Re: [ublas] [bindings][blas] Included ATLAS CLAPACK support

Subject: Re: [ublas] [bindings][blas] Included ATLAS CLAPACK support
From: Rutger ter Borg (rutger_at_[hidden])
Date: 2010-01-11 06:56:57

Thomas Klimpel wrote:

> 
>> Or do you mean CLAPACK, or some other part?
> 
> When I say "and the cblas part is definitively the bulk of them", I refer
> to the CLAPACK part of ATLAS (which is different from clapack, which is
> just the polished f2c version of lapack)
> 
> Regards,
> Thomas

I've just committed support for ATLAS's CLAPACK in the LAPACK bindings. 
Activate with

#define BOOST_NUMERIC_BINDINGS_LAPACK_CLAPACK

and link against the appropriate library (in my case, g++ ... -llapack, or 
g++ ... -latlas_lapack). The lapack regressions ublas_gesv and ublas_posv 
both pass for both back-ends.

Notable changes to all lapack bindings is the propagation of the info 
result. This used to be passed by reference, right now it is being returned 
by the overloaded functions. This is because of integer type used by fortran 
(fortran_int_t), which is not defined / applicable when the ATLAS lapack 
routines are selected. Conversion of the return type (if applicable) now 
occurs in the overloads as well.

Still to do for finalize CLAPACK support:

* Dynamically define the Order of the function to be called (should be easy 
given the examples in blas generator)
* Figure out the exceptional case getri, which needs a workspace in the 
Fortran case, but not in the clapack case.

Questions and/or comments are welcome.

Cheers,

Rutger