Hi all
I'm trying to solve a large scale sparse least square problem by using normal equations A'Ax=A'b
And i am stuck with constructing the matrix A'A .
I'm using sparse blas function mkl_dcsrmultcsr for A'A(is this the right choice ?)
the problem is that i have to pre-allocate the resulting matrix in memory but this is too large.
what are the solutions in this situation ?
i remember there is a techinque for not constructing the A'A explicitely. but i can't find it anymore.
otherwise is there a mkl sparse function that multiplies two sparse matrix and return another sparse matrix ?
thanks in advance
Q. otherwise is there a mkl sparse function that multiplies two sparse matrix and return another sparse matrix ?
-The mkl_dcsrmultcsr takes two sparse matrces as input and returns a sparse matrix.
