Re: convn with real and complex data

[Søren Hauberg]

fre, 28 03 2008 kl. 08:57 -0400, skrev John W. Eaton:
> On 28-Mar-2008, Søren Hauberg wrote:
> 
> | Seems to me like a lot of extra code to avoid the extra template
> | parameter.
> 
> Yes, I agree, but we might also move the traits class to some other
> location so that it could be used by more than just convn.  It seems
> common that most binary operations between real and complex arguments
> would have these argument and return types, and so might be written as
> templates using a traits class like this
I guess that would make sense.

> , and even the body of your
> F__convn__ function might be something we can generalize and share as
> I think that might be a fairly common structure in Octave.
The reason I implemented 'convn' was because I had just implemented
N-dimensional ordered filtering in the 'image' package. This problem
almost has the same structure as convolution, so I figured I might as
well implement convolution. So yes, this might be a fairly common
structure.

Søren