mpv doesn't respect display primaries on Wayland on SDR wide gamut displays, is this a bug or intentional?

[fililip]

As the title suggests, mpv doesn't seem to respect wide gamut display primaries in SDR mode on Wayland with a color profile unless target-trc=pq.

There are three things I tried:

1. Leave mpv.conf alone
   This results in bt.709/gamma2.2 being set:
   

2. Set target-trc=pq
   This looks like it works as intended, every display gets its own primaries retrieved and some test JXL images exported from Krita with cICP data get correctly displayed as Rec.2020.
   

3. Set target-trc=gamma2.2 or target-trc=linear
   This one is weird, I get display-p3 primaries, which are narrower than the display, but content appears correct with P3 colorimetry.
   

Video metadata:

My display's primaries (from the EDID):

• Red: 0.7050, 0.2958
• Green: 0.1923, 0.7548
• Blue: 0.1435, 0.0400
• White: 0.3134, 0.3291

In HDR mode everything seems to just work automatically with all content types from my testing. The only question I have is whether SDR wide gamut color management is supposed to default to bt.709 like that. Is there something I'm missing perhaps?

Specs:
Operating System: Arch Linux
KDE Plasma Version: 6.6.4
KDE Frameworks Version: 6.24.0
Qt Version: 6.11.0
Kernel Version: 6.19.11-arch1-1 (64-bit)
Graphics Platform: Wayland
Processors: 16 × AMD Ryzen 7 5800X 8-Core Processor
Memory: 64 GiB of RAM (62.7 GiB usable)
Graphics Processor: AMD Radeon RX 9070

Thanks!

[kasper93]

Could you post what is preferred colorspace, you can see that in verbose log. Also post what are the available vulkan colorspaces.

I would guess you are using mpv stable 0.41, which uses Vulkan API for colorspaces. And depending on selected transfer different primaries are available. For example for G2.2, you only have bt.709, so the real display primaries are clipped to this. For linear only display-p3 is available, so again primaries are clipped. For PQ, bt.2020 are available, and in this case we use real display primaries. This is a guess, but likely what is happening. Likely NVIDIA drivers, mesa gives all combinations.

With master version of mpv, we use Wayland protocol directly, so we have more flexibility, and likely this issue is resolved.

[kasper93]

What was 0.41 output stats (or log)? Generally for PQ_2020 output it shouldn't change much, unless primaries were not set before. I think we also not clamp primaries to container, so might get clipping a bit.

[fililip]

Here's the log from mpv 0.41 (and older system-wide libplacebo) with target-trc=pq: log.txt. In this case, the colors are as they should appear for Rec.2020.

[fililip]

I apologize, it looks like I patched/cherry-picked something wrong before (or KWin night light was producing some differences in luminance), after a fresh repo reclone (+ no night light) everything appears to work and look correctly (SDR and HDR) with commit 701860c applied and master libplacebo now. I re-checked with multiple images that contain embedded ICC profiles (P3, Rec.2020, Rec.709). Thanks for fixing it so quickly!

Here's the log file from the now-working configuration in case you need it: log.txt.

[kasper93]

Should be fixed on master now. Thanks for reporting this.

[fililip]

Probably not related to this issue, but should different target TRCs work? When I select target-trc=gamma2.4 for instance, mpv falls back to bt.709/gamma2.2 again.

Edit: It looks like some are supported and some aren't, so I guess it depends (probably on the color management protocol)?

[kasper93]

Yes, if some combination of requested parameters is not supported, we fallback to bt.709/gamma2.2.

[kasper93]

You can use --target-colorspace-hint-strict=no to do the conversion to requested params, but it will not be compatible with what is configured in output to the compositor.