How can I most efficiently multiply two matrixes together when I know it will produce a symmetric matrix?

25 Views Asked by At

I'm working with a number of matrix libraries (cuBLAS, OpenBLAS, and MKL BLAS) and am wondering if, in any of these matrix libraries, there exist optimized matrix multiplication functions for when the result is known to be symmetric. It seems there are lots of functions optimized for when an input matrix is symmetric (such as cblas_dsymm), but it seems like one could also get a big speedup when the output matrix is known to be symmetric, since roughly half of the elements don't require any multiplications, they can just be copied from their symmetric counterpart in the other triangle. Does anyone know if such a function exists anywhere?

0

There are 0 best solutions below