RADV's ACO Back-End Is Helping Radeon Navi Linux Gaming Performance

[phoronix]

Phoronix: RADV's ACO Back-End Is Helping Radeon Navi Linux Gaming Performance

It's been almost two months since last looking at the RADV ACO performance for this shader compiler back-end alternative to the AMDGPU LLVM code. ACO is making its debut in the upcoming Mesa 19.3 release while since the last round of testing have been more optimizations and fixes as well as getting the Navi/GFX10 support in place. In this article are some fresh benchmarks of the Vulkan RADV ACO support for not only Polaris and Vega but also the Radeon RX 5700 Navi graphics cards.

http://www.phoronix.com/vr.php?view=28527

[shmerl]

When comparing llvm and aco for The Witcher 3 (Wine+dxvk) on RX 5700 XT, llvm is actually slightly ahead in raw framerate. So ACO has still catching up to do.

I didn't build the very latest master for a few days though, so may be more optimizations rolled in already.

[Termy]

The biggest boost i've seen with ACO is with Star Citizen. Not really with the max FPS, but the game suffers under heavy studdering when shaders are compiling first time under linux. And that is greatly reduced with ACO

[M@GOid]

The results for Shadow of Tomb Raider are higher on 1440p than the 1080p test of last week, even without ACO.

[Michael]

The results for Shadow of Tomb Raider are higher on 1440p than the 1080p test of last week, even without ACO.

Keep in mind earlier Shadow tests were on Linux 5.3 (not 5.4), slightly older Mesa Git rev, etc.

[R41N3R]

Pretty impressive, even though I use ACO mainly for its much better user experience and frame times. And not to mention that llvm release time are not aligned to GPU driver releases and it suffered quite often from serious bugs too.

[Strunkenbold]

My theory why ACO yields better framerates:
llvm has a big problem with some shaders, resulting in crazy stuttering and even noticible stalls. Second and third runs are usually better. But because first run counts to the result, in the end ACO wins. Of course it's totally correct that ACO takes the win overall.
I really wish, that AMD sees that advantage, both of radv and ACO and focus on them. Llvm is probably still needed for Opencl. But no one needs amdgpu-vlk. Some say it's a race between multiple backends and it's good for competition. But I see that none of both sides have enough man power to provide a satisfied user experience.

[agd5f]

Llvm is probably still needed for Opencl. But no one needs amdgpu-vlk. Some say it's a race between multiple backends and it's good for competition. But I see that none of both sides have enough man power to provide a satisfied user experience.

If you are going to make that argument, then why do we need ACO? Why not put that effort into LLVM? LLVM is used for OpenCL, radeonsi OpenGL, amdvlk, as well as all of the HPC and ML stuff that is built on top of ROCm. It doesn't really make sense to invest in two compilers if resources are limited.

[smitty3268]

If you are going to make that argument, then why do we need ACO? Why not put that effort into LLVM? LLVM is used for OpenCL, radeonsi OpenGL, amdvlk, as well as all of the HPC and ML stuff that is built on top of ROCm. It doesn't really make sense to invest in two compilers if resources are limited.

Do you believe the fundamental architecture of the LLVM compiler allows it to possibly reach the compile speed that ACO provides? Or do you think that is not a worthwhile goal to pursue?

[SethDusek]

Has anyone tried ACO with GTA V on an rx580? For some reason all the colors during daytime are overblown (super white), to the point that it can be impossible to see. This issue does not occur with the llvm driver.