[CostModel][X86][XOP] Improve costs for XOP shuffles

VPPERM/VPERMIL2PD/VPERMIL2PS all provide more effective 2-input shuffles than regular AVX instructions

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@311005 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86TargetTransformInfo.cpp b/lib/Target/X86/X86TargetTransformInfo.cpp
index ed56e4512f68e09780dc245cfb6a6165592b5f5b..2f44d610840913daf6880cfa51884f41ca1b23f0 100644 (file)
--- a/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -861,6 +861,28 @@ int X86TTIImpl::getShuffleCost(TTI::ShuffleKind Kind, Type *Tp, int Index,
     if (const auto *Entry = CostTableLookup(AVX2ShuffleTbl, Kind, LT.second))
       return LT.first * Entry->Cost;
 
+  static const CostTblEntry XOPShuffleTbl[] = {
+    { TTI::SK_PermuteSingleSrc, MVT::v4f64,   2 }, // vperm2f128 + vpermil2pd
+    { TTI::SK_PermuteSingleSrc, MVT::v8f32,   2 }, // vperm2f128 + vpermil2ps
+    { TTI::SK_PermuteSingleSrc, MVT::v4i64,   2 }, // vperm2f128 + vpermil2pd
+    { TTI::SK_PermuteSingleSrc, MVT::v8i32,   2 }, // vperm2f128 + vpermil2ps
+    { TTI::SK_PermuteSingleSrc, MVT::v16i16,  4 }, // vextractf128 + 2*vpperm
+                                                   // + vinsertf128
+    { TTI::SK_PermuteSingleSrc, MVT::v32i8,   4 }, // vextractf128 + 2*vpperm
+                                                   // + vinsertf128
+
+    { TTI::SK_PermuteTwoSrc,    MVT::v16i16,  9 }, // 2*vextractf128 + 6*vpperm
+                                                   // + vinsertf128
+    { TTI::SK_PermuteTwoSrc,    MVT::v8i16,   1 }, // vpperm
+    { TTI::SK_PermuteTwoSrc,    MVT::v32i8,   9 }, // 2*vextractf128 + 6*vpperm
+                                                   // + vinsertf128
+    { TTI::SK_PermuteTwoSrc,    MVT::v16i8,   1 }, // vpperm
+  };
+
+  if (ST->hasXOP())
+    if (const auto *Entry = CostTableLookup(XOPShuffleTbl, Kind, LT.second))
+      return LT.first * Entry->Cost;
+
   static const CostTblEntry AVX1ShuffleTbl[] = {
     { TTI::SK_Broadcast, MVT::v4f64,  2 }, // vperm2f128 + vpermilpd
     { TTI::SK_Broadcast, MVT::v8f32,  2 }, // vperm2f128 + vpermilps

diff --git a/test/Analysis/CostModel/X86/shuffle-single-src.ll b/test/Analysis/CostModel/X86/shuffle-single-src.ll
index 590f80d42539fc3d398b4a53fc9e2ed19046e8ed..96f94f4b2887b6c9db34c743718d31bab2927b7d 100644 (file)
--- a/test/Analysis/CostModel/X86/shuffle-single-src.ll
+++ b/test/Analysis/CostModel/X86/shuffle-single-src.ll
@@ -1,6 +1,7 @@
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE2
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+ssse3 | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSSE3
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+sse4.2 | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE42
+; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+xop | FileCheck %s -check-prefix=CHECK -check-prefix=XOP
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+avx | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512F
@@ -17,6 +18,7 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
   ; SSE2: cost of 1 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 1 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 1 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -25,6 +27,7 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
   ; SSE2: cost of 2 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 2 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 2 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 2 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 3 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V256 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V256 = shufflevector
@@ -33,6 +36,7 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
   ; SSE2: cost of 12 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 12 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 12 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 8 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 8 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 6 {{.*}} %V512 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V512 = shufflevector
@@ -41,6 +45,7 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
   ; SSE2: cost of 56 {{.*}} %V1024 = shufflevector
   ; SSSE3: cost of 56 {{.*}} %V1024 = shufflevector
   ; SSE42: cost of 56 {{.*}} %V1024 = shufflevector
+  ; XOP: cost of 48 {{.*}} %V1024 = shufflevector
   ; AVX1: cost of 48 {{.*}} %V1024 = shufflevector
   ; AVX2: cost of 36 {{.*}} %V1024 = shufflevector
   ; AVX512: cost of 2 {{.*}} %V1024 = shufflevector
@@ -55,6 +60,7 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512)
   ; SSE2: cost of 1 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 1 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 1 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -63,6 +69,7 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512)
   ; SSE2: cost of 2 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 2 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 2 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 2 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 3 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V256 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V256 = shufflevector
@@ -71,6 +78,7 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512)
   ; SSE2: cost of 12 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 12 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 12 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 8 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 8 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 6 {{.*}} %V512 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V512 = shufflevector
@@ -85,6 +93,7 @@ define void @test_vXf32(<4 x float> %src128, <8 x float> %src256, <16 x float> %
   ; SSE2: cost of 1 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 1 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 1 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -93,6 +102,7 @@ define void @test_vXf32(<4 x float> %src128, <8 x float> %src256, <16 x float> %
   ; SSE2: cost of 4 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 4 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 4 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 2 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V256 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V256 = shufflevector
@@ -101,6 +111,7 @@ define void @test_vXf32(<4 x float> %src128, <8 x float> %src256, <16 x float> %
   ; SSE2: cost of 24 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 24 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 24 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 8 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 8 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 6 {{.*}} %V512 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V512 = shufflevector
@@ -115,6 +126,7 @@ define void @test_vXi32(<4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512
   ; SSE2: cost of 1 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 1 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 1 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -123,6 +135,7 @@ define void @test_vXi32(<4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512
   ; SSE2: cost of 4 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 4 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 4 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 2 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V256 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V256 = shufflevector
@@ -131,6 +144,7 @@ define void @test_vXi32(<4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512
   ; SSE2: cost of 24 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 24 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 24 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 8 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 8 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 6 {{.*}} %V512 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V512 = shufflevector
@@ -139,6 +153,7 @@ define void @test_vXi32(<4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512
   ; SSE2: cost of 112 {{.*}} %V1024 = shufflevector
   ; SSSE3: cost of 112 {{.*}} %V1024 = shufflevector
   ; SSE42: cost of 112 {{.*}} %V1024 = shufflevector
+  ; XOP: cost of 48 {{.*}} %V1024 = shufflevector
   ; AVX1: cost of 48 {{.*}} %V1024 = shufflevector
   ; AVX2: cost of 36 {{.*}} %V1024 = shufflevector
   ; AVX512: cost of 2 {{.*}} %V1024 = shufflevector
@@ -152,6 +167,7 @@ define void @test_vXi16(<8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src51
   ; SSE2: cost of 5 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 1 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 1 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX512F: cost of 1 {{.*}} %V128 = shufflevector
@@ -162,6 +178,7 @@ define void @test_vXi16(<8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src51
   ; SSE2: cost of 16 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 6 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 6 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 8 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX512F: cost of 4 {{.*}} %V256 = shufflevector
@@ -172,6 +189,7 @@ define void @test_vXi16(<8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src51
   ; SSE2: cost of 96 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 36 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 36 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 18 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 30 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 14 {{.*}} %V512 = shufflevector
   ; AVX512F: cost of 14 {{.*}} %V512 = shufflevector
@@ -182,6 +200,7 @@ define void @test_vXi16(<8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src51
   ; SSE2: cost of 448 {{.*}} %V1024 = shufflevector
   ; SSSE3: cost of 168 {{.*}} %V1024 = shufflevector
   ; SSE42: cost of 168 {{.*}} %V1024 = shufflevector
+  ; XOP: cost of 108 {{.*}} %V1024 = shufflevector
   ; AVX1: cost of 180 {{.*}} %V1024 = shufflevector
   ; AVX2: cost of 84 {{.*}} %V1024 = shufflevector
   ; AVX512F: cost of 84 {{.*}} %V1024 = shufflevector
@@ -196,6 +215,7 @@ define void @test_vXi8(<16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512)
   ; SSE2: cost of 10 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 1 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 1 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -204,6 +224,7 @@ define void @test_vXi8(<16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512)
   ; SSE2: cost of 26 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 6 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 6 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 8 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX512F: cost of 4 {{.*}} %V256 = shufflevector
@@ -214,6 +235,7 @@ define void @test_vXi8(<16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512)
   ; SSE2: cost of 156 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 36 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 36 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 18 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 30 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 14 {{.*}} %V512 = shufflevector
   ; AVX512F: cost of 14 {{.*}} %V512 = shufflevector

diff --git a/test/Analysis/CostModel/X86/shuffle-two-src.ll b/test/Analysis/CostModel/X86/shuffle-two-src.ll
index 1967397c3717bf25c2922d328880251c1c3e8db1..628e7cabc1e1d800d9095df54c38d53109c97fd0 100644 (file)
--- a/test/Analysis/CostModel/X86/shuffle-two-src.ll
+++ b/test/Analysis/CostModel/X86/shuffle-two-src.ll
@@ -1,6 +1,7 @@
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2 | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE2
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+ssse3 | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSSE3
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+sse4.2 | FileCheck %s -check-prefix=CHECK -check-prefix=SSE -check-prefix=SSE42
+; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+xop | FileCheck %s -check-prefix=CHECK -check-prefix=XOP
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+avx | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX1
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+avx2 | FileCheck %s -check-prefix=CHECK -check-prefix=AVX -check-prefix=AVX2
 ; RUN: opt < %s -cost-model -analyze -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f | FileCheck %s --check-prefix=CHECK --check-prefix=AVX512 --check-prefix=AVX512F
@@ -17,6 +18,7 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
   ; SSE2: cost of 1 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 1 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 1 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -25,6 +27,7 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
   ; SSE2: cost of 6 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 6 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 6 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 3 {{.*}} %V256 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V256 = shufflevector
@@ -33,6 +36,7 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
   ; SSE2: cost of 28 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 28 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 28 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 24 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 24 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 18 {{.*}} %V512 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V512 = shufflevector
@@ -41,6 +45,7 @@ define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double>
   ; SSE2: cost of 120 {{.*}} %V1024 = shufflevector
   ; SSSE3: cost of 120 {{.*}} %V1024 = shufflevector
   ; SSE42: cost of 120 {{.*}} %V1024 = shufflevector
+  ; XOP: cost of 112 {{.*}} %V1024 = shufflevector
   ; AVX1: cost of 112 {{.*}} %V1024 = shufflevector
   ; AVX2: cost of 84 {{.*}} %V1024 = shufflevector
   ; AVX512: cost of 6 {{.*}} %V1024 = shufflevector
@@ -55,6 +60,7 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512,
   ; SSE2: cost of 1 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 1 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 1 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -63,6 +69,7 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512,
   ; SSE2: cost of 6 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 6 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 6 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 3 {{.*}} %V256 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V256 = shufflevector
@@ -71,6 +78,7 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512,
   ; SSE2: cost of 28 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 28 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 28 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 24 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 24 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 18 {{.*}} %V512 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V512 = shufflevector
@@ -79,6 +87,7 @@ define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512,
   ; SSE2: cost of 120 {{.*}} %V1024 = shufflevector
   ; SSSE3: cost of 120 {{.*}} %V1024 = shufflevector
   ; SSE42: cost of 120 {{.*}} %V1024 = shufflevector
+  ; XOP: cost of 112 {{.*}} %V1024 = shufflevector
   ; AVX1: cost of 112 {{.*}} %V1024 = shufflevector
   ; AVX2: cost of 84 {{.*}} %V1024 = shufflevector
   ; AVX512: cost of 6 {{.*}} %V1024 = shufflevector
@@ -93,6 +102,7 @@ define void @test_vXf32(<4 x float> %src128, <8 x float> %src256, <16 x float> %
   ; SSE2: cost of 2 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 2 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 2 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 2 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 2 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 2 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -101,6 +111,7 @@ define void @test_vXf32(<4 x float> %src128, <8 x float> %src256, <16 x float> %
   ; SSE2: cost of 12 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 12 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 12 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 3 {{.*}} %V256 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V256 = shufflevector
@@ -109,6 +120,7 @@ define void @test_vXf32(<4 x float> %src128, <8 x float> %src256, <16 x float> %
   ; SSE2: cost of 56 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 56 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 56 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 24 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 24 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 18 {{.*}} %V512 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V512 = shufflevector
@@ -117,6 +129,7 @@ define void @test_vXf32(<4 x float> %src128, <8 x float> %src256, <16 x float> %
   ; SSE2: cost of 240 {{.*}} %V1024 = shufflevector
   ; SSSE3: cost of 240 {{.*}} %V1024 = shufflevector
   ; SSE42: cost of 240 {{.*}} %V1024 = shufflevector
+  ; XOP: cost of 112 {{.*}} %V1024 = shufflevector
   ; AVX1: cost of 112 {{.*}} %V1024 = shufflevector
   ; AVX2: cost of 84 {{.*}} %V1024 = shufflevector
   ; AVX512: cost of 6 {{.*}} %V1024 = shufflevector
@@ -131,6 +144,7 @@ define void @test_vXi32(<4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512
   ; SSE2: cost of 2 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 2 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 2 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 2 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 2 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 2 {{.*}} %V128 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V128 = shufflevector
@@ -139,6 +153,7 @@ define void @test_vXi32(<4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512
   ; SSE2: cost of 12 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 12 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 12 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 4 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 3 {{.*}} %V256 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V256 = shufflevector
@@ -147,6 +162,7 @@ define void @test_vXi32(<4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512
   ; SSE2: cost of 56 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 56 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 56 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 24 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 24 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 18 {{.*}} %V512 = shufflevector
   ; AVX512: cost of 1 {{.*}} %V512 = shufflevector
@@ -155,6 +171,7 @@ define void @test_vXi32(<4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512
   ; SSE2: cost of 240 {{.*}} %V1024 = shufflevector
   ; SSSE3: cost of 240 {{.*}} %V1024 = shufflevector
   ; SSE42: cost of 240 {{.*}} %V1024 = shufflevector
+  ; XOP: cost of 112 {{.*}} %V1024 = shufflevector
   ; AVX1: cost of 112 {{.*}} %V1024 = shufflevector
   ; AVX2: cost of 84 {{.*}} %V1024 = shufflevector
   ; AVX512: cost of 6 {{.*}} %V1024 = shufflevector
@@ -169,6 +186,7 @@ define void @test_vXi16(<8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src51
   ; SSE2: cost of 8 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 3 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 3 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 3 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 3 {{.*}} %V128 = shufflevector
   ; AVX512F: cost of 3 {{.*}} %V128 = shufflevector
@@ -179,6 +197,7 @@ define void @test_vXi16(<8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src51
   ; SSE2: cost of 48 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 18 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 18 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 9 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 15 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 7 {{.*}} %V256 = shufflevector
   ; AVX512F: cost of 7 {{.*}} %V256 = shufflevector
@@ -189,6 +208,7 @@ define void @test_vXi16(<8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src51
   ; SSE2: cost of 224 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 84 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 84 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 54 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 90 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 42 {{.*}} %V512 = shufflevector
   ; AVX512F: cost of 42 {{.*}} %V512 = shufflevector
@@ -199,6 +219,7 @@ define void @test_vXi16(<8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src51
   ; SSE2: cost of 960 {{.*}} %V1024 = shufflevector
   ; SSSE3: cost of 360 {{.*}} %V1024 = shufflevector
   ; SSE42: cost of 360 {{.*}} %V1024 = shufflevector
+  ; XOP: cost of 252 {{.*}} %V1024 = shufflevector
   ; AVX1: cost of 420 {{.*}} %V1024 = shufflevector
   ; AVX2: cost of 196 {{.*}} %V1024 = shufflevector
   ; AVX512F: cost of 196 {{.*}} %V1024 = shufflevector
@@ -215,6 +236,7 @@ define void @test_vXi8(<16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512,
   ; SSE2: cost of 13 {{.*}} %V128 = shufflevector
   ; SSSE3: cost of 3 {{.*}} %V128 = shufflevector
   ; SSE42: cost of 3 {{.*}} %V128 = shufflevector
+  ; XOP: cost of 1 {{.*}} %V128 = shufflevector
   ; AVX1: cost of 3 {{.*}} %V128 = shufflevector
   ; AVX2: cost of 3 {{.*}} %V128 = shufflevector
   ; AVX512F: cost of 3 {{.*}} %V128 = shufflevector
@@ -225,6 +247,7 @@ define void @test_vXi8(<16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512,
   ; SSE2: cost of 78 {{.*}} %V256 = shufflevector
   ; SSSE3: cost of 18 {{.*}} %V256 = shufflevector
   ; SSE42: cost of 18 {{.*}} %V256 = shufflevector
+  ; XOP: cost of 9 {{.*}} %V256 = shufflevector
   ; AVX1: cost of 15 {{.*}} %V256 = shufflevector
   ; AVX2: cost of 7 {{.*}} %V256 = shufflevector
   ; AVX512F: cost of 7 {{.*}} %V256 = shufflevector
@@ -235,6 +258,7 @@ define void @test_vXi8(<16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512,
   ; SSE2: cost of 364 {{.*}} %V512 = shufflevector
   ; SSSE3: cost of 84 {{.*}} %V512 = shufflevector
   ; SSE42: cost of 84 {{.*}} %V512 = shufflevector
+  ; XOP: cost of 54 {{.*}} %V512 = shufflevector
   ; AVX1: cost of 90 {{.*}} %V512 = shufflevector
   ; AVX2: cost of 42 {{.*}} %V512 = shufflevector
   ; AVX512F: cost of 42 {{.*}} %V512 = shufflevector