Re: [ublas] performance of sparse matrices

From: Gunter Winkler (guwi17_at_[hidden])
Date: 2005-03-14 07:22:11

On Monday 14 March 2005 12:07, Paul C. Leopardi wrote:
> Hi all,
> I have updated my test code and my output. Code is attached.
> Output and comments below. Perhaps someone could explain why my results
> should be expected to be different from Gunter's and why they should be the
> order that they are?
> Best regards

I don't think it is a good idea to change the matrix size and the number of 
non zeros simultaneously. These are two different things. So I tried to 
detect the particular dependencies and I found out that the size has much 
more influence than the nnz. But by design the size should have _no_ 
influence only the nnz!

PS: I'll check your numbers later. They look ok. But I think you should also 
try a few more nonzeros (2*size, 4*size, ...)
Btw. the time of sprod should depend on the sparsity pattern of both matrices.

mfg
Gunter