Add stats tracking for available native OpenGL (and EGL) extensions |
||
Issue descriptionCurrently we don't have great records of our user's driver's exposed extensions. In some cases when we're considering implementing a new extension, we're not sure how broad support it could have. We should be able to track this with a sparse histogram. We could implement this in the command buffer after we initialize a context (or EGL display), and/or in ANGLE itself when it is initializing. We could even record some back-end specific stats for back-ends which expose more info.
,
Mar 1 2018
http://www.gpuinfo.org/ seems to already have bunch of stats about each extensions, % of use and also hardwares they are supported on. We can use that information directly and seems like UMA is not needed.
,
Mar 2 2018
I think we were interested in the capabilities of Chrome customers rather than identifying extensions exposed on a particular device. Geoff can you elaborate?
,
Mar 2 2018
I am also a bit confused about the need here. Some elaboration would be appreciated.
,
Mar 2 2018
GPU info database are aimed at developers that need to know what features are supported by a driver/GPU combination OR to check how widespread certain features are. So it not only tells the extensions exposed/available on a particular device but also tells how much % of people are using it. This % estimate should give us a fairly close understanding how important a particular extension is. I am guessing that chrome customer data will also report very similar numbers. But if the requirement of getting only chrome user data has some additional purpose which gpuinfo cant provide, we should look into it.
,
Mar 2 2018
There are two things we can do: * Log the extensions that ANGLE is exposing to Chrome to get a picture of what Chrome's clients/WebGL are capable of doing. This can be done within the GPU process in Chrome's code. * Log native extensions/info from within ANGLE to get a picture of what D3D/OpenGL/VK features are present to help ANGLE/Chrome developers know if it's worth exposing new extensions. We can do both but I think #2 is more useful.
,
Mar 2 2018
For #2, there might be a slight difference between http://www.gpuinfo.org/ sample and Chrome users, however, we probably don't care about that difference in order to make decisions.
,
Mar 2 2018
http://www.gpuinfo.org/ gives us info about what each GPU supports but not about the proportion of users on that GPU. We would have to do some queries to find out what percentage of users have an extension. It may or may not be worth it but it would certainly give us more useful information than the site.
,
Mar 2 2018
It actually does, assuming you don't actually care about how many users on a GPU, but rather, how widely a GPU capability is supported. For example, here is the list of each extension's support percentage https://opengl.gpuinfo.org/gl_extensions.php
,
Mar 2 2018
OK, I'm thinking that this is a less useful feature for Chrome then. It would probably only give us D3D stats which don't vary much and we already report some.
,
Mar 2 2018
What might be worth reporting, could be if a combination of extensions are supported. For example, we need to check a few extensions to turn on DirectComposition. Such data might be valuable. Or feature levels, etc. Does Vulkan has a similar concept?
,
Mar 2 2018
The data is from all the gpu (both mobile and desktop) and OS (win,lin,android,mac etc). so i guess it can give us good estimate of what we want.
,
Mar 2 2018
I think people have talked about Vulkan feature levels but I'm not aware of any such concept being adopted.
,
Aug 1
|
||
►
Sign in to add a comment |
||
Comment 1 by ericrk@chromium.org
, Feb 23 2018Owner: vikassoni@chromium.org