[Mlir-commits] [mlir] [mlir][xegpu] Add vector layout conflict handling in XeGPU layout propagation pass. (PR #182402)

[Jianhui Li]

================
@@ -1132,7 +1133,7 @@ void LayoutInfoPropagation::visitStoreMatrixOp(
     VectorType srcVecTy =
         llvm::cast<VectorType>(storeMatrix.getData().getType());
     assert(srcVecTy.getRank() == 2 && "Expecting 2D vector for store matrix.");
-    auto uArch = getUArch(getChipStr(storeMatrix).value_or(""));
+    const auto *uArch = getUArch(getChipStr(storeMatrix).value_or(""));
----------------
Jianhui-Li wrote:

It is not related to this PR, but is there a way to save uArch info upfront like pass's parameter so each pattern doesn't have to walk the IR to getUarch?

https://github.com/llvm/llvm-project/pull/182402