From: eLinux.org
Benchmark DirectFB
Benchmarks
The DirectFB example suites include benchmark ‘df_dok’. We have ran
this benchmark on the following platform:
Platform | CPU | Clock | I/F | System RAM | Graphics Card | Kernel Version |
---|---|---|---|---|---|---|
A | Renesas SH-4 | 240MHz | CPU | 64MB | SMI SM501 software drawing | 2.4.20 CELF |
A | SMI SM501 enable 2D acceleration | 2.4.20 CELF | ||||
B | Renesas SH-4 | 240MHz | PCI | 64MB | Matrox Millenium | 2.4.19 |
C | Intel Celeron | 450MHz | PCI | 128MB | Matrox Mystique | 2.4.20 |
D | Intel Celeron | 450MHz | PCI | 128MB | Matrox Millenium | 2.4.20 |
F | Intel Celeron | 450MHz | AGP | 128MB | Matrox G450 | 2.4.20 |
E | Intel Pentium4 | 2.4GHz | AGP | 1GB | Matrox G450 | 2.4.20 |
G | Intel CE2110 | 1Ghz | CPU | 100Mb | software drawing | 2.6.16 |
G | Intel GDL accelerated | 2.6.16 |
Benchmarks | Platform | ||||||||
---|---|---|---|---|---|---|---|---|---|
A | A | B | C | D | E | F | G | G | |
Anti-aliased Text [KChars/sec] | 29.22 | 29.41 | 20.40 | 24.83 | 23.96 | 607.39 | 750.00 | 84 | 45 |
Anti-aliased Text (blend) [KChars/sec] | 7.52 | 7.47 | 6.12 | 16.52 | 16.66 | 613.00 | 752.85 | 25 | 19 |
Fill Rectangles [MPixel/sec] | 14.07 | 221.49 | 63.63 | 116.37 | 53.25 | 892.88 | 849.22 | 34 | 105 |
Fill Rectangles (blend) [MPixel/sec] | 1.64 | 1.67 | 1.20 | 3.18 | 3.26 | 236.38 | 225.84 | 3 | 6 |
Fill Triangles [MPixel/sec] | 12.25 | 93.87 | 93.87 | 108.79 | 50.51 | 748.55 | 730.24 | 32 | 1 |
Fill Triangles (blend) [MPixel/sec] | 1.63 | 1.61 | 1.17 | 3.13 | 3.17 | 223.32 | 218.24 | 4 | 0.026 |
Draw Rectangles [KRects/sec] | 1.81 | 16.01 | 10.67 | 12.95 | 8.57 | 59.53 | 36.27 | 11 | 3.6 |
Draw Rectangles (blend) [KRects/sec] | 0.52 | 0.57 | 0.43 | 0.83 | 0.84 | 25.40 | 17.09 | 2.142 | 0.194 |
Draw Lines [KLines/sec] | 7.10 | 66.98 | 61.33 | 62.60 | 48.84 | 281.86 | 162.40 | 41 | 16 |
Draw Lines (blend) [KLines/sec] | 2.33 | 2.43 | 1.94 | 3.69 | 3.70 | 127.29 | 80.04 | 10 | 7 |
Blit [MPixel/sec] | 8.12 | 100.49 | 38.68 | 53.75 | 32.56 | 435.02 | 398.84 | 25 | 64 |
Blit colorkeyed [MPixel/sec] | 1.63 | 102.16 | 39.19 | 58.69 | 32.54 | 445.20 | 421.97 | 28 | 36 |
Blit with format conversion [MPixel/sec] | 4.04 | 4.13 | 3.59 | 18.11 | 17.79 | 203.38 | 193.26 | 13 | 38 |
Blit from 32bit (alphachannel blend) [MPixel/sec] | 1.05 | 1.05 | 0.82 | 2.71 | 2.71 | 171.32 | 158.10 | 3 | 36 |
Blit from 8bit palette [MPixel/sec] | 3.54 | 2.35 | 3.20 | 17.40 | 17.38 | 17.28 | 95.17 | - | - |
Blit from 8bit palette (alphachannel blend) [MPixel/sec] | 1.02 | 0.90 | 0.81 | 2.67 | 2.71 | 3.55 | 5.53 | - | - |
Stretch Blit [MPixel/sec] | 12.17 | 83.50 | 7.06 | 83.50 | 47.61 | 210.32 | 220.77 | 25 | 21 |
Stretch Blit colorkeyed [MPixel/sec] | 7.46 | 7.43 | 4.20 | 46.17 | 46.30 | 211.97 | 221.64 | 26 | 20 |
After enabling 2D acceleration in A* environment, the graphics
benchmark score improves. It looks like even embedded processor like
SH-4 can be used when DirectFB can use hardware acceleration. So a
graphics library would be essential in a graphical system.
For G* (Intel CE2110) environment, just basic operations like opaque
rectangle and blits are really optimized, some operations are much
slower and we can notice rendering artifacts/bugs. So take care,
possible you’ll have to fallback to software rendering on such platform.
Cyan-colored cells are supposed to be accelerated.