[Mlir-commits] [mlir] [mlir][XeGPU][VectorToXeGPU] Propagate vector layouts to xegpu ops (PR #163071)

[Jianhui Li]

================
@@ -616,16 +636,34 @@ struct GatherLowering : public OpRewritePattern<vector::GatherOp> {
         computeOffsets(rewriter, gatherOp, meta.first, meta.second);
     Value flatMemref = memrefToIndexPtr(gatherOp, rewriter);
 
+    auto layoutRes = xegpu::getDistributeLayoutAttr(gatherOp.getResult());
----------------
Jianhui-Li wrote:

>We can take the layout from Result vector, and set to the store op's layout attribute (need to expand the op definition to make the attribute persistent, not lost after optimization passes).

one correction: I think we should not take the layout from the result or operand, since they are temporary so should not serve as our interface.  Instead, we should ask user to set the permanent layout associated with anchor op. 

https://github.com/llvm/llvm-project/pull/163071