From: Andrea Di Biagio <Andrea_DiBiagio@sn.scee.net>
Date: Thu, 3 Jul 2014 22:24:18 +0000 (+0000)
Subject: [CostModel][x86] Improved cost model for alternate shuffles.
X-Git-Url: http://plrg.eecs.uci.edu/git/?a=commitdiff_plain;h=60e9a53c21bef26ed81c1395b569cb48f7b4ef8e;p=oota-llvm.git

[CostModel][x86] Improved cost model for alternate shuffles.

This patch:
 1) Improves the cost model for x86 alternate shuffles (originally
added at revision 211339);
 2) Teaches the Cost Model Analysis pass how to analyze alternate shuffles.

Alternate shuffles are a special kind of blend; on x86, we can often
easily lowered alternate shuffled into single blend
instruction (depending on the subtarget features).

The existing cost model didn't take into account subtarget features.
Also, it had a couple of "dead" entries for vector types that are never
legal (example: on x86 types v2i32 and v2f32 are not legal; those are
always either promoted or widened to 128-bit vector types).

The new x86 cost model takes into account what target features we have
before returning the shuffle cost (i.e. the number of instructions
after the blend is lowered/expanded).

This patch also teaches the Cost Model Analysis how to identify and analyze
alternate shuffles (i.e. 'SK_Alternate' shufflevector instructions):
 - added function 'isAlternateVectorMask';
 - added some logic to check if an instruction is a alternate shuffle and, in
   case, call the target specific TTI to get the corresponding shuffle cost;
 - added a test to verify the cost model analysis on alternate shuffles.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@212296 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Analysis/CostModel.cpp b/lib/Analysis/CostModel.cpp
index 780b1aaa820..1b74f8c19c5 100644
--- a/lib/Analysis/CostModel.cpp
+++ b/lib/Analysis/CostModel.cpp
@@ -95,6 +95,31 @@ static bool isReverseVectorMask(SmallVectorImpl<int> &Mask) {
   return true;
 }
 
+static bool isAlternateVectorMask(SmallVectorImpl<int> &Mask) {
+  bool isAlternate = true;
+  unsigned MaskSize = Mask.size();
+
+  // Example: shufflevector A, B, <0,5,2,7>
+  for (unsigned i = 0; i < MaskSize && isAlternate; ++i) {
+    if (Mask[i] < 0)
+      continue;
+    isAlternate = Mask[i] == (int)((i & 1) ? MaskSize + i : i);
+  }
+
+  if (isAlternate)
+    return true;
+
+  isAlternate = true;
+  // Example: shufflevector A, B, <4,1,6,3>
+  for (unsigned i = 0; i < MaskSize && isAlternate; ++i) {
+    if (Mask[i] < 0)
+      continue;
+    isAlternate = Mask[i] == (int)((i & 1) ? i : MaskSize + i);
+  }
+
+  return isAlternate;
+}
+
 static TargetTransformInfo::OperandValueKind getOperandInfo(Value *V) {
   TargetTransformInfo::OperandValueKind OpInfo =
     TargetTransformInfo::OK_AnyValue;
@@ -466,9 +491,15 @@ unsigned CostModelAnalysis::getInstructionCost(const Instruction *I) const {
     unsigned NumVecElems = VecTypOp0->getVectorNumElements();
     SmallVector<int, 16> Mask = Shuffle->getShuffleMask();
 
-    if (NumVecElems == Mask.size() && isReverseVectorMask(Mask))
-      return TTI->getShuffleCost(TargetTransformInfo::SK_Reverse, VecTypOp0, 0,
-                                 nullptr);
+    if (NumVecElems == Mask.size()) {
+      if (isReverseVectorMask(Mask))
+        return TTI->getShuffleCost(TargetTransformInfo::SK_Reverse, VecTypOp0,
+                                   0, nullptr);
+      if (isAlternateVectorMask(Mask))
+        return TTI->getShuffleCost(TargetTransformInfo::SK_Alternate,
+                                   VecTypOp0, 0, nullptr);
+    }
+
     return -1;
   }
   case Instruction::Call:
diff --git a/lib/Target/X86/X86TargetTransformInfo.cpp b/lib/Target/X86/X86TargetTransformInfo.cpp
index 9fa911967e7..be9caba506d 100644
--- a/lib/Target/X86/X86TargetTransformInfo.cpp
+++ b/lib/Target/X86/X86TargetTransformInfo.cpp
@@ -417,29 +417,99 @@ unsigned X86TTI::getShuffleCost(ShuffleKind Kind, Type *Tp, int Index,
   }
 
   if (Kind == SK_Alternate) {
-    static const CostTblEntry<MVT::SimpleValueType> X86AltShuffleTbl[] = {
-        // Alt shuffle cost table for X86. Cost is the number of instructions
-        // required to create the shuffled vector.
+    // 64-bit packed float vectors (v2f32) are widened to type v4f32.
+    // 64-bit packed integer vectors (v2i32) are promoted to type v2i64.
+    std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Tp);
+
+    // The backend knows how to generate a single VEX.256 version of
+    // instruction VPBLENDW if the target supports AVX2.
+    if (ST->hasAVX2() && LT.second == MVT::v16i16)
+      return LT.first;
 
-        {ISD::VECTOR_SHUFFLE, MVT::v2f32, 1},
-        {ISD::VECTOR_SHUFFLE, MVT::v2i64, 1},
-        {ISD::VECTOR_SHUFFLE, MVT::v2f64, 1},
+    static const CostTblEntry<MVT::SimpleValueType> AVXAltShuffleTbl[] = {
+      {ISD::VECTOR_SHUFFLE, MVT::v4i64, 1},  // vblendpd
+      {ISD::VECTOR_SHUFFLE, MVT::v4f64, 1},  // vblendpd
 
-        {ISD::VECTOR_SHUFFLE, MVT::v2i32, 2},
-        {ISD::VECTOR_SHUFFLE, MVT::v4i32, 2},
-        {ISD::VECTOR_SHUFFLE, MVT::v4f32, 2},
+      {ISD::VECTOR_SHUFFLE, MVT::v8i32, 1},  // vblendps
+      {ISD::VECTOR_SHUFFLE, MVT::v8f32, 1},  // vblendps
 
-        {ISD::VECTOR_SHUFFLE, MVT::v4i16, 8},
-        {ISD::VECTOR_SHUFFLE, MVT::v8i16, 8},
+      // This shuffle is custom lowered into a sequence of:
+      //  2x  vextractf128 , 2x vpblendw , 1x vinsertf128
+      {ISD::VECTOR_SHUFFLE, MVT::v16i16, 5},
 
-        {ISD::VECTOR_SHUFFLE, MVT::v16i8, 49}};
+      // This shuffle is custom lowered into a long sequence of:
+      //  2x vextractf128 , 4x vpshufb , 2x vpor ,  1x vinsertf128
+      {ISD::VECTOR_SHUFFLE, MVT::v32i8, 9}
+    };
 
-    std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Tp);
+    if (ST->hasAVX()) {
+      int Idx = CostTableLookup(AVXAltShuffleTbl, ISD::VECTOR_SHUFFLE, LT.second);
+      if (Idx != -1)
+        return LT.first * AVXAltShuffleTbl[Idx].Cost;
+    }
+
+    static const CostTblEntry<MVT::SimpleValueType> SSE41AltShuffleTbl[] = {
+      // These are lowered into movsd.
+      {ISD::VECTOR_SHUFFLE, MVT::v2i64, 1},
+      {ISD::VECTOR_SHUFFLE, MVT::v2f64, 1},
+
+      // packed float vectors with four elements are lowered into BLENDI dag
+      // nodes. A v4i32/v4f32 BLENDI generates a single 'blendps'/'blendpd'.
+      {ISD::VECTOR_SHUFFLE, MVT::v4i32, 1},
+      {ISD::VECTOR_SHUFFLE, MVT::v4f32, 1},
+
+      // This shuffle generates a single pshufw.
+      {ISD::VECTOR_SHUFFLE, MVT::v8i16, 1},
+
+      // There is no instruction that matches a v16i8 alternate shuffle.
+      // The backend will expand it into the sequence 'pshufb + pshufb + or'.
+      {ISD::VECTOR_SHUFFLE, MVT::v16i8, 3}
+    };
+
+    if (ST->hasSSE41()) {
+      int Idx = CostTableLookup(SSE41AltShuffleTbl, ISD::VECTOR_SHUFFLE, LT.second);
+      if (Idx != -1)
+        return LT.first * SSE41AltShuffleTbl[Idx].Cost;
+    }
+
+    static const CostTblEntry<MVT::SimpleValueType> SSSE3AltShuffleTbl[] = {
+      {ISD::VECTOR_SHUFFLE, MVT::v2i64, 1},  // movsd
+      {ISD::VECTOR_SHUFFLE, MVT::v2f64, 1},  // movsd
+
+      // SSE3 doesn't have 'blendps'. The following shuffles are expanded into
+      // the sequence 'shufps + pshufd'
+      {ISD::VECTOR_SHUFFLE, MVT::v4i32, 2},
+      {ISD::VECTOR_SHUFFLE, MVT::v4f32, 2},
+
+      {ISD::VECTOR_SHUFFLE, MVT::v8i16, 3}, // pshufb + pshufb + or
+      {ISD::VECTOR_SHUFFLE, MVT::v16i8, 3}  // pshufb + pshufb + or
+    };
+ 
+    if (ST->hasSSSE3()) {
+      int Idx = CostTableLookup(SSSE3AltShuffleTbl, ISD::VECTOR_SHUFFLE, LT.second);
+      if (Idx != -1)
+        return LT.first * SSSE3AltShuffleTbl[Idx].Cost;
+    }
 
-    int Idx = CostTableLookup(X86AltShuffleTbl, ISD::VECTOR_SHUFFLE, LT.second);
-    if (Idx == -1)
-      return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
-    return LT.first * X86AltShuffleTbl[Idx].Cost;
+    static const CostTblEntry<MVT::SimpleValueType> SSEAltShuffleTbl[] = {
+      {ISD::VECTOR_SHUFFLE, MVT::v2i64, 1},  // movsd
+      {ISD::VECTOR_SHUFFLE, MVT::v2f64, 1},  // movsd
+
+      {ISD::VECTOR_SHUFFLE, MVT::v4i32, 2}, // shufps + pshufd
+      {ISD::VECTOR_SHUFFLE, MVT::v4f32, 2}, // shufps + pshufd
+ 
+      // This is expanded into a long sequence of four extract + four insert.
+      {ISD::VECTOR_SHUFFLE, MVT::v8i16, 8}, // 4 x pextrw + 4 pinsrw.
+
+      // 8 x (pinsrw + pextrw + and + movb + movzb + or)
+      {ISD::VECTOR_SHUFFLE, MVT::v16i8, 48}
+    };
+
+    // Fall-back (SSE3 and SSE2). 
+    int Idx = CostTableLookup(SSEAltShuffleTbl, ISD::VECTOR_SHUFFLE, LT.second);
+    if (Idx != -1)
+      return LT.first * SSEAltShuffleTbl[Idx].Cost;
+    return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
   }
 
   return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
diff --git a/test/Analysis/CostModel/X86/alternate-shuffle-cost.ll b/test/Analysis/CostModel/X86/alternate-shuffle-cost.ll
new file mode 100644
index 00000000000..2e162f0f000
--- /dev/null
+++ b/test/Analysis/CostModel/X86/alternate-shuffle-cost.ll
@@ -0,0 +1,347 @@
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2,-ssse3 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE2
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+sse2,+sse3,+ssse3 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSSE3
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=SSE41
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7-avx -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX
+; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=core-avx2 -cost-model -analyze | FileCheck %s -check-prefix=CHECK -check-prefix=AVX2
+
+
+; Verify the cost model for alternate shuffles.
+
+; shufflevector instructions with illegal 64-bit vector types.
+; 64-bit packed integer vectors (v2i32) are promoted to type v2i64.
+; 64-bit packed float vectors (v2f32) are widened to type v4f32.
+
+define <2 x i32> @test_v2i32(<2 x i32> %a, <2 x i32> %b) {
+  %1 = shufflevector <2 x i32> %a, <2 x i32> %b, <2 x i32> <i32 0, i32 3>
+  ret <2 x i32> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v2i32':
+; SSE2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+define <2 x float> @test_v2f32(<2 x float> %a, <2 x float> %b) {
+  %1 = shufflevector <2 x float> %a, <2 x float> %b, <2 x i32> <i32 0, i32 3>
+  ret <2 x float> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v2f32':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+define <2 x i32> @test_v2i32_2(<2 x i32> %a, <2 x i32> %b) {
+  %1 = shufflevector <2 x i32> %a, <2 x i32> %b, <2 x i32> <i32 2, i32 1>
+  ret <2 x i32> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v2i32_2':
+; SSE2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+define <2 x float> @test_v2f32_2(<2 x float> %a, <2 x float> %b) {
+  %1 = shufflevector <2 x float> %a, <2 x float> %b, <2 x i32> <i32 2, i32 1>
+  ret <2 x float> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v2f32_2':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+; Test shuffles on packed vectors of two elements.
+
+define <2 x i64> @test_v2i64(<2 x i64> %a, <2 x i64> %b) {
+  %1 = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 0, i32 3>
+  ret <2 x i64> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v2i64':
+; SSE2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+define <2 x double> @test_v2f64(<2 x double> %a, <2 x double> %b) {
+  %1 = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 0, i32 3>
+  ret <2 x double> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v2f64':
+; SSE2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <2 x i64> @test_v2i64_2(<2 x i64> %a, <2 x i64> %b) {
+  %1 = shufflevector <2 x i64> %a, <2 x i64> %b, <2 x i32> <i32 2, i32 1>
+  ret <2 x i64> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v2i64_2':
+; SSE2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <2 x double> @test_v2f64_2(<2 x double> %a, <2 x double> %b) {
+  %1 = shufflevector <2 x double> %a, <2 x double> %b, <2 x i32> <i32 2, i32 1>
+  ret <2 x double> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v2f64_2':
+; SSE2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+; Test shuffles on packed vectors of four elements.
+
+define <4 x i32> @test_v4i32(<4 x i32> %a, <4 x i32> %b) {
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x i32> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v4i32':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <4 x i32> @test_v4i32_2(<4 x i32> %a, <4 x i32> %b) {
+  %1 = shufflevector <4 x i32> %a, <4 x i32> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  ret <4 x i32> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v4i32_2':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <4 x float> @test_v4f32(<4 x float> %a, <4 x float> %b) {
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v4f32':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <4 x float> @test_v4f32_2(<4 x float> %a, <4 x float> %b) {
+  %1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  ret <4 x float> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v4f32_2':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+define <4 x i64> @test_v4i64(<4 x i64> %a, <4 x i64> %b) {
+  %1 = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x i64> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v4i64':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <4 x i64> @test_v4i64_2(<4 x i64> %a, <4 x i64> %b) {
+  %1 = shufflevector <4 x i64> %a, <4 x i64> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  ret <4 x i64> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v4i64_2':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <4 x double> @test_v4f64(<4 x double> %a, <4 x double> %b) {
+  %1 = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x double> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v4f64':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <4 x double> @test_v4f64_2(<4 x double> %a, <4 x double> %b) {
+  %1 = shufflevector <4 x double> %a, <4 x double> %b, <4 x i32> <i32 4, i32 1, i32 6, i32 3>
+  ret <4 x double> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v4f64_2':
+; SSE2: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+; Test shuffles on packed vectors of eight elements.
+define <8 x i16> @test_v8i16(<8 x i16> %a, <8 x i16> %b) {
+  %1 = shufflevector <8 x i16> %a, <8 x i16> %b, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x i16> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v8i16':
+; SSE2: Cost Model: {{.*}} 8 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <8 x i16> @test_v8i16_2(<8 x i16> %a, <8 x i16> %b) {
+  %1 = shufflevector <8 x i16> %a, <8 x i16> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
+  ret <8 x i16> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v8i16_2':
+; SSE2: Cost Model: {{.*}} 8 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <8 x i32> @test_v8i32(<8 x i32> %a, <8 x i32> %b) {
+  %1 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x i32> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v8i32':
+; SSE2: Cost Model: {{.*}} 4 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 4 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <8 x i32> @test_v8i32_2(<8 x i32> %a, <8 x i32> %b) {
+  %1 = shufflevector <8 x i32> %a, <8 x i32> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
+  ret <8 x i32> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v8i32_2':
+; SSE2: Cost Model: {{.*}} 4 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 4 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <8 x float> @test_v8f32(<8 x float> %a, <8 x float> %b) {
+  %1 = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x float> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v8f32':
+; SSE2: Cost Model: {{.*}} 4 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 4 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <8 x float> @test_v8f32_2(<8 x float> %a, <8 x float> %b) {
+  %1 = shufflevector <8 x float> %a, <8 x float> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7>
+  ret <8 x float> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v8f32_2':
+; SSE2: Cost Model: {{.*}} 4 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 4 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+; Test shuffles on packed vectors of sixteen elements.
+define <16 x i8> @test_v16i8(<16 x i8> %a, <16 x i8> %b) {
+  %1 = shufflevector <16 x i8> %a, <16 x i8> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  ret <16 x i8> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v16i8':
+; SSE2: Cost Model: {{.*}} 48 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+
+
+define <16 x i8> @test_v16i8_2(<16 x i8> %a, <16 x i8> %b) {
+  %1 = shufflevector <16 x i8> %a, <16 x i8> %b, <16 x i32> <i32 16, i32 1, i32 18, i32 3, i32 20, i32 5, i32 22, i32 7, i32 24, i32 9, i32 26, i32 11, i32 28, i32 13, i32 30, i32 15>
+  ret <16 x i8> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v16i8_2':
+; SSE2: Cost Model: {{.*}} 48 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 3 for instruction:   %1 = shufflevector
+
+
+define <16 x i16> @test_v16i16(<16 x i16> %a, <16 x i16> %b) {
+  %1 = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  ret <16 x i16> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v16i16':
+; SSE2: Cost Model: {{.*}} 16 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 6 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 5 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+
+define <16 x i16> @test_v16i16_2(<16 x i16> %a, <16 x i16> %b) {
+  %1 = shufflevector <16 x i16> %a, <16 x i16> %b, <16 x i32> <i32 16, i32 1, i32 18, i32 3, i32 20, i32 5, i32 22, i32 7, i32 24, i32 9, i32 26, i32 11, i32 28, i32 13, i32 30, i32 15>
+  ret <16 x i16> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v16i16_2':
+; SSE2: Cost Model: {{.*}} 16 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 6 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 2 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 5 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 1 for instruction:   %1 = shufflevector
+
+define <32 x i8> @test_v32i8(<32 x i8> %a, <32 x i8> %b) {
+  %1 = shufflevector <32 x i8> %a, <32 x i8> %b, <32 x i32> <i32 0, i32 33, i32 2, i32 35, i32 4, i32 37, i32 6, i32 39, i32 8, i32 41, i32 10, i32 43, i32 12, i32 45, i32 14, i32 47, i32 16, i32 49, i32 18, i32 51, i32 20, i32 53, i32 22, i32 55, i32 24, i32 57, i32 26, i32 59, i32 28, i32 61, i32 30, i32 63>
+  ret <32 x i8> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v32i8':
+; SSE2: Cost Model: {{.*}} 96 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 6 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 6 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 9 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 9 for instruction:   %1 = shufflevector
+
+
+define <32 x i8> @test_v32i8_2(<32 x i8> %a, <32 x i8> %b) {
+  %1 = shufflevector <32 x i8> %a, <32 x i8> %b, <32 x i32> <i32 32, i32 1, i32 34, i32 3, i32 36, i32 5, i32 38, i32 7, i32 40, i32 9, i32 42, i32 11, i32 44, i32 13, i32 46, i32 15, i32 48, i32 17, i32 50, i32 19, i32 52, i32 21, i32 54, i32 23, i32 56, i32 25, i32 58, i32 27, i32 60, i32 29, i32 62, i32 31>
+  ret <32 x i8> %1
+}
+; CHECK: Printing analysis 'Cost Model Analysis' for function 'test_v32i8_2':
+; SSE2: Cost Model: {{.*}} 96 for instruction:   %1 = shufflevector
+; SSSE3: Cost Model: {{.*}} 6 for instruction:   %1 = shufflevector
+; SSE41: Cost Model: {{.*}} 6 for instruction:   %1 = shufflevector
+; AVX: Cost Model: {{.*}} 9 for instruction:   %1 = shufflevector
+; AVX2: Cost Model: {{.*}} 9 for instruction:   %1 = shufflevector
+