Testing The Open-Source "RADV" Radeon Vulkan Driver vs. AMDGPU-PRO

[grigi]

Wow. This is impressive. Well done David & Bas!
Do you have plans to backport it to CI and SI?

[jrch2k8]

irrelevant, intel does not use llvmlol, that compiler will generate code for intel gpus

Nir is an IR and in theory SPIR-V should in theory be portable to the mesa compiler that intel uses. Since AMD Gallium driver don't use NIR i had doubts which parts they used. but whatever B already answered that NIR->LLVM.

My point was on SPIR-V to NIR IR, since both intel and RADV use it, it should help having more eyes optimizing this stage then easying the job of their respective compilers and layers after it.

[lethalwp]

Btw. an article about the state of HEVC Main 10 decode would be interesting as vdpauinfo/vainfo show support for it with Polaris GPUs.

For what i've tested, hardware decoding of hevc in main10 in polaris works like a charm, in vaapi. For that i used ffmpeg-git + this homemade patch: http://ffmpeg.org/pipermail/ffmpeg-d...st/197623.html then use mpv with -hwdec=vaapi and opengl renderer.

But i've also noticed some problems when hardware decoding a simple mpeg2 from a dvd on a rx480. That i don't know where the problem is
(bad looking blocks, sometimes even flickering blocks)

[jrch2k8]

Wow. This is impressive. Well done David & Bas!
Do you have plans to backport it to CI and SI?

Well if the GCN 1.0 AMDGPU support lands on 4.9, it should be mostly doable on all GCN able hardware except when you need hardware specific fixes that could take more time

[nadro]

Those results look really nice! Great work! Thanks Michael for this benchmar.

BTW. I think that you should run benchmarks in fullscreen mode, because in windowed/borderless mode compositor affects performance.

[bridgman]

it's not just legal review. somebody has to turn windows vuklan driver to opensourceable form. they have no resources for that, so do not expect they will magically find resources to maintain another driver

Right... we knew at the start that any IP/Legal review would fail dramatically (and it did)... the work involved rewriting big chunks of the driver, and the driver itself is already much larger because it shares code across multiple OSes and multiple APIs.

BTW. I think that you should run benchmarks in fullscreen mode, because in windowed/borderless mode compositor affects performance.

In the article Michael said that the driver wasn't working in fullscreen, so he had to run in windowed mode.

[Mystro256]

Those results look really nice! Great work! Thanks Michael for this benchmar.

BTW. I think that you should run benchmarks in fullscreen mode, because in windowed/borderless mode compositor affects performance.

Straight from the article:

When trying Dota 2 with fullscreen mode at various resolutions, the screen basically appeared locked and wouldn't update past the loading screen. But when alt-tabbing out of the game, a few new frames of the game's demo would proceed to appear... Alt-tabbing again, a few more frames. Or if leaving the game minimized so it would be off-screen, the demo would complete but would get just a few frames per second.

I tried enabling/disabling the Steam overlay, switching between the AMDGPU and modesetting DDX drivers, switching from Compiz/Unity to Xfce with/without compositing, and other steps to try to figure out why the screen wouldn't update in the full-screen mode. But the easiest approach for now was just running Dota 2 in windowed mode. When in windowed mode, there were no issues.

[Mystro256]

Right... we knew at the start that any IP/Legal review would fail dramatically (and it did)... the work involved rewriting big chunks of the driver, and the driver itself is already much larger because it shares code across multiple OSes and multiple APIs.

Does this mean the release is just stalled but is still coming? or do you have no idea/not allowed to say?

[Evil Penguin]

Right... we knew at the start that any IP/Legal review would fail dramatically (and it did)... the work involved rewriting big chunks of the driver, and the driver itself is already much larger because it shares code across multiple OSes and multiple APIs.

Would it then just make more sense to support the RADV driver?
It seems to be off to a really good start.
Sigh, pesky legal BS...

[pal666]

Would it then just make more sense to support the RADV driver?

if you will be happier by adding zero developers to radv, then sure, let them add zero developers to radv