Hello,
What is the best matrix format for the adjacency matrix of a power-law graph when we perform sparse matrix vector multiplication? I would like to use a format that has both memory efficiency and performance efficiency. CSR and CSC are both memory efficient, but these formats will cause many CPU cache misses in multiplication when the matrix is very large. BSR doesn't seem very memory efficient for the adjacency matrix of a power-law graph. Ellpack sparse block (ESB) seems a much better format (https://software.intel.com/en-us/articles/the-intel-math-kernel-library-...), but I can't find it in the document (https://software.intel.com/sites/products/documentation/doclib/iss/2013/...). Is CSR or CSC the only best option for an adjacency matrix?
Thanks,
Da