GDC Parallel SPU Physics slides available

Discussion in 'CellPerformance@B3D' started by erwincoumans, Feb 23, 2008.

  1. erwincoumans

    Newcomer

    Joined:
    May 24, 2007
    Messages:
    9
    Likes Received:
    0
    Location:
    San Francisco
    The slides for the Game Developers Conference Physics Tutorial presentation on parallelizing the collision detection and rigid body dynamics pipeline for Cell SPUs are available here:

    http://www.bulletphysics.com/Bullet...ics-tutorial-on-parallel-game-physics-for-spu

    Note that most of the optimizations have been made available to the open source cross-platform Bullet physics library. We are working on making more SPU optimizations available both for Playstation 3 SPURS and the IBM Cell SDK (for batched ray cast, batch collision queries, constraint solver etc).

    Feedback is welcome,
    Erwin Coumans
     
  2. Carl B

    Carl B Friends call me xbd
    Moderator Legend

    Joined:
    Feb 20, 2005
    Messages:
    6,266
    Likes Received:
    63
    Thanks for the link Erwin... I've been meaning to read a bunch of these GDC presentations over the last several days and have only now gotten around to it. Not specific to your effort here alone, but I think that the resources SCE has put into parallelism in general has been really helpful in moving the industry forward, and we are starting to see a shift in how dev houses prioritize their code development.

    That shift is premised somewhat on the fact that Cell semi-forces one to think threaded, but the benefits to doing so are there for all platforms and these cross-platform tools/libraries help to ease that burden.

    Specific to the context of Cell, I'm interested to see how your work progresses in optimizing for workstation/HPC-centric apps. Unrelated to the preceding, with the middleware consolidation underway in the industry - first Havok then Ageia - does that shift any of the emphasis for the team in terms of direction/priorities? Which is to say, does it shift emphasis away from SPE-specific optimizations in order to fill a possible void in what will become an increasingly architecture-optimized environment, or does it do the opposite and lead you to consider further SPE optimizations at the possible expense of cross-platform portability?
     
    #2 Carl B, Feb 28, 2008
    Last edited by a moderator: Feb 28, 2008
  3. one

    one Unruly Member
    Veteran

    Joined:
    Jul 26, 2004
    Messages:
    4,823
    Likes Received:
    153
    Location:
    Minato-ku, Tokyo
    Phyre Engine also supports Bullet.

    [​IMG]
     
  4. erwincoumans

    Newcomer

    Joined:
    May 24, 2007
    Messages:
    9
    Likes Received:
    0
    Location:
    San Francisco
    Disclaimer first: the following are only my personal opinions, not my employer Sony Computer Entertainment.

    Indeed, Cell SPU/SPE development benefits work on other platforms too. An important message is to first refactor for Cell SPE, which makes it easier to move to multi-core/XBox360, instead of the other way around. My experience so far is that coarse granular workloads, each taking about ~0.3ms to 1ms, helped cross-platform portability. While continuing to optimize for Cell SPE, I want Bullet to support NVidia CUDA (just like Ageia PhysX) and future parallel environments like Intel Larrabee, AMD stream computing efforts etc.

    The complexity developing for parallel hardware justifies availability of good parallelized sample code. Hopefully a cross-platform open source framework such as Bullet can help this collaborative process.
    Erwin
     
Loading...

Share This Page

  • About Us

    Beyond3D has been around for over a decade and prides itself on being the best place on the web for in-depth, technically-driven discussion and analysis of 3D graphics hardware. If you love pixels and transistors, you've come to the right place!

    Beyond3D is proudly published by GPU Tools Ltd.
Loading...