10:08 shelter: Curious, maybe you can't or don't want to answer this but will the upcoming descriptor changes help radv too?
11:23 Venemo: shelter: you can watch Faith's talk here: https://video.tuwien.ac.at/events/xdc/v/OlwauRVEIGa it starts 12:30 in the video. that will give you all the details you need
11:44 shelter: Venemo: i did that, its seems very good for Nvidia, but I was not sure about the amd-side of things. from what i understood tho', is that it's a nice "change" overall.
11:45 shelter: I mean, amd does pretty well as-is.
11:48 shelter: Btw, thanks for helping me with the perf tracking in indy.
12:02 Venemo: shelter: I am not working on that feature so I can't comment on its perf. the way I understood Faith's talk is that the same perf issue didn't exist on AMD HW to begin with.
12:03 hakzsam: shelter: shouldn't change anything for RADV
12:07 shelter: hakzsam: ok, thanks!
14:49 Venemo: johnny0: what happened ultimately with your resizable bar issue on Hawaii?
19:39 johnny0: Venemo: it's still there -- I noticed the VCE issue while gathering information on the memory pressure failures seen during regression testing
19:47 johnny0: basically if you hammer UVD and/or VCE on CIK, even without rebar, lots of failures appear
19:48 johnny0: most consistently if there's a lack of vram (either because most of it is used or there just isn't much)
19:49 johnny0: on 5.19.17 (pre-drm-buddy) these don't seem to appear
20:10 Venemo: johnny0: are you still working on a patch to deal with it?
20:51 johnny0: yes, but I certainly won't mind if someone takes a fresh look at it
20:53 airlied: is this the some buffers need to be below 256MB VRAM?
20:54 Venemo: johnny0: do you have a branch somewhere?
20:55 johnny0: just locally
20:55 Venemo: I can try and take a look if you don't mind pushing it somewhere
20:55 johnny0: airlied: seemingly, or at least within the same 256MB segment
21:16 airlied: ah yes one segment sounds familiar
21:19 johnny0: Venemo: I'll try to wrap up the non-rebar [re-]testing tonight and get some issues posted to the tracker
21:19 johnny0: it's important context for whatever branch I'd make available
21:19 Venemo: thanks
21:20 Venemo: sounds really bad, especially that it happens even without rebar
21:20 johnny0: yep
21:21 johnny0: radeon isn't it much better shape... there's a reason I asked about the vram size in #4601
21:22 Venemo: honestly I don't care much about radeon, but I care about not regressing users when we switch amdgpu to the default