[Mlir-commits] [mlir] [MLIR] Setting MemorySpace During Bufferization + Fixes (PR #78484)

[Matthias Springer]

================
@@ -1622,7 +1623,20 @@ CollapseShapeOp::inferCollapsedType(RankedTensorType type,
     currentDim += dim;
   }
 
-  return RankedTensorType::get(newShape, type.getElementType());
+  auto encoding = type.getEncoding();
+  if (auto v = encoding.dyn_cast_or_null<VerifiableTensorEncoding>()) {
+    auto ignoreError = [&] {
+      auto emitter = mlir::emitError(UnknownLoc::get(type.getContext()));
+      emitter.abandon();
----------------
matthias-springer wrote:

I'm not sure about this part. This looks like a hack and just removing the encoding could be problematic. Can you put this (and all encoding-related changes) in a separate PR? Add @aartbik to the new PR, the sparse team heavily depends on tensor encodings, they may have an oppinion about this.


https://github.com/llvm/llvm-project/pull/78484