This repository contains an extensive benchmarking suite for the new implementations of PyLops sliding and patching operators. Notebooks and summary tables of timings and memory uses of the old and new implementations are provided.
These numbers have been computed using an Intel(R) Xeon(R) Gold 6230R CPU @ 2.10GHz with an NVIDIA GeForce RTX 3090 GPU.
NumPy
| Operator | Old (ms) | New (ms) | New broad. (ms) |
|---|---|---|---|
| Sliding1D | 24.2 | 12.6 | 1.5 |
| Sliding1D.H | 13.2 | 7.65 | 0.353 |
| Sliding2D | 200 | 20.7 | 16.2 |
| Sliding2D.H | 10.8 | 8 | 5.1 |
| Sliding3D | 1620 | 20.7 | 16.2 |
| Sliding3D.H | 674 | 446 | 205 |
| Patching2D | 108 | 50 | 64.8 |
| Patching2D.H | 59.7 | 41.3 | 81 |
| Patching3D | 1880 | 587 | 759 |
| Patching3D.H | 1170 | 796 | 491 |
CuPy (only GPU time using cupyx.profiler.benchmark)
| Operator | Old (ms) | New (ms) | New broad. (ms) |
|---|---|---|---|
| Sliding1D | X | X | X |
| Sliding1D.H | X | X | X |
| Sliding2D | 66.8 | 60.2 | 5 |
| Sliding2D.H | 47 | 32 | 0.5 |
| Sliding3D | 338 | 303 | 14 |
| Sliding3D.H | 216 | 189 | 2.3 |
| Patching2D | 219 | 173 | 17.8 |
| Patching2D.H | 160 | 96 | 0.67 |
| Patching3D | 104 | 80.8 | 7.2 |
| Patching3D.H | 67.1 | 41.3 | 6.9 |
| Operator | Old (Mb) | New (Mb) | New broad. (Mb) |
|---|---|---|---|
| Sliding1D | 1.2 | 0.13/0.016 | 0.13/0.016 |
| Sliding2D | |||
| Sliding3D | 1.1 | 207/2.89 | 207/2.89 |
| Patching2D | |||
| Patching3D |