee8f · b96ee8f
1 parent dda88bf
commit b96ee8f
Show file tree

Hide file tree

Showing 2 changed files with 59 additions and 1 deletion.
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -35823,7 +35823,7 @@ static SDValue combineX86ShuffleChain(ArrayRef<SDValue> Inputs, SDValue Root,
 
   // See if the shuffle is a hidden identity shuffle - repeated args in HOPs
   // etc. can be simplified.
-  if (VT1 == VT2 && VT1.getSizeInBits() == RootSizeInBits) {
+  if (VT1 == VT2 && VT1.getSizeInBits() == RootSizeInBits && VT1.isVector()) {
     SmallVector<int> ScaledMask, IdentityMask;
     unsigned NumElts = VT1.getVectorNumElements();
     if (BaseMask.size() <= NumElts &&

diff --git a/llvm/test/CodeGen/X86/vector-reduce-mul.ll b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
@@ -2344,6 +2344,64 @@ define i8 @illegal_v8i8(i8 %a0, <8 x i8>* %a1) {
   ret i8 %mul
 }
 
+define i8 @PR51858(i128 %arg) {
+; SSE2-LABEL: PR51858:
+; SSE2:       # %bb.0:
+; SSE2-NEXT:    movq %rdi, %xmm0
+; SSE2-NEXT:    movq %rsi, %xmm1
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
+; SSE2-NEXT:    pmullw %xmm1, %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; SSE2-NEXT:    pmullw %xmm0, %xmm1
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
+; SSE2-NEXT:    pmullw %xmm1, %xmm0
+; SSE2-NEXT:    movdqa %xmm0, %xmm1
+; SSE2-NEXT:    psrld $16, %xmm1
+; SSE2-NEXT:    pmullw %xmm0, %xmm1
+; SSE2-NEXT:    movd %xmm1, %eax
+; SSE2-NEXT:    # kill: def $al killed $al killed $eax
+; SSE2-NEXT:    retq
+;
+; SSE41-LABEL: PR51858:
+; SSE41:       # %bb.0:
+; SSE41-NEXT:    movq %rdi, %xmm0
+; SSE41-NEXT:    movq %rsi, %xmm1
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; SSE41-NEXT:    pmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; SSE41-NEXT:    pmullw %xmm1, %xmm0
+; SSE41-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[1,1,1,1]
+; SSE41-NEXT:    pmullw %xmm1, %xmm0
+; SSE41-NEXT:    movdqa %xmm0, %xmm1
+; SSE41-NEXT:    psrld $16, %xmm1
+; SSE41-NEXT:    pmullw %xmm0, %xmm1
+; SSE41-NEXT:    movd %xmm1, %eax
+; SSE41-NEXT:    # kill: def $al killed $al killed $eax
+; SSE41-NEXT:    retq
+;
+; AVX-LABEL: PR51858:
+; AVX:       # %bb.0:
+; AVX-NEXT:    vmovq %rdi, %xmm0
+; AVX-NEXT:    vmovq %rsi, %xmm1
+; AVX-NEXT:    vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
+; AVX-NEXT:    vpmovzxbw {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero,xmm0[4],zero,xmm0[5],zero,xmm0[6],zero,xmm0[7],zero
+; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[2,3,2,3]
+; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,1,1]
+; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vpsrld $16, %xmm0, %xmm1
+; AVX-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX-NEXT:    vmovd %xmm0, %eax
+; AVX-NEXT:    # kill: def $al killed $al killed $eax
+; AVX-NEXT:    retq
+  %vec = bitcast i128 %arg to <16 x i8>
+  %red = tail call i8 @llvm.vector.reduce.mul.v16i8(<16 x i8> %vec)
+  ret i8 %red
+}
+
 declare i64 @llvm.vector.reduce.mul.v2i64(<2 x i64>)
 declare i64 @llvm.vector.reduce.mul.v4i64(<4 x i64>)
 declare i64 @llvm.vector.reduce.mul.v8i64(<8 x i64>)