proposed implementation of interpn

[Alexander Barth]

Hi all,

I wrote an implementation of interpn for n-dimensional interpolation.
It supports only linear interpolation. I tweaked the C/C++ code to
optimize the speed and it runs now pretty fast.
For large problems (see benchmark_interpn.m) it is about 9 - 10 times
faster than matlab R2006a (on an AMD opteron; both binaries are
64-bit). OK, I didn't check if matlab's interpn is implemented as a
script. This is probably the case and explains the difference in
speed.

$ octave -q benchmark_interpn.m
Elapsed time is 0.523849 seconds.
ans =  0.25853

$ matlab64 < benchmark_interpn.m
Elapsed time is 5.527463 seconds.
ans =
   0.2585

I noticed also that C-arrays can be significantly faster than octave's
NDArray. Does NDArray performs bound-checking per default?

Cheers,
Alex

benchmark_interpn.m
Description: Text Data

interpn.cc
Description: Text Data