From MozillaWiki
< Platform‎ | GFX
Jump to: navigation, search

Urgent Bug Triage

Bugs tracked for upcoming Firefox releases, in decreasing order of urgency:


Record minutes below, then copy back to the wiki after the meeting (BEFORE ERASING!) Team Sync:

  • Double buffering WebGl sync up (jgilbert)
    • EGLImage is being done by romaxa
    • Gives us improved performance but to get better we need to add double buffering
    • Done on the design phase of double buffering. Not sure when it's going to be done. Working with GFX drivers. Not sure doing double buffering but it also requires refactor the code right now with the layer manager and the canvas layer implementation. Moving to unifying the code instead. This will also make it easier to add gralloc and nvidia directx interr-opt for winxp.
    • Double-buffering is a bit of a misnomer. Romaxa's EGLImage patches do add double-buffering


  • Skia content rendering on windows. On GUImark2 on HTML5 Text Test resizing div it's 20% faster then cairo unaccelerated. Going to run some talos test for more complete results. Still working on font stuff on linux/android. Patches should be ready to land soon.


  • Cario azure canvas is working, working on some obsure bugs with dom binding on android. Waiting on some reviews.


  • Landed progressive tile drawing. Working on fixing the Talos measurement. After will work low resolution cache of the page to replace the screnshot.
  • Talked with Luke about scheduling GC while the GPU is busy, and continuing doing incremental GC until the GPU is done (polling).

bjacob: Now that JS people seemed to have warmed up to allowing gecko to control/hint GC to improve the WebGL experience, we should reopen the other questions we had about doing that in other parts of the WebGL impl (e.g. right before WebGL context creation). Nical

  • Async video
    • The first part of the work has been done for a week and is taking a while to be reviewed. It should land shortly. Some optimizations can be made by removing copies of the video frame. The current patch are to move the video swap off the main thread. Working on removing the extra video frame and copied directly into shared memory. B2G may not benefit from this because they are going to be using a different gralloc code path.
    • Need to turn on XINIT_THREAD compositing and we can't do it in a pref because it's too early on startup. Currently we have an environment variable to check this early. This environment is a problem because it makes testing harder and if we want users or automated testing this is difficult. Nical recommended that use thread x compositing regardless if we're going to use it not because of this. Nical pushed to Talos to measure the impact. The only thing he can see is differences because of noise. There doesn't seem to be any clear performance regression. It appears to be safe and/or may cause very minor performance regression.


  • Student open source project similar to GSOC
  • Looking for a project proposal this week. The project would start next year. It's not clear if this is interesting to us yet but we should make a proposal. Jeff made a proposal. An 8 to 10 week projects.