[Mlir-commits] [mlir] [mlir][tensor] `tensor.generate`: do not verify dynamic sizes (PR #74568)

[Matthias Springer]

================
@@ -2018,3 +2018,23 @@ func.func @invalid_slice_ops(%t: tensor<?xf32>, %t2: tensor<?xf32>) -> tensor<?x
   %1 = tensor.insert_slice %0 into %t2[2][%c][1] : tensor<?xf32> into tensor<?xf32>
   return %1 : tensor<?xf32>
 }
+
+// -----
+
+// Negative dynamic dimension sizes should not be turned into static sizes.
+// They should also not prevent the op from verifying.
----------------
matthias-springer wrote:

Personally, I'd be happy to see more comments in unit test cases. We have quite a few of those in the bufferization dialect unit tests.


https://github.com/llvm/llvm-project/pull/74568