py/runtime: Remove unnecessary smallint handling.

[Jongy]

This logic is done in mp_obj_new_int anyway. Removing this decreases
code size.

[dpgeorge]

I think the reason it was originally done the way it is is for performance: this is a very hot code path for small-int arithmetic (which accounts for a lot of the int arithmetic) so benefits from the small optimisation done here. If anything, the mp_obj_new_int() call could be replaced with mp_obj_new_int_from_ll(), which essentially does what the comment suggests, namely to inline the mp_obj_new_int() call.

[dpgeorge]

If anything, the mp_obj_new_int() call could be replaced with mp_obj_new_int_from_ll()

I've now done that in e94d644 . So this PR can be closed, but thanks for prompting this nice change!

[Jongy]

Hm. The code duplication remains with mp_obj_new_int, so maybe it could be defined as inline and be inlined here and in every objint implementation.

[dpgeorge]

The code duplication remains with mp_obj_new_int, so maybe it could be defined as inline and be inlined here and in every objint implementation.

Yes it is duplicated. But it would likely increase code size by a lot if inlined unconditionally everywhere. The point with this particular case in runtime.c:mp_binary_op is that this is the primary code that is executed whenever you do any arithmetic on small integers. This is quite a common thing to do, so worth spending a few bytes of code to make it faster. And not really worth spending the bytes in all the other places that it's used.

[Jongy]

Well, of course not to inline all invocations, but perhaps a inline __mp_obj_new_int and then mp_obj_new_int just calls it, so it is inlined there and in mp_binary_op.

[dpgeorge]

perhaps a inline __mp_obj_new_int and then mp_obj_new_int just calls it, so it is inlined there and in mp_binary_op.

Feel free to try it out and submit a PR so it can be discussed in detail. But you may be interested in first reading #293.