Implementation of convn

[Søren Hauberg]

Hi,
  Attached is a fairly simple implementation of 'convn'. The
implementation consists of two files '__convn__.cc' and 'convn.m'. The
'__convn__' function computes the valid part of the convolution, while
the 'convn' function (the user interface) pads the data with zeros,
depending on the 'shape' argument. This split keeps the code fairly
simple, but it isn't as fast as humanly possible (fast enough though
IMHO). Is this approach acceptable?
  I should mention that the current code is missing
    * documentation
    * some optimisations to handle when B is larger than A
I'll implement this later this evening, and resend the code.

Søren

__convn__.cc
Description: Text Data

convn.m
Description: Text Data