Re: [Gnash-dev] youtube profiling

[strk]

On Fri, May 04, 2007 at 01:32:17PM +0200, Udo Giacomozzi wrote:
> Hello strk,
> 
> Friday, May 4, 2007, 11:36:26 AM, you wrote:
> s> Based on the profiling, the bottlneck is still in the renderer,
> s> at least for AGG (not for opengl!!).
> 
> I expect AGG not to perform extremely well for video, but a modern CPU
> should be fast enough to never drop any frames for a 30fps video,
> hmmm.
> 
> I fixed invalidated bounds for video. It now re-renders only the
> relevant part *and* avoids any rendering when there is no new video
> frame <- could improve performance.
> 
> Please profile again...

No visible difference:

[1]     92.0    0.00   20.49                 
gnash::Gui::advance_movie(gnash::Gui*) [1]
                0.00   19.31     502/502         
gnash::Gui::display(gnash::movie_root*) [2]
                0.00    1.17     503/503         
gnash::movie_root::advance(float) [28]

I wouldn't have expected a big difference actually, since the video takes up 
almost
the full visible area.

FYI, here's the top of the flat profile:

Each sample counts as 0.01 seconds.
  %   cumulative   self              self     total
 time   seconds   seconds    calls  ms/call  ms/call  name
 13.39      2.98     2.98 62817300     0.00     0.00  
agg::image_accessor_clip<agg::pixfmt_alpha_blend_rgb<agg::blender_rgb_pre<agg::rgba8,
 agg::order_rgb>, ag
g::row_accessor<unsigned char> > >::span(int, int, unsigned int)
 11.79      5.61     2.62 128599340     0.00     0.00  
agg::dda2_line_interpolator::operator++()
 10.42      7.92     2.32   139594     0.02     0.10  
agg::span_image_filter_rgb_nn<agg::image_accessor_clip<agg::pixfmt_alpha_blend_rgb<agg::blender_rgb_pre<a
gg::rgba8, agg::order_rgb>, agg::row_accessor<unsigned char> > >, 
agg::span_interpolator_linear<agg::trans_affine, 8u> >::generate(agg::rgba8*, 
int, int, unsig
ned int)

--strk;