Re: Qt Vs Cairo performance comparison

["Mark Howard" <mh tildemh com>]

On 10/23/06, Luis Menina <liberforce fr st> wrote:
> http://zrusin.blogspot.com/2006/10/benchmarks.html

So, they claim that QT is between 5 and 7 times faster than cairo.

The test appear to be measuring frames per second when repeatedly
drawing four polygons with a large number of vertices. This has been
tested apparently with the main cvs branches, but no exact details
were given. The source code is not available.

Here's my take on the benchmarks. I'd love for any gtk/cairo
developers to add their comments, or even just say if they
agree/disagree with these:

- Such a small number of tests is highly suspicious. Why not
automatically generate polygons? this should be relaitvely
straightforward. Can three tests really be considered to be
statistically significant?
- the graphs show 3 polygons, but the sample points are for 4
polygons. Is the fourth 50x faster with cairo?
- The first (and supposedly worst for Gtk) polygon is a text rendered
string. Gtk uses pango for text, so would not generate large polygons
like this.
- The third test has 100 000 vertices. I'd assume that such polygons
are hardly ever seen in screen output, so claiming QT as 7* faster
should really be 7* faster in a use case which virtually never exists
in time-critical areas.
- why not release the source code, despite requests in comments on the
blog post? Is something being hidden? Even if these results do show a
genuine bug in cairo, why not give the developers a chance to see
where the problem is?
- The author of the benchmarks noted in a comment that he also tested
the new cairo tesselator, but the results 'were a lot worse'. With the
application, this perhaps could be investigated.

- an interesting comment from the same page:
Would you consider publishing numbers for how it looks on something
other than the nvidia proprietary driver? I would tend to expect the
proprietary drivers to focus on OpenGL at the expense of 2D rendering
capabilities. How about the Intel or Radeon drivers?

To me, this seems like a silly FUD benchmark designed to highlight
some optimisations in specific sections of Qt. What a waste of time.
Qt probably is faster than Cairo, but without real benchmarks we're
still none the wiser and no closer to improving anything.