Have you considered using type parameters to indicate the symmetry of the
filter object? That would allow you to use multiple dispatch to make a
separate method for each kind of symmetry.

-- Leah


On Sun, Jul 20, 2014 at 10:03 AM, Jay Kickliter <[email protected]>
wrote:

> I'm writing some polyphase resampling
> <http://nbviewer.ipython.org/github/jaykickliter/Notebooks/blob/master/Polyphase.ipynb>
> code. Sometimes there are symmetries that cut down on the number of
> required multiplies. I'm still learning how to use Julia's metaprogramming
> facilities, but assuming it's possible to recognize the symmetry and
> generate some optimized code for a particular instance of a filter object,
> how would I execute it? The filters will be composite types that store
> previous state, filter taps, resampling rations, and other information. Is
> it possible to store the code in the object itself, and execute it
> efficiently?
>
> This more of a curiosity. If I could get the code to actually compile to
> vectorized SIMD code, the "optimization" might be slower.
>
> Thanks!
>

Reply via email to