[X86][AVX] Add SimplifyMultipleUseDemandedBits VBROADCAST handling to SimplifyDemande...

As suggested on D79987.

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 89559ad..7edce21 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -36830,6 +36830,12 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
     if (SimplifyDemandedVectorElts(Src, SrcElts, SrcUndef, SrcZero, TLO,
                                    Depth + 1))
       return true;
+    // Aggressively peek through src to get at the demanded elt.
+    // TODO - we should do this for all target/faux shuffles ops.
+    APInt SrcBits = APInt::getAllOnesValue(SrcVT.getScalarSizeInBits());
+    if (SDValue NewSrc = SimplifyMultipleUseDemandedBits(Src, SrcBits, SrcElts,
+                                                         TLO.DAG, Depth + 1))
+      return TLO.CombineTo(Op, TLO.DAG.getNode(Opc, SDLoc(Op), VT, NewSrc));
     break;
   }
   case X86ISD::VPERMV: {

diff --git a/llvm/test/CodeGen/X86/oddshuffles.ll b/llvm/test/CodeGen/X86/oddshuffles.ll
index e182008..910c40d 100644 (file)
--- a/llvm/test/CodeGen/X86/oddshuffles.ll
+++ b/llvm/test/CodeGen/X86/oddshuffles.ll
@@ -2014,18 +2014,16 @@ define <16 x i32> @splat_v3i32(<3 x i32>* %ptr) {
 ;
 ; AVX2-SLOW-LABEL: splat_v3i32:
 ; AVX2-SLOW:       # %bb.0:
-; AVX2-SLOW-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX2-SLOW-NEXT:    vpinsrd $2, 8(%rdi), %xmm0, %xmm1
-; AVX2-SLOW-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm2[0],ymm0[1],ymm2[2,3,4,5,6,7]
-; AVX2-SLOW-NEXT:    vpbroadcastd %xmm1, %ymm1
-; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vmovsd {{.*#+}} xmm1 = mem[0],zero
+; AVX2-SLOW-NEXT:    vxorps %xmm2, %xmm2, %xmm2
+; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm0 = ymm2[0],ymm1[1],ymm2[2,3,4,5,6,7]
+; AVX2-SLOW-NEXT:    vbroadcastss %xmm1, %ymm1
+; AVX2-SLOW-NEXT:    vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2],ymm2[3,4,5,6,7]
 ; AVX2-SLOW-NEXT:    retq
 ;
 ; AVX2-FAST-LABEL: splat_v3i32:
 ; AVX2-FAST:       # %bb.0:
-; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
-; AVX2-FAST-NEXT:    vpinsrd $2, 8(%rdi), %xmm0, %xmm1
+; AVX2-FAST-NEXT:    vmovq {{.*#+}} xmm1 = mem[0],zero
 ; AVX2-FAST-NEXT:    vpxor %xmm0, %xmm0, %xmm0
 ; AVX2-FAST-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2,3,4,5,6,7]
 ; AVX2-FAST-NEXT:    vpshufb {{.*#+}} ymm1 = zero,zero,zero,zero,zero,zero,zero,zero,ymm1[0,1,2,3],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero