Sparse matrix-vector multiplication benchmark
Figure 9 shows the performance achieved on the final benchmark, sparse matrix-vector multiplication. The ordering of the results is the same as for the previous two benchmarks. Even though this is a nested data-parallel algorithm that uses segmented VCODE operations, the shapes of the graphs and the performance ratios are similar to those for the non-nested line-fit benchmark, which uses mostly unsegmented operations. Note that there is less variance in the results than for line-fit because sparse-matrix vector multiplication uses fewer temporary vectors, and hence less garbage collection occurs.
Figure 9: Performance of NESL sparse matrix-vector multiplication benchmark using different intermediate language implementations on a 120 MHz 486 (left) and a SPARCstation 5/85 (right).