On 09/22/2011 09:23 AM, Tarek Elsayed wrote:

Hello,
I noticed that ublas is very slow in sparse matrix matrix and matrix vector multiplications compared to Mathematica, Eigen and MTL4.

Well I recently did some tests on systems 20748, 35775, 115155, and 220k and uBlas does not seem slow to me at least it is faster than MATLAB( which is hell optimized for architecture and related) for sparse matrix-vector multiplication. Do you use DNDEBUG flag as well, which could make a significant difference?

Matrix-matrix, dunno ...