[Mlir-commits] [mlir] [MLIR] Preserve Encoding During TensorOp Creation (PR #80871)

[Mehdi Amini]

================
@@ -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>()) {
----------------
joker-eph wrote:

You're basically saying we should do:

> 1. We don't do the transformation (because we risk making an incorrect transformation)

I don't disagree, I'm not sure we have a good documentation of the intent with respect to this tensor encoding.
(many folks, including me, were a bit concerned when this was introduced that we may be better introducing new types because this seems incredibly unsafe and it makes the codebase hard to get right: this is where we're at now)

https://github.com/llvm/llvm-project/pull/80871