{}, indices_type.getElementType()));
    // Note on zero_attr integer type: DynamicUpdateSlice op start_indices are
    // required to have matching types. This rewrite rule creates
    // DynamicUpdateSlice ops where the first "start index" is always i32 and
    // subsequent ones are constructed based on zero_attr. Thus the type

  // Assert that the tensor list operation is lowered to something that doesn't
  // get stuck on a broken dynamic update slice.
  ASSERT_THAT(compilation_result,
              Not(ComputationProtoContains("%.*=.*DynamicUpdateSlice")));
}

TEST(LegalizeTFTest, LegalizesFunctionWithBoundedDynamicArg) {
  static constexpr char kMlirModuleWithBoundedDynamicArgStr[] = R"(

        op, input.getType(), scalar_zero,
        ArrayRef<Value>({slice1, expanded_item, slice2}));
    return success();
  }

  // This function rewrites the original op into a XLA DynamicUpdateSlice op.
  // |item| is expanded to have the same dimension as input_handle and
  // |index| is expanded to [index, 0, 0, ...] as the indices to input_handle.
  // On a high level, it's doing something like:

}

def TF_XlaDynamicUpdateSliceOp : TF_Op<"XlaDynamicUpdateSlice", [Pure, TF_NoConstantFold]> {
  let summary = "Wraps the XLA DynamicUpdateSlice operator, documented at";

  let description = [{
https://www.tensorflow.org/performance/xla/operation_semantics#dynamicupdateslice
.

XlaDynamicUpdateSlice generates a result which is the value of the `input`

        ML authoring is generally discouraged.
    *   Add support for unsigned 16-bit integer tensor types in cast op.
    *   Experimental support for lowering `list_ops.tensor_list_set_item` with
        `DynamicUpdateSlice`.
    *   Enabled a new MLIR-based dynamic range quantization backend by default
        *   The new backend is used for post-training int8 dynamic range
            quantization and post-training float16 quantization.