01:14 DemiMarie: agd5f: why does it require firmware changes? I thought Windows already used user queues.
02:13 kode54: Windows uses different firmware anyway
02:13 kode54: or so I heard?
02:26 DemiMarie: different firmware is what Nvidia has moved away from
02:54 kode54: what about AMD and Intel?
09:47 MrCooper: melissawen: https://gitlab.freedesktop.org/drm/amd/-/issues/4144 might need your cursor gamma magic again :)
09:52 fililip: kode54: I'm certain it's different on Windows since I remember the difference in handling ROCm on gfx10+; it used to hang on Linux with multi-queue due to missing out of order dispatch and fixing it required firmware changes in addition to a kernel patch, but it worked fine on Windows with Blender
12:43 melissawen: MrCooper, ouch! yes, it seems so... thanks for pointing it out
12:44 agd5f: fililip, it's the same firmware for windows and Linux
12:46 agd5f: DemiMarie, we asked for it. a secure fence packet where userspace can't mess with the fence value. It's not strictly necessary in retrospect, but when we were designing the solution, that's what we came up with
13:50 MrCooper: melissawen: no worries, let me know if there's anything I can do to help; I just bought a 9070, so I can reproduce and test fixes
13:52 melissawen: MrCooper, nice, I don't have the hw indeed, but I checked the code and I think they should just revert that commit and everything will work as expected on Linux... and they can set the ENABLE_CURSOR_DEGAMMA flag by default in other OSes
13:53 melissawen: I'll send the patch reverting it asap
13:53 MrCooper: awesome, thanks!
13:53 melissawen: thank you!
15:23 MrCooper: melissawen: FWIW, https://gitlab.freedesktop.org/drm/amd/-/issues/4144#note_2865522 claims that change didn't fully fix it for one user, might be something else going on there though
15:42 melissawen: MrCooper, so, this second issue I need to think more on how to solve, I'm anot aware of any implicit HDR in the driver, so I suspect cursor and other planes are blended in different color spaces.
15:43 melissawen: I wonder if the driver should implicitly linearize those planes and delinearize again after blending because I don't think there is a way to do color mgmt on hw cursor besides the sRGB degamma ROM :/
15:45 melissawen: probably better if there is an issue specific to deal with the cursor on hdr problem
15:45 melissawen: hwentlan_, ^ any thoughts?
15:48 MrCooper: FWIW, mutter just scales down the cursor pixel values to match the reference white level in BT.2020/PQ, which seems to work fine with older AMD HW, maybe something changed in RDNA4 though
15:50 MrCooper: to be more precise, it converts the pixel values from sRGB to BT.2020/PQ
16:13 melissawen: MrCooper, but that cursor on HDR issue happens on mutter? I'm confused by some reporting that the cursor gamma issue only happens for HDR and SDR is correct, and others saying that happens for both
16:18 MrCooper: melissawen: the issue without your revert affects both, it's much more noticeable with HDR though (cursor appears mostly black), whereas it's barely noticeable with SDR, at least with default mostly black/white cursor shapes
16:19 MrCooper: presumably because the former requires pixel values smaller than 0xff for reference white
16:20 melissawen: ok, so it means the reporter doesn't noticed the issue when they say it "does not happen in SDR mode" here: https://bugs.kde.org/show_bug.cgi?id=501254
16:23 melissawen: anyway, I'll check the behavior with the hw I have while I examine the dcn4 code to see if there is something different
16:25 MrCooper: sounds plausible indeed
16:26 MrCooper: the drop shadow is too dark, easy to miss though
16:29 MrCooper: AFAICT it's very similar if not identical to the corresponding issue on older GCN
17:26 DemiMarie: agd5f: will the firmware changes be done on all non-EOL GPUs?
17:31 agd5f: all that can support the functionality
18:39 DottorLeo: hi!
18:40 DottorLeo: is there any command for the terminal to print the "technical" information about the gpu like CU, UVD,VCN version etc...?
18:43 agd5f: DottorLeo, /sys/class/drm/cardX/device/ip_discovery
18:44 agd5f: dmesg also print some of that
18:45 DottorLeo: agd5f: thanks, if i have multiple cards?
18:45 DottorLeo: card1?
18:47 agd5f: card0, card1, etc.
18:49 DottorLeo: File or directory not existing ^^"
18:49 DottorLeo: just pasting that on the terminal?
18:58 agd5f: ls /sys/class/drm/card0/device/ip_discovery
19:10 DottorLeo: @agd5f sorry need to restart
19:10 DottorLeo: that command is needed on the terminal like that? just paste it?
19:24 agd5f: DottorLeo, it's a directory structure
19:24 agd5f: you can walk through it to see the IPs and their information
19:24 DottorLeo: ah ok!
19:25 agd5f: cat the files to see their values
19:48 DemiMarie: agd5f: which devices will be able to support it?
19:51 agd5f: gfx11 and newer