[TargetLowering] SimplifyDemandedVectorElts - support MUL and ANY_EXTEND_VECTOR_INREG

Also fold ANY_EXTEND_VECTOR_INREG -> BITCAST if we only need the bottom element.

Fixes temporary regression introduced in rL363693.

llvm-svn: 363694

diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 0a688b79357381e2e9fe3631c33865e95d6676e3..e27a88baf411f4399e1dde9d0e4b2b69721c2025 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -2156,6 +2156,7 @@ bool TargetLowering::SimplifyDemandedVectorElts(
     }
     break;
   }
+  case ISD::ANY_EXTEND_VECTOR_INREG:
   case ISD::SIGN_EXTEND_VECTOR_INREG:
   case ISD::ZERO_EXTEND_VECTOR_INREG: {
     APInt SrcUndef, SrcZero;
@@ -2168,6 +2169,13 @@ bool TargetLowering::SimplifyDemandedVectorElts(
     KnownZero = SrcZero.zextOrTrunc(NumElts);
     KnownUndef = SrcUndef.zextOrTrunc(NumElts);
 
+    if (Op.getOpcode() == ISD::ANY_EXTEND_VECTOR_INREG &&
+        Op.getValueSizeInBits() == Src.getValueSizeInBits() &&
+        DemandedSrcElts == 1 && TLO.DAG.getDataLayout().isLittleEndian()) {
+      // aext - if we just need the bottom element then we can bitcast.
+      return TLO.CombineTo(Op, TLO.DAG.getBitcast(VT, Src));
+    }
+
     if (Op.getOpcode() == ISD::ZERO_EXTEND_VECTOR_INREG) {
       // zext(undef) upper bits are guaranteed to be zero.
       if (DemandedElts.isSubsetOf(KnownUndef))
@@ -2201,6 +2209,7 @@ bool TargetLowering::SimplifyDemandedVectorElts(
     KnownUndef = getKnownUndefForVectorBinop(Op, TLO.DAG, UndefLHS, UndefRHS);
     break;
   }
+  case ISD::MUL:
   case ISD::AND: {
     APInt SrcUndef, SrcZero;
     if (SimplifyDemandedVectorElts(Op.getOperand(1), DemandedElts, SrcUndef,

diff --git a/llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll b/llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll
index ac3286282015e08ef90362f439f0d8d2158ee6c9..573f9c836c21f2fcee451388769c396eddbc0ae0 100644 (file)
--- a/llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll
@@ -1837,7 +1837,6 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpackuswb %xmm0, %xmm0, %xmm1
 ; AVX2-NEXT:    vpsrlw $8, %xmm1, %xmm1
-; AVX2-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
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax