gnash-dev
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Gnash-dev] Cairo & Mesa performance tests


From: W Lila
Subject: [Gnash-dev] Cairo & Mesa performance tests
Date: Tue, 05 Sep 2006 09:40:55 +0200

Hi all,

Some while ago my colleague Breght Boschker mentioned interest in Cairo for Gnash. We did some measurements to get some insight of rendering performance with the Cairo backend (and especially compared to the Mesa backend). We used the following test cases:
- horizontal lines of width 1, 2, 10 and 50
- vertical lines of width 1, 10 and 50
- diagonal lines of width 1, 2, 10 and 50
- curved lines of width 1, 10 and 50
- overlapping (filled) + not overlapping (filled) circles
- (filled) squares
- (filled) stars
In our test, we compared the cairo-enabled backend as well as the Mesa-backend with and without DRI (i.e. software-only).
Please find the numbers below:

General settings:
100 fps
550 w, 400h
All tests are automatically stopped after 100 sec.
Gnash is build for profiling and code is not optimized.

                                         Gnash
                           -------------------------------------
                            Cairo     Mesa            Mesa
                                      (without DRI)   (with DRI)
Test                         fps (avg) fps (avg)       fps (avg)
147 horizontal lines
lines_100fps.swf               36         4            100
lines_100fps_2pt.swf           32         3            100
lines_100fps_10pt.swf          21         0.78         100
lines_100fps_50pt.swf*          7.75      0.73         100

135 vertical lines
lines_v_100fps.swf             21.55      6.5          100
lines_v_100fps_10pt.swf        16.1       1            100
lines_v_100fps_50pt.swf*       14.4       1            100

576 (small) diagonal lines
lines_d_100fps_1pt.swf          8.8       5.7           59
lines_d_100fps_2pt.swf          8         3.8           59
lines_d_100fps_10pt.swf         5.9       0.7           58.82
lines_d_100fps_50pt.swf*        2.8       0.7           59

96 curved lines
curved_100fps_1pt.swf          33        21            100
curved_100fps_10pt.swf         26         3.8          100
curved_100fps_50.swf*          15         3.8          100

192 overlapping circles
circles_100fps.swf              0.81      8.9           96.7
circles_100fps_filled.swf       0.8       8.8           95.8

88 not overlapping circles
circles_100fps.swf              1.9      13            100
circles_100fps_filled.swf       1.4       8.9           62

204 squares
squares_100fps.swf             12.8       6.2          100
squares_100fps_filled.swf      13         6.2           98.5

62 stars
stars_100fps_2pt.swf           17.9      10.4          100
stars_100fps_2pt_filled.swf     2         8.5           99.9

* 50pt not supported for Mesa (?)

What is interesting in these numbers, is that line drawing is always *faster* using Cairo than using software-only OpenGL. We attribute the apparent slowdown when rendering non-line primitives to the tesselation step (which is not needed for Cairo).

Regards,
Widjai.






reply via email to

[Prev in Thread] Current Thread [Next in Thread]