[Mlir-commits] [mlir] [mlir][bufferization] Introduce reconcileBufferTypeMismatchFn hook (PR #202667)

[Matthias Springer]

================
@@ -576,27 +557,24 @@ static FailureOr<BufferLikeType> computeLoopRegionIterArgBufferType(
     return yieldedValueBufferType;
 
   // If there is a mismatch between the yielded buffer type and the init_arg
-  // buffer type, the buffer type must be promoted to a fully dynamic layout
-  // map.
-  auto yieldedBufferType = cast<BaseMemRefType>(yieldedValueBufferType);
-  auto iterTensorType = cast<TensorType>(iterArg.getType());
-  auto initBufferType = llvm::cast<BaseMemRefType>(*initArgBufferType);
-  if (initBufferType.getMemorySpace() != yieldedBufferType.getMemorySpace())
-    return loopOp->emitOpError(
-        "init_arg and yielded value bufferize to inconsistent memory spaces");
+  // buffer type, the buffer type must be reconciled.
 #ifndef NDEBUG
-  if (auto yieldedRankedBufferType = dyn_cast<MemRefType>(yieldedBufferType)) {
-    assert(
-        llvm::all_equal({yieldedRankedBufferType.getShape(),
-                         cast<MemRefType>(initBufferType).getShape(),
-                         cast<RankedTensorType>(iterTensorType).getShape()}) &&
-        "expected same shape");
+  if (auto iterTensorType = dyn_cast<TensorType>(iterArg.getType())) {
----------------
matthias-springer wrote:

Same comment about the `NDEBUG` section as above. It would be great to keep some of this error checking, as it could guard against incorrect interface method implementation from the user's side. But I'm not sure how to do it. If you can't find a way, let's just delete it.


https://github.com/llvm/llvm-project/pull/202667