Seven Changes We've Been Waiting On That You Will Not Find In Linux 5.7

[Zico]

I'm just about asking questions, (...) I'm not trying to take a position on anything I don't know.

Relax, while I express my hopes. You asked good questions and I answered as much as I know. I'm not expert, but I think some drivers/instructions (AMD/INTEL/AnotherOne) may be in the kernel.

[coder]

Do you agree with me?

I'm just about asking questions, such as what relevant hardware assist is known to exist and how much drivers actually have to do with harnessing it. I'm not trying to take a position on anything I don't know.

As for AV1, more support -- and especially by fixed-function units -- is definitely better.

[Zico]

(...) which could be entirely a userspace affair.

I agree with You, but better performance is possible with hardware encoding & decoding capabilities (Tiger Lake will get AV1 decoding - Hardware lvl). Without hardware solution, "soft" acceleration overloads cpu/gpu processors ..
Do you agree with me?

[jacob]

Is that really what's holding it up? I missed that, if so.

AFAIK yes, that's currently the main problem.

[coder]

I know that Linus doesn't want any form of in-kernel dbus until the current performance problems have all been addressed in userland,

Is that really what's holding it up? I missed that, if so.

[coder]

Exactly, chip drivers (at kernel level). I hope AMD will have this chip (Hardware AV1) in the APU. I'd like a ready kernel with GPU drivers supporting "hardware AV1". so Are you not?

Do we know whether, how much, and which generations of AMD hardware can support it with fixed-function units? Otherwise, I think you're looking at something like OpenCL "soft" acceleration, which could be entirely a userspace affair.

[jacob]

I know that Linus doesn't want any form of in-kernel dbus until the current performance problems have all been addressed in userland, but IMHO that's a short sighted approach for once. We need a kernel-based dbus implementation sooner rather than later, not just for performance reasons (although it would help) but also for security, auditing and monitoring. Ideally current DAC and MAC mechanisms would naturally apply to it.

[Zico]

Hmmm... to what extent is that even dependent on (in-tree) GPU drivers? Because that'd be the only thing it would have to do with the Kernel.
I think most of AV1 support would be implemented in userspace libraries.

Exactly, chip drivers (at kernel level). I hope AMD will have this chip (Hardware AV1) in the APU. I'd like a ready kernel with GPU drivers supporting "hardware AV1". so Are you not?

[starshipeleven]

Me, and all the other people that up-voted the post. We don't want any of this foolishness. Get off my lawn.

Fuchsia OS for the win.

[coder]

full AV1 support (for better displays e.g notebooks)

Hmmm... to what extent is that even dependent on (in-tree) GPU drivers? Because that'd be the only thing it would have to do with the Kernel.

I think most of AV1 support would be implemented in userspace libraries.