[Mlir-commits] [mlir] [MLIR][XeGPU] Enhance multi-reduction layout propagation rules (PR #186308)

[Charitha Saumya]

================
@@ -437,75 +437,118 @@ xegpu::SliceAttr xegpu::setupMultiReductionResultLayout(
     return DenseI32ArrayAttr::get(context, vec32);
   };
 
-  // Extract original plain layout for workgroup/subgroup size recovery
-  xegpu::SliceAttr consumerSliceLayout =
-      dyn_cast<xegpu::SliceAttr>(consumerLayout);
-  DistributeLayoutAttr plainLayout =
-      consumerSliceLayout ? consumerSliceLayout.flatten().getParent()
-                          : consumerLayout;
+  // Helper lambda to check if the layout from consumer can be reused for the
----------------
charithaintc wrote:

please update the comment of this function `setupMultiReductionResultLayout` if the original comment is outdated after this change. 

https://github.com/llvm/llvm-project/pull/186308