From c4c427ffea861bf35955fe653827bc51bca1846c Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Thu, 11 Apr 2019 14:04:19 +0000
Subject: [PATCH] [X86] SimplifyDemandedVectorElts - add X86ISD::VPERMIL2 mask
 support

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@358167 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp            |  4 ++--
 .../X86/vector-shuffle-combining-xop.ll       | 24 ++++++-------------
 2 files changed, 9 insertions(+), 19 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 223d0923c7f..a896f24c365 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -33234,8 +33234,8 @@ bool X86TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
       return true;
     break;
   }
-  case X86ISD::VPPERM: {
-    // TODO - simplify other variable shuffle masks.
+  case X86ISD::VPPERM:
+  case X86ISD::VPERMIL2: {
     SDValue Mask = Op.getOperand(2);
     APInt MaskUndef, MaskZero;
     if (SimplifyDemandedVectorElts(Mask, DemandedElts, MaskUndef, MaskZero, TLO,
diff --git a/test/CodeGen/X86/vector-shuffle-combining-xop.ll b/test/CodeGen/X86/vector-shuffle-combining-xop.ll
index 3d57a09ffb5..e785530ccd5 100644
--- a/test/CodeGen/X86/vector-shuffle-combining-xop.ll
+++ b/test/CodeGen/X86/vector-shuffle-combining-xop.ll
@@ -153,23 +153,13 @@ define <4 x double> @demandedelts_vpermil2pd256_as_shufpd(<4 x double> %a0, <4 x
 ; X86-AVX2-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,1,2,3]
 ; X86-AVX2-NEXT:    retl
 ;
-; X64-AVX-LABEL: demandedelts_vpermil2pd256_as_shufpd:
-; X64-AVX:       # %bb.0:
-; X64-AVX-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,4,2,7>
-; X64-AVX-NEXT:    vpinsrq $0, %rdi, %xmm2, %xmm2
-; X64-AVX-NEXT:    vblendpd {{.*#+}} ymm2 = ymm2[0,1],mem[2,3]
-; X64-AVX-NEXT:    vpermil2pd $0, %ymm2, %ymm1, %ymm0, %ymm0
-; X64-AVX-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,1,2,3]
-; X64-AVX-NEXT:    retq
-;
-; X64-AVX2-LABEL: demandedelts_vpermil2pd256_as_shufpd:
-; X64-AVX2:       # %bb.0:
-; X64-AVX2-NEXT:    vmovdqa {{.*#+}} xmm2 = <u,4,2,7>
-; X64-AVX2-NEXT:    vpinsrq $0, %rdi, %xmm2, %xmm2
-; X64-AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],mem[4,5,6,7]
-; X64-AVX2-NEXT:    vpermil2pd $0, %ymm2, %ymm1, %ymm0, %ymm0
-; X64-AVX2-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,1,2,3]
-; X64-AVX2-NEXT:    retq
+; X64-LABEL: demandedelts_vpermil2pd256_as_shufpd:
+; X64:       # %bb.0:
+; X64-NEXT:    vmovapd {{.*#+}} xmm2 = <u,4,2,7>
+; X64-NEXT:    vblendpd {{.*#+}} ymm2 = ymm2[0,1],mem[2,3]
+; X64-NEXT:    vpermil2pd $0, %ymm2, %ymm1, %ymm0, %ymm0
+; X64-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,1,2,3]
+; X64-NEXT:    retq
   %res0 = insertelement <4 x i64> <i64 0, i64 4, i64 2, i64 7>, i64 %a2, i32 0
   %res1 = call <4 x double> @llvm.x86.xop.vpermil2pd.256(<4 x double> %a0, <4 x double> %a1, <4 x i64> %res0, i8 0)
   %res2 = shufflevector <4 x double> %res1, <4 x double> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 3>
-- 
2.40.0