0.0.3:
- Exposing GLcoef to Python
- Added
GLthread: New version of Grünwald-Letnikov operator using FFTW multithreading for parallel FFT. Best for very large arrays. - Optimized
GL: Efficient direct convolution. Now the default and fastest for most problem sizes (N < 1e6). - Restored
GLfull: Original implementation with full convolution and padding, included for completeness and possible future optimization. - Updated documentation and examples for all versions.
- docs were moved to github wikis
Benchmark Result for GL:
| count | 0.0.2.3 | new GL | new Glthread |
|---|---|---|---|
| 1+e2 | 0.0389 | 0.0052 | 0.0072 |
| 1+e3 | 0.9822 ms | 1.6205 | 2.5485 |
| 1+e4 | 5.2982 ms | 2.8592 | 6.2919 |
| 1+e5 | 104.1740 ms | 16.63965 | 16.8091 |
| 1+e6 | 667.7444 ms | 131.1534 | 151.7773 |
| 1+e7 | * | 1237.50445 | 1052.473 |
| 1+e8 | * | 13476.0071 | 12980.2857 |