[webkit-help] Capture rendering buffer (windows)

[vizshala]

Hi chris,

I also noticed that in windowless mode ( transparent or opaque ), the CPU
comsumption is more than window mode.
In my machine with amd phenom quad-core 2.5GHz, playing youtube video in the
window mode will comsume about 5% CPU resource while the windowless mode
will double it to about 10%.

At first I tried to capture the web view in windowless mode at the frame
rate of 30 per second. The steps I took are as follows:

1. invalidate the backingstore :
    according to the NPAPI rule, the windowless plugin will call
NPN_InvalidateRect to notify the browser that the plugin area is dirty.
After PluginView got notified, it will invalidate the rect, finally WebView
will add this rect to dirty region. So I use a UI Delegate to receive the
dirty rect.

2. fetch the backingstore handle

3. copy bits by bitblt

But the cpu usage is about 15%, performance is not acceptable. After some
search, I found that the step1 is the performance hit. I think this is due
to the updateBackingStore will go through the layout again.

Is it possible to improve the performance?  thx

Regards,
vizshala at gmail.com


2010/7/22 Chris Hatko <chatko at gmail.com>

> Hey,
>
> I'm still not fully done. I've implemented a solution which forces
> every flash plugin to run in windowless mode (force WMODE param).
> Doing this gives the render control over painting and as long as i
> invalidate the plugin area, and replicate some of the WM_PAINT
> functionality the  m_backingStoreBitmap contains the flash drawing.
> (even when the flash window is hidden)
>
> I think a better solution would be to use WM_PAINT hijacking in
> PluginViewWin.cpp  and insert your own HDC. ( note that in windowed
> mode, all drawing is done by the plugin app through
> PluginView::wndProc). I'm going to trail a solution this week which
> does this. It should also support other plugins (reader, quicktime
> etc). I'll likely post my findings to the group.
>
> Chris
>
>
>
> On Thu, Jul 22, 2010 at 9:26 AM, vizshala <vizshala at gmail.com> wrote:
> > hi Chris,
> >
> > In your original post, you mentioned that it may be possible to let flash
> > paint directly into the m_backingStoreBitmap or the dc we created. Did
> you
> > use api hijacking like the webkit did in pluginview or something else?
> >
> > thx
> >
> > 2010/7/13 Chris Hatko <chatko at gmail.com>
> >>
> >> Thanks Alex,
> >>
> >> I tried using paintDocumentRectToContext, but I was asking for the
> >> paint 30/sec and found that providing a rectangle for the entire
> >> window was too slow. I've made some progress with my original method
> >> which only invalidates the flash/plugin areas  and uses the
> >> backingstore bitmap as the 30/sec snapshot. Below is some code I found
> >> to search for plugin areas. I Hope it helps someone ...
> >>
> >>     // get the plugin (flash) areas and invalidate them
> >>     IntRect pluginArea;
> >>
> >>     for (Frame* frame = (this->m_page)->mainFrame(); frame; frame =
> >> frame->tree()->traverseNext())
> >>         {
> >>                FrameView* view = frame->view();
> >>                if (!view)
> >>                   return 0;
> >>
> >>               const HashSet<RefPtr<Widget> >* children =
> view->children();
> >>               ASSERT(children);
> >>
> >>               HashSet<RefPtr<Widget> >::const_iterator end =
> >> children->end();
> >>               for (HashSet<RefPtr<Widget> >::const_iterator it =
> >> children->begin(); it != end; ++it)
> >>               {
> >>                   Widget* widget = (*it).get();
> >>                   if (widget->isPluginView())  // works for flash at
> least
> >>                   {
> >>                                pluginArea =
> >> (static_cast<PluginView*>(widget))->windowClipRect();
> >>                                this->addToDirtyRegion(pluginArea);
> >>                   }
> >>              }
> >>     }
> >>
> >>
> >> One item i have noticed and that I've wasted many hours debugging is
> >> that the latest Flash (10.1) doesn't render at full speed when it is
> >> in a hidden window. I'll send another question to the group detailing
> >> this.
> >>
> >> Thanks,
> >>
> >> Chris
> >>
> >>
> >> On Mon, Jul 12, 2010 at 8:34 AM, Alexander Shulgin
> >> <alexander.shulgin at yessoftware.com> wrote:
> >> > On 09.07.2010 22:18, Chris Hatko wrote:
> >> >>
> >> >> Hi,
> >> >>
> >> >> My goal is to get the cario webkit port to render to a buffer instead
> >> >> of a window (i'll be making the window invisible), with as little
> >> >> change to the port source as possible.
> >> >
> >> > I'm using IWebFramePrivate::paintDocumentRectToContext(RECT, HDC) for
> >> > this.
> >> >  It works with pretty reasonable results and w/o the need to modify
> the
> >> > source code.
> >> >
> >> > --
> >> > Alex
> >> > _______________________________________________
> >> > webkit-help mailing list
> >> > webkit-help at lists.webkit.org
> >> > http://lists.webkit.org/mailman/listinfo.cgi/webkit-help
> >> >
> >> _______________________________________________
> >> webkit-help mailing list
> >> webkit-help at lists.webkit.org
> >> http://lists.webkit.org/mailman/listinfo.cgi/webkit-help
> >
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-help/attachments/20100803/3c2f561e/attachment.html>