Research Article
Efficient CSR-Based Sparse Matrix-Vector Multiplication on GPU
Table 3
Speedups of IPCSR over other implementation techniques in single precision.
| Matrix | Speedups on C2050 | Speedups on K20c | CUSPARSE | CUSP | clSpMV | PCSR | CSR5 | CSR-Adaptive | CUSPARSE | CUSP | clSpMV | PCSR | CSR5 | CSR-Adaptive |
| Dense | 1.04 | 1.00 | 3.16 | 3.30 | 1.05 | 1.03 | 1.24 | 1.03 | 3.09 | 3.25 | 1.07 | 1.04 | Protein | 1.24 | 1.01 | 1.22 | 1.47 | 1.08 | 0.97 | 1.55 | 1.21 | 1.22 | 1.48 | 1.15 | 1.00 | FEM/Spheres | 1.79 | 1.34 | 1.04 | 1.43 | 1.27 | 1.00 | 2.24 | 1.61 | 1.04 | 1.44 | 1.35 | 1.07 | FEM/Cantilever | 1.71 | 1.31 | 1.09 | 1.48 | 1.09 | 1.00 | 2.14 | 1.65 | 1.09 | 1.50 | 1.16 | 1.08 | Wind Tunnel | 1.98 | 1.23 | 0.95 | 1.06 | 1.04 | 1.00 | 2.49 | 1.59 | 0.95 | 1.07 | 1.11 | 1.07 | FEM/Harbor | 1.84 | 1.44 | 1.78 | 1.69 | 1.09 | 1.02 | 2.31 | 1.59 | 1.78 | 1.70 | 1.16 | 1.08 | QCD | 2.42 | 0.96 | 1.07 | 1.34 | 1.05 | 1.03 | 3.04 | 1.03 | 1.07 | 1.35 | 1.12 | 1.05 | FEM/Ship | 1.89 | 1.62 | 1.12 | 1.57 | 1.11 | 1.03 | 2.38 | 1.92 | 1.12 | 1.59 | 1.18 | 1.10 | Economics | 2.91 | 1.86 | 1.80 | 1.66 | 1.34 | 1.02 | 3.83 | 2.38 | 1.89 | 1.75 | 1.50 | 1.04 | Epidemiology | 1.71 | 0.65 | 0.76 | 1.03 | 1.03 | 1.00 | 2.15 | 0.79 | 0.76 | 1.04 | 1.10 | 1.02 | FEM/Accelerator | 2.32 | 2.14 | 1.29 | 1.52 | 1.16 | 1.01 | 2.91 | 1.77 | 1.29 | 1.53 | 1.24 | 1.07 | Circuit | 2.66 | 2.16 | 1.63 | 1.58 | 1.58 | 1.04 | 3.51 | 2.76 | 1.71 | 1.67 | 1.76 | 1.07 | Webbase | 7.43 | 1.29 | 1.27 | 1.11 | 1.09 | 1.01 | 9.64 | 1.62 | 1.31 | 1.15 | 1.20 | 1.03 | LP | 1.58 | 1.20 | 1.45 | 1.60 | 1.21 | 1.05 | 1.99 | 1.54 | 1.45 | 1.61 | 1.29 | 1.10 | circuit5M | 12.55 | 1.63 | 1.78 | 1.78 | 1.53 | 1.06 | 16.16 | 1.56 | 1.61 | 1.83 | 1.67 | 1.14 | eu-2005 | 2.89 | 1.95 | 2.20 | 2.86 | 1.31 | 1.02 | 3.71 | 2.44 | 2.25 | 2.96 | 1.43 | 1.11 | Ga41As41H72 | 1.60 | 1.41 | 1.71 | 1.98 | 1.34 | 1.02 | 2.01 | 2.02 | 1.71 | 2.00 | 1.43 | 1.09 | in-2004 | 2.51 | 1.76 | 2.14 | 2.75 | 1.35 | 1.01 | 3.23 | 2.59 | 2.20 | 2.84 | 1.47 | 1.08 | mip1 | 1.64 | 1.37 | 1.91 | 2.04 | 1.30 | 0.97 | 2.06 | 1.95 | 1.91 | 2.06 | 1.39 | 1.00 | Si41Ge41H72 | 1.45 | 1.29 | 1.80 | 1.97 | 1.17 | 1.02 | 1.87 | 1.88 | 1.84 | 2.03 | 1.27 | 1.09 |
|
|