As part of testing the ACO shader compiler from Valve, I'm going to benchmark games prior to adding the Ubuntu PPA. But Talos Principle frustrates me a great deal. Can someone PLEASE tell me how to toggle Vulkan on in the game? I've followed instructions, carefully I'd say, from a pinned thread on game's official discussion forum on Steam. The results end up with 86 FPS for OpenGL and about 75 for Vulkan. I could not find a way to verify Vulkan was running through game settings (or be able to toggle between two APIs). All I could do was search for Vulkan in games Talo.ini file, which includes a vulkan variable when you follow the instructions I mentioned. But 75 for Vulkan and 86 for OpenGL - what's going on??

thx good to know !

Updated my script with workaround for Meson bug, and it can now also handle customizing Mesa repo to build both regular Mesa and Mesa-aco using the same script. See example in the inline documentation.

Testing X4: Foundations, using Mesa-aco build with llvm9.

llvm:


aco:


ACO performs worse actually.

As part of testing the ACO shader compiler from Valve, I'm going to benchmark games prior to adding the Ubuntu PPA. But Talos Principle frustrates me a great deal. Can someone PLEASE tell me how to toggle Vulkan on in the game? I've followed instructions, carefully I'd say, from a pinned thread on game's official discussion forum on Steam. The results end up with 86 FPS for OpenGL and about 75 for Vulkan. I could not find a way to verify Vulkan was running through game settings (or be able to toggle between two APIs). All I could do was search for Vulkan in games Talo.ini file, which includes a vulkan variable when you follow the instructions I mentioned. But 75 for Vulkan and 86 for OpenGL - what's going on??

I think that in the last update of the engine they probably removed the option to switch the graphic api (i.e. will always use vulkan). I'm 100% sure of this for Serious Sam (I definitely made the question in their forums).

Regarding the performance, I remember that TP always had better fps with OGL

If you find the time, could you by chance adopt the script so that the output wents to /opt/mesa-aco directly? It always lands in $HOME/mnt/vmshare/opt/mesa-aco/opt/mesa-aco . I tried adopting it myself but then it always lands in /opt/mesa-aco/opt/mesa-aco

I can take a look how to do it. The reason I put it in $HOME/mnt/vmshare in general, is that I like to build it in VM, that has a share with the host Linux (both through $HOME/mnt/vmshare). So I simply use another small script to publish the result on the host from that to /opt/mesa-master proper.

However, it should be possible to override the destination with the env var. Let me add that.

If you find the time, could you by chance adopt the script so that the output wents to /opt/mesa-aco directly?

I updated the script.

Try doing:

dest_dir=/ mesa_debian_build.sh

I don't recommend such method though, since it will try to write directly to /opt/mesa-<branch>/* and delete stuff there as well. It's preferable to build in some local isolation, and deploy as a separate step.

Supraland

llvm:

aco:

Both are pretty bad, due to Unreal Engine not using Vulkan effectively and frametimes being completely messed up.

Both are pretty bad, due to Unreal Engine not using Vulkan effectively and frametimes being completely messed up.

The biggest part seems to be Supraland itself. The erratic frame timings show that it's clearly doing something it's not supposed to. Nothing that can be fixed by the shader compiler.

Heh, compiling llvm itslef is a major pain, even on 3900X I assume. I just use the nightly snapshot form their repo: http://apt.llvm.org/unstable/pool/main/l/llvm-toolchain-snapshot/

Since we are discussing llvm, I'm looking for a way to add a suffix to the name of the targets.
For example, clang would be renamed clang-32. To do this with autotools you would add the
option, --program-suffix=-32. But I haven't figured out how to do this with cmake.

Since we are discussing llvm, I'm looking for a way to add a suffix to the name of the targets.
For example, clang would be renamed clang-32. To do this with autotools you would add the
option, --program-suffix=-32. But I haven't figured out how to do this with cmake.

Why do you need clang though? For Mesa you only need libllvm.

off-topic, but anyone tried AMDVLK from github recently?? with the latest pull I got a 5 fps increment in Shadow of the TR which is a lot considering the total average fps is 45

I haven't used it in a while, since radv became very good. But I can try building it again and compare with aco.

Interesting results. Using aco, vs Mesa master vs latest amdvlk in TW3 with dxvk master, aco was actually behind this time, but slightly.

radv/aco:

radv/llvm:

amdvlk:

That's just framerates, I didn't really analyze stutter.

Just tested Mesa-aco with Star Citizen (see the thread. It works, but the game crashes during quantum flight. Does it work for anyone?

How did you get Star Citizen? Did you pay for it with no contracts?

How did you get Star Citizen? Did you pay for it with no contracts?

Just pledged for the starter Aurora package.

It runs shamelessly brilliant now, I am totally intrigued. After the initial very low shader stutters, the game runs like a charm. FPS is still borked by the game itself, lacking optimization but that is for most windows users as well.

https://www.youtube.com/watch?v=zw7FQNMsbxw

Is it just as slow in Windows?

Yes, amdvlk has improved. but that's just for framerate, I didn't measure stutter levels vs radv/aco.

you guys should try AMDGPU pro driver 19.20, shaders compile so fast but some games has graphical bugs like unreal engine 4 games, its almost stutter free

My small tests:

TALOS Principle: 75.6 - 76.4(ACO)
The Witcher 3: 77 - 82 (ACO)
PREY (demo) : 69 - 73 (ACO)

General note: Games stuttered alot without ACO when running for first time or when they had not been played for a while. To a point games would freeze a frame briefly. With ACO the stuttering was noticeably less to say the least.

Please post hardware specs and screenshots with HUDs for better visibility. Thanks!

Yes sir. I will repost soon :) Deleted initial post because I want to redo whole process again.