[Mlir-commits] [mlir] [mlir][XeGPU] add WgToSg distribution pattern for load_matrix and store_matrix. (PR #154403)

[Artem Kroviakov]

================
@@ -1213,12 +1278,17 @@ def XeGPU_StoreMatrixOp: XeGPU_Op<"store_matrix", [MemoryEffects<[MemWrite]>,
   }];
   let builders = [
     OpBuilder<(ins "Value" : $data, "TypedValue<MemDescType>": $mem_desc,
-                   "llvm::ArrayRef<OpFoldResult>": $offsets, "LayoutTrait": $layout)>,
+                   "llvm::ArrayRef<OpFoldResult>": $offsets, "DistributeLayoutAttr": $layout)>,
   ];
   let extraClassDeclaration = [{
     SmallVector<OpFoldResult> getMixedOffsets() {
       return getMixedValues(getConstOffsets(), getOffsets(), getContext());
     }
+
+    ArrayRef<int64_t> getDistributeShape() {
----------------
akroviakov wrote:

Do we need to mention distribution in the op's methods names? I guess this utility could be reused elsewhere as well, so `getShape` would be good enough.

https://github.com/llvm/llvm-project/pull/154403