[Mlir-commits] [mlir] [MLIR][XeGPU] Add Layout Propagation support for multi-reduction/reduction op with scalar result (PR #189133)

[Jianhui Li]

================
@@ -482,9 +474,17 @@ xegpu::SliceAttr xegpu::setupMultiReductionResultLayout(
           srcLayout = srcLayout.setDimData(dim, srcSgData.value()[dim], -1, -1);
         }
     } else {
-
+      SmallVector<int64_t> consumerSgLayout =
+          consumerLayout ? consumerLayout.getEffectiveSgLayoutAsInt()
+                         : SmallVector<int64_t>();
+      SmallVector<int64_t> consumerOrder =
+          consumerLayout ? consumerLayout.getEffectiveOrderAsInt()
+                         : SmallVector<int64_t>();
+      DenseI32ArrayAttr orderAttr =
+          consumerLayout ? consumerLayout.getOrder() : nullptr;
       SmallVector<int64_t> sgLayout(srcRank), sgData(srcRank), order(srcRank);
-      int remainingSgCount = workgroupSize;
+      int remainingSgCount =
----------------
Jianhui-Li wrote:

The algorithm allocates certain number of subgroup for each dimension and remaining tracks the subgroup number not allocated. 

https://github.com/llvm/llvm-project/pull/189133