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

[Artem Kroviakov]

================
@@ -527,17 +528,18 @@ xegpu::SliceAttr xegpu::setupMultiReductionResultLayout(
     SmallVector<int64_t> instData(srcRank, 1);
     instData[srcRank - 2] =
         std::min(maxReduceVectorSize, srcShape[srcRank - 2]);
-    instData[srcRank - 1] =
-        std::min(static_cast<int64_t>(subgroupSize), srcShape[srcRank - 1]);
+    if (srcRank >= 2)
----------------
akroviakov wrote:

This check suggests that src rank can be `1`, but then the unpredicated `instData[srcRank - 2] =` above would fail? Looks like a bug, since the lane case has it predicated.

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