Created attachment 134765 [details] good perf gallium hud on 4.14 kernel. R9285 Tonga, Compared to older kernels I have a perf regression with latest 4.14.0-rc3 based and previous 4.13-rc5 based agd5f 4.15-wip kernels. This doesn't seem to affect other things like xonotic or unigine demos. Issue = on fixes 4.14 and other older kernels I've tried, the game is OK ish perf wise, 25-60 fps. On 4.15-wip its 10fps. Using HUD one obvious difference is that the slow case is only mostly using one cpu core, on a good kernel all 4 of my old phenom 2 x4 are used - possible red herring of course, I'll upload some screenshots. The game is pre-alpha and doesn't always render properly - but it was vaguely playable fps wise until this kernel.
Created attachment 134766 [details] bad on previous 4.15 kernel cpu load different. All tests done with both CPUs and GPU forced to perf.
Created attachment 134767 [details] bad on latest 4.15-wip cpu still different but not as obvious.
According to the HUD, the GPU load is actually higher in the bad cases. Have you checked what clocks the GPU runs at in each case?
https://cgit.freedesktop.org/~agd5f/linux/commit/?h=amd-staging-drm-next&id=e5f6a57e350a7921e4edc30874679bdff11b13f4 might help.
I had that fix and a couple of others on the first 4.15-wip I tried. They are in now AFAICT. I see Alex updated again since yesterday so I'll try that one later. I did notice but forgot to highlight the GPU difference. Given all the powerplay changes in 4.15-wip I do wonder if all is well but I would have expected Unigine/Xonotic to show it if something was wrong there. One thing different about UT is it asks for more vram than I have, maybe that makes it stand out from other tests. After posting this I did wonder whether the CPU difference is just because the fps is low and not the cause. 4.15-wip doesn't work for me going away from head (will throw some vmfaults and then lock soon after startx) so bisecting is not an easy option.
I tried latest 4.15-wip and it's still low perf. On this test the perf was OK ish initially but degraded over the first 20 seconds. For added hassle it seems that between 4.13 and 4.14-rc1 something changed that means I may hard lock when configuring ffmpeg! I am trying to find that one on linus git, but now agd5f have been rebased they are not stable for me.
This is OK now with amd-staging-drm-next, I'll close when 4.15s get updated assuming they get whatever fixed it as well.
It's not OK, I don't know what happened there, maybe I wasn't in the kernel I thought, though I am not convinced. Whatever the reason, I can't get it to work now.
possibly this is https://bugs.freedesktop.org/show_bug.cgi?id=103100 Though I am not sure as I lock soon after startx if on the commit called there On the commit before I am OK and UT perf is good.
Testing from head of drm-next-4.15 I can change minor to 19 and 20 in drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c and am stable so can test. 19 gives good perf 20 bad, so setting as duplicate. *** This bug has been marked as a duplicate of bug 103100 ***
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.