[Mlir-commits] [mlir] [mlir][bufferization] Refine tensor-buffer compatibility checks (PR #167705)

[Andrei Golubev]

================
@@ -569,11 +569,17 @@ TestTensorType::getBufferType(
 ::mlir::LogicalResult TestTensorType::verifyCompatibleBufferType(
     ::mlir::bufferization::BufferLikeType bufferType,
     ::llvm::function_ref<::mlir::InFlightDiagnostic()> emitError) {
-  auto testMemref = dyn_cast<TestMemrefType>(bufferType);
-  if (!testMemref)
-    return emitError() << "expected TestMemrefType";
+  if (auto testMemref = dyn_cast<TestMemrefType>(bufferType)) {
+    const bool valid = getShape() == testMemref.getShape() &&
+                       getElementType() == testMemref.getElementType();
+    return mlir::success(valid);
+  }
+
+  if (auto builtinMemref = dyn_cast<MemRefType>(bufferType)) {
+    const bool valid = getShape() == builtinMemref.getShape() &&
+                       getElementType() == builtinMemref.getElementType();
+    return mlir::success(valid);
+  }
 
-  const bool valid = getShape() == testMemref.getShape() &&
-                     getElementType() == testMemref.getElementType();
-  return mlir::success(valid);
+  return emitError() << "expected MemRefType or TestMemrefType";
----------------
andrey-golubev wrote:

this specifically - not really.
the bulk of the change is for checking for builtin memref and for test memref now (this is what the test above exercises). the emit-error message is changed just to better align with the new logic (so that the message is non-ambiguous in case someone will start writing new tests on top, etc. etc.).

https://github.com/llvm/llvm-project/pull/167705