From b743e94cdca985cb676049290af4f49b6e49572f Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Sat, 14 Sep 2019 16:38:26 +0000
Subject: [PATCH] [TargetLowering] SimplifyDemandedBits - add EXTRACT_SUBVECTOR
 support.

Call SimplifyDemandedBits on the source vector.

llvm-svn: 371923
---
 llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp | 15 +++++++++++++++
 llvm/test/CodeGen/X86/vector-reduce-mul.ll       | 18 +++++++-----------
 2 files changed, 22 insertions(+), 11 deletions(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 4fc9f24..41b3ba1 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -907,6 +907,21 @@ bool TargetLowering::SimplifyDemandedBits(
     }
     break;
   }
+  case ISD::EXTRACT_SUBVECTOR: {
+    // If index isn't constant, assume we need all the source vector elements.
+    SDValue Src = Op.getOperand(0);
+    ConstantSDNode *SubIdx = dyn_cast<ConstantSDNode>(Op.getOperand(1));
+    unsigned NumSrcElts = Src.getValueType().getVectorNumElements();
+    APInt SrcElts = APInt::getAllOnesValue(NumSrcElts);
+    if (SubIdx && SubIdx->getAPIntValue().ule(NumSrcElts - NumElts)) {
+      // Offset the demanded elts by the subvector index.
+      uint64_t Idx = SubIdx->getZExtValue();
+      SrcElts = DemandedElts.zextOrSelf(NumSrcElts).shl(Idx);
+    }
+    if (SimplifyDemandedBits(Src, DemandedBits, SrcElts, Known, TLO, Depth + 1))
+      return true;
+    break;
+  }
   case ISD::CONCAT_VECTORS: {
     Known.Zero.setAllBits();
     Known.One.setAllBits();
diff --git a/llvm/test/CodeGen/X86/vector-reduce-mul.ll b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
index d42a7f8..88b0f0b 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-mul.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-mul.ll
@@ -1800,17 +1800,13 @@ define i8 @test_v16i8(<16 x i8> %a0) {
 ; AVX2-NEXT:    vpunpckhbw {{.*#+}} xmm1 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15]
 ; AVX2-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
 ; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vmovdqa {{.*#+}} xmm1 = [255,255,255,255,255,255,255,255]
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[2,3,2,3]
-; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrlq $32, %xmm0, %xmm2
-; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpsrld $16, %xmm0, %xmm2
-; AVX2-NEXT:    vpmullw %xmm2, %xmm0, %xmm0
-; AVX2-NEXT:    vpand %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
+; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
+; AVX2-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
+; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm0[2],zero,xmm0[2],zero,xmm0[4],zero,xmm0[6],zero,xmm0[8],zero,xmm0[10],zero,xmm0[12],zero,xmm0[14],zero
+; AVX2-NEXT:    vpmullw %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpextrb $0, %xmm0, %eax
 ; AVX2-NEXT:    # kill: def $al killed $al killed $eax
 ; AVX2-NEXT:    retq
-- 
2.7.4