Those benchmarks have been done with all data already in cache using a Intel Xeon E5520 @ 2.27GHz running Debian 7.
| FabIO module | Filename | Size | Py2.7 v0.1.4 | Py2.7 v0.2.0 | Py3.2 v0.2.0 |
| adscimage | mb_LP_1_001.img | 9 Mpix | 1.16 s | 1.12 s | 2.67 s |
| brukerimage | Cr8F8140k103.0026 | 256kpix | 796 µs | 795 µs | 1.43 ms |
| cbfimage | run2_1_00148.cbf | 6 Mpix | 173 ms | 70.8 ms | 70.8 ms |
| edfimage | F2K_Seb_Lyso0675.edf | 4 Mpix | 512 µs | 597 µs | 595 µs |
| fit2dmaskimage | fit2d_click.msk | 1 Mpix | 30.4 ms | 30.5 ms | 28.4 ms |
| GEimage | GE_aSI_detector_image_1529 | 4 Mpix | 5.37 ms | 5.44 ms | 4.3 ms |
| kcdimage | i01f0001.kcd | 360kpix | 130 ms | 121 ms | 292 ms |
| mar345image | example.mar2300 | 5 Mpix | 78 ms | 77 ms | 77 ms |
| marccdimage | corkcont2_H_0089.mccd | 4 Mpix | 9.28 ms | 9.01 ms | 17.2 ms |
| OXDimage | b191_1_9_1.img | 256kpix | 5.75 ms | 5.67 ms | 8.35 ms |
| pnmimage | image0001.pgm | 1 Mpix | 3.29 ms | 3.25 ms | 5.71 ms |
| raxisimage | mgzn-20hpt.img | 3 Mpix | 53.5 ms | 57.1 ms | 69.2 ms |
| tifimage | oPPA_5grains_0001.tif | 4 Mpix | 59.9 ms | 58.4 ms | 105 ms |
| xsdimage | XSDataImage.xml | 256kpix | 13.3 ms | 12.9 ms | 18.4 ms |
The Python3 version is sometimes twice slower then the Python2 version. As the codebase is the same this regression is not due to FabIO but to the programming language itself. The Performances obtained using Python3.4 is now again close to the one of Python2.7.