00:07Mangix: Venemo: unfortunate
00:22Mangix: interesting. so with waylandvk, vaapi does not work. vaapi-copy does though.
03:31johnny0: Venemo: when you were testing VCE on SI, did the throughput drop when no display was connected?
03:34johnny0: (with 'auto' dpm)
09:56Venemo: johnny0: I haven't really tested VCE1 with no display connected
09:56Venemo: johnny0: I would say probably the drop is expected, because the DPM code sends SI to a lower power state when no display is connected
10:05Venemo: Mangix: as far as I know, from what the video experts tell me, there are some features of Vulkan video that aren't supportable well on old GPUs. I am not familiar with the details though. That said, all the relevant Mesa code is open source so someone can try if they want to.
10:05Venemo: Mangix: at the moment my personal focus is to just get SI and CIK to feature parity and switch them to amdgpu by default, in order to improve the out of the box gaming experience
10:32KitsuWhooa: Venemo: I just want to say that the work towards SI and CIK is extremely appreciated
10:32KitsuWhooa: thanks :)
10:33Venemo: you're welcome, KitsuWhooa
10:39johnny0: Venemo: ahh gotcha, thanks anyway -- the drop on Hawaii is significant enough to at least see if you ran into it
10:45Venemo: johnny0: I guess you can combat it by manually setting a higher power state
10:46Venemo: or we could tell the SMU not to do it when there is some load
10:47Venemo: honestly, IDK why that was added, but I haven't touched it thus far
10:49johnny0: so far what I've found is that the sclk ramps up properly but not the mclk
10:50johnny0: setting the dpm state set manually works, but resets whenever a display changes power state
10:54Venemo: johnny0: is that Hawaii or SI?
10:55johnny0: for Hawaii
10:55Venemo: I see
10:55johnny0: my worry is that the behavior may also be present on the older cards
10:55Venemo: maybe
10:56Venemo: tbh, I haven't looked much into VCE on Hawaii becuase I trusted the people who told me that it had been already working
10:57Venemo: for SI, see the comment here: https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/amd/pm/legacy-dpm/si_dpm.c#L4231
10:57johnny0: i'll check it out, thanks
10:58Venemo: the comment makes me think that it's a completely bad idea to ever use PPSMC_MSG_NoDisplay, but IDK why it was added so didn't dare try to remove it
10:59Venemo: amdgpu doesn't use this message on CIK, but radeon does
11:14johnny0: Venemo: wow yeah, this is definitely worth taking a look at, thanks again
11:15Venemo: once things stabilize, I have plans to improve the SI power management code further, but for now I am just happy to have feature parity with radeon
21:20Mangix: Venemo: no worries. hardware acceleration is not important for this platform. VCE would be more interesting.
21:20Mangix: In other news, looks like 6.17.8 has some amdgpu backports. Hopefully they fix DC on SI.
21:54Mangix: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=v6.17.8&id=93392aba88ede413c7872d90eb5afbf4b12229f7 :(
21:55Mangix: I abused this back in the day to overclock a dual link DVI monitor past its limits. I remember when turning DC on it wouldn't let me get the clocks as high.
21:56Mangix: on Windows I could go even higher: https://www.monitortests.com/forum/Thread-AMD-ATI-Pixel-Clock-Patcher
21:59Venemo: Mangix: you can clock it as high as you want, but not more than the chip can do. it just doesn't work