Re: [ublas] atlas::potrf and symmetric_matrix

From: Kresimir Fresl (fresl_at_[hidden])
Date: 2005-09-26 09:18:15

Karl Meerbergen wrote:

> On Thursday 22 September 2005 16:41, Nico Galoppo wrote:

>> the documentation of atlas::potrf(SA) suggests using symmetric_adaptor<>(A)
>> to let potrf() know which part of the matrix (upper,lower) is being
>> represented. But what about if I don't want to store the original matrix A
>> as a dense matrix? Say i have a symmetric_matrix<> A, shouldn't I be able
>> to call atlas::potrf(A)?

> potrf assumes a matrix has dense storage, for computational efficiency 
> reasons.
> 
> For the packed format, there is pptrf()
> 
> Bye the way, it is lapack::potrf() and not atlas, and the matrix should be a 
> hermitian_adaptor<>(A) and not symmetric_adaptor<>(A) (except when A is 
> real).

Well, yes and no. Atlas and Atlas bindings contain few LAPACK routines: 
gesv, getrf, getrs, getri and posv, potrf, potrs, potri (see 
http://tinyurl.com/5ty8f).

Regards,

fres