Re: [ublas] lapack bindings: ppsv returns info=2

Subject: Re: [ublas] lapack bindings: ppsv returns info=2
From: Rutger ter Borg (rutger_at_[hidden])
Date: 2011-04-22 16:59:53

On 2011-04-22 22:50, George Slavov wrote:
> I'm not. The matrix is positive-definite, which should allow me to use
> ppsv, right? Octave can find the Cholesky decomposition of this matrix.
> One exists. I tried using the LAPACK routine for Cholesky, but it also
> errored out with info = 2.
>
> George
>

You're right, the pp-prefix should be fine in this case. I sounds like 
something isn't passed right -- it has been a while since I've checked 
the details of packed storage, but have you tried column_major ordering 
for the symmetric matrix?

ublas::symmetric_matrix<double, ..., column_major >

If this happens to make things work, it is a bug in the bindings. In 
that case, it sounds like it is missing a static assertion / check for 
this case.

Cheers,

Rutger