[pypy-dev] Slow int code

[Armin Rigo]

Hi Roger,

On Fri, Mar 1, 2013 at 10:13 PM, Roger Flores <aidembb at yahoo.com> wrote:
> I see both BINARY_XOR and BINARY_AND call a function instead of xor and and.  Why?  Is there something I can change in my code to let those instructions be used instead?  Can xoring two ints really cause an exception?

If it becomes calls to, I guess, rbigint.xor and rbigint.and, it's
because these are objects of the Python type "long".  PyPy must
respect the Python semantics: if you do for example "(1 << 80) >> 80",
the result is not 1, but 1L.  This said, I guess that the code is
expected to manipulate 32-bit unsigned integers, and so should not
overflow to long on 64-bit machines.  You may want to investigate, in
the normal Python way, why you get "long" objects there --- this is
not a particularity of PyPy; you can investigate on CPython if you
prefer.

If you're running this code on 32-bit machines, then no chance: you're
getting "long"s.  The only solution would be along the lines I
described before, with a W_TwoIntsLongObject, or even a
W_UnsignedIntLongObject that can store all long objects between 0 and
2**32-1.


A bientôt,

Armin.