[SciPy-user] Calling C code (Newbie Question)
Robert Kern
rkern at ucsd.edu
Tue Sep 28 00:09:01 CDT 2004
Tom Kornack wrote:
[snip]
> If you would like to see just what kind of computation is being done,
> the C and python codes can be found here:
>
> http://listera.org/pub/lomb/
>
> The main loops look like this, where om and time are arrays:
>
> for i in range(numf):
> s2[i] = sum( sin(2.*om[i]*time) )
> c2[i] = sum( cos(2.*om[i]*time) )
>
> and I tried to speed things up by doing:
>
> s2 = sum( sin(2. * outerproduct(om,time) ), 1)
> c2 = sum( cos(2. * outerproduct(om,time) ), 1)
>
> but found that was too memory intensive and still slow. So that's why I
> wanted to do something purely in C.
If you want to do things in C, look into the SciPy package called weave.
It allows you to inline C code. If you are reasonably good at FORTRAN,
I've found that writing a small FORTRAN routine and wrapping it with
F2PY is often easier than using weave when dealing with Numeric arrays.
Another benefit of using FORTRAN this way is that one can easily use
LAPACK or BLAS subroutines.
Another thing you might want to try in pure Numeric is using dot() as
follows:
Your code:
sh = sum( cn*sin(outerproduct(om,time) ), 1)
My code:
sh = dot(sin(outerproduct(time, om)), cn)
That is what I did when I implemented a version of the Lomb-Scargle
periodogram. If you have ATLAS or some other optimized BLAS installed,
you can install Numeric such that dot() uses BLAS. You may still have
memory problems, though. To reduce time at the expense of memory, you
can compute phase=outerproduct(time,om) once and call sin() and cos() on
phase.
<silly_suggestion>
cossin = power.outer(exp(1j*time), om)
projection = dot(cossin, cn)
ch = projection.real
sh = projection.imag
</silly_suggestion>
--
Robert Kern
rkern at ucsd.edu
"In the fields of hell where the grass grows high
Are the graves of dreams allowed to die."
-- Richard Harter
More information about the SciPy-user
mailing list