[webkit-dev] Canvas performance and memory usage

[Oliver Hunt]

The reason ImageBuffer::image() makes a copy (be it a deep copy, or CoW) is almost exclusively for the purpose of ensuring correct behaviour in the case where a canvas is drawn onto itself, eg.

context = myCanvas.getContext("2d");
context.drawImage(myCanvas, 0, 0);

Off hand I can think of no other case where this would be necessary, so it seems like the best solution would be to make the default behaviour for image() be to return a reference to a potentially mutable image, and to give the canvas a distinct method for getting a copied image in the case where it's actually necessary.

--Oliver

On Aug 4, 2010, at 2:45 PM, Christophe Public wrote:

> Hi,
> 
> Actually, I was questioning also the necessity of this extra buffer.
> I'll just give an update and some numbers. We have a very large canvas (few MB) and our updates are very frequent but very small (clip area is for example 3x3 pixels), it takes about 25ms for our system to handle the expose. Once the double buffer is removed, the expose takes less than 1ms.
> 
> We applied our change to image(). When the BitmapImage is created, we now pass the m_data.m_surface after increasing its reference count through cairo_surface_reference and removed the copy.
> 
> Our tests didn't detect any issue after this change.
> 
> Christophe
> 
> On Wed, Aug 4, 2010 at 11:07 AM, David Hyatt 
> 
> <hyatt at apple.com> wrote:
> I'm confused why a special method is needed though.  Can't image() just avoid the full copy?  Given how we use image() in WebKit, I don't think there's any reason to be concerned if image() continues to reflect the contents of the ImageBuffer.  I think we should just switch to that model for the CG port also anyway, since I'm unconvinced we're truly avoiding a copy, and return an image with a custom data provider that feeds the current contents of the ImageBuffer to the image.
> 
> dave
> (hyatt at apple.com)
> 
> On Aug 3, 2010, at 8:55 PM, Martin Robinson wrote:
> 
> > Resent from the proper address:
> >
> > On Tue, Aug 3, 2010 at 6:00 PM, Martin Robinson
> > <martin.james.robinson at gmail.com> wrote:
> >>> I notice that Qt added imageForRendering() and felt they could not use
> >>> image() for some reason.  I'd be curious if a Qt expert could weigh in on
> >>> that, since maybe with a redesign a separate call would not be needed.
> >
> > I'm not a Qt expert, but just based on a quick look, it seems that
> > imageForRendering  avoids the full QPixmap copy. Christophe,
> > when you open a bug for this issue,  please CC me, as I have a
> > small patch in my tree which has the same imageForRendering
> > pecialization, but for Cairo.
> >
> > Martin
> > _______________________________________________
> > webkit-dev mailing list
> > webkit-dev at lists.webkit.org
> > http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
> 
> _______________________________________________
> webkit-dev mailing list
> webkit-dev at lists.webkit.org
> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev
> 
> _______________________________________________
> webkit-dev mailing list
> webkit-dev at lists.webkit.org
> http://lists.webkit.org/mailman/listinfo.cgi/webkit-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-dev/attachments/20100804/8ebbc274/attachment.html>