Results 1 to 5 of 5

Thread: LLVM May Get A TGSI Gallium3D Compiler Back-End

  1. #1
    Join Date
    Jan 2007
    Posts
    14,335

    Default LLVM May Get A TGSI Gallium3D Compiler Back-End

    Phoronix: LLVM May Get A TGSI Gallium3D Compiler Back-End

    A proposal has been made to develop a new LLVM compiler back-end that would generate TGSI instructions, the intermediate representation used by Mesa's Gallium3D drivers...

    http://www.phoronix.com/vr.php?view=MTM0NDA

  2. #2
    Join Date
    Jun 2011
    Posts
    824

    Default

    I'm not quite sure I understand the implications of this, does this mean that I can compile say C++ code to run on the GPU perhaps with some caveats as to how things must be coded, or is this just for openCL and DirectCompute and such?

  3. #3
    Join Date
    Sep 2010
    Posts
    646

    Default

    Quote Originally Posted by Luke_Wolf View Post
    I'm not quite sure I understand the implications of this, does this mean that I can compile say C++ code to run on the GPU perhaps with some caveats as to how things must be coded, or is this just for openCL and DirectCompute and such?
    I up that question.

    Are we talking about all frontends of LLVM to one back end producing TSGI. Or its targeted for OpenCL?

  4. #4
    Join Date
    Oct 2012
    Location
    Cologne, Germany
    Posts
    308

    Default

    Quote Originally Posted by przemoli View Post
    I up that question.

    Are we talking about all frontends of LLVM to one back end producing TSGI. Or its targeted for OpenCL?
    TGSI is just an intermediate representation, an interface for an uniform calling-infrastructure. This means, that implementing OpenCL and especially GLSL will become much easier, because TSGI would be the only interface that would have to be developed for in this specified case.

    Of course, in regards to LLVM, this makes only sense as a backend when we are talking about Gallium opcodes and libraries using them and not about C++ itself.
    Last edited by frign; 04-05-2013 at 06:47 AM.

  5. #5
    Join Date
    Nov 2007
    Posts
    1,024

    Default

    Quote Originally Posted by Luke_Wolf View Post
    I'm not quite sure I understand the implications of this, does this mean that I can compile say C++ code to run on the GPU perhaps with some caveats as to how things must be coded, or is this just for openCL and DirectCompute and such?
    With some additional work, sure. The C/C++ frontend has some requirements of the backend, but a modified C/C++ frontend could be made to work, similar to CUDA. It would also allow for an easier AMP implementation when using Gallium3D, which would be fucking amazing.

Posting Permissions

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