Update to Discrete Hankel Transform

[Felipe Attanasio]

Greetings,

After noticing the "FIXME" in line 205 of dht/dht.c, suggesting a 
potential optimisation, I looked further into it.
I have changed the size of the array Jjj (to size^2, instead of 
size*(size+1)/2 and re-arranged its entries, such
that it can be read sequentially when 'gsl_dht_apply' is called. This 
has the effect of reducing the number of
memory reads and thus makes 'gsl_dht_apply' run more efficiently. I've 
ran some tests with 'size = 4000' and
5000 and have seen a speed-up of 2-3 times compared to the original GSL 
implementation.
I've also ran 'make check' and no errors have been reported.

In order to perform the aforementioned re-arrangement, when 
'gsl_dht_init' is called a temporary array is
allocated and populated with the elements that would originally be in 
Jjj, then the code loops over the
temporary array and populates Jjj.

I am attaching to this email the patch containing the differences. Feel 
free to reach out to me if there are any
questions.

Cheers,

Felipe Attanasio

dht.patch
Description: Text Data