Hello,

I am new to uBlas and I would like to know if what I am doing make sense. Basically, I want to use uBlas with Intel MKL - I realize I could get Atlas for free but I believe MKL should be faster on Intel processors than Atlas.

This is what I do - when I can I use uBlas. When I need some functionality that is not in uBlas (like Cholesky decomp), I would do something like this:

const int dim = 3;
// need to make sure that the matrix is stored in row order
matrix<double,column_major> varCovar(dim, dim);
....
// calling Intel MKL function
dpotrf( &uplo, (int *) (&dim), (double *) &varCovar(0,0), (int *) (&dim), &info );
 
 
Does this seem like a reasonable approach ? Is there a better way ? What is all that talk about binding, what kind of advantage would it give me ?
 
Thank you,
Bumbrlik