From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Fri, 28 Jun 2019 17:57:32 +0000 (+0000)
Subject: [X86] CombineShuffleWithExtract - recurse through EXTRACT_SUBVECTOR chain
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=08a4b34e50b6e2c0b2bdfdcd77025f188dcfb842;p=llvm

[X86] CombineShuffleWithExtract - recurse through EXTRACT_SUBVECTOR chain

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@364667 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index c31e452ed7f..2aa9f50933c 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -32047,15 +32047,15 @@ static SDValue combineX86ShuffleChain(ArrayRef<SDValue> Inputs, SDValue Root,
 
     SDValue Src1 = V1, Src2 = V2;
     unsigned Offset1 = 0, Offset2 = 0;
-    if (V1.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
-        isa<ConstantSDNode>(V1.getOperand(1))) {
-      Src1 = V1.getOperand(0);
-      Offset1 = V1.getConstantOperandVal(1);
-    }
-    if (V2.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
-        isa<ConstantSDNode>(V2.getOperand(1))) {
-      Src2 = V2.getOperand(0);
-      Offset2 = V2.getConstantOperandVal(1);
+    while (Src1.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
+           isa<ConstantSDNode>(Src1.getOperand(1))) {
+      Offset1 += Src1.getConstantOperandVal(1);
+      Src1 = Src1.getOperand(0);
+    }
+    while (Src2.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
+           isa<ConstantSDNode>(Src2.getOperand(1))) {
+      Offset2 += Src2.getConstantOperandVal(1);
+      Src2 = Src2.getOperand(0);
     }
     if (Offset1 == 0 && Offset2 == 0)
       return false;
diff --git a/test/CodeGen/X86/shuffle-vs-trunc-512-widen.ll b/test/CodeGen/X86/shuffle-vs-trunc-512-widen.ll
index 246ed3e6f54..d6648da025e 100644
--- a/test/CodeGen/X86/shuffle-vs-trunc-512-widen.ll
+++ b/test/CodeGen/X86/shuffle-vs-trunc-512-widen.ll
@@ -716,11 +716,8 @@ define <16 x i8> @trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_
 ;
 ; AVX512VBMI-LABEL: trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_57_61:
 ; AVX512VBMI:       # %bb.0:
-; AVX512VBMI-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VBMI-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512VBMI-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VBMI-NEXT:    vpshufb %xmm2, %xmm0, %xmm2
-; AVX512VBMI-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX512VBMI-NEXT:    vpbroadcastq {{.*#+}} zmm1 = [2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785]
+; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm1, %zmm1
 ; AVX512VBMI-NEXT:    vpbroadcastq {{.*#+}} zmm2 = [4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321]
 ; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm2, %zmm0
 ; AVX512VBMI-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
@@ -804,11 +801,8 @@ define <16 x i8> @trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_
 ;
 ; AVX512VBMI-LABEL: trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_57_62:
 ; AVX512VBMI:       # %bb.0:
-; AVX512VBMI-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VBMI-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512VBMI-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VBMI-NEXT:    vpshufb %xmm2, %xmm0, %xmm2
-; AVX512VBMI-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX512VBMI-NEXT:    vpbroadcastq {{.*#+}} zmm1 = [2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785]
+; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm1, %zmm1
 ; AVX512VBMI-NEXT:    vpbroadcastq {{.*#+}} zmm2 = [4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257]
 ; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm2, %zmm0
 ; AVX512VBMI-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
diff --git a/test/CodeGen/X86/shuffle-vs-trunc-512.ll b/test/CodeGen/X86/shuffle-vs-trunc-512.ll
index bf704873d6f..a98dbfd317b 100644
--- a/test/CodeGen/X86/shuffle-vs-trunc-512.ll
+++ b/test/CodeGen/X86/shuffle-vs-trunc-512.ll
@@ -707,11 +707,8 @@ define <16 x i8> @trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_
 ;
 ; AVX512VBMI-LABEL: trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_57_61:
 ; AVX512VBMI:       # %bb.0:
-; AVX512VBMI-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VBMI-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512VBMI-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VBMI-NEXT:    vpshufb %xmm2, %xmm0, %xmm2
-; AVX512VBMI-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX512VBMI-NEXT:    vpbroadcastq {{.*#+}} zmm1 = [2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785]
+; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm1, %zmm1
 ; AVX512VBMI-NEXT:    vpbroadcastq {{.*#+}} zmm2 = [4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321,4411615795313452321]
 ; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm2, %zmm0
 ; AVX512VBMI-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]
@@ -795,11 +792,8 @@ define <16 x i8> @trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_
 ;
 ; AVX512VBMI-LABEL: trunc_shuffle_v64i8_01_05_09_13_17_21_25_29_33_37_41_45_49_53_57_62:
 ; AVX512VBMI:       # %bb.0:
-; AVX512VBMI-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX512VBMI-NEXT:    vmovdqa {{.*#+}} xmm2 = <1,5,9,13,u,u,u,u,u,u,u,u,u,u,u,u>
-; AVX512VBMI-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
-; AVX512VBMI-NEXT:    vpshufb %xmm2, %xmm0, %xmm2
-; AVX512VBMI-NEXT:    vpunpckldq {{.*#+}} xmm1 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
+; AVX512VBMI-NEXT:    vpbroadcastq {{.*#+}} zmm1 = [2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785,2096730264494998785]
+; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm1, %zmm1
 ; AVX512VBMI-NEXT:    vpbroadcastq {{.*#+}} zmm2 = [4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257,4483673389351380257]
 ; AVX512VBMI-NEXT:    vpermb %zmm0, %zmm2, %zmm0
 ; AVX512VBMI-NEXT:    vpblendd {{.*#+}} xmm0 = xmm1[0,1],xmm0[2,3]