Re: [Help-gsl] About invalid parameters in cblas implementation

[Brian Gough]

At Sat, 4 Jul 2009 21:46:39 +0200,
José Luis García Pallero wrote:
> I don't understand clearly the point (ii). It can be implemented a
> tuned cblas_xerbla() for the cblas functions in gsl, but when you
> link your program with other blas implementation, the called xerbla
> won't be the gsl one. 

Yes that's true, but I think there's nothing we can really do about
that.  Other libraries probably have their own way to set the behavior
of xerbla.  We do need to provide the user with a way to override the
abort() in xerbla in the GSL CBLAS at least.

> About the gsl_cblas* wrapper, all test errors must be implemented
> for a correct error code return in all cases. The problem (probably
> not is a real problem) is that when gsl_cblas* is called the error
> checking is performed twice: in the gsl_cblas* routine and in the
> final cblas_* function.

I don't think it's a major problem if the checking is done twice, the
overhead of checking should still be small for any non-trivial
operation.

>  I can try to implement the error checking for cblas_* functions in
> my spare time, but problably will be a slow task.

Given the similarity between the functions, perhaps it is possible the
generate some of the checking code automatically with a script or at
least reduce it to a few standard C macros.

-- 
Brian Gough
(GSL Maintainer)

Support freedom by joining the FSF 
http://www.fsf.org/associate/support_freedom/join_fsf?referrer=37