How does the performance (in GFLOP/s) for sparse-matrix by vector product compare to what you previously achieved for dense-matrix by dense-matrix product? Explain, and quantify if you can, (e.g., using the roofline model).
How does the performance (in GFLOP/s) for sparse-matrix by vector product for COO compare to CSR? Explain, and quantify if you can, (e.g., using the roofline model).
How does the performance (in GFLOP/s) for sparse-matrix by dense matrix product (SPMM) compare to sparse-matrix by vector product (SPMV)? The performance for SPMM should be about the same as for SPMV in the case of a 1 column dense matrix. What is the trend with increasing numbers of columns? Explain, and quantify if you can, using the roofline model.
How does the performance of sparse matrix by dense matrix product (in GFLOP/s) compare to the results you got dense matrix-matrix product in previous assignments? Explain, and quantify if you can, using the roofline model.