[SciPy-User] RZ Factorization

[Sturla Molden]

Thomas Unterthiner <thomas_unterthiner at web.de> wrote:

> I also often to run into the situation of needing BLAS/LAPACK 
> functionality that is not exposed by SciPy. It would be nice if more/all 
> the calls would be wrapped. However, I find it very annoying that the 
> f2py wrappers often silently copy and/or transpose data. I understand 
> that this is a F-order vs C-order issue, but sometimes e.g. I know that 
> my matrices are PSD, yet f2py will copy/transpose anyway. Such behavior 
> is annoying and often makes the SciPy wrappers useless for my purposes.

That is one of the reasons I prefer my own LAPACK wrappers.


> Which I think is fairly straight forward. I think it would be easily 
> doable to wrap all of LAPACKE/CBLAS this way (thus getting rid of C- vs 
> F-order, since that can be specified by the user) and replace the 
> current f2py wrappers with that. If time permits and there is enough 
> interested, I could see if I can hack together a prototype/PR for this.

LAPACKE has the same issue as f2py. It creates copies and transposes C
order arrays.


Sturla