Research Article
A Strategy for Automatic Performance Tuning of Stencil Computations on GPUs
Table 10
Heuristic performance on various input shapes.
| | Z size | NVIDIA | AMD | 8 | 16 | 32 | 64 | 128 | 8 | 16 | 32 | 64 | 128 |
| Speedup | | | | | | | | | | | Hybrid oracle | 1.11 | 1.20 | 1.20 | 1.17 | 1.16 | 1.27 | 1.26 | 1.34 | 1.52 | 1.50 | Hybrid | 1.08 | 1.20 | 1.20 | 1.17 | 1.16 | 0.95 | 1.10 | 1.26 | 1.48 | 1.50 | Random sample | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
| Correct predictions | 5 | 7 | 7 | 7 | 7 | 4 | 4 | 4 | 4 | 6 |
|
|