Not for now, but we should check if we want to support operators having multiple outputs.

Better to do the clamping inside the update_memory_buffer_partial(it)
no need to iterate twice over the same data when the operators are small.
It makes sense when the operator itself are bigger.

Math operators branch are predictable and clamping is a predictable branch.

For convenience you could add a header only method that does the clamping (float clamp_when_enabled(float))