I have a large matrix of size 10 Million × 1000 and I want to compute the result of its multiplication with a square 1000 × 1000 matrix in place.
Currently the level 3 BLAS provides *trmm functions which allow us to overwrite the input matrix during multiplication but only if one of the input matrices is triangular. There does not seem to be any such level 3 functionality when the input matrix is a general square matrix instead of triangular.
Now to do my inplace matrix multiplication, I can either do 10Million level 2 BLAS calls to matrix vector multiplication, or I can do a LU decomposition of A and then two level 3 BLAS calls to *trmm. I was interested in discussing if there was a third solution that I had overlooked.