[Mlir-commits] [mlir] [MLIR][XeGPU] Scattered ops sg-to-wi distribution (PR #154949)

[Artem Kroviakov]

================
@@ -541,21 +551,27 @@ void LayoutInfoPropagation::visitVectorBitcastOp(
   propagateIfChanged(operands[0], operands[0]->meet(resultLayout));
 }
 
-/// Propagate the layout of the result to the tensor descriptor and mask
+/// Propagate the layout of the result to the tensor descriptor, mask and offset
 /// operands in LoadGatherOp.
 void LayoutInfoPropagation::visitLoadGatherOp(
     xegpu::LoadGatherOp load, ArrayRef<LayoutInfoLattice *> operands,
     ArrayRef<const LayoutInfoLattice *> results) {
-  // The layout is strictly determined by the tensor descriptor type.
-  LayoutInfo layout = getDefaultSIMTLayoutInfo(load.getTensorDescType());
+  // The layout is strictly determined by the payload type.
+  auto payloadTy = dyn_cast<VectorType>(load.getValueType());
+  assert(payloadTy && "Only vector payload distribution is supported");
----------------
akroviakov wrote:

Changed to a warning and propagation stop

https://github.com/llvm/llvm-project/pull/154949