1 files changed, 11 insertions, 6 deletions

diff --git a/contrib/llvm-project/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp b/contrib/llvm-project/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
index 8f1d0181ee5b..296becb31e8f 100644
--- a/contrib/llvm-project/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
+++ b/contrib/llvm-project/llvm/lib/Transforms/Scalar/LowerMatrixIntrinsics.cpp
@@ -1339,16 +1339,21 @@ public:
 
     // Copy load operand to new alloca.
     Builder.SetInsertPoint(Copy, Copy->begin());
-    AllocaInst *NewLd =
-        Builder.CreateAlloca(Load->getType(), Load->getPointerAddressSpace());
-    Builder.CreateMemCpy(NewLd, NewLd->getAlign(),
-                         Load->getPointerOperand(), Load->getAlign(),
-                         LoadLoc.Size.getValue());
+    auto *VT = cast<FixedVectorType>(Load->getType());
+    // Use an array type for the alloca, to avoid potentially huge alignment
+    // requirements for large vector types.
+    auto *ArrayTy = ArrayType::get(VT->getElementType(), VT->getNumElements());
+    AllocaInst *Alloca =
+        Builder.CreateAlloca(ArrayTy, Load->getPointerAddressSpace());
+    Value *BC = Builder.CreateBitCast(Alloca, VT->getPointerTo());
+
+    Builder.CreateMemCpy(BC, Alloca->getAlign(), Load->getPointerOperand(),
+                         Load->getAlign(), LoadLoc.Size.getValue());
     Builder.SetInsertPoint(Fusion, Fusion->begin());
     PHINode *PHI = Builder.CreatePHI(Load->getPointerOperandType(), 3);
     PHI->addIncoming(Load->getPointerOperand(), Check0);
     PHI->addIncoming(Load->getPointerOperand(), Check1);
-    PHI->addIncoming(NewLd, Copy);
+    PHI->addIncoming(BC, Copy);
 
     // Adjust DT.
     DTUpdates.push_back({DT->Insert, Check0, Check1});