[AArch64] Break up larger shuffle-masks into legal sizes in getShuffleCost

[dmgreen]

Summary

Given a larger-than-legal shuffle mask, the final codegen will split into multiple sub-vectors. This attempts to model that in AArch64TTIImpl::getShuffleCost, splitting masks up according to the size of the legalized vectors. If the sub-masks have at most 2 input sources we can call getShuffleCost on them and sum the costs, to get a more accurate final cost for the entire shuffle. The call to improveShuffleKindFromMask helps to improve the shuffle kind for the sub-mask cost call.

[samtebbs]

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

2654

I might be misunderstanding this part of the code but I don't see how Source != Source1 will ever be true, since Source1 is assigned to Source when NumSources == 0, which is always the case in the first iteration. Same for the other else-if below.

[RKSimon]

@dmgreen Might llvm::processShuffleMasks from D115653 be of use here?

[dmgreen]

Update and rebase.

[dmgreen]

In D123414#3448091, @RKSimon wrote:

@dmgreen Might llvm::processShuffleMasks from D115653 be of use here?

Hmm. I'm not sure. I remember thinking I should give it a try, I remember deciding it wasn't a good idea, but I don't remember why now. I think it won't give us the information we really need. We would need to process the mask going in, and then process ">=2 source" back into "2 source" and ">2 source" masks, as it will currently build >2 source masks out of 2 source masks. Plus going into and out of lambdas can make the structure difficult to follow, as I had this code it seemed simpler. Perhaps I've misunderstood how it should be used.

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

2654

The first unique Sources we find will get into the first if (NumSources == 0). The next time we find a Source that isn't == Source1 we will get into this if, and if we find more than 2 unique Sources, we just start to accumulate the total number of sources. If the Source is already one of Source1 or Source2 then we have already counted it and don't need to process. it in these checks again. Because NumSources increases it is kind of like a state machine.

[samtebbs]

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

2654

I think I somehow missed the assignment to Source when I looked over it the first time. Thanks for the explanation.

[dmgreen]

From looking at D100486, it appears that processShuffleMasks only handles single-input shuffles. I think I would recommend going with this version for the moment, and re-evaluating to use processShuffleMasks if it changes in the future to handle multiple-input shuffles.

[samtebbs]

LGTM

[thakis]

This makes clang crash building aarch64 bare metal builtins one windows: http://45.33.8.238/win/57078/step_4.txt

Please take a look and revert for now if it takes a while to fix.

[dmgreen]

In D123414#3477245, @thakis wrote:

This makes clang crash building aarch64 bare metal builtins one windows: http://45.33.8.238/win/57078/step_4.txt

Please take a look and revert for now if it takes a while to fix.

Thanks for the report.

You got here before any the bots reported it. https://github.com/llvm/llvm-project/commit/46cef9a82df8d4029ef409cf1fd0329e8c1696e3 had the fix, which looks like it was down to fp128 types. I'll add a test too.

llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp

InstructionCost AArch64TTIImpl::getSpliceCost(VectorType *Tp, int Index) {
   return LegalizationCost * LT.first;
 }
 
 InstructionCost AArch64TTIImpl::getShuffleCost(TTI::ShuffleKind Kind,
                                                VectorType *Tp,
                                                ArrayRef<int> Mask, int Index,
                                                VectorType *SubTp,
                                                ArrayRef<Value *> Args) {
+  std::pair<InstructionCost, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);
+  // If we have a Mask, and the LT is being legalized somehow, split the Mask
+  // into smaller vectors and sum the cost of each shuffle.
+  if (!Mask.empty() && isa<FixedVectorType>(Tp) &&
+      Tp->getScalarSizeInBits() == LT.second.getScalarSizeInBits() &&
+      Tp->getElementCount() != LT.second.getVectorElementCount() && !Index &&
+      !SubTp) {
+    unsigned TpNumElts = cast<FixedVectorType>(Tp)->getNumElements();
+    assert(Mask.size() == TpNumElts && "Expected Mask and Tp size to match!");
+    unsigned LTNumElts = LT.second.getVectorNumElements();
+    unsigned NumVecs = (TpNumElts + LTNumElts - 1) / LTNumElts;
+    VectorType *NTp =
+        VectorType::get(Tp->getScalarType(), LT.second.getVectorElementCount());
+    InstructionCost Cost;
+    for (unsigned N = 0; N < NumVecs; N++) {
+      SmallVector<int> NMask;
+      // Split the existing mask into chunks of size LTNumElts. Track the source
+      // sub-vectors to ensure the result has at most 2 inputs.
+      unsigned Source1, Source2;
+      unsigned NumSources = 0;
+      for (unsigned E = 0; E < LTNumElts; E++) {
+        int MaskElt =
+            N * LTNumElts + E < TpNumElts ? Mask[N * LTNumElts + E] : -1;
+        if (MaskElt < 0) {
+          NMask.push_back(-1);
+          continue;
+        }
+
+        unsigned Source = MaskElt / LTNumElts;
+        if (NumSources == 0) {
+          Source1 = Source;
+          NumSources = 1;
+        } else if (NumSources == 1 && Source != Source1) {
+          Source2 = Source;
+          NumSources = 2;
+        } else if (NumSources >= 2 && Source != Source1 && Source != Source2) {
+          NumSources++;
+        }
+        if (Source == Source1)
+          NMask.push_back(MaskElt % LTNumElts);
+        else if (Source == Source2)
+          NMask.push_back(MaskElt % LTNumElts + LTNumElts);
+        else
+          // These are not correct, but are usused for anything other than the
+          // modular lane number.
+          NMask.push_back(MaskElt % LTNumElts);
+      }
+      // If the sub-mask has at most 2 input sub-vectors then re-cost it using
+      // getShuffleCost. If not then cost it using the worst case.
+      if (NumSources <= 2)
+        Cost += getShuffleCost(NumSources <= 1 ? TTI::SK_PermuteSingleSrc
+                                               : TTI::SK_PermuteTwoSrc,
+                               NTp, NMask, 0, nullptr, Args);
+      else if (any_of(enumerate(NMask), [&](const auto &ME) {
+                 return ME.value() % LTNumElts == ME.index();
+               }))
+        Cost += LTNumElts - 1;
+      else
+        Cost += LTNumElts;
+    }
+    return Cost;
+  }
+
   // If we have 4 elements for the shuffle and a Mask, get the cost straight
   // from the perfect shuffle tables.
   if (!Mask.empty() && Tp->getElementCount() == ElementCount::getFixed(4) &&
       (Tp->getScalarSizeInBits() == 16 || Tp->getScalarSizeInBits() == 32))
     return getPerfectShuffleCost(Mask);
 
   Kind = improveShuffleKindFromMask(Kind, Mask);
-  std::pair<InstructionCost, MVT> LT = TLI->getTypeLegalizationCost(DL, Tp);
   if (Kind == TTI::SK_Broadcast || Kind == TTI::SK_Transpose ||
       Kind == TTI::SK_Select || Kind == TTI::SK_PermuteSingleSrc ||
       Kind == TTI::SK_Reverse) {
     static const CostTblEntry ShuffleTbl[] = {
       // Broadcast shuffle kinds can be performed with 'dup'.
       { TTI::SK_Broadcast, MVT::v8i8,  1 },
       { TTI::SK_Broadcast, MVT::v16i8, 1 },
       { TTI::SK_Broadcast, MVT::v4i16, 1 },

llvm/test/Analysis/CostModel/AArch64/shuffle-other.ll

 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v16i8_4_05 = shufflevector <16 x i8> undef, <16 x i8> undef, <16 x i32> <i32 0, i32 1, i32 16, i32 17, i32 18, i32 19, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i16_2_0 = shufflevector <4 x i16> undef, <4 x i16> undef, <4 x i32> <i32 0, i32 1, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i16_2_1 = shufflevector <4 x i16> undef, <4 x i16> undef, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v8i16_2_0 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 8, i32 9, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_2_1 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 8, i32 9, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_2_2 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i16_2_3 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8i16_2_05 = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 8, i32 9, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v16i16_4_0 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v16i16_4_0 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_4_1 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_4_2 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i16_4_3 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 21 for instruction: %v16i16_4_05 = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 1, i32 16, i32 17, i32 18, i32 19, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i32_2_0 = shufflevector <4 x i32> undef, <4 x i32> undef, <4 x i32> <i32 0, i32 1, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v4i32_2_1 = shufflevector <4 x i32> undef, <4 x i32> undef, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v8i32_2_0 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 8, i32 9, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v8i32_2_0 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 8, i32 9, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_2_1 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 8, i32 9, i32 4, i32 5, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_2_2 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 6, i32 7>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v8i32_2_3 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 8, i32 9>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %v8i32_2_05 = shufflevector <8 x i32> undef, <8 x i32> undef, <8 x i32> <i32 0, i32 8, i32 9, i32 3, i32 4, i32 5, i32 6, i32 7>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16i32_4_0 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %v16i32_4_0 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_4_1 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 16, i32 17, i32 18, i32 19, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_4_2 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v16i32_4_3 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 16, i32 17, i32 18, i32 19>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %v16i32_4_05 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 16, i32 17, i32 18, i32 19, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %v4i8_2_0 = shufflevector <4 x i8> undef, <4 x i8> undef, <4 x i32> <i32 0, i32 1, i32 6, i32 7>
   %v4i8_2_1 = shufflevector <4 x i8> undef, <4 x i8> undef, <4 x i32> <i32 4, i32 5, i32 0, i32 1>
;
 
   ret void
 }
 
 define void @multipart() {
 ; CHECK-LABEL: 'multipart'
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 42 for instruction: %v16a = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 8 for instruction: %v16b = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v16c = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 84 for instruction: %v16d = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v16c = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 50 for instruction: %v16d = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 2, i32 18, i32 3, i32 19, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v32a = shufflevector <4 x i32> undef, <4 x i32> undef, <4 x i32> <i32 2, i32 3, i32 0, i32 1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v32a4 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v32idrev = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 15, i32 14, i32 13, i32 12, i32 16, i32 17, i32 18, i32 19, i32 31, i32 30, i32 29, i32 28>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v32many = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %v32many2 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 1, i32 4, i32 8, i32 12, i32 17, i32 20, i32 24, i32 28, i32 2, i32 6, i32 11, i32 14, i32 18, i32 22, i32 27, i32 30>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v323 = shufflevector <3 x i32> undef, <3 x i32> undef, <3 x i32> <i32 2, i32 3, i32 0>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32a4 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 0, i32 1>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %v32idrev = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 15, i32 14, i32 13, i32 12, i32 16, i32 17, i32 18, i32 19, i32 31, i32 30, i32 29, i32 28>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v32many = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 0, i32 4, i32 8, i32 12, i32 16, i32 20, i32 24, i32 28, i32 2, i32 6, i32 10, i32 14, i32 18, i32 22, i32 26, i32 30>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %v32many2 = shufflevector <16 x i32> undef, <16 x i32> undef, <16 x i32> <i32 1, i32 4, i32 8, i32 12, i32 17, i32 20, i32 24, i32 28, i32 2, i32 6, i32 11, i32 14, i32 18, i32 22, i32 27, i32 30>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v323 = shufflevector <3 x i32> undef, <3 x i32> undef, <3 x i32> <i32 2, i32 3, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64a = shufflevector <2 x i64> undef, <2 x i64> undef, <2 x i32> <i32 1, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %v64b = shufflevector <2 x i64> undef, <2 x i64> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v64ab = shufflevector <4 x i64> undef, <4 x i64> undef, <4 x i32> <i32 1, i32 1, i32 0, i32 0>
-; CHECK-NEXT:  Cost Model: Found an estimated cost of 12 for instruction: %v64d = shufflevector <4 x i64> undef, <4 x i64> undef, <4 x i32> <i32 1, i32 1, i32 4, i32 4>
+; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %v64d = shufflevector <4 x i64> undef, <4 x i64> undef, <4 x i32> <i32 1, i32 1, i32 4, i32 4>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64a = shufflevector <2 x double> undef, <2 x double> undef, <2 x i32> <i32 1, i32 1>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %f64b = shufflevector <2 x double> undef, <2 x double> undef, <2 x i32> zeroinitializer
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %f64ab = shufflevector <4 x double> undef, <4 x double> undef, <4 x i32> <i32 1, i32 1, i32 0, i32 0>
 ; CHECK-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: ret void
 ;
   %v16a = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11>
   %v16b = shufflevector <8 x i16> undef, <8 x i16> undef, <8 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>
   %v16c = shufflevector <16 x i16> undef, <16 x i16> undef, <16 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>