Subject: Re: [ublas] about symmetric matrix handling in bindings
From: Nizar Khalifa Sallem (nksallem_at_[hidden])
Date: 2010-11-08 10:18:38
At Mon, 08 Nov 2010 16:01:43 +0100,
Rutger ter Borg wrote:
> On 11/08/2010 03:28 PM, Nizar Khalifa Sallem wrote:
> > hi all,
> > I have some questions regarding the current binding state of blas and
> > lapack.
> > I can't find any symmetric matrixes special product in the new
> > blasX.hpp ( with X[1-3] ;) ) and I was wondering if the general
> > purpose products i.e. gemm, gemv and ger can handle
> > boost::symmetric_matrix in an optimized fashion or not?
> > Best,
> > --
> > Nizar
> Hello Nizar,
> algorithms geared towards symmetric matrices usually start with 'sy'.
> Please recheck the blas/levelN/ subdirectories.
> ublas mailing list
> Sent to: nksallem_at_[hidden]
This is exactly the point: I am running the last svn checkout of
boost-sandbox and when checking boost/numeric/bindings/blas/blas3.hpp
for instance you can only find gemm and the only symmetric special is
syrk similarly when you have a look to
boost/numeric/bindings/blas/blas2.hpp you only find gemv and ger.
The sy** doesn't exist which seems to be really curious for me since
all of the sy** stuff do exist inside blas.h.
That is why I asked if there is some magic somewhere else that do
detect the matrix type (i.e. symmetric or triangular for instance) and
call the appropriate routines.
I hope my message is clearer
-- Best, Nizar