[Mlir-commits] [mlir] [mlir][bufferization] Use TensorLike, BufferLike type interfaces (PR #136736)

[Andrei Golubev]

================
@@ -1386,3 +1387,26 @@ TestMultiSlotAlloca::handleDestructuringComplete(
     const DestructurableMemorySlot &slot, OpBuilder &builder) {
   return createNewMultiAllocaWithoutSlot(slot, builder, *this);
 }
+
+::mlir::LogicalResult test::TestDummyTensorOp::bufferize(
+    ::mlir::RewriterBase &rewriter,
+    const ::mlir::bufferization::BufferizationOptions &options) {
+  const auto inType = getInput().getType();
+  const auto bufferizedInType = test::TestMemrefType::get(
+      getContext(), inType.getShape(), inType.getElementType(), nullptr);
+  const auto outType = getOutput().getType();
+  const auto bufferizedOutType = test::TestMemrefType::get(
+      getContext(), outType.getShape(), outType.getElementType(), nullptr);
+
+  // replace op with memref analogy, preserve correct types at the boundaries
+  auto toMemref = rewriter.create<::mlir::bufferization::ToMemrefOp>(
+      getLoc(), bufferizedInType, getInput());
+  auto dummyMemrefOp = rewriter.create<test::TestDummyMemrefOp>(
----------------
andrey-golubev wrote:

self-review: to memref / to tensor calls must come from the infrastructure itself. here, one should just create the `TestDummyMemrefOp`?

https://github.com/llvm/llvm-project/pull/136736