Dear Andreas, Mohsen,
Thank you for the pointers. I will try the scikits approach as you
Andreas, PyCUDA looks like a great project. I saw one of your talks online
Andreas and it sparked my interest. At the moment the performance of my
code is not really an issue, but I anticipate I will want to spend a bit of
time looking into optimisation at some point in the near future. I was
just curious if significant speed-up was possible with relatively little
changes to the existing code. It seems this may be possible with scikits.
Although with my specific problem being sparse and tri-diagonal probably
better performance can be had using PyCUDA with CUSP or cuSPARSE wrappers
as you suggested.
So right now I don't have any specific questions regarding the SparseSolve
code, when I revisit this and have invested a bit of time learning CUDA I
Thank you for the help.
On 12 March 2013 00:56, Andreas Kloeckner <lists(a)informa.tiker.net> wrote:
"boyfarrell(a)gmail.com" <boyfarrell(a)gmail.com> writes:
Is it possible to solve a linear system of the
form Ax=d using PyCUDA?
The actual system I wish to solve is sparse and tridiagonal.
As long as it's positive definite and you don't mind installing pymetis
and Cython, the answer is yes.
That said, writing a CUSP wrapper is likely a saner approach than using
this code. It's undocumented for a reason. :)
I noticed the sparse module in the Docs,
, but I could
understand the code (I'm very familiar with Python, but not with
CUDA). Is there an example with comments or a tutorial would be even
better. At the moment my code is using scipy and numpy.
What about it is not making sense? If you've got a specific question,
I'd be happy to try and answer.