[Mlir-commits] [mlir] [MLIR][XeGPU] Add unrolling/blocking support for 3D+ batched operations (PR #201725)

[Charitha Saumya]

================
@@ -44,36 +44,57 @@ resolveUnrealizedConversionCastOp(UnrealizedConversionCastOp castOp) {
   ValueRange inputs = castOp.getInputs();
   ValueRange outputs = castOp.getOutputs();
 
-  auto hasIdenticalVectorTypes = [](ValueRange values) {
+  auto hasIdenticalVectorOrTdescTypes = [](ValueRange values) {
     auto types = values.getTypes();
     return llvm::all_of(types, [&](Type type) {
-      return isa<VectorType>(type) && type == types.front();
+      return (isa<VectorType>(type) || isa<xegpu::TensorDescType>(type)) &&
+             type == types.front();
     });
   };
 
   // We only interest in the case where all inputs and outputs have the
   // identical VectorTypes
----------------
charithaintc wrote:

nit: comment only talks about vector?

https://github.com/llvm/llvm-project/pull/201725