On Wed, Jan 14, 2015 at 4:29 AM, Mike de Boer <mdeb...@mozilla.com> wrote:

>
> On 13 Jan 2015, at 21:52, Jeff Muizelaar <jmuizel...@mozilla.com> wrote:
>
>
>
> On Tue, Jan 13, 2015 at 10:56 AM, Mike de Boer <mdeb...@mozilla.com>
> wrote:
>
>>
>> 2. Optionally bypass the browser compositor when a WebGL context is in
>> fullscreen mode. In this mode, WebGL draw calls would write to the  OS back
>> buffer directly, increasing performance. Of course, this would never be
>> possible if the WebGL context has to be rendered amidst other HTML elements
>> on a web page, so that’s why the proposition here is for fullscreen mode
>> only.
>>
>
> There was a thread on this on the public webgl mailing list recently:
>
> https://www.khronos.org/webgl/public-mailing-list/archives/1412/msg00062.html
>
>
> This is exactly what I’m talking about, Robert, and it’d be used primarily
> to reduce latency and improve frame rates for fullscreen games.
>
>
> Interestingly enough, I believe Safari and IE already avoid the problem
> this would solve because they use the system compositor to composite there
> layers instead of a built-in one.
>
>
> So how do they deal with compositing adjecent HTML into a single texture?
> Regardless, I'd believe it right away that these browser targeted to a
> single platform can optimise their rendering pipeline.
> I did hear that Blink (perhaps Webkit too?) _does_ have this problem as
> well.
>

That content is just in a different layer and is composited on top by the
system compositor.

Safari uses CoreAnimation as compositor. CoreAnimation supports doing the
composition in the WindowServer process directly instead of into an
application backbuffer. DirectComposition gives the same benefit on Windows
and we actually have a similar thing on FirefoxOS.

Firefox and Chrome both composite everything into an application backbuffer
that is composited by the window manager and thus have an additional copy.

-Jeff
_______________________________________________
dev-platform mailing list
dev-platform@lists.mozilla.org
https://lists.mozilla.org/listinfo/dev-platform

Reply via email to