AMD Sends Out Revised DAL Patches, Various Cleanups

[cRaZy-bisCuiT]

juno Thanks for the information. I assume the quality of code might get better by that. Better than fglrx always seemed to be. (; Is it likely to see AMD contributing more to Mesa when all the needed features are implemented in amdgpu/dal? I think there're still a lot of potential for optimization. (:

[juno]

@cRaZy-bisCuiT: I doubt that people working on DAL will work on mesa parts. Those devs are responsible for the display stack on multiple platforms, not 3D stuff only for Linux. But I'm not involved at all, so I can't tell for sure. Maybe bridgman, agd5f or someone else can. However, it doesn't really matter to me. There are updates regularly, though, just look at the front page. Marek for example is a great contributor, others as well work on improving the experience with Mesa software.

[bridgman]

So the patches are sent out to be tested now? I don't understand why they wouldn't be integrated in 4.10 otherwise. Some RCs before?

The patches were sent out for review, not for testing. We already have public branches with DAL enabled (eg agd5f's amd-staging-4.7) for testing.

If this is the plan, I wonder what would happen if somebody from the community gets Vega working with minor changes/extensions to the current display code 😁

Then you would get to keep running without the features that DAL brings. I don't see how that would be a big win.

If we can't find a common ground to get DAL upstream in time then we would probably implement the non-DAL support ourselves, but that would be throwaway work using time taken away from getting DAL upstream so hoping to avoid it.

Better than fglrx always seemed to be. (; Is it likely to see AMD contributing more to Mesa when all the needed features are implemented in amdgpu/dal? I think there're still a lot of potential for optimization. (:

I don't really understand the question, particularly the "when all of the needed features are implemented in amdgpu/dal" part - Mesa and amdgpu/DAL implement different parts of the driver stack functionality and both of them will continue to need to be updated.

[juno]

Then you would get to keep running without the features that DAL brings. I don't see how that would be a big win.

If we can't find a common ground to get DAL upstream in time then we would probably implement the non-DAL support ourselves, but that would be throwaway work using time taken away from getting DAL upstream so hoping to avoid it.

I didn't say it would be a win, I just wondered. So this is actually the plan. And any "enthusiast" could possibly prevent us from having dal after all...

[Adarion]

Well, I hope it'll make it into Kernel / mesa / ... releases before Vega hits the streets. Would be nice to have launch day support (rather than using vesa drivers and the likes).
I wonder how all this (Zen CPU, Vega GPU and HBM2 memory) is going to fit on one AM4-chip - fit by size and TDP and how a stacked memory will deal with the heat dissipation...

[Veto]

4101 insertions, 8868 deletions

Well, it seems to be right on track
Is this statistic relative to mainline or relative to the previous DAL code?
If it actually manages to reduce the size of the current mainline code AND bring new features, it should be a no-brainer to merge it.

[smitty3268]

Well, it seems to be right on track
Is this statistic relative to mainline or relative to the previous DAL code?
If it actually manages to reduce the size of the current mainline code AND bring new features, it should be a no-brainer to merge it.

Yeah, -5000 lines of code seemed like a no-brainer to get included immediately.

Then i realized that this was obviously just the changes to the existing DAL patches that had cleanups over the last month. I don't know what the initial DAL patches were, but they were pretty huge IIRC.

[GreatEmerald]

Would be nice to have launch day support (rather than using vesa drivers and the likes).

Hm, speaking of which, vesa drivers do not exist (or are never used) any more, right? Since it's now the modesetting DDX + llvmpipe.

[Adarion]

x11-drivers/xf86-video-vesa does still exist and may sometimes be the only (nearly failsafe) option to get an image on the screen. There is also a kernel VESA console driver and some usermode vesa thingy. Would be really neat if we had at least KMS and stuff for all these chips not yet supported.