[Axiom-developer] Blas in axiom (was Re: Axiom trunk failure)

[Vanuxem Grégory]

Hi Tim,

Le jeudi 04 mai 2006 à 00:59 -0400, root a écrit :

[...]

> i've joined axiom to the numerical mathematics consortium
> (http://www.nmconsortium.org/FldRte/?id=72&page=Associate+Members)
> because i have an interest in recovering the numerical library
> facility for axiom. i have rewritten the BLAS library into literate
> form and have gotten permission from a BLAS person to use his research
> work as documentation for the routines. i'm in the process of
> documenting the BLAS code now. when it completes i'll release a
> numeric library for axiom that is literate BLAS. then i'll move on to
> the next numerical piece. along the way i'm learning about sensitivity
> analysis, methods of graphing branch cuts, etc. and trying to add what
> i've learned to the documentation for the code.

Humm... I have a lot of questions about your work on a Blas library.

Is it coded in spad ? Is the interface fixed ?  If so did you use the
naming and calling convention of Blas ? For example the
<grep>[s|d|c|z]axpy</grep> blas routine are named axpy in your
implementation ? Same question apply for routines arguments
(<grep>inc.</grep> for vector and <grep>ld.</grep> for leading dimension
of matrices). Is the Float domain supported or only the DoubleFloat one?

In fact if I have all these questions it's because my version of Axiom
contains an interface to Blas and part of Lapack libraries but only for
reals (i.e DoubleFloat). This is probably a different implementation
since, in my work, the interface is coded in C and Lisp and needed a
complete new <grep>\<.*Matrix.*\></grep> tree in spad. It needs working
Blas and Lapack libraries too, either in C or in Fortran (in my case GCL
is linked to AMD's ACML).

The complex case is more difficult to solve (the internals of complex
long-float in GCL does not facilitate this task) and I actually don't
know if I will add an interface to it or how I will achieve this goal.
So if you can briefly explain your work I would be very happy.

I worked on this several months or years ago and re-begin to work on it
recently.


And yes, I know, I'm too curious.

Cheers,

Greg

PS : My implementation still require a lot of work but there is always
work to be done.