Boost logo

Ublas :

From: Paul C. Leopardi (paul.leopardi_at_[hidden])
Date: 2008-01-11 18:19:43

On Sat, 12 Jan 2008, nisha kannookadan wrote:
> Hi
> The feedback was real helpful.
> > Another solution is to use sparse storage, if your matrix sparse.
> > According to my experience with ublas and linear solvers, sparse storage
> > may be faster than dense storage even for small sizes (n=20) and low
> > percentage of non-zeros (60%). In my simulation software based on ublas,
> > sparse computations run 10-100 times faster than Matlab sparse
> > computations.
> I tried to use sparse matrix, but it made it for some reason even slower,
> so I discarded it. ANd I guess the vectors are the problem, which are quite
> full anyways.

As of January 2007 there was a known problem with addition of both uBLAS
compressed and coordinate matrices which made the time complexity far worse
than it could be. See timings here:

Start of thread is here:

Original test code is here

Best, Paul