Subject: [ublas] sparse matrix and its conjugate transpose
From: Sebastian Birk (birk_at_[hidden])
Date: 2010-08-18 06:02:45
I noticed a huge impact on the performance of a matrix-vector product
when computing something like
B = prod(herm(A),X);
for a sparse matrix A, which is not really surprising, if herm(A) is
formed explicitly. This seems to be exactly what uBLAS does.
Is there an efficient method and intuitive way to compute a sparse
hermitan matrix-vector product? In MATLAB it seems to make no difference
if A' or A is multiplied by X.
The obvious workaround would be to precompute herm(A) and pass it to
every function that needs to multiply with it. But this would bloat the
function calls and is not my preferred solution.
Thanks in advance.