[Mlir-commits] [mlir] [MLIR][XeGPU] Clean up the temporary layout usage in XeGPU test (PR #195739)

[Jianhui Li]

================
@@ -1598,6 +1680,32 @@ xegpu::inferSourceLayoutFromResult(OpOperand &operand,
       return resLayout;
   }
 
+  // For vector::Insert Op, infer source layout from result layout using
+  // shapes.
+  if (auto insert = dyn_cast<vector::InsertOp>(op)) {
+    VectorType resVecTy = dyn_cast<VectorType>(insert.getResult().getType());
+    VectorType valueToStoreTy =
----------------
Jianhui-Li wrote:

You are right that the valueToInsert could be a scalar. Added a check to defend. 

https://github.com/llvm/llvm-project/pull/195739