SpMV_Struct_Benchmarks

Structured SpMV algorithms performance

This page gathers instructions and references to conduct performance tests of the spmv_struct algorithms in Kokkos-Kernels.

GPU performance

K80

The first table shows the comparison of performance on 2D problems using FD (5pt) and FE (9pt) stencils

Matrix size	FD Struct	FD Unstr	FE Struct	FD Unstr
50	0.03615	0.02654	0.03738	0.02720
100	0.04266	0.03411	0.04668	0.04069
150	0.05323	0.05426	0.06753	0.07118
225	0.07749	0.09979	0.10876	0.12682
300	0.10972	0.15491	0.16904	0.20675
400	0.16881	0.25213	0.26955	0.33847
1,000	0.88039	1.42716	1.55607	1.86464
2,000	3.30934	4.35175	5.36563	5.63246

P100

The first table shows the comparison of performance on 2D problems using FD (5pt) and FE (9pt) stencils

Matrix size	FD Struct	FD Unstr	FE Struct	FD Unstr
50	0.02639	0.02654	0.02062	0.01521
100	0.02071	0.01825	0.02106	0.01716
150	0.02959	0.04015	0.02340	0.02078
225	0.02780	0.02477	0.03321	0.05163
300	0.03654	0.06123	0.04925	0.07147
400	0.04875	0.09654	0.06846	0.12177
1,000	0.20233	0.53389	0.32596	0.66484
2,000	0.74472	2.10274	1.23610	2.48697

V100