Results 1 to 10 of 23

Thread: Radeon Gallium3D OpenCL Is Coming Close

Hybrid View

  1. #1
    Join Date
    Jan 2007
    Posts
    15,635

    Default Radeon Gallium3D OpenCL Is Coming Close

    Phoronix: Radeon Gallium3D OpenCL Is Coming Close

    Following the OpenCL Gallium3D state tracker having been merged into Mesa earlier this week, the open-source Radeon OpenCL support is coming close...

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

  2. #2
    Join Date
    Jan 2009
    Posts
    1,765

    Default

    any apps that work with it or is it tooo early for that???? (ie darktable i think uses OpenCL)


    awesome work anyway

  3. #3
    Join Date
    Jul 2011
    Location
    florida usa
    Posts
    81

    Default

    it would be really good to know how we are going to move forward from here as far as what ir language will be used in gallium. right now we have tgsi being used everywhere except radeon compute, but, there were all these plans o replace the entire ir stack with llvm. this would be a start i guess, but is that what we really want? we need to stick with something though, that was kinda the entire point of gallium to try and unify as much of the graphics stack between different drivers as we can.

  4. #4
    Join Date
    Oct 2007
    Location
    Toronto-ish
    Posts
    7,571

    Default

    Quote Originally Posted by benjamin545 View Post
    it would be really good to know how we are going to move forward from here as far as what ir language will be used in gallium. right now we have tgsi being used everywhere except radeon compute, but, there were all these plans o replace the entire ir stack with llvm. this would be a start i guess, but is that what we really want? we need to stick with something though, that was kinda the entire point of gallium to try and unify as much of the graphics stack between different drivers as we can.
    I'm not sure "in gallium" is the right boundary. There's also the GLSL IR vs TGSI question.

    At the time we started working on GPU compute clover used LLVM exclusively so that was the obvious choice, particularly since we were looking at LLVM for SI shader compiler...

    Francisco then refactored clover as a Gallium3D state tracker so TGSI was the obvious choice, particularly since the nouveau folks were *not* looking at LLVM at the time....

    I imagine the Intel devs are looking at compute with IB and for them GLSL IR will presumably be the obvious choice, since they just finished designing and implementing it...

    At last year's AFDS we talked about a proposed common IR for GPU/CPU compute we were going to support as part of our Heterogeneous System Architecture (HSA, previously FSA) initiative, so *that* might seem like an obvious choice if we get it right....

    When we discussed this at the last XDS it seemed there hadn't been enough work done with compute to have a clear consensus on what IR should be used between GLSL IR, TGSI, LLVM, FSAIL/HSAIL or something else. We concluded that the best thing for now would be for all of us to get some experience running real world compute workloads on GPUs using whatever IR was most convenient then get back together in 12-18 months and try to converge on a single IR for graphics and compute.

    There was probably an unspoken hope that we would get convergence on graphics IR first
    Last edited by bridgman; 05-13-2012 at 12:57 PM.

  5. #5
    Join Date
    May 2012
    Location
    Missouri
    Posts
    59

    Default

    Quote Originally Posted by 89c51 View Post
    any apps that work with it or is it tooo early for that???? (ie darktable i think uses OpenCL)


    awesome work anyway
    I'm curious too. I moved my A8-3850 to my HTPC because it was too finickey on my Ubuntu machine. I'd love to leverage the GPU on it without resorting to AMDs drivers, all I'd really like to see is some better video encoding.

    Then again, Quicksync on Linux would be nice for my i3 too.

  6. #6
    Join Date
    Feb 2008
    Location
    Linuxland
    Posts
    5,332

    Default

    Quote Originally Posted by MonkeyPaw View Post
    I'm curious too. I moved my A8-3850 to my HTPC because it was too finickey on my Ubuntu machine. I'd love to leverage the GPU on it without resorting to AMDs drivers, all I'd really like to see is some better video encoding.

    Then again, Quicksync on Linux would be nice for my i3 too.
    Every single OpenCL/dedicated block encoder so far have terrible quality, including QuickSync and what have we. Just curious, what are you doing that needs the speed more than the quality?

  7. #7
    Join Date
    May 2012
    Location
    Missouri
    Posts
    59

    Default

    Quote Originally Posted by curaga View Post
    Every single OpenCL/dedicated block encoder so far have terrible quality, including QuickSync and what have we. Just curious, what are you doing that needs the speed more than the quality?
    I do various encodes from HD home videos to upload, as well as taking recorded TV (in HD) and encoding to x264. Last I heard, Quicksync had the best IQ when compared to other GPU off-loads, and it's ridiculously fast.

  8. #8
    Join Date
    Jul 2011
    Location
    florida usa
    Posts
    81

    Default

    @ bridgeman

    Ah, so, basicaly, we( we as in you, dont you just love when people say we but really they themselves aren't part of the "we") have no clue how demanding compute will be on the ir and in what way the ir will need to bend to effectivly operate.

    can you tell me this then. i know we often hear about the ir languages probobly more so than any other componant of the graphics stack below the actual end user api's, but how inconviniant is it really to switch from one to the other? in a game engine it would be a real job to change out from ogl to dx, even from ogl to ogl es if ddone certain ways, but how much of a bother would it be for you to change amd compute back end frrom the llvm over to tgsi if that was the more unified aproch?

    also, whats the chances someone will start slinging gcc ir in there as an option what with their plans to try and make a competing ir more like what llvm has?

Posting Permissions

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