Page 1 of 3 123 LastLast
Results 1 to 10 of 24

Thread: Assembly Shader Rework Hitting Mesa Today

Hybrid View

  1. #1
    Join Date
    Jan 2007
    Posts
    14,315

    Default Assembly Shader Rework Hitting Mesa Today

    Phoronix: Assembly Shader Rework Hitting Mesa Today

    Intel's Ian Romanick began last year working on a new GLSL compiler for Mesa, but those plans were abandoned due to a lack of time and the overwhelming challenge. Instead, Ian began focusing on optimizing the IR (Intermediate Representation) used by Mesa in its OpenGL support and then generating the final machine code. Ian has been working in a branch of Mesa with what he calls the "Assembly Shader Rework" and rewriting the ARB program parser and adding in support for NVIDIA layered extensions...

    http://www.phoronix.com/vr.php?view=NzQ2OQ

  2. #2
    Join Date
    Mar 2008
    Posts
    571

    Default

    Great... can't wait for Ubuntu 10.04 or even better 10.10

  3. #3
    Join Date
    Aug 2008
    Location
    Finland
    Posts
    1,577

    Default

    Quote Originally Posted by bulletxt View Post
    Great... can't wait for Ubuntu 10.04 or even better 10.10
    If not, just install from git.

  4. #4
    Join Date
    Oct 2008
    Location
    Sweden
    Posts
    983

    Default

    "One of our internal tests runs about 89,000 vertex programs. This test takes about 30 minutes (1,800 seconds) on current Mesa master. On the new
    code it takes about 25 seconds."


    Wow, talk about speed up!

  5. #5
    Join Date
    Jul 2008
    Location
    Greece
    Posts
    3,776

    Default

    Quote Originally Posted by whizse View Post
    "One of our internal tests runs about 89,000 vertex programs. This test takes about 30 minutes (1,800 seconds) on current Mesa master. On the new
    code it takes about 25 seconds."


    Wow, talk about speed up!
    Well, it's not like it was fast before As they fix the code and implement needed features, the speed gets to acceptable levels. It's starting to come together at last.

  6. #6
    Join Date
    May 2007
    Posts
    230

    Default

    Quote Originally Posted by RealNC View Post
    Well, it's not like it was fast before As they fix the code and implement needed features, the speed gets to acceptable levels. It's starting to come together at last.
    I think there is misunderstanding here, GL applications won't get faster with this code. What get faster is shader parsing. GL applications submit shader in a textual form the GL implementation has to parse it and translate it into some IR. The new merged code speedup this transformation. Thing is a proper GL application will submit its shader once and then reuse it so that the shader transformation shouldn't happen much in any proper GL applications in the end it won't speedup Doom or any games. Still it's a valuable improvement.

  7. #7

    Default

    Something keep escaping from me...

    Gallium will be used to expose a simple API to graphic card drivers, Mesa will be the softawre-side API (as for now). So the IR Ian have rework is where/what ?

    A representation computed by mesa after the softaware call and send to the driver for machine code generation ?

  8. #8
    Join Date
    Jul 2009
    Posts
    2

    Default Just merged


  9. #9
    Join Date
    Apr 2008
    Location
    /dev/random
    Posts
    218

    Default

    Quote Originally Posted by Xheyther View Post
    Something keep escaping from me...

    Gallium will be used to expose a simple API to graphic card drivers, Mesa will be the softawre-side API (as for now). So the IR Ian have rework is where/what ?

    A representation computed by mesa after the softaware call and send to the driver for machine code generation ?
    GL1-2 will still be handled through mesa(for now anyway) with gallium's mesa state tracker

  10. #10
    Join Date
    Jan 2009
    Location
    Vienna, Austria; Germany; hello world :)
    Posts
    620

    Default

    mesa fails to install here (it compiled fine):

    * QA Notice: Package has poor programming practices which may compile
    * fine but exhibit random runtime failures.
    * shader/hash_table.c:71: warning: implicit declaration of function ‘_mesa_malloc’
    * shader/hash_table.c:91: warning: implicit declaration of function ‘_mesa_free’
    * shader/hash_table.c:140: warning: implicit declaration of function ‘_mesa_calloc’


    *
    * QA Notice: Package has poor programming practices which may compile
    * but will almost certainly crash on 64bit architectures.
    *
    * Function `_mesa_calloc' implicitly converted to pointer at shader/hash_table.c:140

    *
    * Please file a bug about this at http://bugs.gentoo.org/
    * with the maintaining herd of the package.
    *
    *
    * ERROR: media-libs/mesa-9999 failed.
    * Call stack:
    * misc-functions.sh, line 730: Called install_qa_check
    * misc-functions.sh, line 422: Called die
    * The specific snippet of code:
    * die "install aborted due to" \
    * The die message:
    * install aborted due to poor programming practices shown above
    just to let you devs know about this "problem"

    this way perhaps crashes might be prevented right from the start ...

    yellow == warning, but merges / installs

    red == warning & blocker, won't merge / install

    thanks for the merge (and the hard work contained in it)

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •