From f7b978a71bc057d19966b16945eb5fd5068b309e Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Wed, 1 May 2019 10:58:38 +0000
Subject: [PATCH] [X86][SSE] Move extract_subvector(pshufb) fold to
 SimplifyDemandedVectorEltsForTargetNode

This lets us hit more cases than combineExtractSubvector and allows us reuse more code.

llvm-svn: 359669
---
 llvm/lib/Target/X86/X86ISelLowering.cpp  | 15 +++------------
 llvm/test/CodeGen/X86/var-permute-128.ll | 31 +++++++++++++------------------
 2 files changed, 16 insertions(+), 30 deletions(-)

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 53823e2..c4f4b03 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -33406,6 +33406,9 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
   // TODO: Handle 512-bit -> 128/256-bit ops as well.
   if (VT.is256BitVector() && DemandedElts.lshr(NumElts / 2) == 0) {
     switch (Opc) {
+      // Target Shuffles.
+    case X86ISD::PSHUFB:
+      // Horizontal Ops.
     case X86ISD::HADD:
     case X86ISD::HSUB:
     case X86ISD::FHADD:
@@ -42817,18 +42820,6 @@ static SDValue combineExtractSubvector(SDNode *N, SelectionDAG &DAG,
         InVec.getOperand(0).getSimpleValueType().is128BitVector()) {
       return DAG.getNode(InOpcode, SDLoc(N), VT, InVec.getOperand(0));
     }
-    if (InOpcode == ISD::BITCAST) {
-      // TODO - do this for target shuffles in general.
-      SDValue InVecBC = peekThroughOneUseBitcasts(InVec);
-      if (InVecBC.getOpcode() == X86ISD::PSHUFB && VT.is128BitVector()) {
-        SDLoc DL(N);
-        SDValue SubPSHUFB =
-            DAG.getNode(X86ISD::PSHUFB, DL, MVT::v16i8,
-                        extract128BitVector(InVecBC.getOperand(0), 0, DAG, DL),
-                        extract128BitVector(InVecBC.getOperand(1), 0, DAG, DL));
-        return DAG.getBitcast(VT, SubPSHUFB);
-      }
-    }
   }
 
   return SDValue();
diff --git a/llvm/test/CodeGen/X86/var-permute-128.ll b/llvm/test/CodeGen/X86/var-permute-128.ll
index e557b7c..4d48527 100644
--- a/llvm/test/CodeGen/X86/var-permute-128.ll
+++ b/llvm/test/CodeGen/X86/var-permute-128.ll
@@ -1027,13 +1027,11 @@ define <16 x i8> @var_shuffle_v16i8_from_v32i8_v16i8(<32 x i8> %v, <16 x i8> %in
 ; AVX2-LABEL: var_shuffle_v16i8_from_v32i8_v16i8:
 ; AVX2:       # %bb.0:
 ; AVX2-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
-; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
-; AVX2-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
-; AVX2-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
+; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX2-NEXT:    vpshufb %xmm1, %xmm2, %xmm2
+; AVX2-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX2-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
-; AVX2-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
+; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -1041,13 +1039,11 @@ define <16 x i8> @var_shuffle_v16i8_from_v32i8_v16i8(<32 x i8> %v, <16 x i8> %in
 ; AVX512-LABEL: var_shuffle_v16i8_from_v32i8_v16i8:
 ; AVX512:       # %bb.0:
 ; AVX512-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
-; AVX512-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
-; AVX512-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
-; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; AVX512-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
-; AVX512-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
+; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX512-NEXT:    vpshufb %xmm1, %xmm2, %xmm2
+; AVX512-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX512-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
-; AVX512-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
+; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512-NEXT:    vzeroupper
 ; AVX512-NEXT:    retq
@@ -1055,13 +1051,12 @@ define <16 x i8> @var_shuffle_v16i8_from_v32i8_v16i8(<32 x i8> %v, <16 x i8> %in
 ; AVX512VLBW-LABEL: var_shuffle_v16i8_from_v32i8_v16i8:
 ; AVX512VLBW:       # %bb.0:
 ; AVX512VLBW-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
-; AVX512VLBW-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
-; AVX512VLBW-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
-; AVX512VLBW-NEXT:    vextracti128 $1, %ymm0, %xmm3
-; AVX512VLBW-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
+; AVX512VLBW-NEXT:    vextracti128 $1, %ymm0, %xmm2
+; AVX512VLBW-NEXT:    vpshufb %xmm1, %xmm2, %xmm2
+; AVX512VLBW-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
 ; AVX512VLBW-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %k1
-; AVX512VLBW-NEXT:    vpshufb %ymm1, %ymm0, %ymm2 {%k1}
-; AVX512VLBW-NEXT:    vmovdqa %xmm2, %xmm0
+; AVX512VLBW-NEXT:    vmovdqu8 %ymm2, %ymm0 {%k1}
+; AVX512VLBW-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX512VLBW-NEXT:    vzeroupper
 ; AVX512VLBW-NEXT:    retq
 ;
-- 
2.7.4