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

[Jianhui Li]

================
@@ -61,6 +67,16 @@ struct TestXeGPUUnrollingPatterns
               tdescTy = loadNdOp.getTensorDescType();
             } else if (auto storeNdOp = dyn_cast<xegpu::StoreNdOp>(op)) {
               tdescTy = storeNdOp.getTensorDescType();
+            } else if (auto createOp = dyn_cast<xegpu::CreateDescOp>(op)) {
----------------
Jianhui-Li wrote:

It is hard to use TypeSwitch, since al these conditions need to do a common code section at the end. 

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