[Mlir-commits] [mlir] [MLIR][XeGPU] Allow load/store/prefetch uses [memref+offset] instead of tdesc (PR #150576)

Adam Siemieniuk llvmlistbot at llvm.org
Wed Jul 30 05:40:46 PDT 2025


================
@@ -575,7 +577,8 @@ struct UnrollStoreScatterOp : public UnrollPattern<xegpu::StoreScatterOp> {
     VectorType valueTy = llvm::dyn_cast<VectorType>(op.getValue().getType());
     xegpu::TensorDescType tdescTy = op.getTensorDescType();
 
-    if (!tdescTy.isScattered())
+    // TODO: handle the unstructure source case (!tdesTy)
+    if (!tdescTy || !tdescTy.isScattered())
----------------
adam-smnk wrote:

nit: I think the `isScattered` is already checked by op verifier and can be taken as an invariant in the passes

https://github.com/llvm/llvm-project/pull/150576


More information about the Mlir-commits mailing list