[Matrix] Factor and distribute transposes across multiplies

[anemet]

Summary

Now that we can fold some transposes into multiplies (CM: A * B^t and RM:
A^t * B), we want to move them around to create the optimal expressions:

• fold away double transposes while still using them to assert the shape
• sink transposes hoping they cancel out
• lift transposes when both operands are transposed

This also modifies the matrix remarks to include the number of exposed
transposes (i.e. transposes that we couldn't fold into a multiply).

The adjustment to the test remarks-inlining is a bit subtle: I am changing the
double transpose to a single transpose so that we don't remove it completely.
More importantly this changes some of the total instruction count, most
notable stores because we can no longer use a vector store.

[fhahn]

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

683

For the later transforms, we collect a worklist once which contains all matrix instructions. Could we use the same here to avoid having to iterate over each function again?

752

If we have a TT matmul, lift the transpose until we have a non-TT situation.

Is this comment accurate? IIUC we only convert TT multiplies to versions where we can fold one transpose into the multiply?

llvm/test/Transforms/LowerMatrixIntrinsics/transpose-and-multiply-fold.ll

11 ↗

(On Diff #346295)

I think it would also be good to have tests that check the generated IR, together with some combinations with non-square matrixes.

[fhahn]

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

692

It should be enough to capture &II

[anemet]

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

683

Unless we really think this is a performance issue, I'd like to avoid the extra bookkeeping and just represent everything in the IR and no on-the-side data structure that needs updating. As I was saying offline I think we already have too much bookkeeping going on (e.g. for the remarks) so it's hard to know what to update at times (). Having a backward and a forward matrix algebraic simplification pass (which is what optimizeTransposes is) that is logically separated from the lowering pass I think makes a good sense in terms of "separation of concerns". What do you think?

752

Rephrased the comment.

[anemet]

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

683

(The only state that is live across optimizeTransposes is the shape-info so that we have gather as much shape info as possible before removing shape-carrying operations like a double-transpose. I should probably add a comment about this.)

[fhahn]

We also have a stripped down version of the pass to run in the backend pipelines (LowerMatrixIntrinsicsMinimalLegacyPass) We should probably no perform the optimizations there.

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

683

Having a backward and a forward matrix algebraic simplification pass (which is what optimizeTransposes is) that is logically separated from the lowering pass I think makes a good sense in terms of "separation of concerns". What do you think?

I think the worklist for matrix instructions is slightly different, as it would not directly impact any code related to transformations, just the way we visit them. It's not needed right now, but I think we will need it sooner or later as add more simplficiations that may depend on each other.

Compile-time wise it should be fine, given that it is only run when the pass is explicitly enabled. But we should keep an eye on it, to avoid increasing compile time for our adopters too much.

[fhahn]

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

683

While looking at this, I realized that we can do an easy early exit if there are no matrix intrinsics at all D102931. That should alleviate the impact on functions without matrix code :)

[anemet]

Address Florian's comments.

[fhahn]

LGTM, thanks!

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

207

nit: /// for doc-comment as the other members?

[int3]

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

820

I am getting a link-time failure when doing a local ninja lld:

Undefined symbols for architecture x86_64:
  "llvm::Value::dump() const", referenced from:
      (anonymous namespace)::LowerMatrixIntrinsics::Visit() in libLLVMScalarOpts.a(LowerMatrixIntrinsics.cpp.o)
ld: symbol(s) not found for architecture x86_64

Not sure if/how it passes the buildbots, but it does indeed look like Value.cpp doesn't implement dump().

820

(and commenting out this line does indeed fix my local build)

[int3]

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

820

nevermind, I see a fix has just been pushed :)

llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp

 #include "llvm/Analysis/VectorUtils.h"
 #include "llvm/IR/CFG.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/DebugInfoMetadata.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/IRBuilder.h"
 #include "llvm/IR/Instructions.h"
 #include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/MatrixBuilder.h"
 #include "llvm/IR/PatternMatch.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/Alignment.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"

 static cl::opt<MatrixLayoutTy> MatrixLayout(
     "matrix-default-layout", cl::init(MatrixLayoutTy::ColumnMajor),
     cl::desc("Sets the default matrix layout"),
     cl::values(clEnumValN(MatrixLayoutTy::ColumnMajor, "column-major",
                           "Use column-major layout"),
                clEnumValN(MatrixLayoutTy::RowMajor, "row-major",
                           "Use row-major layout")));
 
+static cl::opt<bool> PrintAfterTransposeOpt("matrix-print-after-transpose-opt",
+                                            cl::init(false));
+
 /// Helper function to either return Scope, if it is a subprogram or the
 /// attached subprogram for a local scope.
 static DISubprogram *getSubprogram(DIScope *Scope) {
   if (auto *Subprogram = dyn_cast<DISubprogram>(Scope))
     return Subprogram;
   return cast<DILocalScope>(Scope)->getSubprogram();
 }
 
class LowerMatrixIntrinsics {
   /// Contains estimates of the number of operations (loads, stores, compute) required to lower a matrix operation.
   struct OpInfoTy {
     /// Number of stores emitted to generate this matrix.
     unsigned NumStores = 0;
     /// Number of loads emitted to generate this matrix.
     unsigned NumLoads = 0;
     /// Number of compute operations emitted to generate this matrix.
     unsigned NumComputeOps = 0;
+    // Most of the time transposes can be fused with matrix multiplies or can be
+    // folded away via algebraic simplifications.  This is the number of
+    // transposes that we failed to make "free" via such optimizations.
+    unsigned NumExposedTransposes = 0;
 
     OpInfoTy &operator+=(const OpInfoTy &RHS) {
       NumStores += RHS.NumStores;
       NumLoads += RHS.NumLoads;
       NumComputeOps += RHS.NumComputeOps;
+      NumExposedTransposes += RHS.NumExposedTransposes;
       return *this;
     }
   };
 
   /// Wrapper class representing a matrix as a set of vectors, either in row or
   /// column major layout. All vectors must have the same vector type.
   class MatrixTy {
     SmallVector<Value *, 16> Vectors;
public:
 
     void setNumLoads(unsigned N) { OpInfo.NumLoads = N; }
 
     MatrixTy &addNumStores(unsigned N) {
       OpInfo.NumStores += N;
       return *this;
     }
 
+    MatrixTy &addNumExposedTransposes(unsigned N) {
+      OpInfo.NumExposedTransposes += N;
+      return *this;
+    }
+
     MatrixTy &addNumComputeOps(unsigned N) {
       OpInfo.NumComputeOps += N;
       return *this;
     }
 
     unsigned getNumStores() const { return OpInfo.NumStores; }
     unsigned getNumLoads() const { return OpInfo.NumLoads; }
     unsigned getNumComputeOps() const { return OpInfo.NumComputeOps; }
struct ShapeInfo {
     }
   };
 
   /// Maps instructions to their shape information. The shape information
   /// describes the shape to be used while lowering. This matches the shape of
   /// the result value of the instruction, with the only exceptions being store
   /// instructions and the matrix_column_major_store intrinsics. For those, the
   /// shape information indicates that those instructions should be lowered
-  /// using shape information as well.
-  DenseMap<Value *, ShapeInfo> ShapeMap;
+  /// using shape information as well.  A ValueMap is used so that when
+  /// sub-passes like optimizeTransposes performs RAUW the map stays
+  /// up-to-date.
+  ValueMap<Value *, ShapeInfo> ShapeMap;
 
   /// List of instructions to remove. While lowering, we are not replacing all
   /// users of a lowered instruction, if shape information is available and
   /// those need to be removed after we finished lowering.
   SmallVector<Instruction *, 16> ToRemove;
 
   /// Map from instructions to their produced column matrix.
   MapVector<Value *, MatrixTy> Inst2ColumnMatrix;
 
 public:
   LowerMatrixIntrinsics(Function &F, TargetTransformInfo &TTI,
                         AliasAnalysis *AA, DominatorTree *DT, LoopInfo *LI,
                         OptimizationRemarkEmitter *ORE)
       : Func(F), DL(F.getParent()->getDataLayout()), TTI(TTI), AA(AA), DT(DT),
         LI(LI), ORE(ORE) {}
 
   unsigned getNumOps(Type *VT) {
     assert(isa<VectorType>(VT) && "Expected vector type");
     return getNumOps(VT->getScalarType(),
                      cast<FixedVectorType>(VT)->getNumElements());
   }
 
-  //
+  /// Is this the minimal version executed in the backend pipelines.
+  bool isMinimal() const {
+    return !DT;
+  }
+
   /// Return the estimated number of vector ops required for an operation on
   /// \p VT * N.
   unsigned getNumOps(Type *ST, unsigned N) {
     return std::ceil((ST->getPrimitiveSizeInBits() * N).getFixedSize() /
                      double(TTI.getRegisterBitWidth(
                                    TargetTransformInfo::RGK_FixedWidthVector)
                                 .getFixedSize()));
   }
while (!WorkList.empty()) {
       for (size_t I = BeforeProcessingV; I != WorkList.size(); I++)
         for (User *U : WorkList[I]->users())
           if (isa<Instruction>(U) && V != U)
             NewWorkList.push_back(cast<Instruction>(U));
     }
     return NewWorkList;
   }
 
+  /// Try moving transposes in order to fold them away or into multiplies.
+  void optimizeTransposes() {
+    // First sink all transposes inside matmuls, hoping that we end up with NN,
+    // NT or TN variants.
+    for (BasicBlock &BB : reverse(Func)) {
+      for (auto II = BB.rbegin(); II != BB.rend();) {
+        Instruction &I = *II;
+        // We may remove II.  By default continue on the next/prev instruction.
+        ++II;
+        // If we were to erase II, move again.
+        auto EraseFromParent = [&II](Value *V) {
+          auto *Inst = cast<Instruction>(V);
+          if (Inst->use_empty()) {
+            if (Inst == &*II) {
+              ++II;
+            }
+            Inst->eraseFromParent();
+          }
+        };
+
+        // If we're creating a new instruction, continue from there.
+        Instruction *NewInst = nullptr;
+
+        IRBuilder<> IB(&I);
+        MatrixBuilder<IRBuilder<>> Builder(IB);
+
+        Value *TA, *TAMA, *TAMB;
+        ConstantInt *R, *K, *C;
+        if (match(&I, m_Intrinsic<Intrinsic::matrix_transpose>(m_Value(TA)))) {
+
+          // Transpose of a transpose is a nop
+          Value *TATA;
+          if (match(TA,
+                    m_Intrinsic<Intrinsic::matrix_transpose>(m_Value(TATA)))) {
+            I.replaceAllUsesWith(TATA);
+            EraseFromParent(&I);
+            EraseFromParent(TA);
+          }
+
+          // (A * B)^t -> B^t * A^t
+          // RxK KxC      CxK   KxR
+          else if (match(TA, m_Intrinsic<Intrinsic::matrix_multiply>(
+                                 m_Value(TAMA), m_Value(TAMB), m_ConstantInt(R),
+                                 m_ConstantInt(K), m_ConstantInt(C)))) {
+            Value *T0 = Builder.CreateMatrixTranspose(TAMB, K->getZExtValue(),
+                                                      C->getZExtValue(),
+                                                      TAMB->getName() + "_t");
+            // We are being run after shape prop, add shape for newly created
+            // instructions so that we lower them later.
+            setShapeInfo(T0, {C, K});
+            Value *T1 = Builder.CreateMatrixTranspose(TAMA, R->getZExtValue(),
+                                                      K->getZExtValue(),
+                                                      TAMA->getName() + "_t");
+            setShapeInfo(T1, {K, R});
+            NewInst = Builder.CreateMatrixMultiply(T0, T1, C->getZExtValue(),
+                                                   K->getZExtValue(),
+                                                   R->getZExtValue(), "mmul");
+            setShapeInfo(NewInst, {C, R});
+            I.replaceAllUsesWith(NewInst);
+            EraseFromParent(&I);
+            EraseFromParent(TA);
+          }
+        }
+
+        // If we replaced I with a new instruction, continue from there.
+        if (NewInst)
+          II = std::next(BasicBlock::reverse_iterator(NewInst));
+      }
+    }
+
+    // If we have a TT matmul, lift the transpose.  We may be able to fold into
+    // consuming multiply.
+    for (BasicBlock &BB : Func) {
+      for (BasicBlock::iterator II = BB.begin(); II != BB.end();) {
+        Instruction *I = &*II;
+        // We may remove I.
+        ++II;
+        Value *A, *B, *AT, *BT;
+        ConstantInt *R, *K, *C;
+        if (match(&*I, m_Intrinsic<Intrinsic::matrix_multiply>(
+                           m_Value(A), m_Value(B), m_ConstantInt(R),
+                           m_ConstantInt(K), m_ConstantInt(C))) &&
+            match(A, m_Intrinsic<Intrinsic::matrix_transpose>(m_Value(AT))) &&
+            match(B, m_Intrinsic<Intrinsic::matrix_transpose>(m_Value((BT))))) {
+          IRBuilder<> IB(&*I);
+          MatrixBuilder<IRBuilder<>> Builder(IB);
+          Value *M = Builder.CreateMatrixMultiply(
+              BT, AT, C->getZExtValue(), K->getZExtValue(), R->getZExtValue());
+          setShapeInfo(M, {C, R});
+          Value *NewInst = Builder.CreateMatrixTranspose(M, R->getZExtValue(),
+                                                         C->getZExtValue());
+          setShapeInfo(NewInst, {C, R});
+          I->replaceAllUsesWith(NewInst);
+          if (I->use_empty())
+            I->eraseFromParent();
+          if (A->use_empty())
+            cast<Instruction>(A)->eraseFromParent();
+          if (B->use_empty())
+            cast<Instruction>(B)->eraseFromParent();
+        }
+      }
+    }
+  }
+
   bool Visit() {
     if (EnableShapePropagation) {
       SmallVector<Instruction *, 32> WorkList;
 
       // Initially only the shape of matrix intrinsics is known.
       // Initialize the work list with ops carrying shape information.
       for (BasicBlock &BB : Func)
         for (Instruction &Inst : BB) {
if (EnableShapePropagation) {
         }
       // Propagate shapes until nothing changes any longer.
       while (!WorkList.empty()) {
         WorkList = propagateShapeForward(WorkList);
         WorkList = propagateShapeBackward(WorkList);
       }
     }
 
+    if (!isMinimal()) {
+      optimizeTransposes();
+      if (PrintAfterTransposeOpt) {
+        dbgs() << "Dump after matrix transpose optimization:\n";
+        Func.dump();
+      }
+    }
+
     bool Changed = false;
     SmallVector<CallInst *, 16> MaybeFusableInsts;
     SmallVector<Instruction *, 16> MatrixInsts;
 
     // First, collect all instructions with shape information and candidates for
     // fusion (currently only matrix multiplies).
     ReversePostOrderTraversal<Function *> RPOT(&Func);
     for (auto *BB : RPOT)
for (unsigned I = 0; I < NewNumVecs; ++I) {
       Result.addVector(ResultVector);
     }
 
     // TODO: Improve estimate of operations needed for transposes. Currently we
     // just count the insertelement/extractelement instructions, but do not
     // account for later simplifications/combines.
     finalizeLowering(
         Inst,
-        Result.addNumComputeOps(2 * ArgShape.NumRows * ArgShape.NumColumns),
+        Result.addNumComputeOps(2 * ArgShape.NumRows * ArgShape.NumColumns)
+            .addNumExposedTransposes(1),
         Builder);
   }
 
   /// Lower load instructions, if shape information is available.
   bool VisitLoad(LoadInst *Inst, Value *Ptr, IRBuilder<> &Builder) {
     auto I = ShapeMap.find(Inst);
     if (I == ShapeMap.end())
       return false;
void emitRemarks() {
 
           OptimizationRemark Rem(DEBUG_TYPE, "matrix-lowered", Loc,
                                  cast<Instruction>(L)->getParent());
 
           Rem << "Lowered with ";
           Rem << ore::NV("NumStores", Counts.NumStores) << " stores, "
               << ore::NV("NumLoads", Counts.NumLoads) << " loads, "
               << ore::NV("NumComputeOps", Counts.NumComputeOps)
-              << " compute ops";
+              << " compute ops, "
+              << ore::NV("NumExposedTransposes", Counts.NumExposedTransposes)
+              << " exposed transposes";
 
           if (SharedCounts.NumStores > 0 || SharedCounts.NumLoads > 0 ||
               SharedCounts.NumComputeOps > 0) {
             Rem << ",\nadditionally "
                 << ore::NV("NumStores", SharedCounts.NumStores) << " stores, "
                 << ore::NV("NumLoads", SharedCounts.NumLoads) << " loads, "
                 << ore::NV("NumFPOps", SharedCounts.NumComputeOps)
                 << " compute ops"

llvm/test/Transforms/LowerMatrixIntrinsics/remarks-inlining.ll

 ; void test(double *A, double *B, double *C) {
 ;   store(add(load<double, 3, 5>(A), load<double, 3, 5>(B)), C);
 ; }
 ;
 
 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
 target triple = "aarch64-apple-ios"
 
-; CHECK-LABEL: remark: load.h:41:43: Lowered with 0 stores, 10 loads, 0 compute ops
+; CHECK-LABEL: remark: load.h:41:43: Lowered with 0 stores, 10 loads, 0 compute ops, 0 exposed transposes
 ; CHECK-NEXT:  load(addr %A)
 
-; CHECK-LABEL: remark: load.h:41:43: Lowered with 0 stores, 10 loads, 0 compute ops
+; CHECK-LABEL: remark: load.h:41:43: Lowered with 0 stores, 10 loads, 0 compute ops, 0 exposed transposes
 ; CHECK-NEXT:  column.major.load.3x5.double(addr %B, 5)
 
-; CHECK-LABEL: remark: load.h:41:11: Lowered with 0 stores, 1 loads, 0 compute ops
+; CHECK-LABEL: remark: load.h:41:11: Lowered with 0 stores, 1 loads, 0 compute ops, 0 exposed transposes
 ; CHECK-NEXT: load(addr %D)
 
-; CHECK-LABEL: remark: assign.h:32:43: Lowered with 0 stores, 10 loads, 0 compute ops
+; CHECK-LABEL: remark: assign.h:32:43: Lowered with 0 stores, 10 loads, 0 compute ops, 0 exposed transposes
 ; CHECK-NEXT:  load(addr %A)
 
-; CHECK-LABEL: remark: assign.h:32:43: Lowered with 0 stores, 10 loads, 0 compute ops
+; CHECK-LABEL: remark: assign.h:32:43: Lowered with 0 stores, 10 loads, 0 compute ops, 0 exposed transposes
 ; CHECK-NEXT:  column.major.load.3x5.double(addr %B, 5)
 
-; CHECK-LABEL: remark: toplevel.c:410:0: Lowered with 10 stores, 20 loads, 10 compute ops
+; CHECK-LABEL: remark: toplevel.c:410:0: Lowered with 10 stores, 20 loads, 10 compute ops, 0 exposed transposes
 ; CHECK-NEXT:  store(
 ; CHECK-NEXT:   fadd(
 ; CHECK-NEXT:    load(addr %A),
 ; CHECK-NEXT:    column.major.load.3x5.double(addr %B, 5)),
 ; CHECK-NEXT:   addr %C)
 
-; CHECK-LABEL: remark: toplevel.c:510:0: Lowered with 1 stores, 1 loads, 8 compute ops
+; CHECK-LABEL: remark: toplevel.c:510:0: Lowered with 2 stores, 1 loads, 4 compute ops, 1 exposed transposes
 ; CHECK-NEXT:  store(
-; CHECK-NEXT:   transpose.1x2.float(transpose.2x1.float(load(addr %D))),
+; CHECK-NEXT:   transpose.2x1.float(load(addr %D)),
 ; CHECK-NEXT:   addr %D)
 
-; CHECK-LABEL: remark: add.h:66:11: Lowered with 0 stores, 0 loads, 10 compute ops
+; CHECK-LABEL: remark: add.h:66:11: Lowered with 0 stores, 0 loads, 10 compute ops, 0 exposed transposes
 ; CHECK-NEXT:  fadd(
 ; CHECK-NEXT:   addr %A,
 ; CHECK-NEXT:   scalar)
 
-; CHECK-LABEL: remark: store.h:10:11: Lowered with 10 stores, 0 loads, 0 compute ops
+; CHECK-LABEL: remark: store.h:10:11: Lowered with 10 stores, 0 loads, 0 compute ops, 0 exposed transposes
 ; CHECK-NEXT:  store(
 ; CHECK-NEXT:   scalar,
 ; CHECK-NEXT:   addr %C)
 
-; CHECK-LABEL: remark: store.h:66:11: Lowered with 1 stores, 0 loads, 0 compute ops
+; CHECK-LABEL: remark: store.h:66:11: Lowered with 2 stores, 0 loads, 0 compute ops, 0 exposed transposes
 ; CHECK-NEXT:  store(
 ; CHECK-NEXT:  scalar,
 ; CHECK-NEXT:  addr %D)
 
-; CHECK-LABEL: remark: transpose.h:13:11: Lowered with 0 stores, 0 loads, 8 compute ops
-; CHECK-NEXT:  transpose.1x2.float(transpose.2x1.float(addr %D))
+; CHECK-LABEL: remark: transpose.h:13:11: Lowered with 0 stores, 0 loads, 4 compute ops, 1 exposed transposes
+; CHECK-NEXT:  transpose.2x1.float(addr %D)
 
 define void @toplevel(<15 x double>* %A, double* %B, <15 x double>* %C, <2 x float>* %D) !dbg !16 {
 entry:
   %a = load <15 x double>, <15 x double> *%A, align 16, !dbg !3791
   %b = call <15 x double> @llvm.matrix.column.major.load(double* %B, i64 5, i1 false, i32 3, i32 5), !dbg !3793
   %c  = fadd <15 x double> %a, %b, !dbg !100
   store <15 x double> %c, <15 x double> *%C, align 16, !dbg !102
 
   %load = load <2 x float>, <2 x float>* %D, !dbg !104
   %t1 = call <2 x float> @llvm.matrix.transpose(<2 x float> %load, i32 2, i32 1), !dbg !106
-  %t2 = call <2 x float> @llvm.matrix.transpose(<2 x float> %t1, i32 1, i32 2), !dbg !106
-  store <2 x float> %t2, <2 x float>* %D, !dbg !108
+  store <2 x float> %t1, <2 x float>* %D, !dbg !108
   ret void
 }
 
 declare <15 x double> @llvm.matrix.column.major.load(double*, i64, i1, i32, i32)
 declare <2 x float> @llvm.matrix.transpose(<2 x float>, i32, i32)
 
 !llvm.dbg.cu = !{!0}
 !llvm.module.flags = !{!3, !4}

llvm/test/Transforms/LowerMatrixIntrinsics/remarks-shared-subtrees.ll

 ; YAML-NEXT:  Function:        test_2leafs
 ; YAML-NEXT:  Args:
 ; YAML-NEXT:    - String:          'Lowered with '
 ; YAML-NEXT:    - NumStores:       '4'
 ; YAML-NEXT:    - String:          ' stores, '
 ; YAML-NEXT:    - NumLoads:        '0'
 ; YAML-NEXT:    - String:          ' loads, '
 ; YAML-NEXT:    - NumComputeOps:   '0'
-; YAML-NEXT:    - String:          ' compute ops'
+; YAML-NEXT:    - String:          ' compute ops, '
+; YAML-NEXT:   - NumExposedTransposes: '0'
+; YAML-NEXT:   - String:          ' exposed transposes'
 ; YAML-NEXT:    - String:          ",\nadditionally "
 ; YAML-NEXT:    - NumStores:       '0'
 ; YAML-NEXT:    - String:          ' stores, '
 ; YAML-NEXT:    - NumLoads:        '4'
 ; YAML-NEXT:    - String:          ' loads, '
 ; YAML-NEXT:    - NumFPOps:        '16'
 ; YAML-NEXT:    - String:          ' compute ops'
 ; YAML-NEXT:    - String:          ' are shared with other expressions'

 ; YAML-NEXT:  Function:        test_2leafs
 ; YAML-NEXT:  Args:
 ; YAML-NEXT:    - String:          'Lowered with '
 ; YAML-NEXT:    - NumStores:       '30'
 ; YAML-NEXT:    - String:          ' stores, '
 ; YAML-NEXT:    - NumLoads:        '45'
 ; YAML-NEXT:    - String:          ' loads, '
 ; YAML-NEXT:    - NumComputeOps:   '120'
-; YAML-NEXT:    - String:          ' compute ops'
+; YAML-NEXT:    - String:          ' compute ops, '
+; YAML-NEXT:   - NumExposedTransposes: '0'
+; YAML-NEXT:   - String:          ' exposed transposes'
 ; YAML-NEXT:    - String:          ",\nadditionally "
 ; YAML-NEXT:    - NumStores:       '0'
 ; YAML-NEXT:    - String:          ' stores, '
 ; YAML-NEXT:    - NumLoads:        '4'
 ; YAML-NEXT:    - String:          ' loads, '
 ; YAML-NEXT:    - NumFPOps:        '16'
 ; YAML-NEXT:    - String:          ' compute ops'
 ; YAML-NEXT:    - String:          ' are shared with other expressions'

llvm/test/Transforms/LowerMatrixIntrinsics/transpose-opts.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; REQUIRES: aarch64-registered-target
+
+; This test needs to be target specific due to the cost estimate in the output.
+
+; RUN: opt -lower-matrix-intrinsics -S < %s | FileCheck %s
+; RUN: opt -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s
+; RUN: opt -lower-matrix-intrinsics -S -o /dev/null -pass-remarks-output=%t  < %s && FileCheck --input-file %t --check-prefix=REMARK %s
+; RUN: opt -passes='lower-matrix-intrinsics' -S -o /dev/null -pass-remarks-output=%t < %s && FileCheck --input-file %t --check-prefix=REMARK %s
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "aarch64-apple-ios"
+
+define void @double_transpose(<9 x double>* %A, <9 x double>* %B) {
+; REMARK:      Pass:            lower-matrix-intrinsics
+; REMARK-NEXT: Name:            matrix-lowered
+; REMARK-NEXT: Function:        double_transpose
+; REMARK-NEXT: Args:
+; REMARK-NEXT:   - String:          'Lowered with '
+; REMARK-NEXT:   - NumStores:       '6'
+; REMARK-NEXT:   - String:          ' stores, '
+; REMARK-NEXT:   - NumLoads:        '6'
+; REMARK-NEXT:   - String:          ' loads, '
+; REMARK-NEXT:   - NumComputeOps:   '0'
+; REMARK-NEXT:   - String:          ' compute ops, '
+; REMARK-NEXT:   - NumExposedTransposes: '0'
+; REMARK-NEXT:   - String:          ' exposed transposes'
+; REMARK-NEXT:   - String:           |
+; REMARK:            store(
+; REMARK-NEXT:        load(addr %A),
+; REMARK-NEXT:        addr %B)
+; CHECK-LABEL: @double_transpose(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = bitcast <9 x double>* [[A:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast double* [[TMP0]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST]], align 16
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, double* [[TMP0]], i64 3
+; CHECK-NEXT:    [[VEC_CAST1:%.*]] = bitcast double* [[VEC_GEP]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD2:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST1]], align 8
+; CHECK-NEXT:    [[VEC_GEP3:%.*]] = getelementptr double, double* [[TMP0]], i64 6
+; CHECK-NEXT:    [[VEC_CAST4:%.*]] = bitcast double* [[VEC_GEP3]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD5:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST4]], align 16
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <9 x double>* [[B:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST6:%.*]] = bitcast double* [[TMP1]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[COL_LOAD]], <3 x double>* [[VEC_CAST6]], align 16
+; CHECK-NEXT:    [[VEC_GEP7:%.*]] = getelementptr double, double* [[TMP1]], i64 3
+; CHECK-NEXT:    [[VEC_CAST8:%.*]] = bitcast double* [[VEC_GEP7]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[COL_LOAD2]], <3 x double>* [[VEC_CAST8]], align 8
+; CHECK-NEXT:    [[VEC_GEP9:%.*]] = getelementptr double, double* [[TMP1]], i64 6
+; CHECK-NEXT:    [[VEC_CAST10:%.*]] = bitcast double* [[VEC_GEP9]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[COL_LOAD5]], <3 x double>* [[VEC_CAST10]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %a = load <9 x double>, <9 x double>* %A, align 16
+  %at = call <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double> %a, i32 3, i32 3)
+  %att = call <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double> %at, i32 3, i32 3)
+  store <9 x double> %att, <9 x double>* %B, align 16
+  ret void
+}
+
+define void @multiply_ntt(<6 x double>* %A, <6 x double>* %B, <8 x double>* %C, <8 x double>* %R) {
+; REMARK:      Pass:            lower-matrix-intrinsics
+; REMARK-NEXT: Name:            matrix-lowered
+; REMARK-NEXT: Function:        multiply_ntt
+; REMARK-NEXT: Args:
+; REMARK-NEXT:   - String:          'Lowered with '
+; REMARK-NEXT:   - NumStores:       '4'
+; REMARK-NEXT:   - String:          ' stores, '
+; REMARK-NEXT:   - NumLoads:        '10'
+; REMARK-NEXT:   - String:          ' loads, '
+; REMARK-NEXT:   - NumComputeOps:   '38'
+; REMARK-NEXT:   - String:          ' compute ops, '
+; REMARK-NEXT:   - NumExposedTransposes: '0'
+; REMARK-NEXT:   - String:          ' exposed transposes'
+; REMARK-NEXT:   - String:           |
+; REMARK:            store(
+; REMARK-NEXT:        multiply.2x3.3x4.double(
+; REMARK-NEXT:         load(addr %A),
+; REMARK-NEXT:         transpose.4x3.double(multiply.4x2.2x3.double(
+; REMARK-NEXT:           load(addr %C),
+; REMARK-NEXT:           load(addr %B)))),
+; REMARK-NEXT:        addr %R)
+; CHECK-LABEL: @multiply_ntt(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = bitcast <6 x double>* [[A:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast double* [[TMP0]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 16
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, double* [[TMP0]], i64 2
+; CHECK-NEXT:    [[VEC_CAST39:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD40:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST39]], align 16
+; CHECK-NEXT:    [[VEC_GEP41:%.*]] = getelementptr double, double* [[TMP0]], i64 4
+; CHECK-NEXT:    [[VEC_CAST42:%.*]] = bitcast double* [[VEC_GEP41]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD43:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST42]], align 16
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> [[COL_LOAD40]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x double> [[COL_LOAD43]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x double> [[TMP1]], <4 x double> [[TMP2]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <6 x double>* [[B:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST44:%.*]] = bitcast double* [[TMP4]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD45:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST44]], align 16
+; CHECK-NEXT:    [[VEC_GEP46:%.*]] = getelementptr double, double* [[TMP4]], i64 2
+; CHECK-NEXT:    [[VEC_CAST47:%.*]] = bitcast double* [[VEC_GEP46]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD48:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST47]], align 16
+; CHECK-NEXT:    [[VEC_GEP49:%.*]] = getelementptr double, double* [[TMP4]], i64 4
+; CHECK-NEXT:    [[VEC_CAST50:%.*]] = bitcast double* [[VEC_GEP49]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD51:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST50]], align 16
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <8 x double>* [[C:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST52:%.*]] = bitcast double* [[TMP5]] to <4 x double>*
+; CHECK-NEXT:    [[COL_LOAD53:%.*]] = load <4 x double>, <4 x double>* [[VEC_CAST52]], align 16
+; CHECK-NEXT:    [[VEC_GEP54:%.*]] = getelementptr double, double* [[TMP5]], i64 4
+; CHECK-NEXT:    [[VEC_CAST55:%.*]] = bitcast double* [[VEC_GEP54]] to <4 x double>*
+; CHECK-NEXT:    [[COL_LOAD56:%.*]] = load <4 x double>, <4 x double>* [[VEC_CAST55]], align 16
+; CHECK-NEXT:    [[BLOCK57:%.*]] = shufflevector <4 x double> [[COL_LOAD53]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x double> [[COL_LOAD45]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT58:%.*]] = insertelement <2 x double> poison, double [[TMP6]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT59:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT58]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = fmul <2 x double> [[BLOCK57]], [[SPLAT_SPLAT59]]
+; CHECK-NEXT:    [[BLOCK60:%.*]] = shufflevector <4 x double> [[COL_LOAD56]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x double> [[COL_LOAD45]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT61:%.*]] = insertelement <2 x double> poison, double [[TMP8]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT62:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT61]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = fmul <2 x double> [[BLOCK60]], [[SPLAT_SPLAT62]]
+; CHECK-NEXT:    [[TMP10:%.*]] = fadd <2 x double> [[TMP7]], [[TMP9]]
+; CHECK-NEXT:    [[TMP11:%.*]] = shufflevector <2 x double> [[TMP10]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP12:%.*]] = shufflevector <4 x double> undef, <4 x double> [[TMP11]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
+; CHECK-NEXT:    [[BLOCK63:%.*]] = shufflevector <4 x double> [[COL_LOAD53]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x double> [[COL_LOAD45]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT64:%.*]] = insertelement <2 x double> poison, double [[TMP13]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT65:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT64]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = fmul <2 x double> [[BLOCK63]], [[SPLAT_SPLAT65]]
+; CHECK-NEXT:    [[BLOCK66:%.*]] = shufflevector <4 x double> [[COL_LOAD56]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP15:%.*]] = extractelement <2 x double> [[COL_LOAD45]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT67:%.*]] = insertelement <2 x double> poison, double [[TMP15]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT68:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT67]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = fmul <2 x double> [[BLOCK66]], [[SPLAT_SPLAT68]]
+; CHECK-NEXT:    [[TMP17:%.*]] = fadd <2 x double> [[TMP14]], [[TMP16]]
+; CHECK-NEXT:    [[TMP18:%.*]] = shufflevector <2 x double> [[TMP17]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP19:%.*]] = shufflevector <4 x double> [[TMP12]], <4 x double> [[TMP18]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+; CHECK-NEXT:    [[BLOCK69:%.*]] = shufflevector <4 x double> [[COL_LOAD53]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <2 x double> [[COL_LOAD48]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT70:%.*]] = insertelement <2 x double> poison, double [[TMP20]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT71:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT70]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = fmul <2 x double> [[BLOCK69]], [[SPLAT_SPLAT71]]
+; CHECK-NEXT:    [[BLOCK72:%.*]] = shufflevector <4 x double> [[COL_LOAD56]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <2 x double> [[COL_LOAD48]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT73:%.*]] = insertelement <2 x double> poison, double [[TMP22]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT74:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT73]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP23:%.*]] = fmul <2 x double> [[BLOCK72]], [[SPLAT_SPLAT74]]
+; CHECK-NEXT:    [[TMP24:%.*]] = fadd <2 x double> [[TMP21]], [[TMP23]]
+; CHECK-NEXT:    [[TMP25:%.*]] = shufflevector <2 x double> [[TMP24]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP26:%.*]] = shufflevector <4 x double> undef, <4 x double> [[TMP25]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
+; CHECK-NEXT:    [[BLOCK75:%.*]] = shufflevector <4 x double> [[COL_LOAD53]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <2 x double> [[COL_LOAD48]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT76:%.*]] = insertelement <2 x double> poison, double [[TMP27]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT77:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT76]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP28:%.*]] = fmul <2 x double> [[BLOCK75]], [[SPLAT_SPLAT77]]
+; CHECK-NEXT:    [[BLOCK78:%.*]] = shufflevector <4 x double> [[COL_LOAD56]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <2 x double> [[COL_LOAD48]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT79:%.*]] = insertelement <2 x double> poison, double [[TMP29]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT80:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT79]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP30:%.*]] = fmul <2 x double> [[BLOCK78]], [[SPLAT_SPLAT80]]
+; CHECK-NEXT:    [[TMP31:%.*]] = fadd <2 x double> [[TMP28]], [[TMP30]]
+; CHECK-NEXT:    [[TMP32:%.*]] = shufflevector <2 x double> [[TMP31]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP33:%.*]] = shufflevector <4 x double> [[TMP26]], <4 x double> [[TMP32]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+; CHECK-NEXT:    [[BLOCK81:%.*]] = shufflevector <4 x double> [[COL_LOAD53]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP34:%.*]] = extractelement <2 x double> [[COL_LOAD51]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT82:%.*]] = insertelement <2 x double> poison, double [[TMP34]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT83:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT82]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP35:%.*]] = fmul <2 x double> [[BLOCK81]], [[SPLAT_SPLAT83]]
+; CHECK-NEXT:    [[BLOCK84:%.*]] = shufflevector <4 x double> [[COL_LOAD56]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP36:%.*]] = extractelement <2 x double> [[COL_LOAD51]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT85:%.*]] = insertelement <2 x double> poison, double [[TMP36]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT86:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT85]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP37:%.*]] = fmul <2 x double> [[BLOCK84]], [[SPLAT_SPLAT86]]
+; CHECK-NEXT:    [[TMP38:%.*]] = fadd <2 x double> [[TMP35]], [[TMP37]]
+; CHECK-NEXT:    [[TMP39:%.*]] = shufflevector <2 x double> [[TMP38]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP40:%.*]] = shufflevector <4 x double> undef, <4 x double> [[TMP39]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
+; CHECK-NEXT:    [[BLOCK87:%.*]] = shufflevector <4 x double> [[COL_LOAD53]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <2 x double> [[COL_LOAD51]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT88:%.*]] = insertelement <2 x double> poison, double [[TMP41]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT89:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT88]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP42:%.*]] = fmul <2 x double> [[BLOCK87]], [[SPLAT_SPLAT89]]
+; CHECK-NEXT:    [[BLOCK90:%.*]] = shufflevector <4 x double> [[COL_LOAD56]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP43:%.*]] = extractelement <2 x double> [[COL_LOAD51]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT91:%.*]] = insertelement <2 x double> poison, double [[TMP43]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT92:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT91]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP44:%.*]] = fmul <2 x double> [[BLOCK90]], [[SPLAT_SPLAT92]]
+; CHECK-NEXT:    [[TMP45:%.*]] = fadd <2 x double> [[TMP42]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = shufflevector <2 x double> [[TMP45]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP47:%.*]] = shufflevector <4 x double> [[TMP40]], <4 x double> [[TMP46]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+; CHECK-NEXT:    [[TMP48:%.*]] = shufflevector <4 x double> [[TMP19]], <4 x double> [[TMP33]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP49:%.*]] = shufflevector <4 x double> [[TMP47]], <4 x double> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP50:%.*]] = shufflevector <8 x double> [[TMP48]], <8 x double> [[TMP49]], <12 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>
+; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <6 x double> [[TMP3]], <6 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[SPLIT1:%.*]] = shufflevector <6 x double> [[TMP3]], <6 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[SPLIT2:%.*]] = shufflevector <6 x double> [[TMP3]], <6 x double> poison, <2 x i32> <i32 4, i32 5>
+; CHECK-NEXT:    [[SPLIT3:%.*]] = shufflevector <12 x double> [[TMP50]], <12 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[SPLIT4:%.*]] = shufflevector <12 x double> [[TMP50]], <12 x double> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[SPLIT5:%.*]] = shufflevector <12 x double> [[TMP50]], <12 x double> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
+; CHECK-NEXT:    [[BLOCK:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP51:%.*]] = extractelement <4 x double> [[SPLIT3]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <2 x double> poison, double [[TMP51]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP52:%.*]] = fmul <2 x double> [[BLOCK]], [[SPLAT_SPLAT]]
+; CHECK-NEXT:    [[BLOCK6:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP53:%.*]] = extractelement <4 x double> [[SPLIT4]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT7:%.*]] = insertelement <2 x double> poison, double [[TMP53]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT8:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT7]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP54:%.*]] = fmul <2 x double> [[BLOCK6]], [[SPLAT_SPLAT8]]
+; CHECK-NEXT:    [[TMP55:%.*]] = fadd <2 x double> [[TMP52]], [[TMP54]]
+; CHECK-NEXT:    [[BLOCK9:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP56:%.*]] = extractelement <4 x double> [[SPLIT5]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT10:%.*]] = insertelement <2 x double> poison, double [[TMP56]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT11:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT10]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP57:%.*]] = fmul <2 x double> [[BLOCK9]], [[SPLAT_SPLAT11]]
+; CHECK-NEXT:    [[TMP58:%.*]] = fadd <2 x double> [[TMP55]], [[TMP57]]
+; CHECK-NEXT:    [[TMP59:%.*]] = shufflevector <2 x double> [[TMP58]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <2 x double> undef, <2 x double> [[TMP59]], <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[BLOCK12:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP61:%.*]] = extractelement <4 x double> [[SPLIT3]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT13:%.*]] = insertelement <2 x double> poison, double [[TMP61]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT14:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT13]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP62:%.*]] = fmul <2 x double> [[BLOCK12]], [[SPLAT_SPLAT14]]
+; CHECK-NEXT:    [[BLOCK15:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP63:%.*]] = extractelement <4 x double> [[SPLIT4]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT16:%.*]] = insertelement <2 x double> poison, double [[TMP63]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT17:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT16]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP64:%.*]] = fmul <2 x double> [[BLOCK15]], [[SPLAT_SPLAT17]]
+; CHECK-NEXT:    [[TMP65:%.*]] = fadd <2 x double> [[TMP62]], [[TMP64]]
+; CHECK-NEXT:    [[BLOCK18:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP66:%.*]] = extractelement <4 x double> [[SPLIT5]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT19:%.*]] = insertelement <2 x double> poison, double [[TMP66]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT20:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT19]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP67:%.*]] = fmul <2 x double> [[BLOCK18]], [[SPLAT_SPLAT20]]
+; CHECK-NEXT:    [[TMP68:%.*]] = fadd <2 x double> [[TMP65]], [[TMP67]]
+; CHECK-NEXT:    [[TMP69:%.*]] = shufflevector <2 x double> [[TMP68]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP70:%.*]] = shufflevector <2 x double> undef, <2 x double> [[TMP69]], <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[BLOCK21:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <4 x double> [[SPLIT3]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT22:%.*]] = insertelement <2 x double> poison, double [[TMP71]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT23:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT22]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP72:%.*]] = fmul <2 x double> [[BLOCK21]], [[SPLAT_SPLAT23]]
+; CHECK-NEXT:    [[BLOCK24:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP73:%.*]] = extractelement <4 x double> [[SPLIT4]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT25:%.*]] = insertelement <2 x double> poison, double [[TMP73]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT26:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT25]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP74:%.*]] = fmul <2 x double> [[BLOCK24]], [[SPLAT_SPLAT26]]
+; CHECK-NEXT:    [[TMP75:%.*]] = fadd <2 x double> [[TMP72]], [[TMP74]]
+; CHECK-NEXT:    [[BLOCK27:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP76:%.*]] = extractelement <4 x double> [[SPLIT5]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT28:%.*]] = insertelement <2 x double> poison, double [[TMP76]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT29:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT28]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP77:%.*]] = fmul <2 x double> [[BLOCK27]], [[SPLAT_SPLAT29]]
+; CHECK-NEXT:    [[TMP78:%.*]] = fadd <2 x double> [[TMP75]], [[TMP77]]
+; CHECK-NEXT:    [[TMP79:%.*]] = shufflevector <2 x double> [[TMP78]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP80:%.*]] = shufflevector <2 x double> undef, <2 x double> [[TMP79]], <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[BLOCK30:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <4 x double> [[SPLIT3]], i64 3
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT31:%.*]] = insertelement <2 x double> poison, double [[TMP81]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT32:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT31]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP82:%.*]] = fmul <2 x double> [[BLOCK30]], [[SPLAT_SPLAT32]]
+; CHECK-NEXT:    [[BLOCK33:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP83:%.*]] = extractelement <4 x double> [[SPLIT4]], i64 3
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT34:%.*]] = insertelement <2 x double> poison, double [[TMP83]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT35:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT34]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP84:%.*]] = fmul <2 x double> [[BLOCK33]], [[SPLAT_SPLAT35]]
+; CHECK-NEXT:    [[TMP85:%.*]] = fadd <2 x double> [[TMP82]], [[TMP84]]
+; CHECK-NEXT:    [[BLOCK36:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP86:%.*]] = extractelement <4 x double> [[SPLIT5]], i64 3
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT37:%.*]] = insertelement <2 x double> poison, double [[TMP86]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT38:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT37]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP87:%.*]] = fmul <2 x double> [[BLOCK36]], [[SPLAT_SPLAT38]]
+; CHECK-NEXT:    [[TMP88:%.*]] = fadd <2 x double> [[TMP85]], [[TMP87]]
+; CHECK-NEXT:    [[TMP89:%.*]] = shufflevector <2 x double> [[TMP88]], <2 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP90:%.*]] = shufflevector <2 x double> undef, <2 x double> [[TMP89]], <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP91:%.*]] = bitcast <8 x double>* [[R:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST93:%.*]] = bitcast double* [[TMP91]] to <2 x double>*
+; CHECK-NEXT:    store <2 x double> [[TMP60]], <2 x double>* [[VEC_CAST93]], align 16
+; CHECK-NEXT:    [[VEC_GEP94:%.*]] = getelementptr double, double* [[TMP91]], i64 2
+; CHECK-NEXT:    [[VEC_CAST95:%.*]] = bitcast double* [[VEC_GEP94]] to <2 x double>*
+; CHECK-NEXT:    store <2 x double> [[TMP70]], <2 x double>* [[VEC_CAST95]], align 16
+; CHECK-NEXT:    [[VEC_GEP96:%.*]] = getelementptr double, double* [[TMP91]], i64 4
+; CHECK-NEXT:    [[VEC_CAST97:%.*]] = bitcast double* [[VEC_GEP96]] to <2 x double>*
+; CHECK-NEXT:    store <2 x double> [[TMP80]], <2 x double>* [[VEC_CAST97]], align 16
+; CHECK-NEXT:    [[VEC_GEP98:%.*]] = getelementptr double, double* [[TMP91]], i64 6
+; CHECK-NEXT:    [[VEC_CAST99:%.*]] = bitcast double* [[VEC_GEP98]] to <2 x double>*
+; CHECK-NEXT:    store <2 x double> [[TMP90]], <2 x double>* [[VEC_CAST99]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %a = load <6 x double>, <6 x double>* %A, align 16
+  %b = load <6 x double>, <6 x double>* %B, align 16
+  %c = load <8 x double>, <8 x double>* %C, align 16
+  %b_t = call <6 x double> @llvm.matrix.transpose.v6f64.v6f64(<6 x double> %b, i32 2, i32 3)
+  %c_t = call <8 x double> @llvm.matrix.transpose.v8f64.v8f64(<8 x double> %c, i32 4, i32 2)
+  %m1 = call <12 x double> @llvm.matrix.multiply.v12f64.v6f64.v8f64(<6 x double> %b_t, <8 x double> %c_t, i32 3, i32 2, i32 4)
+  %m2 = call <8 x double> @llvm.matrix.multiply.v8f64.v6f64.v12f64(<6 x double> %a, <12 x double> %m1, i32 2, i32 3, i32 4)
+  store <8 x double> %m2, <8 x double>* %R, align 16
+  ret void
+}
+
+define void @multiply_tt_t(<9 x double>* %A, <9 x double>* %B, <9 x double>* %C) {
+; REMARK:      Pass:            lower-matrix-intrinsics
+; REMARK-NEXT: Name:            matrix-lowered
+; REMARK-NEXT: Function:        multiply_tt_t
+; REMARK-NEXT: Args:
+; REMARK-NEXT:   - String:          'Lowered with '
+; REMARK-NEXT:   - NumStores:       '6'
+; REMARK-NEXT:   - String:          ' stores, '
+; REMARK-NEXT:   - NumLoads:        '12'
+; REMARK-NEXT:   - String:          ' loads, '
+; REMARK-NEXT:   - NumComputeOps:   '30'
+; REMARK-NEXT:   - String:          ' compute ops, '
+; REMARK-NEXT:   - NumExposedTransposes: '0'
+; REMARK-NEXT:   - String:          ' exposed transposes'
+; REMARK-NEXT:   - String:           |
+; REMARK:            store(
+; REMARK-NEXT:        multiply.3x3.3x3.double(
+; REMARK-NEXT:         load(addr %B),
+; REMARK-NEXT:         load(addr %A)),
+; REMARK-NEXT:        addr %C)
+; CHECK-LABEL: @multiply_tt_t(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = bitcast <9 x double>* [[A:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast double* [[TMP0]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST]], align 16
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, double* [[TMP0]], i64 3
+; CHECK-NEXT:    [[VEC_CAST1:%.*]] = bitcast double* [[VEC_GEP]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD2:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST1]], align 8
+; CHECK-NEXT:    [[VEC_GEP3:%.*]] = getelementptr double, double* [[TMP0]], i64 6
+; CHECK-NEXT:    [[VEC_CAST4:%.*]] = bitcast double* [[VEC_GEP3]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD5:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST4]], align 16
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <9 x double>* [[B:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST6:%.*]] = bitcast double* [[TMP1]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD7:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST6]], align 16
+; CHECK-NEXT:    [[VEC_GEP8:%.*]] = getelementptr double, double* [[TMP1]], i64 3
+; CHECK-NEXT:    [[VEC_CAST9:%.*]] = bitcast double* [[VEC_GEP8]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD10:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST9]], align 8
+; CHECK-NEXT:    [[VEC_GEP11:%.*]] = getelementptr double, double* [[TMP1]], i64 6
+; CHECK-NEXT:    [[VEC_CAST12:%.*]] = bitcast double* [[VEC_GEP11]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD13:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST12]], align 16
+; CHECK-NEXT:    [[BLOCK:%.*]] = shufflevector <3 x double> [[COL_LOAD7]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <3 x double> [[COL_LOAD]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <2 x double> poison, double [[TMP2]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP3:%.*]] = fmul <2 x double> [[BLOCK]], [[SPLAT_SPLAT]]
+; CHECK-NEXT:    [[BLOCK14:%.*]] = shufflevector <3 x double> [[COL_LOAD10]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <3 x double> [[COL_LOAD]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT15:%.*]] = insertelement <2 x double> poison, double [[TMP4]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT16:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT15]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP5:%.*]] = fmul <2 x double> [[BLOCK14]], [[SPLAT_SPLAT16]]
+; CHECK-NEXT:    [[TMP6:%.*]] = fadd <2 x double> [[TMP3]], [[TMP5]]
+; CHECK-NEXT:    [[BLOCK17:%.*]] = shufflevector <3 x double> [[COL_LOAD13]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <3 x double> [[COL_LOAD]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT18:%.*]] = insertelement <2 x double> poison, double [[TMP7]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT19:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT18]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP8:%.*]] = fmul <2 x double> [[BLOCK17]], [[SPLAT_SPLAT19]]
+; CHECK-NEXT:    [[TMP9:%.*]] = fadd <2 x double> [[TMP6]], [[TMP8]]
+; CHECK-NEXT:    [[TMP10:%.*]] = shufflevector <2 x double> [[TMP9]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP11:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP10]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK20:%.*]] = shufflevector <3 x double> [[COL_LOAD7]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <3 x double> [[COL_LOAD]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT21:%.*]] = insertelement <1 x double> poison, double [[TMP12]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT22:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT21]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP13:%.*]] = fmul <1 x double> [[BLOCK20]], [[SPLAT_SPLAT22]]
+; CHECK-NEXT:    [[BLOCK23:%.*]] = shufflevector <3 x double> [[COL_LOAD10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <3 x double> [[COL_LOAD]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT24:%.*]] = insertelement <1 x double> poison, double [[TMP14]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT25:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT24]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP15:%.*]] = fmul <1 x double> [[BLOCK23]], [[SPLAT_SPLAT25]]
+; CHECK-NEXT:    [[TMP16:%.*]] = fadd <1 x double> [[TMP13]], [[TMP15]]
+; CHECK-NEXT:    [[BLOCK26:%.*]] = shufflevector <3 x double> [[COL_LOAD13]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP17:%.*]] = extractelement <3 x double> [[COL_LOAD]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT27:%.*]] = insertelement <1 x double> poison, double [[TMP17]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT28:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT27]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP18:%.*]] = fmul <1 x double> [[BLOCK26]], [[SPLAT_SPLAT28]]
+; CHECK-NEXT:    [[TMP19:%.*]] = fadd <1 x double> [[TMP16]], [[TMP18]]
+; CHECK-NEXT:    [[TMP20:%.*]] = shufflevector <1 x double> [[TMP19]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP21:%.*]] = shufflevector <3 x double> [[TMP11]], <3 x double> [[TMP20]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[BLOCK29:%.*]] = shufflevector <3 x double> [[COL_LOAD7]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP22:%.*]] = extractelement <3 x double> [[COL_LOAD2]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT30:%.*]] = insertelement <2 x double> poison, double [[TMP22]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT31:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT30]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP23:%.*]] = fmul <2 x double> [[BLOCK29]], [[SPLAT_SPLAT31]]
+; CHECK-NEXT:    [[BLOCK32:%.*]] = shufflevector <3 x double> [[COL_LOAD10]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP24:%.*]] = extractelement <3 x double> [[COL_LOAD2]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT33:%.*]] = insertelement <2 x double> poison, double [[TMP24]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT34:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT33]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP25:%.*]] = fmul <2 x double> [[BLOCK32]], [[SPLAT_SPLAT34]]
+; CHECK-NEXT:    [[TMP26:%.*]] = fadd <2 x double> [[TMP23]], [[TMP25]]
+; CHECK-NEXT:    [[BLOCK35:%.*]] = shufflevector <3 x double> [[COL_LOAD13]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <3 x double> [[COL_LOAD2]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT36:%.*]] = insertelement <2 x double> poison, double [[TMP27]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT37:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT36]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP28:%.*]] = fmul <2 x double> [[BLOCK35]], [[SPLAT_SPLAT37]]
+; CHECK-NEXT:    [[TMP29:%.*]] = fadd <2 x double> [[TMP26]], [[TMP28]]
+; CHECK-NEXT:    [[TMP30:%.*]] = shufflevector <2 x double> [[TMP29]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP31:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP30]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK38:%.*]] = shufflevector <3 x double> [[COL_LOAD7]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP32:%.*]] = extractelement <3 x double> [[COL_LOAD2]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT39:%.*]] = insertelement <1 x double> poison, double [[TMP32]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT40:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT39]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP33:%.*]] = fmul <1 x double> [[BLOCK38]], [[SPLAT_SPLAT40]]
+; CHECK-NEXT:    [[BLOCK41:%.*]] = shufflevector <3 x double> [[COL_LOAD10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP34:%.*]] = extractelement <3 x double> [[COL_LOAD2]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT42:%.*]] = insertelement <1 x double> poison, double [[TMP34]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT43:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT42]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP35:%.*]] = fmul <1 x double> [[BLOCK41]], [[SPLAT_SPLAT43]]
+; CHECK-NEXT:    [[TMP36:%.*]] = fadd <1 x double> [[TMP33]], [[TMP35]]
+; CHECK-NEXT:    [[BLOCK44:%.*]] = shufflevector <3 x double> [[COL_LOAD13]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <3 x double> [[COL_LOAD2]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT45:%.*]] = insertelement <1 x double> poison, double [[TMP37]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT46:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT45]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP38:%.*]] = fmul <1 x double> [[BLOCK44]], [[SPLAT_SPLAT46]]
+; CHECK-NEXT:    [[TMP39:%.*]] = fadd <1 x double> [[TMP36]], [[TMP38]]
+; CHECK-NEXT:    [[TMP40:%.*]] = shufflevector <1 x double> [[TMP39]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP41:%.*]] = shufflevector <3 x double> [[TMP31]], <3 x double> [[TMP40]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[BLOCK47:%.*]] = shufflevector <3 x double> [[COL_LOAD7]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <3 x double> [[COL_LOAD5]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT48:%.*]] = insertelement <2 x double> poison, double [[TMP42]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT49:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT48]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP43:%.*]] = fmul <2 x double> [[BLOCK47]], [[SPLAT_SPLAT49]]
+; CHECK-NEXT:    [[BLOCK50:%.*]] = shufflevector <3 x double> [[COL_LOAD10]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <3 x double> [[COL_LOAD5]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT51:%.*]] = insertelement <2 x double> poison, double [[TMP44]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT52:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT51]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP45:%.*]] = fmul <2 x double> [[BLOCK50]], [[SPLAT_SPLAT52]]
+; CHECK-NEXT:    [[TMP46:%.*]] = fadd <2 x double> [[TMP43]], [[TMP45]]
+; CHECK-NEXT:    [[BLOCK53:%.*]] = shufflevector <3 x double> [[COL_LOAD13]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <3 x double> [[COL_LOAD5]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT54:%.*]] = insertelement <2 x double> poison, double [[TMP47]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT55:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT54]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP48:%.*]] = fmul <2 x double> [[BLOCK53]], [[SPLAT_SPLAT55]]
+; CHECK-NEXT:    [[TMP49:%.*]] = fadd <2 x double> [[TMP46]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = shufflevector <2 x double> [[TMP49]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP51:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP50]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK56:%.*]] = shufflevector <3 x double> [[COL_LOAD7]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <3 x double> [[COL_LOAD5]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT57:%.*]] = insertelement <1 x double> poison, double [[TMP52]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT58:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT57]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP53:%.*]] = fmul <1 x double> [[BLOCK56]], [[SPLAT_SPLAT58]]
+; CHECK-NEXT:    [[BLOCK59:%.*]] = shufflevector <3 x double> [[COL_LOAD10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP54:%.*]] = extractelement <3 x double> [[COL_LOAD5]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT60:%.*]] = insertelement <1 x double> poison, double [[TMP54]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT61:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT60]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP55:%.*]] = fmul <1 x double> [[BLOCK59]], [[SPLAT_SPLAT61]]
+; CHECK-NEXT:    [[TMP56:%.*]] = fadd <1 x double> [[TMP53]], [[TMP55]]
+; CHECK-NEXT:    [[BLOCK62:%.*]] = shufflevector <3 x double> [[COL_LOAD13]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP57:%.*]] = extractelement <3 x double> [[COL_LOAD5]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT63:%.*]] = insertelement <1 x double> poison, double [[TMP57]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT64:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT63]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP58:%.*]] = fmul <1 x double> [[BLOCK62]], [[SPLAT_SPLAT64]]
+; CHECK-NEXT:    [[TMP59:%.*]] = fadd <1 x double> [[TMP56]], [[TMP58]]
+; CHECK-NEXT:    [[TMP60:%.*]] = shufflevector <1 x double> [[TMP59]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <3 x double> [[TMP51]], <3 x double> [[TMP60]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[TMP62:%.*]] = bitcast <9 x double>* [[C:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST65:%.*]] = bitcast double* [[TMP62]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[TMP21]], <3 x double>* [[VEC_CAST65]], align 16
+; CHECK-NEXT:    [[VEC_GEP66:%.*]] = getelementptr double, double* [[TMP62]], i64 3
+; CHECK-NEXT:    [[VEC_CAST67:%.*]] = bitcast double* [[VEC_GEP66]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[TMP41]], <3 x double>* [[VEC_CAST67]], align 8
+; CHECK-NEXT:    [[VEC_GEP68:%.*]] = getelementptr double, double* [[TMP62]], i64 6
+; CHECK-NEXT:    [[VEC_CAST69:%.*]] = bitcast double* [[VEC_GEP68]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[TMP61]], <3 x double>* [[VEC_CAST69]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %a = load <9 x double>, <9 x double>* %A, align 16
+  %at = call <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double> %a, i32 3, i32 3)
+  %b = load <9 x double>, <9 x double>* %B, align 16
+  %bt = call <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double> %b, i32 3, i32 3)
+  %c = call <9 x double> @llvm.matrix.multiply.v9f64.v9f64.v9f64(<9 x double> %at, <9 x double> %bt, i32 3, i32 3, i32 3)
+  %ct = call <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double> %c, i32 3, i32 3)
+  store <9 x double> %ct, <9 x double>* %C, align 16
+  ret void
+}
+
+define void @multiply_nt_t(<6 x double>* %A, <12 x double>* %B, <8 x double>* %C) {
+; REMARK:      Pass:            lower-matrix-intrinsics
+; REMARK-NEXT: Name:            matrix-lowered
+; REMARK-NEXT: Function:        multiply_nt_t
+; REMARK-NEXT: Args:
+; REMARK-NEXT:   - String:          'Lowered with '
+; REMARK-NEXT:   - NumStores:       '4'
+; REMARK-NEXT:   - String:          ' stores, '
+; REMARK-NEXT:   - NumLoads:        '9'
+; REMARK-NEXT:   - String:          ' loads, '
+; REMARK-NEXT:   - NumComputeOps:   '20'
+; REMARK-NEXT:   - String:          ' compute ops, '
+; REMARK-NEXT:   - NumExposedTransposes: '0'
+; REMARK-NEXT:   - String:          ' exposed transposes'
+; REMARK-NEXT:   - String:           |
+; REMARK:            store(
+; REMARK-NEXT:        multiply.4x3.3x2.double(
+; REMARK-NEXT:         load(addr %B),
+; REMARK-NEXT:         transpose.2x3.double(load(addr %A))),
+; REMARK-NEXT:        addr %C)
+; CHECK-LABEL: @multiply_nt_t(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = bitcast <6 x double>* [[A:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast double* [[TMP0]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST]], align 16
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, double* [[TMP0]], i64 2
+; CHECK-NEXT:    [[VEC_CAST39:%.*]] = bitcast double* [[VEC_GEP]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD40:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST39]], align 16
+; CHECK-NEXT:    [[VEC_GEP41:%.*]] = getelementptr double, double* [[TMP0]], i64 4
+; CHECK-NEXT:    [[VEC_CAST42:%.*]] = bitcast double* [[VEC_GEP41]] to <2 x double>*
+; CHECK-NEXT:    [[COL_LOAD43:%.*]] = load <2 x double>, <2 x double>* [[VEC_CAST42]], align 16
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> [[COL_LOAD40]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <2 x double> [[COL_LOAD43]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x double> [[TMP1]], <4 x double> [[TMP2]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <12 x double>* [[B:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST44:%.*]] = bitcast double* [[TMP4]] to <4 x double>*
+; CHECK-NEXT:    [[COL_LOAD45:%.*]] = load <4 x double>, <4 x double>* [[VEC_CAST44]], align 16
+; CHECK-NEXT:    [[VEC_GEP46:%.*]] = getelementptr double, double* [[TMP4]], i64 4
+; CHECK-NEXT:    [[VEC_CAST47:%.*]] = bitcast double* [[VEC_GEP46]] to <4 x double>*
+; CHECK-NEXT:    [[COL_LOAD48:%.*]] = load <4 x double>, <4 x double>* [[VEC_CAST47]], align 16
+; CHECK-NEXT:    [[VEC_GEP49:%.*]] = getelementptr double, double* [[TMP4]], i64 8
+; CHECK-NEXT:    [[VEC_CAST50:%.*]] = bitcast double* [[VEC_GEP49]] to <4 x double>*
+; CHECK-NEXT:    [[COL_LOAD51:%.*]] = load <4 x double>, <4 x double>* [[VEC_CAST50]], align 16
+; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <4 x double> [[COL_LOAD45]], <4 x double> [[COL_LOAD48]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <4 x double> [[COL_LOAD51]], <4 x double> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP7:%.*]] = shufflevector <8 x double> [[TMP5]], <8 x double> [[TMP6]], <12 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>
+; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <12 x double> [[TMP7]], <12 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+; CHECK-NEXT:    [[SPLIT1:%.*]] = shufflevector <12 x double> [[TMP7]], <12 x double> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[SPLIT2:%.*]] = shufflevector <12 x double> [[TMP7]], <12 x double> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>
+; CHECK-NEXT:    [[SPLIT3:%.*]] = shufflevector <6 x double> [[TMP3]], <6 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[SPLIT4:%.*]] = shufflevector <6 x double> [[TMP3]], <6 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[SPLIT5:%.*]] = shufflevector <6 x double> [[TMP3]], <6 x double> poison, <2 x i32> <i32 4, i32 5>
+; CHECK-NEXT:    [[BLOCK:%.*]] = shufflevector <4 x double> [[SPLIT]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <2 x double> poison, double [[TMP8]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = fmul <2 x double> [[BLOCK]], [[SPLAT_SPLAT]]
+; CHECK-NEXT:    [[BLOCK6:%.*]] = shufflevector <4 x double> [[SPLIT1]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT7:%.*]] = insertelement <2 x double> poison, double [[TMP10]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT8:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT7]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = fmul <2 x double> [[BLOCK6]], [[SPLAT_SPLAT8]]
+; CHECK-NEXT:    [[TMP12:%.*]] = fadd <2 x double> [[TMP9]], [[TMP11]]
+; CHECK-NEXT:    [[BLOCK9:%.*]] = shufflevector <4 x double> [[SPLIT2]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP13:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT10:%.*]] = insertelement <2 x double> poison, double [[TMP13]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT11:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT10]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP14:%.*]] = fmul <2 x double> [[BLOCK9]], [[SPLAT_SPLAT11]]
+; CHECK-NEXT:    [[TMP15:%.*]] = fadd <2 x double> [[TMP12]], [[TMP14]]
+; CHECK-NEXT:    [[TMP16:%.*]] = shufflevector <2 x double> [[TMP15]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP17:%.*]] = shufflevector <4 x double> undef, <4 x double> [[TMP16]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
+; CHECK-NEXT:    [[BLOCK12:%.*]] = shufflevector <4 x double> [[SPLIT]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT13:%.*]] = insertelement <2 x double> poison, double [[TMP18]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT14:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT13]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = fmul <2 x double> [[BLOCK12]], [[SPLAT_SPLAT14]]
+; CHECK-NEXT:    [[BLOCK15:%.*]] = shufflevector <4 x double> [[SPLIT1]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP20:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT16:%.*]] = insertelement <2 x double> poison, double [[TMP20]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT17:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT16]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP21:%.*]] = fmul <2 x double> [[BLOCK15]], [[SPLAT_SPLAT17]]
+; CHECK-NEXT:    [[TMP22:%.*]] = fadd <2 x double> [[TMP19]], [[TMP21]]
+; CHECK-NEXT:    [[BLOCK18:%.*]] = shufflevector <4 x double> [[SPLIT2]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT19:%.*]] = insertelement <2 x double> poison, double [[TMP23]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT20:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT19]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP24:%.*]] = fmul <2 x double> [[BLOCK18]], [[SPLAT_SPLAT20]]
+; CHECK-NEXT:    [[TMP25:%.*]] = fadd <2 x double> [[TMP22]], [[TMP24]]
+; CHECK-NEXT:    [[TMP26:%.*]] = shufflevector <2 x double> [[TMP25]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP27:%.*]] = shufflevector <4 x double> [[TMP17]], <4 x double> [[TMP26]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+; CHECK-NEXT:    [[BLOCK21:%.*]] = shufflevector <4 x double> [[SPLIT]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT22:%.*]] = insertelement <2 x double> poison, double [[TMP28]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT23:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT22]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP29:%.*]] = fmul <2 x double> [[BLOCK21]], [[SPLAT_SPLAT23]]
+; CHECK-NEXT:    [[BLOCK24:%.*]] = shufflevector <4 x double> [[SPLIT1]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP30:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT25:%.*]] = insertelement <2 x double> poison, double [[TMP30]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT26:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT25]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP31:%.*]] = fmul <2 x double> [[BLOCK24]], [[SPLAT_SPLAT26]]
+; CHECK-NEXT:    [[TMP32:%.*]] = fadd <2 x double> [[TMP29]], [[TMP31]]
+; CHECK-NEXT:    [[BLOCK27:%.*]] = shufflevector <4 x double> [[SPLIT2]], <4 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP33:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT28:%.*]] = insertelement <2 x double> poison, double [[TMP33]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT29:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT28]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP34:%.*]] = fmul <2 x double> [[BLOCK27]], [[SPLAT_SPLAT29]]
+; CHECK-NEXT:    [[TMP35:%.*]] = fadd <2 x double> [[TMP32]], [[TMP34]]
+; CHECK-NEXT:    [[TMP36:%.*]] = shufflevector <2 x double> [[TMP35]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP37:%.*]] = shufflevector <4 x double> undef, <4 x double> [[TMP36]], <4 x i32> <i32 4, i32 5, i32 2, i32 3>
+; CHECK-NEXT:    [[BLOCK30:%.*]] = shufflevector <4 x double> [[SPLIT]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP38:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT31:%.*]] = insertelement <2 x double> poison, double [[TMP38]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT32:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT31]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP39:%.*]] = fmul <2 x double> [[BLOCK30]], [[SPLAT_SPLAT32]]
+; CHECK-NEXT:    [[BLOCK33:%.*]] = shufflevector <4 x double> [[SPLIT1]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP40:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT34:%.*]] = insertelement <2 x double> poison, double [[TMP40]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT35:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT34]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP41:%.*]] = fmul <2 x double> [[BLOCK33]], [[SPLAT_SPLAT35]]
+; CHECK-NEXT:    [[TMP42:%.*]] = fadd <2 x double> [[TMP39]], [[TMP41]]
+; CHECK-NEXT:    [[BLOCK36:%.*]] = shufflevector <4 x double> [[SPLIT2]], <4 x double> poison, <2 x i32> <i32 2, i32 3>
+; CHECK-NEXT:    [[TMP43:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT37:%.*]] = insertelement <2 x double> poison, double [[TMP43]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT38:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT37]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP44:%.*]] = fmul <2 x double> [[BLOCK36]], [[SPLAT_SPLAT38]]
+; CHECK-NEXT:    [[TMP45:%.*]] = fadd <2 x double> [[TMP42]], [[TMP44]]
+; CHECK-NEXT:    [[TMP46:%.*]] = shufflevector <2 x double> [[TMP45]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP47:%.*]] = shufflevector <4 x double> [[TMP37]], <4 x double> [[TMP46]], <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+; CHECK-NEXT:    [[TMP48:%.*]] = bitcast <8 x double>* [[C:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST52:%.*]] = bitcast double* [[TMP48]] to <4 x double>*
+; CHECK-NEXT:    store <4 x double> [[TMP27]], <4 x double>* [[VEC_CAST52]], align 16
+; CHECK-NEXT:    [[VEC_GEP53:%.*]] = getelementptr double, double* [[TMP48]], i64 4
+; CHECK-NEXT:    [[VEC_CAST54:%.*]] = bitcast double* [[VEC_GEP53]] to <4 x double>*
+; CHECK-NEXT:    store <4 x double> [[TMP47]], <4 x double>* [[VEC_CAST54]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %a = load <6 x double>, <6 x double>* %A, align 16
+  %b = load <12 x double>, <12 x double>* %B, align 16
+  %bt = call <12 x double> @llvm.matrix.transpose.v12f64.v12f64(<12 x double> %b, i32 4, i32 3)
+  %c = call <8 x double> @llvm.matrix.multiply.v8f64.v6f64.v12f64(<6 x double> %a, <12 x double> %bt, i32 2, i32 3, i32 4)
+  %ct = call <8 x double> @llvm.matrix.transpose.v8f64.v8f64(<8 x double> %c, i32 2, i32 4)
+  store <8 x double> %ct, <8 x double>* %C, align 16
+  ret void
+}
+
+define void @multiply_ntt_t(<9 x double>* %A, <9 x double>* %B, <9 x double>* %C, <9 x double>* %R) {
+; REMARK:      Pass:            lower-matrix-intrinsics
+; REMARK-NEXT: Name:            matrix-lowered
+; REMARK-NEXT: Function:        multiply_ntt_t
+; REMARK-NEXT: Args:
+; REMARK-NEXT:   - String:          'Lowered with '
+; REMARK-NEXT:   - NumStores:       '6'
+; REMARK-NEXT:   - String:          ' stores, '
+; REMARK-NEXT:   - NumLoads:        '18'
+; REMARK-NEXT:   - String:          ' loads, '
+; REMARK-NEXT:   - NumComputeOps:   '60'
+; REMARK-NEXT:   - String:          ' compute ops, '
+; REMARK-NEXT:   - NumExposedTransposes: '0'
+; REMARK-NEXT:   - String:          ' exposed transposes'
+; REMARK-NEXT:   - String:           |
+; REMARK:            store(
+; REMARK-NEXT:        multiply.3x3.3x3.double(
+; REMARK-NEXT:         multiply.3x3.3x3.double(
+; REMARK-NEXT:          load(addr %C),
+; REMARK-NEXT:          load(addr %B)),
+; REMARK-NEXT:         transpose.3x3.double(load(addr %A))),
+; REMARK-NEXT:        addr %R)
+; CHECK-LABEL: @multiply_ntt_t(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = bitcast <9 x double>* [[A:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST:%.*]] = bitcast double* [[TMP0]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST]], align 16
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, double* [[TMP0]], i64 3
+; CHECK-NEXT:    [[VEC_CAST58:%.*]] = bitcast double* [[VEC_GEP]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD59:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST58]], align 8
+; CHECK-NEXT:    [[VEC_GEP60:%.*]] = getelementptr double, double* [[TMP0]], i64 6
+; CHECK-NEXT:    [[VEC_CAST61:%.*]] = bitcast double* [[VEC_GEP60]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD62:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST61]], align 16
+; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <3 x double> [[COL_LOAD]], <3 x double> [[COL_LOAD59]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <3 x double> [[COL_LOAD62]], <3 x double> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <6 x double> [[TMP1]], <6 x double> [[TMP2]], <9 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
+; CHECK-NEXT:    [[TMP4:%.*]] = bitcast <9 x double>* [[B:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST63:%.*]] = bitcast double* [[TMP4]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD64:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST63]], align 16
+; CHECK-NEXT:    [[VEC_GEP65:%.*]] = getelementptr double, double* [[TMP4]], i64 3
+; CHECK-NEXT:    [[VEC_CAST66:%.*]] = bitcast double* [[VEC_GEP65]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD67:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST66]], align 8
+; CHECK-NEXT:    [[VEC_GEP68:%.*]] = getelementptr double, double* [[TMP4]], i64 6
+; CHECK-NEXT:    [[VEC_CAST69:%.*]] = bitcast double* [[VEC_GEP68]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD70:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST69]], align 16
+; CHECK-NEXT:    [[TMP5:%.*]] = bitcast <9 x double>* [[C:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST71:%.*]] = bitcast double* [[TMP5]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD72:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST71]], align 16
+; CHECK-NEXT:    [[VEC_GEP73:%.*]] = getelementptr double, double* [[TMP5]], i64 3
+; CHECK-NEXT:    [[VEC_CAST74:%.*]] = bitcast double* [[VEC_GEP73]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD75:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST74]], align 8
+; CHECK-NEXT:    [[VEC_GEP76:%.*]] = getelementptr double, double* [[TMP5]], i64 6
+; CHECK-NEXT:    [[VEC_CAST77:%.*]] = bitcast double* [[VEC_GEP76]] to <3 x double>*
+; CHECK-NEXT:    [[COL_LOAD78:%.*]] = load <3 x double>, <3 x double>* [[VEC_CAST77]], align 16
+; CHECK-NEXT:    [[BLOCK79:%.*]] = shufflevector <3 x double> [[COL_LOAD72]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <3 x double> [[COL_LOAD64]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT80:%.*]] = insertelement <2 x double> poison, double [[TMP6]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT81:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT80]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP7:%.*]] = fmul <2 x double> [[BLOCK79]], [[SPLAT_SPLAT81]]
+; CHECK-NEXT:    [[BLOCK82:%.*]] = shufflevector <3 x double> [[COL_LOAD75]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <3 x double> [[COL_LOAD64]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT83:%.*]] = insertelement <2 x double> poison, double [[TMP8]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT84:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT83]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP9:%.*]] = fmul <2 x double> [[BLOCK82]], [[SPLAT_SPLAT84]]
+; CHECK-NEXT:    [[TMP10:%.*]] = fadd <2 x double> [[TMP7]], [[TMP9]]
+; CHECK-NEXT:    [[BLOCK85:%.*]] = shufflevector <3 x double> [[COL_LOAD78]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP11:%.*]] = extractelement <3 x double> [[COL_LOAD64]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT86:%.*]] = insertelement <2 x double> poison, double [[TMP11]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT87:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT86]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP12:%.*]] = fmul <2 x double> [[BLOCK85]], [[SPLAT_SPLAT87]]
+; CHECK-NEXT:    [[TMP13:%.*]] = fadd <2 x double> [[TMP10]], [[TMP12]]
+; CHECK-NEXT:    [[TMP14:%.*]] = shufflevector <2 x double> [[TMP13]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP15:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP14]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK88:%.*]] = shufflevector <3 x double> [[COL_LOAD72]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <3 x double> [[COL_LOAD64]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT89:%.*]] = insertelement <1 x double> poison, double [[TMP16]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT90:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT89]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP17:%.*]] = fmul <1 x double> [[BLOCK88]], [[SPLAT_SPLAT90]]
+; CHECK-NEXT:    [[BLOCK91:%.*]] = shufflevector <3 x double> [[COL_LOAD75]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP18:%.*]] = extractelement <3 x double> [[COL_LOAD64]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT92:%.*]] = insertelement <1 x double> poison, double [[TMP18]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT93:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT92]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = fmul <1 x double> [[BLOCK91]], [[SPLAT_SPLAT93]]
+; CHECK-NEXT:    [[TMP20:%.*]] = fadd <1 x double> [[TMP17]], [[TMP19]]
+; CHECK-NEXT:    [[BLOCK94:%.*]] = shufflevector <3 x double> [[COL_LOAD78]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP21:%.*]] = extractelement <3 x double> [[COL_LOAD64]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT95:%.*]] = insertelement <1 x double> poison, double [[TMP21]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT96:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT95]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP22:%.*]] = fmul <1 x double> [[BLOCK94]], [[SPLAT_SPLAT96]]
+; CHECK-NEXT:    [[TMP23:%.*]] = fadd <1 x double> [[TMP20]], [[TMP22]]
+; CHECK-NEXT:    [[TMP24:%.*]] = shufflevector <1 x double> [[TMP23]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP25:%.*]] = shufflevector <3 x double> [[TMP15]], <3 x double> [[TMP24]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[BLOCK97:%.*]] = shufflevector <3 x double> [[COL_LOAD72]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP26:%.*]] = extractelement <3 x double> [[COL_LOAD67]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT98:%.*]] = insertelement <2 x double> poison, double [[TMP26]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT99:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT98]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP27:%.*]] = fmul <2 x double> [[BLOCK97]], [[SPLAT_SPLAT99]]
+; CHECK-NEXT:    [[BLOCK100:%.*]] = shufflevector <3 x double> [[COL_LOAD75]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP28:%.*]] = extractelement <3 x double> [[COL_LOAD67]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT101:%.*]] = insertelement <2 x double> poison, double [[TMP28]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT102:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT101]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP29:%.*]] = fmul <2 x double> [[BLOCK100]], [[SPLAT_SPLAT102]]
+; CHECK-NEXT:    [[TMP30:%.*]] = fadd <2 x double> [[TMP27]], [[TMP29]]
+; CHECK-NEXT:    [[BLOCK103:%.*]] = shufflevector <3 x double> [[COL_LOAD78]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <3 x double> [[COL_LOAD67]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT104:%.*]] = insertelement <2 x double> poison, double [[TMP31]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT105:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT104]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP32:%.*]] = fmul <2 x double> [[BLOCK103]], [[SPLAT_SPLAT105]]
+; CHECK-NEXT:    [[TMP33:%.*]] = fadd <2 x double> [[TMP30]], [[TMP32]]
+; CHECK-NEXT:    [[TMP34:%.*]] = shufflevector <2 x double> [[TMP33]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP35:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP34]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK106:%.*]] = shufflevector <3 x double> [[COL_LOAD72]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP36:%.*]] = extractelement <3 x double> [[COL_LOAD67]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT107:%.*]] = insertelement <1 x double> poison, double [[TMP36]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT108:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT107]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP37:%.*]] = fmul <1 x double> [[BLOCK106]], [[SPLAT_SPLAT108]]
+; CHECK-NEXT:    [[BLOCK109:%.*]] = shufflevector <3 x double> [[COL_LOAD75]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP38:%.*]] = extractelement <3 x double> [[COL_LOAD67]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT110:%.*]] = insertelement <1 x double> poison, double [[TMP38]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT111:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT110]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP39:%.*]] = fmul <1 x double> [[BLOCK109]], [[SPLAT_SPLAT111]]
+; CHECK-NEXT:    [[TMP40:%.*]] = fadd <1 x double> [[TMP37]], [[TMP39]]
+; CHECK-NEXT:    [[BLOCK112:%.*]] = shufflevector <3 x double> [[COL_LOAD78]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP41:%.*]] = extractelement <3 x double> [[COL_LOAD67]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT113:%.*]] = insertelement <1 x double> poison, double [[TMP41]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT114:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT113]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP42:%.*]] = fmul <1 x double> [[BLOCK112]], [[SPLAT_SPLAT114]]
+; CHECK-NEXT:    [[TMP43:%.*]] = fadd <1 x double> [[TMP40]], [[TMP42]]
+; CHECK-NEXT:    [[TMP44:%.*]] = shufflevector <1 x double> [[TMP43]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP45:%.*]] = shufflevector <3 x double> [[TMP35]], <3 x double> [[TMP44]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[BLOCK115:%.*]] = shufflevector <3 x double> [[COL_LOAD72]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP46:%.*]] = extractelement <3 x double> [[COL_LOAD70]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT116:%.*]] = insertelement <2 x double> poison, double [[TMP46]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT117:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT116]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP47:%.*]] = fmul <2 x double> [[BLOCK115]], [[SPLAT_SPLAT117]]
+; CHECK-NEXT:    [[BLOCK118:%.*]] = shufflevector <3 x double> [[COL_LOAD75]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP48:%.*]] = extractelement <3 x double> [[COL_LOAD70]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT119:%.*]] = insertelement <2 x double> poison, double [[TMP48]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT120:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT119]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP49:%.*]] = fmul <2 x double> [[BLOCK118]], [[SPLAT_SPLAT120]]
+; CHECK-NEXT:    [[TMP50:%.*]] = fadd <2 x double> [[TMP47]], [[TMP49]]
+; CHECK-NEXT:    [[BLOCK121:%.*]] = shufflevector <3 x double> [[COL_LOAD78]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP51:%.*]] = extractelement <3 x double> [[COL_LOAD70]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT122:%.*]] = insertelement <2 x double> poison, double [[TMP51]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT123:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT122]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP52:%.*]] = fmul <2 x double> [[BLOCK121]], [[SPLAT_SPLAT123]]
+; CHECK-NEXT:    [[TMP53:%.*]] = fadd <2 x double> [[TMP50]], [[TMP52]]
+; CHECK-NEXT:    [[TMP54:%.*]] = shufflevector <2 x double> [[TMP53]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP55:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP54]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK124:%.*]] = shufflevector <3 x double> [[COL_LOAD72]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP56:%.*]] = extractelement <3 x double> [[COL_LOAD70]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT125:%.*]] = insertelement <1 x double> poison, double [[TMP56]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT126:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT125]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP57:%.*]] = fmul <1 x double> [[BLOCK124]], [[SPLAT_SPLAT126]]
+; CHECK-NEXT:    [[BLOCK127:%.*]] = shufflevector <3 x double> [[COL_LOAD75]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP58:%.*]] = extractelement <3 x double> [[COL_LOAD70]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT128:%.*]] = insertelement <1 x double> poison, double [[TMP58]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT129:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT128]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP59:%.*]] = fmul <1 x double> [[BLOCK127]], [[SPLAT_SPLAT129]]
+; CHECK-NEXT:    [[TMP60:%.*]] = fadd <1 x double> [[TMP57]], [[TMP59]]
+; CHECK-NEXT:    [[BLOCK130:%.*]] = shufflevector <3 x double> [[COL_LOAD78]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP61:%.*]] = extractelement <3 x double> [[COL_LOAD70]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT131:%.*]] = insertelement <1 x double> poison, double [[TMP61]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT132:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT131]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP62:%.*]] = fmul <1 x double> [[BLOCK130]], [[SPLAT_SPLAT132]]
+; CHECK-NEXT:    [[TMP63:%.*]] = fadd <1 x double> [[TMP60]], [[TMP62]]
+; CHECK-NEXT:    [[TMP64:%.*]] = shufflevector <1 x double> [[TMP63]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP65:%.*]] = shufflevector <3 x double> [[TMP55]], <3 x double> [[TMP64]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[TMP66:%.*]] = shufflevector <3 x double> [[TMP25]], <3 x double> [[TMP45]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[TMP67:%.*]] = shufflevector <3 x double> [[TMP65]], <3 x double> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 undef, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP68:%.*]] = shufflevector <6 x double> [[TMP66]], <6 x double> [[TMP67]], <9 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
+; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <9 x double> [[TMP68]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; CHECK-NEXT:    [[SPLIT2:%.*]] = shufflevector <9 x double> [[TMP68]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[SPLIT3:%.*]] = shufflevector <9 x double> [[TMP68]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
+; CHECK-NEXT:    [[SPLIT4:%.*]] = shufflevector <9 x double> [[TMP3]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; CHECK-NEXT:    [[SPLIT5:%.*]] = shufflevector <9 x double> [[TMP3]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[SPLIT6:%.*]] = shufflevector <9 x double> [[TMP3]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
+; CHECK-NEXT:    [[BLOCK:%.*]] = shufflevector <3 x double> [[SPLIT]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP69:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <2 x double> poison, double [[TMP69]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP70:%.*]] = fmul <2 x double> [[BLOCK]], [[SPLAT_SPLAT]]
+; CHECK-NEXT:    [[BLOCK7:%.*]] = shufflevector <3 x double> [[SPLIT2]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP71:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT8:%.*]] = insertelement <2 x double> poison, double [[TMP71]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT9:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT8]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP72:%.*]] = fmul <2 x double> [[BLOCK7]], [[SPLAT_SPLAT9]]
+; CHECK-NEXT:    [[TMP73:%.*]] = fadd <2 x double> [[TMP70]], [[TMP72]]
+; CHECK-NEXT:    [[BLOCK10:%.*]] = shufflevector <3 x double> [[SPLIT3]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <3 x double> [[SPLIT6]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT11:%.*]] = insertelement <2 x double> poison, double [[TMP74]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT12:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT11]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP75:%.*]] = fmul <2 x double> [[BLOCK10]], [[SPLAT_SPLAT12]]
+; CHECK-NEXT:    [[TMP76:%.*]] = fadd <2 x double> [[TMP73]], [[TMP75]]
+; CHECK-NEXT:    [[TMP77:%.*]] = shufflevector <2 x double> [[TMP76]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP78:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP77]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK13:%.*]] = shufflevector <3 x double> [[SPLIT]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP79:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT14:%.*]] = insertelement <1 x double> poison, double [[TMP79]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT15:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT14]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP80:%.*]] = fmul <1 x double> [[BLOCK13]], [[SPLAT_SPLAT15]]
+; CHECK-NEXT:    [[BLOCK16:%.*]] = shufflevector <3 x double> [[SPLIT2]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP81:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT17:%.*]] = insertelement <1 x double> poison, double [[TMP81]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT18:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT17]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP82:%.*]] = fmul <1 x double> [[BLOCK16]], [[SPLAT_SPLAT18]]
+; CHECK-NEXT:    [[TMP83:%.*]] = fadd <1 x double> [[TMP80]], [[TMP82]]
+; CHECK-NEXT:    [[BLOCK19:%.*]] = shufflevector <3 x double> [[SPLIT3]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP84:%.*]] = extractelement <3 x double> [[SPLIT6]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT20:%.*]] = insertelement <1 x double> poison, double [[TMP84]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT21:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT20]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP85:%.*]] = fmul <1 x double> [[BLOCK19]], [[SPLAT_SPLAT21]]
+; CHECK-NEXT:    [[TMP86:%.*]] = fadd <1 x double> [[TMP83]], [[TMP85]]
+; CHECK-NEXT:    [[TMP87:%.*]] = shufflevector <1 x double> [[TMP86]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP88:%.*]] = shufflevector <3 x double> [[TMP78]], <3 x double> [[TMP87]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[BLOCK22:%.*]] = shufflevector <3 x double> [[SPLIT]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP89:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT23:%.*]] = insertelement <2 x double> poison, double [[TMP89]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT24:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT23]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP90:%.*]] = fmul <2 x double> [[BLOCK22]], [[SPLAT_SPLAT24]]
+; CHECK-NEXT:    [[BLOCK25:%.*]] = shufflevector <3 x double> [[SPLIT2]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP91:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT26:%.*]] = insertelement <2 x double> poison, double [[TMP91]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT27:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT26]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP92:%.*]] = fmul <2 x double> [[BLOCK25]], [[SPLAT_SPLAT27]]
+; CHECK-NEXT:    [[TMP93:%.*]] = fadd <2 x double> [[TMP90]], [[TMP92]]
+; CHECK-NEXT:    [[BLOCK28:%.*]] = shufflevector <3 x double> [[SPLIT3]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <3 x double> [[SPLIT6]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT29:%.*]] = insertelement <2 x double> poison, double [[TMP94]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT30:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT29]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP95:%.*]] = fmul <2 x double> [[BLOCK28]], [[SPLAT_SPLAT30]]
+; CHECK-NEXT:    [[TMP96:%.*]] = fadd <2 x double> [[TMP93]], [[TMP95]]
+; CHECK-NEXT:    [[TMP97:%.*]] = shufflevector <2 x double> [[TMP96]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP98:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP97]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK31:%.*]] = shufflevector <3 x double> [[SPLIT]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP99:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT32:%.*]] = insertelement <1 x double> poison, double [[TMP99]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT33:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT32]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP100:%.*]] = fmul <1 x double> [[BLOCK31]], [[SPLAT_SPLAT33]]
+; CHECK-NEXT:    [[BLOCK34:%.*]] = shufflevector <3 x double> [[SPLIT2]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP101:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT35:%.*]] = insertelement <1 x double> poison, double [[TMP101]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT36:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT35]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP102:%.*]] = fmul <1 x double> [[BLOCK34]], [[SPLAT_SPLAT36]]
+; CHECK-NEXT:    [[TMP103:%.*]] = fadd <1 x double> [[TMP100]], [[TMP102]]
+; CHECK-NEXT:    [[BLOCK37:%.*]] = shufflevector <3 x double> [[SPLIT3]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP104:%.*]] = extractelement <3 x double> [[SPLIT6]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT38:%.*]] = insertelement <1 x double> poison, double [[TMP104]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT39:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT38]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP105:%.*]] = fmul <1 x double> [[BLOCK37]], [[SPLAT_SPLAT39]]
+; CHECK-NEXT:    [[TMP106:%.*]] = fadd <1 x double> [[TMP103]], [[TMP105]]
+; CHECK-NEXT:    [[TMP107:%.*]] = shufflevector <1 x double> [[TMP106]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP108:%.*]] = shufflevector <3 x double> [[TMP98]], <3 x double> [[TMP107]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[BLOCK40:%.*]] = shufflevector <3 x double> [[SPLIT]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP109:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT41:%.*]] = insertelement <2 x double> poison, double [[TMP109]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT42:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT41]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP110:%.*]] = fmul <2 x double> [[BLOCK40]], [[SPLAT_SPLAT42]]
+; CHECK-NEXT:    [[BLOCK43:%.*]] = shufflevector <3 x double> [[SPLIT2]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP111:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT44:%.*]] = insertelement <2 x double> poison, double [[TMP111]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT45:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT44]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP112:%.*]] = fmul <2 x double> [[BLOCK43]], [[SPLAT_SPLAT45]]
+; CHECK-NEXT:    [[TMP113:%.*]] = fadd <2 x double> [[TMP110]], [[TMP112]]
+; CHECK-NEXT:    [[BLOCK46:%.*]] = shufflevector <3 x double> [[SPLIT3]], <3 x double> poison, <2 x i32> <i32 0, i32 1>
+; CHECK-NEXT:    [[TMP114:%.*]] = extractelement <3 x double> [[SPLIT6]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT47:%.*]] = insertelement <2 x double> poison, double [[TMP114]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT48:%.*]] = shufflevector <2 x double> [[SPLAT_SPLATINSERT47]], <2 x double> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP115:%.*]] = fmul <2 x double> [[BLOCK46]], [[SPLAT_SPLAT48]]
+; CHECK-NEXT:    [[TMP116:%.*]] = fadd <2 x double> [[TMP113]], [[TMP115]]
+; CHECK-NEXT:    [[TMP117:%.*]] = shufflevector <2 x double> [[TMP116]], <2 x double> poison, <3 x i32> <i32 0, i32 1, i32 undef>
+; CHECK-NEXT:    [[TMP118:%.*]] = shufflevector <3 x double> undef, <3 x double> [[TMP117]], <3 x i32> <i32 3, i32 4, i32 2>
+; CHECK-NEXT:    [[BLOCK49:%.*]] = shufflevector <3 x double> [[SPLIT]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP119:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT50:%.*]] = insertelement <1 x double> poison, double [[TMP119]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT51:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT50]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP120:%.*]] = fmul <1 x double> [[BLOCK49]], [[SPLAT_SPLAT51]]
+; CHECK-NEXT:    [[BLOCK52:%.*]] = shufflevector <3 x double> [[SPLIT2]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP121:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT53:%.*]] = insertelement <1 x double> poison, double [[TMP121]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT54:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT53]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP122:%.*]] = fmul <1 x double> [[BLOCK52]], [[SPLAT_SPLAT54]]
+; CHECK-NEXT:    [[TMP123:%.*]] = fadd <1 x double> [[TMP120]], [[TMP122]]
+; CHECK-NEXT:    [[BLOCK55:%.*]] = shufflevector <3 x double> [[SPLIT3]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP124:%.*]] = extractelement <3 x double> [[SPLIT6]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT56:%.*]] = insertelement <1 x double> poison, double [[TMP124]], i32 0
+; CHECK-NEXT:    [[SPLAT_SPLAT57:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT56]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP125:%.*]] = fmul <1 x double> [[BLOCK55]], [[SPLAT_SPLAT57]]
+; CHECK-NEXT:    [[TMP126:%.*]] = fadd <1 x double> [[TMP123]], [[TMP125]]
+; CHECK-NEXT:    [[TMP127:%.*]] = shufflevector <1 x double> [[TMP126]], <1 x double> poison, <3 x i32> <i32 0, i32 undef, i32 undef>
+; CHECK-NEXT:    [[TMP128:%.*]] = shufflevector <3 x double> [[TMP118]], <3 x double> [[TMP127]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[TMP129:%.*]] = bitcast <9 x double>* [[R:%.*]] to double*
+; CHECK-NEXT:    [[VEC_CAST133:%.*]] = bitcast double* [[TMP129]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[TMP88]], <3 x double>* [[VEC_CAST133]], align 16
+; CHECK-NEXT:    [[VEC_GEP134:%.*]] = getelementptr double, double* [[TMP129]], i64 3
+; CHECK-NEXT:    [[VEC_CAST135:%.*]] = bitcast double* [[VEC_GEP134]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[TMP108]], <3 x double>* [[VEC_CAST135]], align 8
+; CHECK-NEXT:    [[VEC_GEP136:%.*]] = getelementptr double, double* [[TMP129]], i64 6
+; CHECK-NEXT:    [[VEC_CAST137:%.*]] = bitcast double* [[VEC_GEP136]] to <3 x double>*
+; CHECK-NEXT:    store <3 x double> [[TMP128]], <3 x double>* [[VEC_CAST137]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %a = load <9 x double>, <9 x double>* %A, align 16
+  %b = load <9 x double>, <9 x double>* %B, align 16
+  %bt = call <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double> %b, i32 3, i32 3)
+  %c = load <9 x double>, <9 x double>* %C, align 16
+  %ct = call <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double> %c, i32 3, i32 3)
+  %btct = call <9 x double> @llvm.matrix.multiply.v9f64.v9f64.v9f64(<9 x double> %bt, <9 x double> %ct, i32 3, i32 3, i32 3)
+  %abtct= call <9 x double> @llvm.matrix.multiply.v9f64.v9f64.v9f64(<9 x double> %a, <9 x double> %btct, i32 3, i32 3, i32 3)
+  %abtct_t = call <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double> %abtct, i32 3, i32 3)
+  store <9 x double> %abtct_t, <9 x double>* %R, align 16
+  ret void
+}
+
+declare <9 x double> @llvm.matrix.multiply.v9f64.v9f64.v9f64(<9 x double>, <9 x double>, i32, i32, i32)
+declare <12 x double> @llvm.matrix.multiply.v12f64.v6f64.v8f64(<6 x double>, <8 x double>, i32, i32, i32)
+declare <8 x double> @llvm.matrix.multiply.v8f64.v6f64.v12f64(<6 x double> %a, <12 x double>, i32, i32, i32)
+declare <9 x double> @llvm.matrix.transpose.v9f64.v9f64(<9 x double>, i32, i32)
+declare <6 x double> @llvm.matrix.transpose.v6f64.v6f64(<6 x double>, i32, i32)
+declare <8 x double> @llvm.matrix.transpose.v8f64.v8f64(<8 x double>, i32, i32)
+declare <12 x double> @llvm.matrix.transpose.v12f64.v12f64(<12 x double>, i32, i32)