[llvm] [Matrix] Propagate shape information through PHI instructions (PR #141681)

Jon Roelofs via llvm-commits llvm-commits at lists.llvm.org
Wed May 28 11:44:39 PDT 2025


https://github.com/jroelofs updated https://github.com/llvm/llvm-project/pull/141681

>From b1b79aad6639f82f4d4f70eed23f6c252dcdd2c6 Mon Sep 17 00:00:00 2001
From: Jon Roelofs <jonathan_roelofs at apple.com>
Date: Tue, 27 May 2025 12:59:31 -0700
Subject: [PATCH 1/5] [Matrix] Propagate shape information through PHI
 instructions

... and split them as we lower themm, avoiding several shuffles in the process.
---
 .../Scalar/LowerMatrixIntrinsics.cpp          |  93 +++++++-
 .../Transforms/LowerMatrixIntrinsics/phi.ll   | 216 ++++++++++++++++++
 .../propagate-backwards-unsupported.ll        | 123 +++++-----
 3 files changed, 366 insertions(+), 66 deletions(-)
 create mode 100644 llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll

diff --git a/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp b/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
index 56d4be513ea6f..c06d08688ab1c 100644
--- a/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
+++ b/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
@@ -30,6 +30,7 @@
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Analysis/VectorUtils.h"
 #include "llvm/IR/CFG.h"
+#include "llvm/IR/Constants.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/DebugInfoMetadata.h"
 #include "llvm/IR/Function.h"
@@ -230,6 +231,7 @@ static bool isUniformShape(Value *V) {
     return true;
 
   switch (I->getOpcode()) {
+  case Instruction::PHI:
   case Instruction::FAdd:
   case Instruction::FSub:
   case Instruction::FMul: // Scalar multiply.
@@ -360,6 +362,33 @@ class LowerMatrixIntrinsics {
         addVector(PoisonValue::get(FixedVectorType::get(
             EltTy, isColumnMajor() ? NumRows : NumColumns)));
     }
+    MatrixTy(ConstantData *Constant, const ShapeInfo &SI)
+        : IsColumnMajor(SI.IsColumnMajor) {
+      Type *EltTy = cast<VectorType>(Constant->getType())->getElementType();
+      Type *RowTy = VectorType::get(EltTy, ElementCount::getFixed(SI.NumRows));
+
+      for (unsigned J = 0, D = SI.getNumVectors(); J < D; ++J) {
+        if (auto *CDV = dyn_cast<ConstantDataVector>(Constant)) {
+          unsigned Width = SI.getStride();
+          size_t EltSize = EltTy->getPrimitiveSizeInBits() / 8;
+          StringRef Data = CDV->getRawDataValues().substr(
+              J * Width * EltSize, Width * EltSize);
+          addVector(ConstantDataVector::getRaw(Data, Width,
+                                               CDV->getElementType()));
+        } else if (isa<PoisonValue>(Constant))
+          addVector(PoisonValue::get(RowTy));
+        else if (isa<UndefValue>(Constant))
+          addVector(UndefValue::get(RowTy));
+        else if (isa<ConstantAggregateZero>(Constant))
+          addVector(ConstantAggregateZero::get(RowTy));
+        else {
+#ifndef NDEBUG
+          Constant->dump();
+          report_fatal_error("unhandled ConstantData type");
+#endif
+        }
+      }
+    }
 
     Value *getVector(unsigned i) const { return Vectors[i]; }
     Value *getColumn(unsigned i) const {
@@ -564,6 +593,27 @@ class LowerMatrixIntrinsics {
       MatrixVal = M.embedInVector(Builder);
     }
 
+    // If it's a PHI, split it now. We'll take care of fixing up the operands
+    // later once we're in VisitPHI.
+    if (auto *PHI = dyn_cast<PHINode>(MatrixVal)) {
+      auto *EltTy = cast<VectorType>(PHI->getType())->getElementType();
+      MatrixTy PhiM{SI.NumRows, SI.NumColumns, EltTy};
+
+      IRBuilder<>::InsertPointGuard IPG(Builder);
+      Builder.SetInsertPoint(PHI);
+      for (unsigned VI = 0, VE = PhiM.getNumVectors(); VI != VE; ++VI)
+        PhiM.setVector(VI, Builder.CreatePHI(PhiM.getVectorTy(),
+                                             PHI->getNumIncomingValues(),
+                                             PHI->getName()));
+
+      Inst2ColumnMatrix[PHI] = PhiM;
+      return PhiM;
+    }
+
+    // If it's a constant, materialize the split version of it with this shape.
+    if (auto *IncomingConst = dyn_cast<ConstantData>(MatrixVal))
+      return MatrixTy(IncomingConst, SI);
+
     // Otherwise split MatrixVal.
     SmallVector<Value *, 16> SplitVecs;
     for (unsigned MaskStart = 0;
@@ -1077,6 +1127,11 @@ class LowerMatrixIntrinsics {
         Changed |= VisitStore(cast<StoreInst>(Inst), Op1, Op2, Builder);
     }
 
+    // Fifth, lower all the PHI's with shape information.
+    for (Instruction *Inst : MatrixInsts)
+      if (auto *PHI = dyn_cast<PHINode>(Inst))
+        Changed |= VisitPHI(PHI);
+
     if (ORE) {
       RemarkGenerator RemarkGen(Inst2ColumnMatrix, *ORE, Func);
       RemarkGen.emitRemarks();
@@ -1349,7 +1404,8 @@ class LowerMatrixIntrinsics {
                         IRBuilder<> &Builder) {
     auto inserted = Inst2ColumnMatrix.insert(std::make_pair(Inst, Matrix));
     (void)inserted;
-    assert(inserted.second && "multiple matrix lowering mapping");
+    assert((inserted.second || isa<PHINode>(Inst)) &&
+           "multiple matrix lowering mapping");
 
     ToRemove.push_back(Inst);
     Value *Flattened = nullptr;
@@ -2133,6 +2189,41 @@ class LowerMatrixIntrinsics {
     return true;
   }
 
+  bool VisitPHI(PHINode *Inst) {
+    auto I = ShapeMap.find(Inst);
+    if (I == ShapeMap.end())
+      return false;
+
+    IRBuilder<> Builder(Inst);
+
+    MatrixTy PhiM = getMatrix(Inst, I->second, Builder);
+
+    for (unsigned IncomingI = 0, IncomingE = Inst->getNumIncomingValues();
+         IncomingI != IncomingE; ++IncomingI) {
+      Value *IncomingV = Inst->getIncomingValue(IncomingI);
+      BasicBlock *IncomingB = Inst->getIncomingBlock(IncomingI);
+
+      // getMatrix() may insert some instructions. The safe place to insert them
+      // is at the end of the parent block, where the register allocator would
+      // have inserted the copies that materialize the PHI.
+      if (auto *IncomingInst = dyn_cast<Instruction>(IncomingV))
+        Builder.SetInsertPoint(IncomingInst->getParent()->getTerminator());
+
+      MatrixTy OpM = getMatrix(IncomingV, I->second, Builder);
+
+      for (unsigned VI = 0, VE = PhiM.getNumVectors(); VI != VE; ++VI) {
+        PHINode *NewPHI = cast<PHINode>(PhiM.getVector(VI));
+        NewPHI->addIncoming(OpM.getVector(VI), IncomingB);
+      }
+    }
+
+    // finalizeLowering() may also insert instructions in some cases. The safe
+    // place for those is at the end of the initial block of PHIs.
+    Builder.SetInsertPoint(*Inst->getInsertionPointAfterDef());
+    finalizeLowering(Inst, PhiM, Builder);
+    return true;
+  }
+
   /// Lower binary operators, if shape information is available.
   bool VisitBinaryOperator(BinaryOperator *Inst) {
     auto I = ShapeMap.find(Inst);
diff --git a/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll b/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll
new file mode 100644
index 0000000000000..d49b4d1112062
--- /dev/null
+++ b/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll
@@ -0,0 +1,216 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt -matrix-allow-contract=false -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s
+
+define void @matrix_phi(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, ptr [[IN1:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN1]], i64 3
+; CHECK-NEXT:    [[COL_LOAD1:%.*]] = load <3 x double>, ptr [[VEC_GEP]], align 8
+; CHECK-NEXT:    [[VEC_GEP2:%.*]] = getelementptr double, ptr [[IN1]], i64 6
+; CHECK-NEXT:    [[COL_LOAD3:%.*]] = load <3 x double>, ptr [[VEC_GEP2]], align 16
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[PHI9:%.*]] = phi <3 x double> [ [[COL_LOAD]], [[ENTRY:%.*]] ], [ [[TMP0:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI10:%.*]] = phi <3 x double> [ [[COL_LOAD1]], [[ENTRY]] ], [ [[TMP1:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI11:%.*]] = phi <3 x double> [ [[COL_LOAD3]], [[ENTRY]] ], [ [[TMP2:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[CTR:%.*]] = phi i32 [ [[COUNT:%.*]], [[ENTRY]] ], [ [[DEC:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[COL_LOAD4:%.*]] = load <3 x double>, ptr [[IN2:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP5:%.*]] = getelementptr double, ptr [[IN2]], i64 3
+; CHECK-NEXT:    [[COL_LOAD6:%.*]] = load <3 x double>, ptr [[VEC_GEP5]], align 8
+; CHECK-NEXT:    [[VEC_GEP7:%.*]] = getelementptr double, ptr [[IN2]], i64 6
+; CHECK-NEXT:    [[COL_LOAD8:%.*]] = load <3 x double>, ptr [[VEC_GEP7]], align 16
+; CHECK-NEXT:    [[TMP0]] = fadd <3 x double> [[PHI9]], [[COL_LOAD4]]
+; CHECK-NEXT:    [[TMP1]] = fadd <3 x double> [[PHI10]], [[COL_LOAD6]]
+; CHECK-NEXT:    [[TMP2]] = fadd <3 x double> [[PHI11]], [[COL_LOAD8]]
+; CHECK-NEXT:    [[DEC]] = sub i32 [[CTR]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[DEC]], 0
+; CHECK-NEXT:    br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
+; CHECK:       exit:
+; CHECK-NEXT:    store <3 x double> [[TMP0]], ptr [[OUT:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP12:%.*]] = getelementptr double, ptr [[OUT]], i64 3
+; CHECK-NEXT:    store <3 x double> [[TMP1]], ptr [[VEC_GEP12]], align 8
+; CHECK-NEXT:    [[VEC_GEP13:%.*]] = getelementptr double, ptr [[OUT]], i64 6
+; CHECK-NEXT:    store <3 x double> [[TMP2]], ptr [[VEC_GEP13]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  %mat = load <9 x double>, ptr %in1
+  br label %loop
+
+loop:
+  %phi = phi <9 x double> [%mat, %entry], [%sum, %loop]
+  %ctr = phi i32 [%count, %entry], [%dec, %loop]
+
+  %in2v = load <9 x double>, ptr %in2
+
+  ; Give in2 the shape: 3 x 3
+  %in2t  = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2v, i32 3, i32 3)
+  %in2tt = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2t, i32 3, i32 3)
+
+  %sum = fadd <9 x double> %phi, %in2tt
+
+  %dec = sub i32 %ctr, 1
+  %cmp = icmp eq i32 %dec, 0
+  br i1 %cmp, label %exit, label %loop
+
+exit:
+  store <9 x double> %sum, ptr %out
+  ret void
+}
+
+define void @matrix_phi_zeroinitializer(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi_zeroinitializer(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[PHI4:%.*]] = phi <3 x double> [ zeroinitializer, [[ENTRY:%.*]] ], [ [[TMP0:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI5:%.*]] = phi <3 x double> [ zeroinitializer, [[ENTRY]] ], [ [[TMP1:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI6:%.*]] = phi <3 x double> [ zeroinitializer, [[ENTRY]] ], [ [[TMP2:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[CTR:%.*]] = phi i32 [ [[COUNT:%.*]], [[ENTRY]] ], [ [[DEC:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, ptr [[IN2:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN2]], i64 3
+; CHECK-NEXT:    [[COL_LOAD1:%.*]] = load <3 x double>, ptr [[VEC_GEP]], align 8
+; CHECK-NEXT:    [[VEC_GEP2:%.*]] = getelementptr double, ptr [[IN2]], i64 6
+; CHECK-NEXT:    [[COL_LOAD3:%.*]] = load <3 x double>, ptr [[VEC_GEP2]], align 16
+; CHECK-NEXT:    [[TMP0]] = fadd <3 x double> [[PHI4]], [[COL_LOAD]]
+; CHECK-NEXT:    [[TMP1]] = fadd <3 x double> [[PHI5]], [[COL_LOAD1]]
+; CHECK-NEXT:    [[TMP2]] = fadd <3 x double> [[PHI6]], [[COL_LOAD3]]
+; CHECK-NEXT:    [[DEC]] = sub i32 [[CTR]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[DEC]], 0
+; CHECK-NEXT:    br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
+; CHECK:       exit:
+; CHECK-NEXT:    store <3 x double> [[TMP0]], ptr [[OUT:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP7:%.*]] = getelementptr double, ptr [[OUT]], i64 3
+; CHECK-NEXT:    store <3 x double> [[TMP1]], ptr [[VEC_GEP7]], align 8
+; CHECK-NEXT:    [[VEC_GEP8:%.*]] = getelementptr double, ptr [[OUT]], i64 6
+; CHECK-NEXT:    store <3 x double> [[TMP2]], ptr [[VEC_GEP8]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %phi = phi <9 x double> [zeroinitializer, %entry], [%sum, %loop]
+  %ctr = phi i32 [%count, %entry], [%dec, %loop]
+
+  %in2v = load <9 x double>, ptr %in2
+
+  ; Give in2 the shape: 3 x 3
+  %in2t  = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2v, i32 3, i32 3)
+  %in2tt = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2t, i32 3, i32 3)
+
+  %sum = fadd <9 x double> %phi, %in2tt
+
+  %dec = sub i32 %ctr, 1
+  %cmp = icmp eq i32 %dec, 0
+  br i1 %cmp, label %exit, label %loop
+
+exit:
+  store <9 x double> %sum, ptr %out
+  ret void
+}
+
+define void @matrix_phi_undef(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi_undef(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[PHI4:%.*]] = phi <3 x double> [ undef, [[ENTRY:%.*]] ], [ [[TMP0:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI5:%.*]] = phi <3 x double> [ undef, [[ENTRY]] ], [ [[TMP1:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI6:%.*]] = phi <3 x double> [ undef, [[ENTRY]] ], [ [[TMP2:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[CTR:%.*]] = phi i32 [ [[COUNT:%.*]], [[ENTRY]] ], [ [[DEC:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, ptr [[IN2:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN2]], i64 3
+; CHECK-NEXT:    [[COL_LOAD1:%.*]] = load <3 x double>, ptr [[VEC_GEP]], align 8
+; CHECK-NEXT:    [[VEC_GEP2:%.*]] = getelementptr double, ptr [[IN2]], i64 6
+; CHECK-NEXT:    [[COL_LOAD3:%.*]] = load <3 x double>, ptr [[VEC_GEP2]], align 16
+; CHECK-NEXT:    [[TMP0]] = fadd <3 x double> [[PHI4]], [[COL_LOAD]]
+; CHECK-NEXT:    [[TMP1]] = fadd <3 x double> [[PHI5]], [[COL_LOAD1]]
+; CHECK-NEXT:    [[TMP2]] = fadd <3 x double> [[PHI6]], [[COL_LOAD3]]
+; CHECK-NEXT:    [[DEC]] = sub i32 [[CTR]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[DEC]], 0
+; CHECK-NEXT:    br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
+; CHECK:       exit:
+; CHECK-NEXT:    store <3 x double> [[TMP0]], ptr [[OUT:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP7:%.*]] = getelementptr double, ptr [[OUT]], i64 3
+; CHECK-NEXT:    store <3 x double> [[TMP1]], ptr [[VEC_GEP7]], align 8
+; CHECK-NEXT:    [[VEC_GEP8:%.*]] = getelementptr double, ptr [[OUT]], i64 6
+; CHECK-NEXT:    store <3 x double> [[TMP2]], ptr [[VEC_GEP8]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %phi = phi <9 x double> [undef, %entry], [%sum, %loop]
+  %ctr = phi i32 [%count, %entry], [%dec, %loop]
+
+  %in2v = load <9 x double>, ptr %in2
+
+  ; Give in2 the shape: 3 x 3
+  %in2t  = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2v, i32 3, i32 3)
+  %in2tt = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2t, i32 3, i32 3)
+
+  %sum = fadd <9 x double> %phi, %in2tt
+
+  %dec = sub i32 %ctr, 1
+  %cmp = icmp eq i32 %dec, 0
+  br i1 %cmp, label %exit, label %loop
+
+exit:
+  store <9 x double> %sum, ptr %out
+  ret void
+}
+
+define void @matrix_phi_poison(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi_poison(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[PHI4:%.*]] = phi <3 x double> [ poison, [[ENTRY:%.*]] ], [ [[TMP0:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI5:%.*]] = phi <3 x double> [ poison, [[ENTRY]] ], [ [[TMP1:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI6:%.*]] = phi <3 x double> [ poison, [[ENTRY]] ], [ [[TMP2:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[CTR:%.*]] = phi i32 [ [[COUNT:%.*]], [[ENTRY]] ], [ [[DEC:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, ptr [[IN2:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN2]], i64 3
+; CHECK-NEXT:    [[COL_LOAD1:%.*]] = load <3 x double>, ptr [[VEC_GEP]], align 8
+; CHECK-NEXT:    [[VEC_GEP2:%.*]] = getelementptr double, ptr [[IN2]], i64 6
+; CHECK-NEXT:    [[COL_LOAD3:%.*]] = load <3 x double>, ptr [[VEC_GEP2]], align 16
+; CHECK-NEXT:    [[TMP0]] = fadd <3 x double> [[PHI4]], [[COL_LOAD]]
+; CHECK-NEXT:    [[TMP1]] = fadd <3 x double> [[PHI5]], [[COL_LOAD1]]
+; CHECK-NEXT:    [[TMP2]] = fadd <3 x double> [[PHI6]], [[COL_LOAD3]]
+; CHECK-NEXT:    [[DEC]] = sub i32 [[CTR]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[DEC]], 0
+; CHECK-NEXT:    br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
+; CHECK:       exit:
+; CHECK-NEXT:    store <3 x double> [[TMP0]], ptr [[OUT:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP7:%.*]] = getelementptr double, ptr [[OUT]], i64 3
+; CHECK-NEXT:    store <3 x double> [[TMP1]], ptr [[VEC_GEP7]], align 8
+; CHECK-NEXT:    [[VEC_GEP8:%.*]] = getelementptr double, ptr [[OUT]], i64 6
+; CHECK-NEXT:    store <3 x double> [[TMP2]], ptr [[VEC_GEP8]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %phi = phi <9 x double> [poison, %entry], [%sum, %loop]
+  %ctr = phi i32 [%count, %entry], [%dec, %loop]
+
+  %in2v = load <9 x double>, ptr %in2
+
+  ; Give in2 the shape: 3 x 3
+  %in2t  = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2v, i32 3, i32 3)
+  %in2tt = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2t, i32 3, i32 3)
+
+  %sum = fadd <9 x double> %phi, %in2tt
+
+  %dec = sub i32 %ctr, 1
+  %cmp = icmp eq i32 %dec, 0
+  br i1 %cmp, label %exit, label %loop
+
+exit:
+  store <9 x double> %sum, ptr %out
+  ret void
+}
diff --git a/llvm/test/Transforms/LowerMatrixIntrinsics/propagate-backwards-unsupported.ll b/llvm/test/Transforms/LowerMatrixIntrinsics/propagate-backwards-unsupported.ll
index 2af2c979f2065..6ed8e46d62892 100644
--- a/llvm/test/Transforms/LowerMatrixIntrinsics/propagate-backwards-unsupported.ll
+++ b/llvm/test/Transforms/LowerMatrixIntrinsics/propagate-backwards-unsupported.ll
@@ -28,9 +28,6 @@ define <9 x double> @unsupported_phi(i1 %cond, <9 x double> %A, <9 x double> %B,
 ; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <3 x double> [[TMP13]], double [[TMP14]], i64 1
 ; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 2
 ; CHECK-NEXT:    [[TMP17:%.*]] = insertelement <3 x double> [[TMP15]], double [[TMP16]], i64 2
-; CHECK-NEXT:    [[TMP18:%.*]] = shufflevector <3 x double> [[TMP5]], <3 x double> [[TMP11]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
-; CHECK-NEXT:    [[TMP19:%.*]] = shufflevector <3 x double> [[TMP17]], <3 x double> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP20:%.*]] = shufflevector <6 x double> [[TMP18]], <6 x double> [[TMP19]], <9 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
 ; CHECK-NEXT:    br label [[IF_END:%.*]]
 ; CHECK:       if.else:
 ; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <9 x double> [[B:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
@@ -54,183 +51,179 @@ define <9 x double> @unsupported_phi(i1 %cond, <9 x double> %A, <9 x double> %B,
 ; CHECK-NEXT:    [[TMP36:%.*]] = insertelement <3 x double> [[TMP34]], double [[TMP35]], i64 1
 ; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <3 x double> [[SPLIT2]], i64 2
 ; CHECK-NEXT:    [[TMP38:%.*]] = insertelement <3 x double> [[TMP36]], double [[TMP37]], i64 2
-; CHECK-NEXT:    [[TMP39:%.*]] = shufflevector <3 x double> [[TMP26]], <3 x double> [[TMP32]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
-; CHECK-NEXT:    [[TMP40:%.*]] = shufflevector <3 x double> [[TMP38]], <3 x double> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP41:%.*]] = shufflevector <6 x double> [[TMP39]], <6 x double> [[TMP40]], <9 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
 ; CHECK-NEXT:    br label [[IF_END]]
 ; CHECK:       if.end:
-; CHECK-NEXT:    [[MERGE:%.*]] = phi <9 x double> [ [[TMP20]], [[IF_THEN]] ], [ [[TMP41]], [[IF_ELSE]] ]
-; CHECK-NEXT:    [[SPLIT6:%.*]] = shufflevector <9 x double> [[C:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
-; CHECK-NEXT:    [[SPLIT7:%.*]] = shufflevector <9 x double> [[C]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
-; CHECK-NEXT:    [[SPLIT8:%.*]] = shufflevector <9 x double> [[C]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
-; CHECK-NEXT:    [[SPLIT9:%.*]] = shufflevector <9 x double> [[MERGE]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; CHECK-NEXT:    [[MERGE9:%.*]] = phi <3 x double> [ [[TMP5]], [[IF_THEN]] ], [ [[TMP26]], [[IF_ELSE]] ]
+; CHECK-NEXT:    [[MERGE10:%.*]] = phi <3 x double> [ [[TMP11]], [[IF_THEN]] ], [ [[TMP32]], [[IF_ELSE]] ]
+; CHECK-NEXT:    [[MERGE11:%.*]] = phi <3 x double> [ [[TMP17]], [[IF_THEN]] ], [ [[TMP38]], [[IF_ELSE]] ]
+; CHECK-NEXT:    [[SPLIT9:%.*]] = shufflevector <9 x double> [[MERGE:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
 ; CHECK-NEXT:    [[SPLIT10:%.*]] = shufflevector <9 x double> [[MERGE]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
 ; CHECK-NEXT:    [[SPLIT11:%.*]] = shufflevector <9 x double> [[MERGE]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
-; CHECK-NEXT:    [[BLOCK:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 0
+; CHECK-NEXT:    [[BLOCK:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x double> poison, double [[TMP42]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP43:%.*]] = fmul <1 x double> [[BLOCK]], [[SPLAT_SPLAT]]
-; CHECK-NEXT:    [[BLOCK12:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 1
+; CHECK-NEXT:    [[BLOCK12:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT13:%.*]] = insertelement <1 x double> poison, double [[TMP44]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT14:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT13]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP45:%.*]] = fmul <1 x double> [[BLOCK12]], [[SPLAT_SPLAT14]]
 ; CHECK-NEXT:    [[TMP46:%.*]] = fadd <1 x double> [[TMP43]], [[TMP45]]
-; CHECK-NEXT:    [[BLOCK15:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 2
+; CHECK-NEXT:    [[BLOCK15:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT16:%.*]] = insertelement <1 x double> poison, double [[TMP47]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT17:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT16]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP48:%.*]] = fmul <1 x double> [[BLOCK15]], [[SPLAT_SPLAT17]]
 ; CHECK-NEXT:    [[TMP49:%.*]] = fadd <1 x double> [[TMP46]], [[TMP48]]
 ; CHECK-NEXT:    [[TMP50:%.*]] = shufflevector <1 x double> [[TMP49]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
 ; CHECK-NEXT:    [[TMP51:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP50]], <3 x i32> <i32 3, i32 1, i32 2>
-; CHECK-NEXT:    [[BLOCK18:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 0
+; CHECK-NEXT:    [[BLOCK18:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT19:%.*]] = insertelement <1 x double> poison, double [[TMP52]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT20:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT19]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP53:%.*]] = fmul <1 x double> [[BLOCK18]], [[SPLAT_SPLAT20]]
-; CHECK-NEXT:    [[BLOCK21:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP54:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 1
+; CHECK-NEXT:    [[BLOCK21:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP54:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT22:%.*]] = insertelement <1 x double> poison, double [[TMP54]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT23:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT22]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP55:%.*]] = fmul <1 x double> [[BLOCK21]], [[SPLAT_SPLAT23]]
 ; CHECK-NEXT:    [[TMP56:%.*]] = fadd <1 x double> [[TMP53]], [[TMP55]]
-; CHECK-NEXT:    [[BLOCK24:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP57:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 2
+; CHECK-NEXT:    [[BLOCK24:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP57:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT25:%.*]] = insertelement <1 x double> poison, double [[TMP57]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT26:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT25]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP58:%.*]] = fmul <1 x double> [[BLOCK24]], [[SPLAT_SPLAT26]]
 ; 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 poison, i32 poison>
 ; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <3 x double> [[TMP51]], <3 x double> [[TMP60]], <3 x i32> <i32 0, i32 3, i32 2>
-; CHECK-NEXT:    [[BLOCK27:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 0
+; CHECK-NEXT:    [[BLOCK27:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT28:%.*]] = insertelement <1 x double> poison, double [[TMP62]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT29:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT28]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP63:%.*]] = fmul <1 x double> [[BLOCK27]], [[SPLAT_SPLAT29]]
-; CHECK-NEXT:    [[BLOCK30:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 1
+; CHECK-NEXT:    [[BLOCK30:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT31:%.*]] = insertelement <1 x double> poison, double [[TMP64]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT32:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT31]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP65:%.*]] = fmul <1 x double> [[BLOCK30]], [[SPLAT_SPLAT32]]
 ; CHECK-NEXT:    [[TMP66:%.*]] = fadd <1 x double> [[TMP63]], [[TMP65]]
-; CHECK-NEXT:    [[BLOCK33:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <3 x double> [[SPLIT9]], i64 2
+; CHECK-NEXT:    [[BLOCK33:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT34:%.*]] = insertelement <1 x double> poison, double [[TMP67]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT35:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT34]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP68:%.*]] = fmul <1 x double> [[BLOCK33]], [[SPLAT_SPLAT35]]
 ; CHECK-NEXT:    [[TMP69:%.*]] = fadd <1 x double> [[TMP66]], [[TMP68]]
 ; CHECK-NEXT:    [[TMP70:%.*]] = shufflevector <1 x double> [[TMP69]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
 ; CHECK-NEXT:    [[TMP71:%.*]] = shufflevector <3 x double> [[TMP61]], <3 x double> [[TMP70]], <3 x i32> <i32 0, i32 1, i32 3>
-; CHECK-NEXT:    [[BLOCK36:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP72:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 0
+; CHECK-NEXT:    [[BLOCK36:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP72:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT37:%.*]] = insertelement <1 x double> poison, double [[TMP72]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT38:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT37]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP73:%.*]] = fmul <1 x double> [[BLOCK36]], [[SPLAT_SPLAT38]]
-; CHECK-NEXT:    [[BLOCK39:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 1
+; CHECK-NEXT:    [[BLOCK39:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT40:%.*]] = insertelement <1 x double> poison, double [[TMP74]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT41:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT40]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP75:%.*]] = fmul <1 x double> [[BLOCK39]], [[SPLAT_SPLAT41]]
 ; CHECK-NEXT:    [[TMP76:%.*]] = fadd <1 x double> [[TMP73]], [[TMP75]]
-; CHECK-NEXT:    [[BLOCK42:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 2
+; CHECK-NEXT:    [[BLOCK42:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT43:%.*]] = insertelement <1 x double> poison, double [[TMP77]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT44:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT43]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP78:%.*]] = fmul <1 x double> [[BLOCK42]], [[SPLAT_SPLAT44]]
 ; CHECK-NEXT:    [[TMP79:%.*]] = fadd <1 x double> [[TMP76]], [[TMP78]]
 ; CHECK-NEXT:    [[TMP80:%.*]] = shufflevector <1 x double> [[TMP79]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
 ; CHECK-NEXT:    [[TMP81:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP80]], <3 x i32> <i32 3, i32 1, i32 2>
-; CHECK-NEXT:    [[BLOCK45:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 0
+; CHECK-NEXT:    [[BLOCK45:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT46:%.*]] = insertelement <1 x double> poison, double [[TMP82]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT47:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT46]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP83:%.*]] = fmul <1 x double> [[BLOCK45]], [[SPLAT_SPLAT47]]
-; CHECK-NEXT:    [[BLOCK48:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP84:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 1
+; CHECK-NEXT:    [[BLOCK48:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP84:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT49:%.*]] = insertelement <1 x double> poison, double [[TMP84]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT50:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT49]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP85:%.*]] = fmul <1 x double> [[BLOCK48]], [[SPLAT_SPLAT50]]
 ; CHECK-NEXT:    [[TMP86:%.*]] = fadd <1 x double> [[TMP83]], [[TMP85]]
-; CHECK-NEXT:    [[BLOCK51:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP87:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 2
+; CHECK-NEXT:    [[BLOCK51:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP87:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT52:%.*]] = insertelement <1 x double> poison, double [[TMP87]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT53:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT52]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP88:%.*]] = fmul <1 x double> [[BLOCK51]], [[SPLAT_SPLAT53]]
 ; CHECK-NEXT:    [[TMP89:%.*]] = fadd <1 x double> [[TMP86]], [[TMP88]]
 ; CHECK-NEXT:    [[TMP90:%.*]] = shufflevector <1 x double> [[TMP89]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
 ; CHECK-NEXT:    [[TMP91:%.*]] = shufflevector <3 x double> [[TMP81]], <3 x double> [[TMP90]], <3 x i32> <i32 0, i32 3, i32 2>
-; CHECK-NEXT:    [[BLOCK54:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 0
+; CHECK-NEXT:    [[BLOCK54:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT55:%.*]] = insertelement <1 x double> poison, double [[TMP92]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT56:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT55]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP93:%.*]] = fmul <1 x double> [[BLOCK54]], [[SPLAT_SPLAT56]]
-; CHECK-NEXT:    [[BLOCK57:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 1
+; CHECK-NEXT:    [[BLOCK57:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT58:%.*]] = insertelement <1 x double> poison, double [[TMP94]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT59:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT58]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP95:%.*]] = fmul <1 x double> [[BLOCK57]], [[SPLAT_SPLAT59]]
 ; CHECK-NEXT:    [[TMP96:%.*]] = fadd <1 x double> [[TMP93]], [[TMP95]]
-; CHECK-NEXT:    [[BLOCK60:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP97:%.*]] = extractelement <3 x double> [[SPLIT10]], i64 2
+; CHECK-NEXT:    [[BLOCK60:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP97:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT61:%.*]] = insertelement <1 x double> poison, double [[TMP97]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT62:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT61]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP98:%.*]] = fmul <1 x double> [[BLOCK60]], [[SPLAT_SPLAT62]]
 ; CHECK-NEXT:    [[TMP99:%.*]] = fadd <1 x double> [[TMP96]], [[TMP98]]
 ; CHECK-NEXT:    [[TMP100:%.*]] = shufflevector <1 x double> [[TMP99]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
 ; CHECK-NEXT:    [[TMP101:%.*]] = shufflevector <3 x double> [[TMP91]], <3 x double> [[TMP100]], <3 x i32> <i32 0, i32 1, i32 3>
-; CHECK-NEXT:    [[BLOCK63:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP102:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 0
+; CHECK-NEXT:    [[BLOCK63:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP102:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT64:%.*]] = insertelement <1 x double> poison, double [[TMP102]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT65:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT64]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP103:%.*]] = fmul <1 x double> [[BLOCK63]], [[SPLAT_SPLAT65]]
-; CHECK-NEXT:    [[BLOCK66:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP104:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 1
+; CHECK-NEXT:    [[BLOCK66:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP104:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT67:%.*]] = insertelement <1 x double> poison, double [[TMP104]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT68:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT67]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP105:%.*]] = fmul <1 x double> [[BLOCK66]], [[SPLAT_SPLAT68]]
 ; CHECK-NEXT:    [[TMP106:%.*]] = fadd <1 x double> [[TMP103]], [[TMP105]]
-; CHECK-NEXT:    [[BLOCK69:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP107:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 2
+; CHECK-NEXT:    [[BLOCK69:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP107:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT70:%.*]] = insertelement <1 x double> poison, double [[TMP107]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT71:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT70]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP108:%.*]] = fmul <1 x double> [[BLOCK69]], [[SPLAT_SPLAT71]]
 ; CHECK-NEXT:    [[TMP109:%.*]] = fadd <1 x double> [[TMP106]], [[TMP108]]
 ; CHECK-NEXT:    [[TMP110:%.*]] = shufflevector <1 x double> [[TMP109]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
 ; CHECK-NEXT:    [[TMP111:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP110]], <3 x i32> <i32 3, i32 1, i32 2>
-; CHECK-NEXT:    [[BLOCK72:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP112:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 0
+; CHECK-NEXT:    [[BLOCK72:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP112:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT73:%.*]] = insertelement <1 x double> poison, double [[TMP112]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT74:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT73]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP113:%.*]] = fmul <1 x double> [[BLOCK72]], [[SPLAT_SPLAT74]]
-; CHECK-NEXT:    [[BLOCK75:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP114:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 1
+; CHECK-NEXT:    [[BLOCK75:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP114:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT76:%.*]] = insertelement <1 x double> poison, double [[TMP114]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT77:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT76]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP115:%.*]] = fmul <1 x double> [[BLOCK75]], [[SPLAT_SPLAT77]]
 ; CHECK-NEXT:    [[TMP116:%.*]] = fadd <1 x double> [[TMP113]], [[TMP115]]
-; CHECK-NEXT:    [[BLOCK78:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP117:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 2
+; CHECK-NEXT:    [[BLOCK78:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP117:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT79:%.*]] = insertelement <1 x double> poison, double [[TMP117]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT80:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT79]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP118:%.*]] = fmul <1 x double> [[BLOCK78]], [[SPLAT_SPLAT80]]
 ; CHECK-NEXT:    [[TMP119:%.*]] = fadd <1 x double> [[TMP116]], [[TMP118]]
 ; CHECK-NEXT:    [[TMP120:%.*]] = shufflevector <1 x double> [[TMP119]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
 ; CHECK-NEXT:    [[TMP121:%.*]] = shufflevector <3 x double> [[TMP111]], <3 x double> [[TMP120]], <3 x i32> <i32 0, i32 3, i32 2>
-; CHECK-NEXT:    [[BLOCK81:%.*]] = shufflevector <3 x double> [[SPLIT6]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP122:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 0
+; CHECK-NEXT:    [[BLOCK81:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP122:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT82:%.*]] = insertelement <1 x double> poison, double [[TMP122]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT83:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT82]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP123:%.*]] = fmul <1 x double> [[BLOCK81]], [[SPLAT_SPLAT83]]
-; CHECK-NEXT:    [[BLOCK84:%.*]] = shufflevector <3 x double> [[SPLIT7]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP124:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 1
+; CHECK-NEXT:    [[BLOCK84:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP124:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT85:%.*]] = insertelement <1 x double> poison, double [[TMP124]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT86:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT85]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP125:%.*]] = fmul <1 x double> [[BLOCK84]], [[SPLAT_SPLAT86]]
 ; CHECK-NEXT:    [[TMP126:%.*]] = fadd <1 x double> [[TMP123]], [[TMP125]]
-; CHECK-NEXT:    [[BLOCK87:%.*]] = shufflevector <3 x double> [[SPLIT8]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP127:%.*]] = extractelement <3 x double> [[SPLIT11]], i64 2
+; CHECK-NEXT:    [[BLOCK87:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP127:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
 ; CHECK-NEXT:    [[SPLAT_SPLATINSERT88:%.*]] = insertelement <1 x double> poison, double [[TMP127]], i64 0
 ; CHECK-NEXT:    [[SPLAT_SPLAT89:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT88]], <1 x double> poison, <1 x i32> zeroinitializer
 ; CHECK-NEXT:    [[TMP128:%.*]] = fmul <1 x double> [[BLOCK87]], [[SPLAT_SPLAT89]]

>From 71b99d3b4da5f0f655dc2f3c46cc7d42214517a1 Mon Sep 17 00:00:00 2001
From: Jon Roelofs <jonathan_roelofs at apple.com>
Date: Tue, 27 May 2025 15:40:31 -0700
Subject: [PATCH 2/5] move formerly unsupported test to new home

---
 .../Transforms/LowerMatrixIntrinsics/phi.ll   | 266 +++++++++++++++++-
 .../propagate-backwards-unsupported.ll        | 254 -----------------
 2 files changed, 258 insertions(+), 262 deletions(-)

diff --git a/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll b/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll
index d49b4d1112062..e2720b86d87ca 100644
--- a/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll
+++ b/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll
@@ -1,8 +1,8 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -matrix-allow-contract=false -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s
 
-define void @matrix_phi(ptr %in1, ptr %in2, i32 %count, ptr %out) {
-; CHECK-LABEL: @matrix_phi(
+define void @matrix_phi_loop(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi_loop(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, ptr [[IN1:%.*]], align 128
 ; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN1]], i64 3
@@ -59,8 +59,8 @@ exit:
   ret void
 }
 
-define void @matrix_phi_zeroinitializer(ptr %in1, ptr %in2, i32 %count, ptr %out) {
-; CHECK-LABEL: @matrix_phi_zeroinitializer(
+define void @matrix_phi_loop_zeroinitializer(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi_loop_zeroinitializer(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
@@ -111,8 +111,8 @@ exit:
   ret void
 }
 
-define void @matrix_phi_undef(ptr %in1, ptr %in2, i32 %count, ptr %out) {
-; CHECK-LABEL: @matrix_phi_undef(
+define void @matrix_phi_loop_undef(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi_loop_undef(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
@@ -163,8 +163,8 @@ exit:
   ret void
 }
 
-define void @matrix_phi_poison(ptr %in1, ptr %in2, i32 %count, ptr %out) {
-; CHECK-LABEL: @matrix_phi_poison(
+define void @matrix_phi_loop_poison(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi_loop_poison(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    br label [[LOOP:%.*]]
 ; CHECK:       loop:
@@ -214,3 +214,253 @@ exit:
   store <9 x double> %sum, ptr %out
   ret void
 }
+
+define <9 x double> @matrix_phi_ifthenelse(i1 %cond, <9 x double> %A, <9 x double> %B, <9 x double> %C) {
+; CHECK-LABEL: @matrix_phi_ifthenelse(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br i1 [[COND:%.*]], label [[IF_THEN:%.*]], label [[IF_ELSE:%.*]]
+; CHECK:       if.then:
+; CHECK-NEXT:    [[SPLIT3:%.*]] = shufflevector <9 x double> [[A:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; CHECK-NEXT:    [[SPLIT4:%.*]] = shufflevector <9 x double> [[A]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[SPLIT5:%.*]] = shufflevector <9 x double> [[A]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
+; CHECK-NEXT:    [[TMP0:%.*]] = extractelement <3 x double> [[SPLIT3]], i64 0
+; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <3 x double> poison, double [[TMP0]], i64 0
+; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 0
+; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <3 x double> [[TMP1]], double [[TMP2]], i64 1
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 0
+; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <3 x double> [[TMP3]], double [[TMP4]], i64 2
+; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <3 x double> [[SPLIT3]], i64 1
+; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <3 x double> poison, double [[TMP6]], i64 0
+; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 1
+; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <3 x double> [[TMP7]], double [[TMP8]], i64 1
+; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 1
+; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <3 x double> [[TMP9]], double [[TMP10]], i64 2
+; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <3 x double> [[SPLIT3]], i64 2
+; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <3 x double> poison, double [[TMP12]], i64 0
+; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 2
+; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <3 x double> [[TMP13]], double [[TMP14]], i64 1
+; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 2
+; CHECK-NEXT:    [[TMP17:%.*]] = insertelement <3 x double> [[TMP15]], double [[TMP16]], i64 2
+; CHECK-NEXT:    br label [[IF_END:%.*]]
+; CHECK:       if.else:
+; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <9 x double> [[B:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; CHECK-NEXT:    [[SPLIT1:%.*]] = shufflevector <9 x double> [[B]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[SPLIT2:%.*]] = shufflevector <9 x double> [[B]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
+; CHECK-NEXT:    [[TMP21:%.*]] = extractelement <3 x double> [[SPLIT]], i64 0
+; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <3 x double> poison, double [[TMP21]], i64 0
+; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <3 x double> [[SPLIT1]], i64 0
+; CHECK-NEXT:    [[TMP24:%.*]] = insertelement <3 x double> [[TMP22]], double [[TMP23]], i64 1
+; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <3 x double> [[SPLIT2]], i64 0
+; CHECK-NEXT:    [[TMP26:%.*]] = insertelement <3 x double> [[TMP24]], double [[TMP25]], i64 2
+; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <3 x double> [[SPLIT]], i64 1
+; CHECK-NEXT:    [[TMP28:%.*]] = insertelement <3 x double> poison, double [[TMP27]], i64 0
+; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <3 x double> [[SPLIT1]], i64 1
+; CHECK-NEXT:    [[TMP30:%.*]] = insertelement <3 x double> [[TMP28]], double [[TMP29]], i64 1
+; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <3 x double> [[SPLIT2]], i64 1
+; CHECK-NEXT:    [[TMP32:%.*]] = insertelement <3 x double> [[TMP30]], double [[TMP31]], i64 2
+; CHECK-NEXT:    [[TMP33:%.*]] = extractelement <3 x double> [[SPLIT]], i64 2
+; CHECK-NEXT:    [[TMP34:%.*]] = insertelement <3 x double> poison, double [[TMP33]], i64 0
+; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <3 x double> [[SPLIT1]], i64 2
+; CHECK-NEXT:    [[TMP36:%.*]] = insertelement <3 x double> [[TMP34]], double [[TMP35]], i64 1
+; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <3 x double> [[SPLIT2]], i64 2
+; CHECK-NEXT:    [[TMP38:%.*]] = insertelement <3 x double> [[TMP36]], double [[TMP37]], i64 2
+; CHECK-NEXT:    br label [[IF_END]]
+; CHECK:       if.end:
+; CHECK-NEXT:    [[MERGE9:%.*]] = phi <3 x double> [ [[TMP5]], [[IF_THEN]] ], [ [[TMP26]], [[IF_ELSE]] ]
+; CHECK-NEXT:    [[MERGE10:%.*]] = phi <3 x double> [ [[TMP11]], [[IF_THEN]] ], [ [[TMP32]], [[IF_ELSE]] ]
+; CHECK-NEXT:    [[MERGE11:%.*]] = phi <3 x double> [ [[TMP17]], [[IF_THEN]] ], [ [[TMP38]], [[IF_ELSE]] ]
+; CHECK-NEXT:    [[SPLIT9:%.*]] = shufflevector <9 x double> [[MERGE:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
+; CHECK-NEXT:    [[SPLIT10:%.*]] = shufflevector <9 x double> [[MERGE]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[SPLIT11:%.*]] = shufflevector <9 x double> [[MERGE]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
+; CHECK-NEXT:    [[BLOCK:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x double> poison, double [[TMP42]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP43:%.*]] = fmul <1 x double> [[BLOCK]], [[SPLAT_SPLAT]]
+; CHECK-NEXT:    [[BLOCK12:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT13:%.*]] = insertelement <1 x double> poison, double [[TMP44]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT14:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT13]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP45:%.*]] = fmul <1 x double> [[BLOCK12]], [[SPLAT_SPLAT14]]
+; CHECK-NEXT:    [[TMP46:%.*]] = fadd <1 x double> [[TMP43]], [[TMP45]]
+; CHECK-NEXT:    [[BLOCK15:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT16:%.*]] = insertelement <1 x double> poison, double [[TMP47]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT17:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT16]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP48:%.*]] = fmul <1 x double> [[BLOCK15]], [[SPLAT_SPLAT17]]
+; CHECK-NEXT:    [[TMP49:%.*]] = fadd <1 x double> [[TMP46]], [[TMP48]]
+; CHECK-NEXT:    [[TMP50:%.*]] = shufflevector <1 x double> [[TMP49]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP51:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP50]], <3 x i32> <i32 3, i32 1, i32 2>
+; CHECK-NEXT:    [[BLOCK18:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT19:%.*]] = insertelement <1 x double> poison, double [[TMP52]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT20:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT19]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP53:%.*]] = fmul <1 x double> [[BLOCK18]], [[SPLAT_SPLAT20]]
+; CHECK-NEXT:    [[BLOCK21:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP54:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT22:%.*]] = insertelement <1 x double> poison, double [[TMP54]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT23:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT22]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP55:%.*]] = fmul <1 x double> [[BLOCK21]], [[SPLAT_SPLAT23]]
+; CHECK-NEXT:    [[TMP56:%.*]] = fadd <1 x double> [[TMP53]], [[TMP55]]
+; CHECK-NEXT:    [[BLOCK24:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP57:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT25:%.*]] = insertelement <1 x double> poison, double [[TMP57]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT26:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT25]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP58:%.*]] = fmul <1 x double> [[BLOCK24]], [[SPLAT_SPLAT26]]
+; 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 poison, i32 poison>
+; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <3 x double> [[TMP51]], <3 x double> [[TMP60]], <3 x i32> <i32 0, i32 3, i32 2>
+; CHECK-NEXT:    [[BLOCK27:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT28:%.*]] = insertelement <1 x double> poison, double [[TMP62]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT29:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT28]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP63:%.*]] = fmul <1 x double> [[BLOCK27]], [[SPLAT_SPLAT29]]
+; CHECK-NEXT:    [[BLOCK30:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT31:%.*]] = insertelement <1 x double> poison, double [[TMP64]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT32:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT31]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP65:%.*]] = fmul <1 x double> [[BLOCK30]], [[SPLAT_SPLAT32]]
+; CHECK-NEXT:    [[TMP66:%.*]] = fadd <1 x double> [[TMP63]], [[TMP65]]
+; CHECK-NEXT:    [[BLOCK33:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT34:%.*]] = insertelement <1 x double> poison, double [[TMP67]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT35:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT34]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP68:%.*]] = fmul <1 x double> [[BLOCK33]], [[SPLAT_SPLAT35]]
+; CHECK-NEXT:    [[TMP69:%.*]] = fadd <1 x double> [[TMP66]], [[TMP68]]
+; CHECK-NEXT:    [[TMP70:%.*]] = shufflevector <1 x double> [[TMP69]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP71:%.*]] = shufflevector <3 x double> [[TMP61]], <3 x double> [[TMP70]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[BLOCK36:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP72:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT37:%.*]] = insertelement <1 x double> poison, double [[TMP72]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT38:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT37]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP73:%.*]] = fmul <1 x double> [[BLOCK36]], [[SPLAT_SPLAT38]]
+; CHECK-NEXT:    [[BLOCK39:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT40:%.*]] = insertelement <1 x double> poison, double [[TMP74]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT41:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT40]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP75:%.*]] = fmul <1 x double> [[BLOCK39]], [[SPLAT_SPLAT41]]
+; CHECK-NEXT:    [[TMP76:%.*]] = fadd <1 x double> [[TMP73]], [[TMP75]]
+; CHECK-NEXT:    [[BLOCK42:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT43:%.*]] = insertelement <1 x double> poison, double [[TMP77]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT44:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT43]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP78:%.*]] = fmul <1 x double> [[BLOCK42]], [[SPLAT_SPLAT44]]
+; CHECK-NEXT:    [[TMP79:%.*]] = fadd <1 x double> [[TMP76]], [[TMP78]]
+; CHECK-NEXT:    [[TMP80:%.*]] = shufflevector <1 x double> [[TMP79]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP81:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP80]], <3 x i32> <i32 3, i32 1, i32 2>
+; CHECK-NEXT:    [[BLOCK45:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT46:%.*]] = insertelement <1 x double> poison, double [[TMP82]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT47:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT46]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP83:%.*]] = fmul <1 x double> [[BLOCK45]], [[SPLAT_SPLAT47]]
+; CHECK-NEXT:    [[BLOCK48:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP84:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT49:%.*]] = insertelement <1 x double> poison, double [[TMP84]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT50:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT49]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP85:%.*]] = fmul <1 x double> [[BLOCK48]], [[SPLAT_SPLAT50]]
+; CHECK-NEXT:    [[TMP86:%.*]] = fadd <1 x double> [[TMP83]], [[TMP85]]
+; CHECK-NEXT:    [[BLOCK51:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP87:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT52:%.*]] = insertelement <1 x double> poison, double [[TMP87]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT53:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT52]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP88:%.*]] = fmul <1 x double> [[BLOCK51]], [[SPLAT_SPLAT53]]
+; CHECK-NEXT:    [[TMP89:%.*]] = fadd <1 x double> [[TMP86]], [[TMP88]]
+; CHECK-NEXT:    [[TMP90:%.*]] = shufflevector <1 x double> [[TMP89]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP91:%.*]] = shufflevector <3 x double> [[TMP81]], <3 x double> [[TMP90]], <3 x i32> <i32 0, i32 3, i32 2>
+; CHECK-NEXT:    [[BLOCK54:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT55:%.*]] = insertelement <1 x double> poison, double [[TMP92]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT56:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT55]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP93:%.*]] = fmul <1 x double> [[BLOCK54]], [[SPLAT_SPLAT56]]
+; CHECK-NEXT:    [[BLOCK57:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT58:%.*]] = insertelement <1 x double> poison, double [[TMP94]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT59:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT58]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP95:%.*]] = fmul <1 x double> [[BLOCK57]], [[SPLAT_SPLAT59]]
+; CHECK-NEXT:    [[TMP96:%.*]] = fadd <1 x double> [[TMP93]], [[TMP95]]
+; CHECK-NEXT:    [[BLOCK60:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP97:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT61:%.*]] = insertelement <1 x double> poison, double [[TMP97]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT62:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT61]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP98:%.*]] = fmul <1 x double> [[BLOCK60]], [[SPLAT_SPLAT62]]
+; CHECK-NEXT:    [[TMP99:%.*]] = fadd <1 x double> [[TMP96]], [[TMP98]]
+; CHECK-NEXT:    [[TMP100:%.*]] = shufflevector <1 x double> [[TMP99]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP101:%.*]] = shufflevector <3 x double> [[TMP91]], <3 x double> [[TMP100]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[BLOCK63:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP102:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT64:%.*]] = insertelement <1 x double> poison, double [[TMP102]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT65:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT64]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP103:%.*]] = fmul <1 x double> [[BLOCK63]], [[SPLAT_SPLAT65]]
+; CHECK-NEXT:    [[BLOCK66:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP104:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT67:%.*]] = insertelement <1 x double> poison, double [[TMP104]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT68:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT67]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP105:%.*]] = fmul <1 x double> [[BLOCK66]], [[SPLAT_SPLAT68]]
+; CHECK-NEXT:    [[TMP106:%.*]] = fadd <1 x double> [[TMP103]], [[TMP105]]
+; CHECK-NEXT:    [[BLOCK69:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP107:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT70:%.*]] = insertelement <1 x double> poison, double [[TMP107]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT71:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT70]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP108:%.*]] = fmul <1 x double> [[BLOCK69]], [[SPLAT_SPLAT71]]
+; CHECK-NEXT:    [[TMP109:%.*]] = fadd <1 x double> [[TMP106]], [[TMP108]]
+; CHECK-NEXT:    [[TMP110:%.*]] = shufflevector <1 x double> [[TMP109]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP111:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP110]], <3 x i32> <i32 3, i32 1, i32 2>
+; CHECK-NEXT:    [[BLOCK72:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP112:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT73:%.*]] = insertelement <1 x double> poison, double [[TMP112]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT74:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT73]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP113:%.*]] = fmul <1 x double> [[BLOCK72]], [[SPLAT_SPLAT74]]
+; CHECK-NEXT:    [[BLOCK75:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP114:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT76:%.*]] = insertelement <1 x double> poison, double [[TMP114]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT77:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT76]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP115:%.*]] = fmul <1 x double> [[BLOCK75]], [[SPLAT_SPLAT77]]
+; CHECK-NEXT:    [[TMP116:%.*]] = fadd <1 x double> [[TMP113]], [[TMP115]]
+; CHECK-NEXT:    [[BLOCK78:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
+; CHECK-NEXT:    [[TMP117:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT79:%.*]] = insertelement <1 x double> poison, double [[TMP117]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT80:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT79]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP118:%.*]] = fmul <1 x double> [[BLOCK78]], [[SPLAT_SPLAT80]]
+; CHECK-NEXT:    [[TMP119:%.*]] = fadd <1 x double> [[TMP116]], [[TMP118]]
+; CHECK-NEXT:    [[TMP120:%.*]] = shufflevector <1 x double> [[TMP119]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP121:%.*]] = shufflevector <3 x double> [[TMP111]], <3 x double> [[TMP120]], <3 x i32> <i32 0, i32 3, i32 2>
+; CHECK-NEXT:    [[BLOCK81:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP122:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT82:%.*]] = insertelement <1 x double> poison, double [[TMP122]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT83:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT82]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP123:%.*]] = fmul <1 x double> [[BLOCK81]], [[SPLAT_SPLAT83]]
+; CHECK-NEXT:    [[BLOCK84:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP124:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT85:%.*]] = insertelement <1 x double> poison, double [[TMP124]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT86:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT85]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP125:%.*]] = fmul <1 x double> [[BLOCK84]], [[SPLAT_SPLAT86]]
+; CHECK-NEXT:    [[TMP126:%.*]] = fadd <1 x double> [[TMP123]], [[TMP125]]
+; CHECK-NEXT:    [[BLOCK87:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
+; CHECK-NEXT:    [[TMP127:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
+; CHECK-NEXT:    [[SPLAT_SPLATINSERT88:%.*]] = insertelement <1 x double> poison, double [[TMP127]], i64 0
+; CHECK-NEXT:    [[SPLAT_SPLAT89:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT88]], <1 x double> poison, <1 x i32> zeroinitializer
+; CHECK-NEXT:    [[TMP128:%.*]] = fmul <1 x double> [[BLOCK87]], [[SPLAT_SPLAT89]]
+; CHECK-NEXT:    [[TMP129:%.*]] = fadd <1 x double> [[TMP126]], [[TMP128]]
+; CHECK-NEXT:    [[TMP130:%.*]] = shufflevector <1 x double> [[TMP129]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP131:%.*]] = shufflevector <3 x double> [[TMP121]], <3 x double> [[TMP130]], <3 x i32> <i32 0, i32 1, i32 3>
+; CHECK-NEXT:    [[TMP132:%.*]] = shufflevector <3 x double> [[TMP71]], <3 x double> [[TMP101]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
+; CHECK-NEXT:    [[TMP133:%.*]] = shufflevector <3 x double> [[TMP131]], <3 x double> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP134:%.*]] = shufflevector <6 x double> [[TMP132]], <6 x double> [[TMP133]], <9 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
+; CHECK-NEXT:    ret <9 x double> [[TMP134]]
+;
+entry:
+  br i1 %cond, label %if.then, label %if.else
+
+if.then:                    ; preds = %entry
+  %A.trans = tail call <9 x double> @llvm.matrix.transpose.v9f64(<9 x double> %A, i32 3, i32 3)
+  br label %if.end
+
+if.else:                                       ; preds = %entry
+  %B.trans = tail call <9 x double> @llvm.matrix.transpose.v9f64(<9 x double> %B, i32 3, i32 3)
+  br label %if.end
+
+if.end:                                        ; preds = %if.then, %if.else
+  %merge = phi <9 x double> [ %A.trans, %if.then], [ %B.trans, %if.else ]
+  %res = tail call <9 x double> @llvm.matrix.multiply.v9f64.v9f64.v9f64(<9 x double> %C, <9 x double> %merge, i32 3, i32 3, i32 3)
+  ret <9 x double> %res
+}
diff --git a/llvm/test/Transforms/LowerMatrixIntrinsics/propagate-backwards-unsupported.ll b/llvm/test/Transforms/LowerMatrixIntrinsics/propagate-backwards-unsupported.ll
index 6ed8e46d62892..f07e1762d404f 100644
--- a/llvm/test/Transforms/LowerMatrixIntrinsics/propagate-backwards-unsupported.ll
+++ b/llvm/test/Transforms/LowerMatrixIntrinsics/propagate-backwards-unsupported.ll
@@ -1,260 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
 ; RUN: opt -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s
 
-; Check that we we use flattened vectors for PHI operands and extract the columns afterwards.
-define <9 x double> @unsupported_phi(i1 %cond, <9 x double> %A, <9 x double> %B, <9 x double> %C) {
-; CHECK-LABEL: @unsupported_phi(
-; CHECK-NEXT:  entry:
-; CHECK-NEXT:    br i1 [[COND:%.*]], label [[IF_THEN:%.*]], label [[IF_ELSE:%.*]]
-; CHECK:       if.then:
-; CHECK-NEXT:    [[SPLIT3:%.*]] = shufflevector <9 x double> [[A:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
-; CHECK-NEXT:    [[SPLIT4:%.*]] = shufflevector <9 x double> [[A]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
-; CHECK-NEXT:    [[SPLIT5:%.*]] = shufflevector <9 x double> [[A]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
-; CHECK-NEXT:    [[TMP0:%.*]] = extractelement <3 x double> [[SPLIT3]], i64 0
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <3 x double> poison, double [[TMP0]], i64 0
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 0
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <3 x double> [[TMP1]], double [[TMP2]], i64 1
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 0
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <3 x double> [[TMP3]], double [[TMP4]], i64 2
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <3 x double> [[SPLIT3]], i64 1
-; CHECK-NEXT:    [[TMP7:%.*]] = insertelement <3 x double> poison, double [[TMP6]], i64 0
-; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 1
-; CHECK-NEXT:    [[TMP9:%.*]] = insertelement <3 x double> [[TMP7]], double [[TMP8]], i64 1
-; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 1
-; CHECK-NEXT:    [[TMP11:%.*]] = insertelement <3 x double> [[TMP9]], double [[TMP10]], i64 2
-; CHECK-NEXT:    [[TMP12:%.*]] = extractelement <3 x double> [[SPLIT3]], i64 2
-; CHECK-NEXT:    [[TMP13:%.*]] = insertelement <3 x double> poison, double [[TMP12]], i64 0
-; CHECK-NEXT:    [[TMP14:%.*]] = extractelement <3 x double> [[SPLIT4]], i64 2
-; CHECK-NEXT:    [[TMP15:%.*]] = insertelement <3 x double> [[TMP13]], double [[TMP14]], i64 1
-; CHECK-NEXT:    [[TMP16:%.*]] = extractelement <3 x double> [[SPLIT5]], i64 2
-; CHECK-NEXT:    [[TMP17:%.*]] = insertelement <3 x double> [[TMP15]], double [[TMP16]], i64 2
-; CHECK-NEXT:    br label [[IF_END:%.*]]
-; CHECK:       if.else:
-; CHECK-NEXT:    [[SPLIT:%.*]] = shufflevector <9 x double> [[B:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
-; CHECK-NEXT:    [[SPLIT1:%.*]] = shufflevector <9 x double> [[B]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
-; CHECK-NEXT:    [[SPLIT2:%.*]] = shufflevector <9 x double> [[B]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
-; CHECK-NEXT:    [[TMP21:%.*]] = extractelement <3 x double> [[SPLIT]], i64 0
-; CHECK-NEXT:    [[TMP22:%.*]] = insertelement <3 x double> poison, double [[TMP21]], i64 0
-; CHECK-NEXT:    [[TMP23:%.*]] = extractelement <3 x double> [[SPLIT1]], i64 0
-; CHECK-NEXT:    [[TMP24:%.*]] = insertelement <3 x double> [[TMP22]], double [[TMP23]], i64 1
-; CHECK-NEXT:    [[TMP25:%.*]] = extractelement <3 x double> [[SPLIT2]], i64 0
-; CHECK-NEXT:    [[TMP26:%.*]] = insertelement <3 x double> [[TMP24]], double [[TMP25]], i64 2
-; CHECK-NEXT:    [[TMP27:%.*]] = extractelement <3 x double> [[SPLIT]], i64 1
-; CHECK-NEXT:    [[TMP28:%.*]] = insertelement <3 x double> poison, double [[TMP27]], i64 0
-; CHECK-NEXT:    [[TMP29:%.*]] = extractelement <3 x double> [[SPLIT1]], i64 1
-; CHECK-NEXT:    [[TMP30:%.*]] = insertelement <3 x double> [[TMP28]], double [[TMP29]], i64 1
-; CHECK-NEXT:    [[TMP31:%.*]] = extractelement <3 x double> [[SPLIT2]], i64 1
-; CHECK-NEXT:    [[TMP32:%.*]] = insertelement <3 x double> [[TMP30]], double [[TMP31]], i64 2
-; CHECK-NEXT:    [[TMP33:%.*]] = extractelement <3 x double> [[SPLIT]], i64 2
-; CHECK-NEXT:    [[TMP34:%.*]] = insertelement <3 x double> poison, double [[TMP33]], i64 0
-; CHECK-NEXT:    [[TMP35:%.*]] = extractelement <3 x double> [[SPLIT1]], i64 2
-; CHECK-NEXT:    [[TMP36:%.*]] = insertelement <3 x double> [[TMP34]], double [[TMP35]], i64 1
-; CHECK-NEXT:    [[TMP37:%.*]] = extractelement <3 x double> [[SPLIT2]], i64 2
-; CHECK-NEXT:    [[TMP38:%.*]] = insertelement <3 x double> [[TMP36]], double [[TMP37]], i64 2
-; CHECK-NEXT:    br label [[IF_END]]
-; CHECK:       if.end:
-; CHECK-NEXT:    [[MERGE9:%.*]] = phi <3 x double> [ [[TMP5]], [[IF_THEN]] ], [ [[TMP26]], [[IF_ELSE]] ]
-; CHECK-NEXT:    [[MERGE10:%.*]] = phi <3 x double> [ [[TMP11]], [[IF_THEN]] ], [ [[TMP32]], [[IF_ELSE]] ]
-; CHECK-NEXT:    [[MERGE11:%.*]] = phi <3 x double> [ [[TMP17]], [[IF_THEN]] ], [ [[TMP38]], [[IF_ELSE]] ]
-; CHECK-NEXT:    [[SPLIT9:%.*]] = shufflevector <9 x double> [[MERGE:%.*]], <9 x double> poison, <3 x i32> <i32 0, i32 1, i32 2>
-; CHECK-NEXT:    [[SPLIT10:%.*]] = shufflevector <9 x double> [[MERGE]], <9 x double> poison, <3 x i32> <i32 3, i32 4, i32 5>
-; CHECK-NEXT:    [[SPLIT11:%.*]] = shufflevector <9 x double> [[MERGE]], <9 x double> poison, <3 x i32> <i32 6, i32 7, i32 8>
-; CHECK-NEXT:    [[BLOCK:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP42:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x double> poison, double [[TMP42]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP43:%.*]] = fmul <1 x double> [[BLOCK]], [[SPLAT_SPLAT]]
-; CHECK-NEXT:    [[BLOCK12:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP44:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT13:%.*]] = insertelement <1 x double> poison, double [[TMP44]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT14:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT13]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP45:%.*]] = fmul <1 x double> [[BLOCK12]], [[SPLAT_SPLAT14]]
-; CHECK-NEXT:    [[TMP46:%.*]] = fadd <1 x double> [[TMP43]], [[TMP45]]
-; CHECK-NEXT:    [[BLOCK15:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP47:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT16:%.*]] = insertelement <1 x double> poison, double [[TMP47]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT17:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT16]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP48:%.*]] = fmul <1 x double> [[BLOCK15]], [[SPLAT_SPLAT17]]
-; CHECK-NEXT:    [[TMP49:%.*]] = fadd <1 x double> [[TMP46]], [[TMP48]]
-; CHECK-NEXT:    [[TMP50:%.*]] = shufflevector <1 x double> [[TMP49]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP51:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP50]], <3 x i32> <i32 3, i32 1, i32 2>
-; CHECK-NEXT:    [[BLOCK18:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP52:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT19:%.*]] = insertelement <1 x double> poison, double [[TMP52]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT20:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT19]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP53:%.*]] = fmul <1 x double> [[BLOCK18]], [[SPLAT_SPLAT20]]
-; CHECK-NEXT:    [[BLOCK21:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP54:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT22:%.*]] = insertelement <1 x double> poison, double [[TMP54]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT23:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT22]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP55:%.*]] = fmul <1 x double> [[BLOCK21]], [[SPLAT_SPLAT23]]
-; CHECK-NEXT:    [[TMP56:%.*]] = fadd <1 x double> [[TMP53]], [[TMP55]]
-; CHECK-NEXT:    [[BLOCK24:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP57:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT25:%.*]] = insertelement <1 x double> poison, double [[TMP57]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT26:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT25]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP58:%.*]] = fmul <1 x double> [[BLOCK24]], [[SPLAT_SPLAT26]]
-; 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 poison, i32 poison>
-; CHECK-NEXT:    [[TMP61:%.*]] = shufflevector <3 x double> [[TMP51]], <3 x double> [[TMP60]], <3 x i32> <i32 0, i32 3, i32 2>
-; CHECK-NEXT:    [[BLOCK27:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP62:%.*]] = extractelement <3 x double> [[MERGE9]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT28:%.*]] = insertelement <1 x double> poison, double [[TMP62]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT29:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT28]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP63:%.*]] = fmul <1 x double> [[BLOCK27]], [[SPLAT_SPLAT29]]
-; CHECK-NEXT:    [[BLOCK30:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP64:%.*]] = extractelement <3 x double> [[MERGE9]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT31:%.*]] = insertelement <1 x double> poison, double [[TMP64]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT32:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT31]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP65:%.*]] = fmul <1 x double> [[BLOCK30]], [[SPLAT_SPLAT32]]
-; CHECK-NEXT:    [[TMP66:%.*]] = fadd <1 x double> [[TMP63]], [[TMP65]]
-; CHECK-NEXT:    [[BLOCK33:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP67:%.*]] = extractelement <3 x double> [[MERGE9]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT34:%.*]] = insertelement <1 x double> poison, double [[TMP67]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT35:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT34]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP68:%.*]] = fmul <1 x double> [[BLOCK33]], [[SPLAT_SPLAT35]]
-; CHECK-NEXT:    [[TMP69:%.*]] = fadd <1 x double> [[TMP66]], [[TMP68]]
-; CHECK-NEXT:    [[TMP70:%.*]] = shufflevector <1 x double> [[TMP69]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP71:%.*]] = shufflevector <3 x double> [[TMP61]], <3 x double> [[TMP70]], <3 x i32> <i32 0, i32 1, i32 3>
-; CHECK-NEXT:    [[BLOCK36:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP72:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT37:%.*]] = insertelement <1 x double> poison, double [[TMP72]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT38:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT37]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP73:%.*]] = fmul <1 x double> [[BLOCK36]], [[SPLAT_SPLAT38]]
-; CHECK-NEXT:    [[BLOCK39:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP74:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT40:%.*]] = insertelement <1 x double> poison, double [[TMP74]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT41:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT40]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP75:%.*]] = fmul <1 x double> [[BLOCK39]], [[SPLAT_SPLAT41]]
-; CHECK-NEXT:    [[TMP76:%.*]] = fadd <1 x double> [[TMP73]], [[TMP75]]
-; CHECK-NEXT:    [[BLOCK42:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP77:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT43:%.*]] = insertelement <1 x double> poison, double [[TMP77]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT44:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT43]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP78:%.*]] = fmul <1 x double> [[BLOCK42]], [[SPLAT_SPLAT44]]
-; CHECK-NEXT:    [[TMP79:%.*]] = fadd <1 x double> [[TMP76]], [[TMP78]]
-; CHECK-NEXT:    [[TMP80:%.*]] = shufflevector <1 x double> [[TMP79]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP81:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP80]], <3 x i32> <i32 3, i32 1, i32 2>
-; CHECK-NEXT:    [[BLOCK45:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP82:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT46:%.*]] = insertelement <1 x double> poison, double [[TMP82]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT47:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT46]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP83:%.*]] = fmul <1 x double> [[BLOCK45]], [[SPLAT_SPLAT47]]
-; CHECK-NEXT:    [[BLOCK48:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP84:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT49:%.*]] = insertelement <1 x double> poison, double [[TMP84]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT50:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT49]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP85:%.*]] = fmul <1 x double> [[BLOCK48]], [[SPLAT_SPLAT50]]
-; CHECK-NEXT:    [[TMP86:%.*]] = fadd <1 x double> [[TMP83]], [[TMP85]]
-; CHECK-NEXT:    [[BLOCK51:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP87:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT52:%.*]] = insertelement <1 x double> poison, double [[TMP87]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT53:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT52]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP88:%.*]] = fmul <1 x double> [[BLOCK51]], [[SPLAT_SPLAT53]]
-; CHECK-NEXT:    [[TMP89:%.*]] = fadd <1 x double> [[TMP86]], [[TMP88]]
-; CHECK-NEXT:    [[TMP90:%.*]] = shufflevector <1 x double> [[TMP89]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP91:%.*]] = shufflevector <3 x double> [[TMP81]], <3 x double> [[TMP90]], <3 x i32> <i32 0, i32 3, i32 2>
-; CHECK-NEXT:    [[BLOCK54:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP92:%.*]] = extractelement <3 x double> [[MERGE10]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT55:%.*]] = insertelement <1 x double> poison, double [[TMP92]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT56:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT55]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP93:%.*]] = fmul <1 x double> [[BLOCK54]], [[SPLAT_SPLAT56]]
-; CHECK-NEXT:    [[BLOCK57:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP94:%.*]] = extractelement <3 x double> [[MERGE10]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT58:%.*]] = insertelement <1 x double> poison, double [[TMP94]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT59:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT58]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP95:%.*]] = fmul <1 x double> [[BLOCK57]], [[SPLAT_SPLAT59]]
-; CHECK-NEXT:    [[TMP96:%.*]] = fadd <1 x double> [[TMP93]], [[TMP95]]
-; CHECK-NEXT:    [[BLOCK60:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP97:%.*]] = extractelement <3 x double> [[MERGE10]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT61:%.*]] = insertelement <1 x double> poison, double [[TMP97]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT62:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT61]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP98:%.*]] = fmul <1 x double> [[BLOCK60]], [[SPLAT_SPLAT62]]
-; CHECK-NEXT:    [[TMP99:%.*]] = fadd <1 x double> [[TMP96]], [[TMP98]]
-; CHECK-NEXT:    [[TMP100:%.*]] = shufflevector <1 x double> [[TMP99]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP101:%.*]] = shufflevector <3 x double> [[TMP91]], <3 x double> [[TMP100]], <3 x i32> <i32 0, i32 1, i32 3>
-; CHECK-NEXT:    [[BLOCK63:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP102:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT64:%.*]] = insertelement <1 x double> poison, double [[TMP102]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT65:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT64]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP103:%.*]] = fmul <1 x double> [[BLOCK63]], [[SPLAT_SPLAT65]]
-; CHECK-NEXT:    [[BLOCK66:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP104:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT67:%.*]] = insertelement <1 x double> poison, double [[TMP104]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT68:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT67]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP105:%.*]] = fmul <1 x double> [[BLOCK66]], [[SPLAT_SPLAT68]]
-; CHECK-NEXT:    [[TMP106:%.*]] = fadd <1 x double> [[TMP103]], [[TMP105]]
-; CHECK-NEXT:    [[BLOCK69:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP107:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT70:%.*]] = insertelement <1 x double> poison, double [[TMP107]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT71:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT70]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP108:%.*]] = fmul <1 x double> [[BLOCK69]], [[SPLAT_SPLAT71]]
-; CHECK-NEXT:    [[TMP109:%.*]] = fadd <1 x double> [[TMP106]], [[TMP108]]
-; CHECK-NEXT:    [[TMP110:%.*]] = shufflevector <1 x double> [[TMP109]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP111:%.*]] = shufflevector <3 x double> poison, <3 x double> [[TMP110]], <3 x i32> <i32 3, i32 1, i32 2>
-; CHECK-NEXT:    [[BLOCK72:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP112:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT73:%.*]] = insertelement <1 x double> poison, double [[TMP112]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT74:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT73]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP113:%.*]] = fmul <1 x double> [[BLOCK72]], [[SPLAT_SPLAT74]]
-; CHECK-NEXT:    [[BLOCK75:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP114:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT76:%.*]] = insertelement <1 x double> poison, double [[TMP114]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT77:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT76]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP115:%.*]] = fmul <1 x double> [[BLOCK75]], [[SPLAT_SPLAT77]]
-; CHECK-NEXT:    [[TMP116:%.*]] = fadd <1 x double> [[TMP113]], [[TMP115]]
-; CHECK-NEXT:    [[BLOCK78:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 1>
-; CHECK-NEXT:    [[TMP117:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT79:%.*]] = insertelement <1 x double> poison, double [[TMP117]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT80:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT79]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP118:%.*]] = fmul <1 x double> [[BLOCK78]], [[SPLAT_SPLAT80]]
-; CHECK-NEXT:    [[TMP119:%.*]] = fadd <1 x double> [[TMP116]], [[TMP118]]
-; CHECK-NEXT:    [[TMP120:%.*]] = shufflevector <1 x double> [[TMP119]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP121:%.*]] = shufflevector <3 x double> [[TMP111]], <3 x double> [[TMP120]], <3 x i32> <i32 0, i32 3, i32 2>
-; CHECK-NEXT:    [[BLOCK81:%.*]] = shufflevector <3 x double> [[SPLIT9]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP122:%.*]] = extractelement <3 x double> [[MERGE11]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT82:%.*]] = insertelement <1 x double> poison, double [[TMP122]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT83:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT82]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP123:%.*]] = fmul <1 x double> [[BLOCK81]], [[SPLAT_SPLAT83]]
-; CHECK-NEXT:    [[BLOCK84:%.*]] = shufflevector <3 x double> [[SPLIT10]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP124:%.*]] = extractelement <3 x double> [[MERGE11]], i64 1
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT85:%.*]] = insertelement <1 x double> poison, double [[TMP124]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT86:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT85]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP125:%.*]] = fmul <1 x double> [[BLOCK84]], [[SPLAT_SPLAT86]]
-; CHECK-NEXT:    [[TMP126:%.*]] = fadd <1 x double> [[TMP123]], [[TMP125]]
-; CHECK-NEXT:    [[BLOCK87:%.*]] = shufflevector <3 x double> [[SPLIT11]], <3 x double> poison, <1 x i32> <i32 2>
-; CHECK-NEXT:    [[TMP127:%.*]] = extractelement <3 x double> [[MERGE11]], i64 2
-; CHECK-NEXT:    [[SPLAT_SPLATINSERT88:%.*]] = insertelement <1 x double> poison, double [[TMP127]], i64 0
-; CHECK-NEXT:    [[SPLAT_SPLAT89:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT88]], <1 x double> poison, <1 x i32> zeroinitializer
-; CHECK-NEXT:    [[TMP128:%.*]] = fmul <1 x double> [[BLOCK87]], [[SPLAT_SPLAT89]]
-; CHECK-NEXT:    [[TMP129:%.*]] = fadd <1 x double> [[TMP126]], [[TMP128]]
-; CHECK-NEXT:    [[TMP130:%.*]] = shufflevector <1 x double> [[TMP129]], <1 x double> poison, <3 x i32> <i32 0, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP131:%.*]] = shufflevector <3 x double> [[TMP121]], <3 x double> [[TMP130]], <3 x i32> <i32 0, i32 1, i32 3>
-; CHECK-NEXT:    [[TMP132:%.*]] = shufflevector <3 x double> [[TMP71]], <3 x double> [[TMP101]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>
-; CHECK-NEXT:    [[TMP133:%.*]] = shufflevector <3 x double> [[TMP131]], <3 x double> poison, <6 x i32> <i32 0, i32 1, i32 2, i32 poison, i32 poison, i32 poison>
-; CHECK-NEXT:    [[TMP134:%.*]] = shufflevector <6 x double> [[TMP132]], <6 x double> [[TMP133]], <9 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
-; CHECK-NEXT:    ret <9 x double> [[TMP134]]
-;
-
-
-
-entry:
-  br i1 %cond, label %if.then, label %if.else
-
-if.then:                    ; preds = %entry
-  %A.trans = tail call <9 x double> @llvm.matrix.transpose.v9f64(<9 x double> %A, i32 3, i32 3)
-  br label %if.end
-
-if.else:                                       ; preds = %entry
-  %B.trans = tail call <9 x double> @llvm.matrix.transpose.v9f64(<9 x double> %B, i32 3, i32 3)
-  br label %if.end
-
-if.end:                                        ; preds = %if.then, %if.else
-  %merge = phi <9 x double> [ %A.trans, %if.then], [ %B.trans, %if.else ]
-  %res = tail call <9 x double> @llvm.matrix.multiply.v9f64.v9f64.v9f64(<9 x double> %C, <9 x double> %merge, i32 3, i32 3, i32 3)
-  ret <9 x double> %res
-}
-
 ; Make sure we use a flattened vector when calling @foo and the use its flat vector result properly.
 define <9 x double> @unsupported_call(i1 %cond, <9 x double> %A, <9 x double> %B) {
 ; CHECK-LABEL: @unsupported_call(

>From 905c1e93506605b841fc5beadf9c1dc94cb4ecd6 Mon Sep 17 00:00:00 2001
From: Jon Roelofs <jonathan_roelofs at apple.com>
Date: Tue, 27 May 2025 15:41:56 -0700
Subject: [PATCH 3/5] clang-format

---
 llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp b/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
index c06d08688ab1c..8db812eacbd48 100644
--- a/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
+++ b/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
@@ -371,10 +371,10 @@ class LowerMatrixIntrinsics {
         if (auto *CDV = dyn_cast<ConstantDataVector>(Constant)) {
           unsigned Width = SI.getStride();
           size_t EltSize = EltTy->getPrimitiveSizeInBits() / 8;
-          StringRef Data = CDV->getRawDataValues().substr(
-              J * Width * EltSize, Width * EltSize);
-          addVector(ConstantDataVector::getRaw(Data, Width,
-                                               CDV->getElementType()));
+          StringRef Data = CDV->getRawDataValues().substr(J * Width * EltSize,
+                                                          Width * EltSize);
+          addVector(
+              ConstantDataVector::getRaw(Data, Width, CDV->getElementType()));
         } else if (isa<PoisonValue>(Constant))
           addVector(PoisonValue::get(RowTy));
         else if (isa<UndefValue>(Constant))

>From 9ee44f03681404f339c9dd8dcc0d4e5729333999 Mon Sep 17 00:00:00 2001
From: Jon Roelofs <jonathan_roelofs at apple.com>
Date: Tue, 27 May 2025 15:56:50 -0700
Subject: [PATCH 4/5] add test for ConstantDataVector lowering

---
 .../Transforms/LowerMatrixIntrinsics/phi.ll   | 52 +++++++++++++++++++
 1 file changed, 52 insertions(+)

diff --git a/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll b/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll
index e2720b86d87ca..2d2125d7b444f 100644
--- a/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll
+++ b/llvm/test/Transforms/LowerMatrixIntrinsics/phi.ll
@@ -215,6 +215,58 @@ exit:
   ret void
 }
 
+define void @matrix_phi_loop_cdv(ptr %in1, ptr %in2, i32 %count, ptr %out) {
+; CHECK-LABEL: @matrix_phi_loop_cdv(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    br label [[LOOP:%.*]]
+; CHECK:       loop:
+; CHECK-NEXT:    [[PHI4:%.*]] = phi <3 x double> [ <double 0.000000e+00, double 1.000000e+00, double 2.000000e+00>, [[ENTRY:%.*]] ], [ [[TMP0:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI5:%.*]] = phi <3 x double> [ <double 3.000000e+00, double 4.000000e+00, double 5.000000e+00>, [[ENTRY]] ], [ [[TMP1:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[PHI6:%.*]] = phi <3 x double> [ <double 6.000000e+00, double 7.000000e+00, double 8.000000e+00>, [[ENTRY]] ], [ [[TMP2:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[CTR:%.*]] = phi i32 [ [[COUNT:%.*]], [[ENTRY]] ], [ [[DEC:%.*]], [[LOOP]] ]
+; CHECK-NEXT:    [[COL_LOAD:%.*]] = load <3 x double>, ptr [[IN2:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP:%.*]] = getelementptr double, ptr [[IN2]], i64 3
+; CHECK-NEXT:    [[COL_LOAD1:%.*]] = load <3 x double>, ptr [[VEC_GEP]], align 8
+; CHECK-NEXT:    [[VEC_GEP2:%.*]] = getelementptr double, ptr [[IN2]], i64 6
+; CHECK-NEXT:    [[COL_LOAD3:%.*]] = load <3 x double>, ptr [[VEC_GEP2]], align 16
+; CHECK-NEXT:    [[TMP0]] = fadd <3 x double> [[PHI4]], [[COL_LOAD]]
+; CHECK-NEXT:    [[TMP1]] = fadd <3 x double> [[PHI5]], [[COL_LOAD1]]
+; CHECK-NEXT:    [[TMP2]] = fadd <3 x double> [[PHI6]], [[COL_LOAD3]]
+; CHECK-NEXT:    [[DEC]] = sub i32 [[CTR]], 1
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[DEC]], 0
+; CHECK-NEXT:    br i1 [[CMP]], label [[EXIT:%.*]], label [[LOOP]]
+; CHECK:       exit:
+; CHECK-NEXT:    store <3 x double> [[TMP0]], ptr [[OUT:%.*]], align 128
+; CHECK-NEXT:    [[VEC_GEP7:%.*]] = getelementptr double, ptr [[OUT]], i64 3
+; CHECK-NEXT:    store <3 x double> [[TMP1]], ptr [[VEC_GEP7]], align 8
+; CHECK-NEXT:    [[VEC_GEP8:%.*]] = getelementptr double, ptr [[OUT]], i64 6
+; CHECK-NEXT:    store <3 x double> [[TMP2]], ptr [[VEC_GEP8]], align 16
+; CHECK-NEXT:    ret void
+;
+entry:
+  br label %loop
+
+loop:
+  %phi = phi <9 x double> [<double 0., double 1., double 2., double 3., double 4., double 5., double 6., double 7., double 8.>, %entry], [%sum, %loop]
+  %ctr = phi i32 [%count, %entry], [%dec, %loop]
+
+  %in2v = load <9 x double>, ptr %in2
+
+  ; Give in2 the shape: 3 x 3
+  %in2t  = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2v, i32 3, i32 3)
+  %in2tt = call <9 x double> @llvm.matrix.transpose(<9 x double> %in2t, i32 3, i32 3)
+
+  %sum = fadd <9 x double> %phi, %in2tt
+
+  %dec = sub i32 %ctr, 1
+  %cmp = icmp eq i32 %dec, 0
+  br i1 %cmp, label %exit, label %loop
+
+exit:
+  store <9 x double> %sum, ptr %out
+  ret void
+}
+
 define <9 x double> @matrix_phi_ifthenelse(i1 %cond, <9 x double> %A, <9 x double> %B, <9 x double> %C) {
 ; CHECK-LABEL: @matrix_phi_ifthenelse(
 ; CHECK-NEXT:  entry:

>From 169960dd3e7a9cb4e25924497c024285fed8a9f2 Mon Sep 17 00:00:00 2001
From: Jon Roelofs <jonathan_roelofs at apple.com>
Date: Tue, 27 May 2025 15:58:04 -0700
Subject: [PATCH 5/5] move report_fatal_error outside of NDEBUG block

---
 llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp b/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
index 8db812eacbd48..f447baf717b87 100644
--- a/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
+++ b/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
@@ -384,8 +384,8 @@ class LowerMatrixIntrinsics {
         else {
 #ifndef NDEBUG
           Constant->dump();
-          report_fatal_error("unhandled ConstantData type");
 #endif
+          report_fatal_error("unhandled ConstantData type");
         }
       }
     }



More information about the llvm-commits mailing list