[Mlir-commits] [mlir] [MLIR][XeGPU] Remove create tdesc & update offset op from xegpu dialect (PR #182804)

[Nishant Patel]

================
@@ -563,7 +495,10 @@ struct UnrollLoadGatherOp : public UnrollPattern<xegpu::LoadGatherOp> {
       return failure();
 
     SmallVector<int64_t> targetMaskShape(*targetShape);
-    int64_t originalChunkSize = tdescTy.getChunkSizeAsInt();
+    // Derive chunk size from the tensor descriptor shape: for 2D descriptors,
----------------
nbpatel wrote:

fixed

https://github.com/llvm/llvm-project/pull/182804