Vsync scheduling improvements for Chrome/Android

[Sami Kyostila]

Here is a design document for the first round of touch input latency -related improvements on Android.

https://docs.google.com/a/chromium.org/document/d/16822du6DLKDZ1vQVNWI3gDVYoSqCSezgEmWZ0arvkP8/edit

Anyone with a @chromium.org account should be able to comment. If you don't have access, let me know and I'll add you.

- Sami

[Sohan Jyoti Ghosh]

Hi Sami,

Can you please add me (sohanjy...@gmail.com) to view this vsync document.

Thanks,

Sohan

[bogdan...@yandex-team.ru]

Hi Sami,

I will also appreciate if you'd gave me access to read this document.

Thanks,

Ilya

четверг, 18 апреля 2013 г., 23:17:47 UTC+4 пользователь Sami Kyostila написал:

[Tom Hudson]

That document now begins with:

Please see the version at chromium.org instead:

https://docs.google.com/a/chromium.org/document/d/16822du6DLKDZ1vQVNWI3gDVYoSqCSezgEmWZ0arvkP8/edit#

... and even the Chromium version, which you ought to be able to read, is AFAIK now obsolete.

Tom

--
You received this message because you are subscribed to the Google Groups "Graphics-dev" group.
To view this discussion on the web visit https://groups.google.com/a/chromium.org/d/msgid/graphics-dev/9ab6e3cb-d212-4da2-aa25-1b201dab79ed%40chromium.org.
 
 

[Ilya Bogdanovich]

I see, thanks!

Ilya

[blurb...@gmail.com]

A very good test of VSYNC support for requestAnimationFrame() is the TestUFO motion site:

http://www.testufo.com
http://www.testufo.com/#test=photo
http://www.testufo.com/#test=framerates-text

A very good debugging time graph of requestAnimationFrame() is:
http://www.testufo.com/#test=animation-time-graph

Works great in Chrome for Windows/Mac, and in other browsers supported at http://www.testufo.com/browser.html ... Also works great on, iPhone, iPad, Win8 Tablets, but very terrible on Android.

But all Android devices I've tried, does a terrible job at requestAnimationFrame() -- even the latest and greatest Samsung Galaxy S3/S4 I've tried.

[Adam Treat]

Why does the renderer process need to rely upon the browser process to be notified of the start of a vsync cycle?  Shouldn't the renderer process be able to retrieve this information from a lower-level system or call?  It seems sad that we rely upon an IPC message inside of chromium for renderer process(es) to be notified of vsync.

[Sami Kyostila]

Hi Adam,

The renderer runs in a sandbox and doesn't have direct access to that
information. Since we also pass input events through IPC, I don't
think it's unreasonable to use the same mechanism for vsync because
the renderer generally has to wait for input before drawing.

That said, the cost of an IPC message is fairly high on Android (~0.2
ms on a Nexus 4), so I think there's room for optimization.

- Sami