From 9aa25be14910c773ec2426a4605eca344646a6f1 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Tue, 18 Jun 2019 15:49:35 +0000
Subject: [PATCH] [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
---
 llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp | 9 +++++++++
 llvm/test/CodeGen/X86/vector-reduce-mul-widen.ll | 1 -
 2 files changed, 9 insertions(+), 1 deletion(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 0a688b793573..e27a88baf411 100644
--- 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 ac3286282015..573f9c836c21 100644
--- 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
-- 
2.34.1