Platform/GFX/2011-August-15

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

Minutes

  • Should the Cairo update be backed out? (jrmuizel) 
    • Chris Jones points out that a lot of mobile bugs and even some X bugs were fixed by the Cairo update.  The biggest set of fennec problems was worked around by bug 637828 though.
    • jrmuizel doesn't want to back out, and cjones strongly doesn't want to back out.
    • It doesn't really seem like the bugs are severe enough to warrant a backout. Jeff will continue fixing the bugs.
    • The crash bug is most severe, but we understand it.
    • dependent bugs: bug 666893
  • Azure update (Bas) 
    • Thebes wrapper depends on the Cairo D2D backend, and therefore will be a little delayed.
    • Joe's going to work on the Cairo D2D backend this week.
    • Azure on Fx7?
      • There's one semi-rare crash that we can't reproduce (bug 676690). We might be able to work around it, though.
      • ClearType isn't supported in Azure. We made this choice on purpose to speed things up for the implementation of Azure-D2D. There are various mitigation strategies, including using the Cairo implementation of canvas in moz-opaque situations.
      • Bas to collect regressions and link them to the original Azure D2D bug (bug 651858).
  • D3D9 TP4 regression (benwa) 
    • bug 660264
    • Two issues:
      • System patch(es) (windows, drivers, directx) should fix the regression.
        • We know the union of all the patches available fixes this Tp4 regression. What we don't know is which specific patch or subset of patches fixes the regression.
        • We took an educated guess that it was the driver update, and we were wrong.
        • Current guess is windows update
        • Will check out a talos machine to confirm
      • The regression isn't seen on tp5
  • Cross-process D3D10 landed (cjones)
    • Context given here: Electrolysis/Short_Terms_Goals
    • bug 648484
    • landed, stuck this time
    • follow-up bugs for proper synchronization linked above; painting is racy right now
    • needs performance analysis
    • security review tentatively proposed for Sept. 13th, 1000 PDT
    • OpenGL sharing on Mac is still an open problem, but /should/ work out of the box.  Needs mac-specific optimizations though
      • Nominations for victim: BenWa or mattwoodrow
  • WebGL Antialiasing update (jgilbert) 
    • Down to one webgl conformace suite regression
  • For the love of all that is holy, book your all-hands travel if you are going to it. (joe)
    • Friday is the last day, though it may be less structured than the other days. 4pm-ish looks to be the last scheduled time.
  • TestPilot DirectWrite font rendering survey (joe/jdaggett)
    • John to email Jono today.
  • Web3D Community Working Group (Bas)
    • There is a group of people from Web3D (the group that originated with VRML) who are working on a declarative 3D web standard.
    • http://www.x3dom.org/x3dom/w3c/charter/ - W3C Incubator group for the nascent standard.
    • Bas thinks we should send a representative to the working group in order to steer it towards something that we feel better with.
    • Bas will go. No one else is very interested/believes in it
  • SIGGRAPH wrap-up (Bas, joe)
    • Scaleform
      • Triangulation with 1 pixel border for AA (sounds the same as D2D)
      • Batching is important (retained-mode API)
    • GPUs continue to execute branches in the "old style" - that is, masking off the ALUs that missed the branch, and executing in lock step across the entire SIMD unit.
      • This has different implications depending on the architecture. NVIDIA is bigger but wider (eg ~16-32 threads executing on the same processor); AMD the opposite.
    • Web3D consortium has implemented a bunch of things on top of WebGL
    • Deferred shading via cached fragments on every triangle; gets shading down to about 1-2x per pixel vs 20+
    • If you have any questions, desire for citations or desire for a paper, please contact Joe <joe@mozilla.com>