[Mlir-commits] [mlir] [mlir][amdgpu] Add tensor load store operations (PR #172686)

[Erick Ochoa Lopez]

================
@@ -3306,6 +3327,30 @@ void mlir::populateAMDGPUTypeAndAttributeConversions(
     Type i32 = IntegerType::get(type.getContext(), 32);
     return typeConverter.convertType(VectorType::get(4, i32));
   });
+  typeConverter.addConversion(
+      [&](TDMDescriptorType type,
+          SmallVectorImpl<Type> &result) -> std::optional<LogicalResult> {
+        Type i32 = IntegerType::get(type.getContext(), 32);
+        Type v4i32 = typeConverter.convertType(VectorType::get(4, i32));
+        Type v8i32 = typeConverter.convertType(VectorType::get(8, i32));
+        llvm::append_values(result, v4i32, v8i32, v4i32, v4i32);
+        return success();
+      });
+
+  auto addUnrealizedCast = [](OpBuilder &builder, TypeRange types,
+                              ValueRange inputs,
+                              Location loc) -> SmallVector<Value> {
+    if (inputs.size() != 1)
+      return {};
+
+    if (!isa<TDMDescriptorType>(inputs[0].getType()))
----------------
amd-eochoalo wrote:

Added comment here : https://github.com/llvm/llvm-project/pull/172686/commits/2f55093c5210964de3e71bab95b1210edb352c67 (I'm not too sure how much to add there, the comment above is a little bit more meta than one suitable for an in code comment in my opinion. Let me know what you think! Happy to make edits.)

https://github.com/llvm/llvm-project/pull/172686