Compute load not showing in fdinfo chart when running LuxMark v3.1 about amdgpu_top

[Umio-Yasuno]

I have installed amdgpu_top from git and run LuxMark v3.1 (GPU benchmark with the default scene - Luxball) to benchmark OpenCL performance with rocm-opencl package v2.0.0-40502-sles152.164 from openSUSE Tumbleweed (20231022), Linux 6.5.8.

I was expecting the compute load on the fdinfo chart to go to 100% during the benchmark, but nothing happened. I have an AMD Radeon RX 6800 XT Midnight card.

Could you try ls -l /proc/$(pidof luxmark.bin)/fd (or ls -l /proc/$(pidof luxmark)/fd ?) and
grep -r "drm-engine" /proc/$(pidof luxmark.bin)/fdinfo (or grep -r "drm-engine" /proc/$(pidof luxmark)/fdinfo ?) when running luxmark?

Same with GravityMark 1.81 Vulkan/OpenGL computer shaders benchmark, but GFX load does go up to 100% in this case.

I do not know what GravityMark is doing internally.
But the compute usage for vkpeak and gamescope is displayed.

Also, is there any chance amdgpu_top could show the decoder currently in use (AVC, HEVC, VP9 ...) ?

There is no way to get it I think.
UMD/KMD does not leak file codec information to the outside world.

[Umio-Yasuno]

As far as I can tell, ROCm uses /dev/kfd, so no usage is recorded in fdinfo.
I will be working on support for the amdkfd (/sys/class/kfd/kfd/proc) driver.

[Umio-Yasuno]

Unfortunately, I couldn't find a way to get gpu usage per process from amdkfd.

[Umio-Yasuno]

I couldn't find a way to get gpu usage per process from amdkfd,
so I recommend that you refer to GFX activity in GPU Metrics or performance counters (GRBM, GRBM2).