Re[2]: [Gnash-dev] youtube profiling

[Udo Giacomozzi]

Hello strk,

Thursday, May 3, 2007, 11:25:19 AM, you wrote:
s> This takes us to AGG, and in particular to:
...

I don't know what the values mean, but it seems that AGG is busy
scaling the frame image, which is what I expected.

Note AGG does also matrix conversions for the pixels. AFAIK there are
even a bunch of methods to *apply* these matrix transformations
(that's the image_accessor), some are more accurate some are faster.
I'll have a look at it but I don't expect a huge improvement.

s> ... render_scanlines_aa ... (375 times, for a total of 12.6 seconds).

That's nearly exactly 30 fps.
Striking...


s> I just see that render_scanlines_aa (or drawVideoFrame) is only called
s> 375 times, over 464 total calls to ::advance (now noticing).

Try to see if area around the video gets updated in each frame (red
rectangle debugging).


s> Now my question was about the number of actual frames in the FLV becase
s> if that number is much smaller then our '375' we are likely updating
s> more then we should. I'm sure invalidated bounds are not implemented in
s> video_frame_instance. Another question would be: would implementing properly
s> invalidated bounds still alwayus call drawVideoFrame ?

I tried to download the FLV and analyze it with ffmpeg, but wget with
error "303 See other" !? Anyway, you could comment out
drawVideoFrame() and see how values change.

Udo