[GRASS5] Re: Numercial stuff in GRASS

Glynn Clements glynn.clements at virgin.net
Sat Jul 7 11:46:20 EDT 2001


Markus Neteler wrote:

> Glynn, please keep the modules above working, even if LAPACK is not present,
> I hope that's possible not to have LAPACK being a strict dependency.

OK. The main issue is how to handle FFTW; either we make it a
dependency, or we bundle it. Note that FFTW is a lot more involved
than the existing FFT code; it contains multiple FFT algorithms, and
selects whichever one is likely to be fastest.

A couple more points about the gmath library generally:

1. G_matrix_{get,set}_element should probably be replaced by macros,
to enable a compiler to perform "loop induction" optimisation
(basically, eliminating the multiplication when accessing elements
within a loop with a fixed increment).

2. It may be worth also having an index array (an array of pointers to
the start of each row) to allow random access without the need for
multiplication.

Also, there seems to be a fair amount of matrix code in libgis, e.g. 
G_tred2, G_tqli (used by gmath/eigen.c), G_ludcmp, alloc/free stuff in
gis/dalloc.c, maybe other stuff.

-- 
Glynn Clements <glynn.clements at virgin.net>



More information about the grass-dev mailing list