[Mlir-commits] [mlir] [MLIR][XeGPU] Updates XeGPU TensorDescAttr and Refine Gather/Scatter definition. (PR #109144)

[Chao Chen]

================
@@ -19,12 +19,18 @@ class XeGPUAttr<string name, string attrMnemonic, list<Trait> traits = [],
   let mnemonic = attrMnemonic;
 }
 
-def XeGPU_TensorDescAttr: XeGPUAttr<"TensorDesc", "tdesc_attr"> {
+class XeGPU_TensorDescAttr<string name, string attrMnemonic, list<Trait> traits = [],
+                         string baseCppClass = "::mlir::Attribute">
+    : XeGPUAttr<name, attrMnemonic, traits, baseCppClass> {
+  let assemblyFormat = "`<` struct(params) `>`";
+}
+
+def XeGPU_BlockTensorDescAttr: XeGPU_TensorDescAttr<"BlockTensorDesc", "block_tdesc_attr"> {
----------------
chencha3 wrote:

They are supposed to go with TensorDescType, since it will be easier for the implementation of lowering pass. 

https://github.com/llvm/llvm-project/pull/109144