Question

3. Compare CSR (Compressed Sparse Row) VS CSC (Compressed Sparse Column) implementation in case of GPU (CUDA). What are the advantages and disadvantages of each format for this problem? Wherever applicable, use hardware counters to back your claim.

Answer 1

CSR and esc a da Compressed Sparse row (CSR) is the most frequently used format to store sparse matrices However, CSR - based Spmv on graphics Processing units (GPUs) has poor performance patterns, due to irregular memory access load imbalance and reduced parallelism, The is a Compressed sparse Raw general sparse matrix CSR) Format format. CSR format consist of three array: row.ptr, column of non zeroes and matrin values. csc is the transpose Counterpart of the Compressed Sparces row format The Csc format is defined analogously to the CSR format, but sorts the entries along the columns instead of the rows.

Advantages and disadvantages The CSR and Cso format reduces the memory requirements at the cost of less flexibility in terms of storage order Instead of storing row ide r each entry CSR enforces and ordering of the values to according the rows such that all entries of a given row lie consecutively in memory . Thus, the raw ids can be replaced by pointers to the first element in each row. Starina a matrix in CSR format reduces the momory requirement by nnz - ninteorol compared to the Coo format. i Ese format ante Storing a matrin in csc format to storing the transposed matrix in format CSR