Platform/GFX/2011-August-22

• Every Monday at 2:30 PM Pacific/5:30 PM Eastern/21:30 UTC/Tuesday at 9:30 AM NZDT
• +1 650 903 0800 x92 Conf# 99366
• +1 800 707 2533 (pin 369) Conf# 99366 (toll free, Skype)
• Video (Vidyo) link: https://v.mozilla.com/flex.html?roomdirect.html&key=DeP4hnHU6nlN
• Vidyo room 9366 (if you have LDAP and can log in at https://v.mozilla.com)
• Minutes will be recorded at http://etherpad.mozilla.com:9000/GFX-meeting then pasted on this wiki page.

Minutes:

• GL layers on X11 (mattwoodrow)
  • Landed, backed out because of a performance regression.
  • texture_from_pixmap - glXBindTexImageEXT is causing huge allocations (Bug 680817).
  • Bas asks: Why are we only noticing this now?
    • Answer: We were more concerned about correctness. Just didn't look.
  • Matt to find some easy way to reproduce this performance regression and send it to Jeff, others for testing.
  • Matt asks: Is there anyone we can contact at nvidia about this?
    • bjacob says yes - we do have various contacts.
• WebGL conformance test suite upgrade and status (bjacob)
  • Tried to land it twice, and it has bounced.
  • Current status: no oranges in the WebGL test suite itself, but when you run it, you get oranges (mostly timeouts) in subsequent mochitests.
  • Maybe the WebGL mochitest is using too much memory?
  • Other issues: some Linux test slaves seemingly don't have OES_float_textures?! Possibly wrong version of driver?
  • Filed tracking bug for WebGL conformance test failures: bug webgl-conformance
• mobile fonts (gal)
  • bug 674909 - we are using synthetic bolding on some devices which apparently lack bold fonts for some faces, which makes bold text look terrible
  • some discussion about various options; further discussion to happen on the bug and/or offline
• Status of Test Pilot font survey (nattokirai)
  • the survey conversion is ready, but Jono's worried about the size of responses (e.g. images) going back to the server
  • We might send images individually rather than in a batch, and we might also send users only a subset of the questions.
  • Only needs Firefox 6. We're still trying to resolve a couple of bugs before it can be used, though.
• GL Layers on Android (ajuma, benwa)
  • https://wiki.mozilla.org/Platform/GFX/GLLayersOnMobile
  • Patches close to landing:
    • Scrolling - Landed but backed out unnecessarily, now non trivial bit rot
    • Orientation change + black flashes and black screen on resume - waiting for re-review
    • URL Bar + surface resize - r+ and green tryserver, will land tonight
  • Next issues:
    • Profiling
    • Context lost
      • We have some reports of lost contexts, but we don't know of any particular steps to reproduce. Some common ones don't seem to cause a context loss.
      • The Android documentation specifically calls out device sleep, but we haven't been able to reproduce that yet.
    • zooming - Slight black flash + artifacts
      • Sadly, not fixed by BenWa's scrolling patch :(
    • reftest - Tegra board received + setup
    • Need to get initial Talos perf numbers so we know whether we're causing any issues that will get us backed out.
    • Chris Jones also suggests we start getting people to turn on OpenGL-accelerated compositing on their devices and get more bug reports.
  • JP has some links for performance tests, specifically some Labs tools:
    • http://people.mozilla.org/~dougt/Dashboard/panel.html
    • http://people.mozilla.org/~dougt/Dashboard.a/panel.html
• WebGL Antialiasing (jgilbert)
  • There's still a hard crash in one of the tests on Windows, but only with WGL.
  • Suggestion: only allow antialiasing using ANGLE, and fix WGL later.
  • Bas has a question: how are we selecting the antialiasing mode?
    • jgilbert has a reply: we always use the highest antialiasing mode available. Bas has some concerns that this could hurt on slower hardware.
    • bjacob says that Chrome does 4x multisampling an we should do the same.
  • Should land now, minus WGL, and have a pref to disable AA so that we can easily turn it off.
    • Bas suggests it be an integer between 0 and 8 to control the AA sample rate.
• Cairo for FF7 (jprmc)
  • Did we find any bugs since last week?
    • No.
  • Everything nommed for beta?
    • Yes.
  • So there continues to be no reason to back out the Cairo update.
• Azure for FF7 (jprmc)
  • No reason to turn off.
  • Maybe want to transplant the moz-opaque ClearType support changeset on to Aurora, though it's not a high priority.
• Assorted bugs needing traction/info
  • bug 675383 - regression from GDI_CLASSIC for certain font families; roc, jfkthame, or Bas could look at this. Bas to own making sure jfkthame looks at this.
  • bug 591358 - mattwoodrow to take a look?
  • bug 671428 - does this one need to be pushed to aurora/beta? Yes, and it will be nominated.