Hello !

I've a new question about eigen

I use band matrix and trigonal matrix and I see that there is nothing is Eigen for these matrix. However can I use the blas function like dgttrs to use the fact that the matrix has a special form. Or is there another solution ?

Thank you !

PS : Congratulation for the Beta 1 of the 3.0, I will test it soon
PS2 : I see that you try to make some parallelisation of the code ? how can i enabled that ?

In Eigen3 we have a BandMatrix class with a BLAS/Lapack compatible storage. However, currently there is only the storage class, and operations such as products have still to be implemented. Since it BLAS compatible, you should be able to use blas/lapack functions working on banded matrices with it.

Note that dgttrs is for general (dense) storage. It is equivalent to Eigen's partial LU decomposition.

For parallelization, simply compile your program with openmp, e.g., -fopenmp for gcc. Note that only a very few features will be accelerated (dense matrix-matrix products, partial LU and LLT).

I have looked a bit but I've not found a page explained how to use the BLAS prototype ? I have to install these by myself or it's already ok ?

Yes I know the storage for dgttrs I have even the fortran code...

And for the the beta version it works great for the matrix matrix product with openmp on Visual Studio 2005 I've 8 processors and I need only 3 seconds to make a 3500*3500 product ! versus 12 seconds for only 2500*2500 with eigen 2 (and 45 seconds for 2500*2500 with my old librairy ^^)