AMD Developer Posts More Patches For RADV

[cj.wijtmans]

Modularity on both is on same level. Don't want a driver? Cool, you don't need to compile it. If you dislike the fact you get all drivers in one package, blame your distro

i use a meta-distro. I already custom-compile my kernel. If you dont understand what i am talking about why evenb other replying?

[cj.wijtmans]

What does "monolithic" mean in this context? Because the AMD or noveau kernel modules are well, modules. But the kernel driver isn't related with 3D rendering code, it only exposes the hardware to Mesa.

I know that. But expansions that are computers on their own should not be in the kernel. iGPU? fine, maybe. Just look at the whole AMDGPU, nouveau, nvidia driver mess.

[starshipeleven]

I know that. But expansions that are computers on their own should not be in the kernel. iGPU? fine, maybe.

Sure. How do you plan to talk with these extensions after you removed the interfaces to them from the kernel/userspace? And how to make sure that graphics tasks that are better executed by the CPU aren't choking the GPU?

Because that's what is in the kernel modules and Mesa, code allowing load sharing and communication with a very specialized piece of hardware that sucks outside of its narrow area of expertise, so the CPU must babysit this thing.

Just look at the whole AMDGPU, nouveau, nvidia driver mess.

I'm not sure with what you mean by "mess" here. The task is complex, the hardware is also quite complex. Sure they can be better, but it's not going to be significantly less complex than now.

(inb4 libv mentions that if they didn't make AtomBIOS it would have been so much better)

[cj.wijtmans]

Sure. How do you plan to talk with these extensions after you removed the interfaces to them from the kernel/userspace? And how to make sure that graphics tasks that are better executed by the CPU aren't choking the GPU?

Because that's what is in the kernel modules and Mesa, code allowing load sharing and communication with a very specialized piece of hardware that sucks outside of its narrow area of expertise, so the CPU must babysit this thing.

I'm not sure with what you mean by "mess" here. The task is complex, the hardware is also quite complex. Sure they can be better, but it's not going to be significantly less complex than now.

(inb4 libv mentions that if they didn't make AtomBIOS it would have been so much better)

How does windows do it?

[andrebrait]

fingers crossed for moving fixes from amd vulkan to radv via radeons lol?
i would like your self-punishing wish to be fulfilled, if it would be possible to do so without affecting me. unfortunately, it is not possible, so no, we will not see idiotic solutions just to please you

Wut?

I know I'm not supposed to feed the trolls, but I'm assuming you might not be one...

I just meant I'd like to see some love given to open source solutions that already work (and do so pretty well).

Is there anything so wrong with RADV? Explaining it might be better than saying "we will not see idiotic solutions just to please you".

[nanonyme]

Apparently with a gigantic unified runtime that handles scheduling and driver restarts and loads and loads of common functionality and smaller drivers (WDDM2)

[starshipeleven]

How does windows do it?

Does it? Because I saw breakage in proprietary drivers (up to causing BSODs) all the times MS touches the video subsystem.

People is still used to Win7 that was pretty fucking stable in this regard as it entered the "only security bugfixes" stage pretty early due to marketing for Win8, but that's just because they didn't touch video subsystem. Easy mode.