[Mlir-commits] [mlir] [MLIR][XeGPU] Extend unrolling support for scatter ops with chunk_size (PR #144447)

[Chao Chen]

================
@@ -490,7 +541,7 @@ struct UnrollPrefetchOp : public UnrollPattern<xegpu::PrefetchOp> {
     xegpu::TensorDescType tdescTy = op.getTensorDescType();
 
     // check if the tensor descriptor type is a 1d vector type
-    if (tdescTy.getRank() > 1)
+    if (tdescTy.getRank() > 2)
----------------
chencha3 wrote:

use !tdescTy.isScattered()?

https://github.com/llvm/llvm-project/pull/144447