If single sparse matrix used in cusparseSpMV(), but with multiple dense vectors x and y, that are obviously of the same size, is it OK to reuse auxiliary buffer of size calculated through cusparseSpMV_bufferSize(), as well as vector descriptors created through cusparseCreateDnVec(), before first cusparseSpMV call? It seems logical to me that it would be OK (this all should depend only on sparsity structure of given matrix, that doesn’t change), and more efficient this way. However, cusparseCreateDnVec() expects vector elements to be passed through its third argument, so this confuses me.
Thanks.