Thanks for being here! I am curious on the relationship between cuGraph and other CUDA libraries. My work now is highly dependent on cuSparse and cuSolver for geometry surface meshes.
Can cuGraph accommodate sparse adjacency matrix as input? Can I use it as an extension of cuSparse?
cuGraph can be compatible with cuSparse and cuSolver. We can accept COO and CSR/CSC inputs and execute graph algorithms against those structures. However, we don't integrate directly with these packages. Our internal format is optimized for our graph primitive library, so we make a copy of the cuSparse and cuSolver input in order to execute these algorithms. Similarly, our output (when appropriate) can be converted into a COO or CSR structure that can be passed back to cuSparse and cuSolver.