Announcement

Collapse
No announcement yet.

Open-Source ATI Evergreen Acceleration Builds Up

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Louise
    replied
    Thanks for clearing that out =)

    Very interesting where all this is going.

    Leave a comment:


  • bridgman
    replied
    Nope, just because we expect the 3D core in the Fusion chips to be able to use existing code without many changes. If you lift the "core" and "uncore" terminology used for CPUs, most of the work is going to be in supporting the "uncore" part of the GPU, and that code lives primarily in the kernel and X drivers (mostly the kernel in a KMS system).

    I wasn't really thinking about the scenario where the X driver uses Gallium3D for 2D and video acceleration. If that happens I think it will be *after* Gallium3D has become the default in the 3D stack, ie the stack at the time of Fusion launch is most likely to use Gallium3D for the Mesa driver but not for the X driver.

    Bottom line, however, is that we are going to support whatever the community is favoring at the time. Anything I say here is really just my collection of guesses re: where the community (including our devs) will be at the time.

    Given the current state of Evergreen support and Gallium3D support for 6xx and higher, I think it's likely that Evergreen support on Gallium3D will be comparable to the support for 6xx and 7xx hardware starting in a couple of months. What's not so clear is where Gallium3D support for 6xx/7xx/Evergreen at the time the Fusion parts start to ship.

    Given that most of the work will be in the "uncore" part anyways, I don't think it matters too much.

    Leave a comment:


  • Louise
    replied
    Originally posted by bridgman View Post
    I think we would update both. Current thinking is that most of the changes will be outside Mesa anyways.
    Because Gallium haven't matured enough when Fusion hits the channel, to go with an exclusive Gallium implementation?

    Or perhaps because the Evergreen implementation haven't got Gallium yet?

    Leave a comment:


  • bridgman
    replied
    I think we would update both. Current thinking is that most of the changes will be outside Mesa anyways.

    Leave a comment:


  • Louise
    replied
    Originally posted by bridgman View Post
    Right now I imagine every dev has a slightly different set of priorities, so it's probably safe to assume that work is happening pretty much everywhere with the possible exception of video decode acceleration. From an AMD point of view our priorities are (a) hardware enablement for Evergreen GPUs (b) getting ready for the upcoming Fusion parts (Evergreen support is a key part of that) and (c) supporting the other community devs in whatever *they* are working on (which is usually different stuff).
    Btw. Is the plan still to first to make a classic implementation and then Gallium? Or will you go straight to Gallium?

    Leave a comment:


  • Louise
    replied
    Originally posted by bridgman View Post
    Right now I imagine every dev has a slightly different set of priorities, so it's probably safe to assume that work is happening pretty much everywhere with the possible exception of video decode acceleration. From an AMD point of view our priorities are (a) hardware enablement for Evergreen GPUs (b) getting ready for the upcoming Fusion parts (Evergreen support is a key part of that) and (c) supporting the other community devs in whatever *they* are working on (which is usually different stuff).
    Cool! I very much like that roadmap

    Does Fusion have a separated DRM (evil) from the 3D stuff to honour open source drivers?

    Leave a comment:


  • Louise
    replied
    Originally posted by bridgman View Post
    The major blocks - "2D features", "Mesa 3D features" and "Gallium features" - seem to be correct, although it's possible that some of them could move to "Done" (we're not very good at deciding when to make that last transition). The big change over the last month or so has been going from "WIP with no visible sign of progress" to "WIP - oh look, code !" but not enough to move any of the WIP cells to Mostly.
    Ah, so AMD is following the "Google Still in Beta" proceedure

    Originally posted by bridgman View Post
    In the last two blocks - "Output" and "Other" - I imagine that all of the "WIP" cells should probably read "Mostly" (but I haven't had a chance to test myself) and the "TODO" cells should read either "WIP" or "Mostly". Maybe someone who actually *has* a dual-link DVI display or a DisplayPort monitor could comment or update the page.

    We probably need to update the definitions to make it clear whether "Done" means "Done in released drivers" or "Done but you need to build from git right now". I would update it but I'm not sure what current thinking is, will ask the IRC folks. Clarifying that would probably help make the transition from Mostly to Done a bit crisper.
    It is very impressive what have been accomplished!

    Btw. I am learning GLSL now and using the closed source drivers. When I upgrade to Fedora 13, and use the open source drivers, that doesn't have a complete GLSL implemention. Will bugs look like they are my code, or will I know that it driver related?


    Btw2. This is very cool program to convert MS shader to GLSL
    http://sourceforge.net/projects/hlsl2glsl/

    Leave a comment:


  • bridgman
    replied
    Not sure right now, but it will definitely be quicker than Evergreen. We need to have Evergreen graphics support solid before the first Fusion parts come out. It's going to depend a what distros and drivers our OEM customers decide to ship with at launch, which in turn will determine who gets priority for engineering hardware and developers.

    I'll know more as we get closer to first customer shipments, but I probably won't be able to say much until the actual launch date.

    Leave a comment:


  • smitty3268
    replied
    Originally posted by bridgman View Post
    Right now I imagine every dev has a slightly different set of priorities, so it's probably safe to assume that work is happening pretty much everywhere with the possible exception of video decode acceleration. From an AMD point of view our priorities are (a) hardware enablement for Evergreen GPUs (b) getting ready for the upcoming Fusion parts (Evergreen support is a key part of that) and (c) supporting the other community devs in whatever *they* are working on (which is usually different stuff).
    Does (b) mean there's a plan to support Fusion by the time it's released like Intel is able to do with their new hardware? Or are you just trying to get into a position where adding support afterward comes quicker than it has with Evergreen?

    Leave a comment:


  • bridgman
    replied
    Right now I imagine every dev has a slightly different set of priorities, so it's probably safe to assume that work is happening pretty much everywhere with the possible exception of video decode acceleration. From an AMD point of view our priorities are (a) hardware enablement for Evergreen GPUs (b) getting ready for the upcoming Fusion parts (Evergreen support is a key part of that) and (c) supporting the other community devs in whatever *they* are working on (which is usually different stuff).

    Leave a comment:

Working...
X