On 12/06/15 00:48, Sturla Molden wrote:
> The code below did not make any difference to SciPy, neither faster nor
> slower, so I am not going to use it. But if you have a lot of different
> distance metrics, it will still be about as fast as inlining the
> distance metric callback at compile time.
Note that computed goto has a special kind of hardware support ("jump
through register"), unlike other means of dynamic dispatch (e.g. a
switch statement or if/else blocks). This makes it generally faster than
other means of dynamic dispatching. This non-standard GNU extension is
the only way of using this hardware functionality without writing
assembly code for it.
Intel has rewritten the CPython interpreter to use computed gotos
instead of a big switch to dispatch on the Python bytecodes. It is
already in Python 3 and coming to Python 2.7.
Sturla
------------------------------------------------------------------------------
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
