AMGX
AMGX copied to clipboard
AMGx setup stage become very slow after I include all the zero in the input sparse matrix .
Dose AMGx provide a mechanism to exclude all the zero in the input matrix.
Can I just double check you are referring to an explicit CSR matrix, but where the data is fully dense?
For AmgX we expect that the user provides a sparse matrix and there isn't any functionality I know of that compresses the matrix. You could potentially use cuSPARSE (https://docs.nvidia.com/cuda/cusparse/index.html#prune-example1).
No, there is no mechanism for pruning input matrix zeros.