[Mlir-commits] [mlir] [mlir][bufferization] Introduce castBufferTypeFn hook and API (PR #205080)

[Matthias Springer]

================
@@ -100,6 +100,20 @@ LogicalResult mlir::bufferization::foldToBufferToTensorPair(
     return success();
   }
 
+  if (!llvm::isa<BaseMemRefType>(srcType) ||
----------------
matthias-springer wrote:

`foldToBufferToTensorPair` is called from a canonicalization pattern. I wouldn't create a new alloc here. The current implementation looks fine: just propagate the error.

https://github.com/llvm/llvm-project/pull/205080