[Mlir-commits] [mlir] [MLIR][XeGPU] Add unrolling/blocking support for 3D+ batched operations (PR #201725)

[Artem Kroviakov]

================
@@ -983,7 +994,11 @@ void LayoutInfoPropagation::visitStoreNdOp(
       if (instHeight == -1)
         store.emitWarning(
             "No suitable instruction multiple found for the given shape.");
-      instData = {instHeight, instWidth};
+      // For rank > 2, prepend unit dims for batch dimensions.
+      for (int64_t d = 0; d < dataTy.getRank() - 2; ++d)
----------------
akroviakov wrote:

Why not initialize instData as
```
SmallVector<int>  instData(dataTy.getRank(), 1);
``` 
? Other branches either override the whole vector (1D) or update the last two elements.

https://github.com/llvm/llvm-project/pull/201725