[TESTING][S905(X)] 10bit/HDR/Dithering Test Builds & Discussion

  • I definitely see vertical lines artifacts in 2160p / 10bit videos in the testing1 build.
    Also, there is some weird macroblocking and flickering, for instance in the Samsung Wonderland demo in some of the skies - look for the top middle of the screen. Those artifacts are gone when I use the TV internal media player.

    I think 1080p content is fine.



    My AVR does not like 4:2:2...
    Well, it would have been for testing only. In the end I need to think about another solution to get HD audio to the AVR.

    The vertical lines have been in all 10bit output I've had on s905x (android and LE), so I'm not sure what can be done about them. However, I don't notice them at normal viewing distances. I just ordered a s912 box, and will see if they're also on it.


    The artifacts I'm not sure of, but they could be related to chroma sampling. I'll work on manual 4:2:0 and see if they're still there.


    Edit: you could try a 25, 30, or 60fps video to see if they're still there.

    Edited once, last by johngalt ().

  • I will look for the artifacts in different framerate demos today.


    Apart from that, everything seems to be very good in "testing1". :)
    BT.2020 and HDR10 are properly passed and look right.

    You are right, the lines are barely visible at normal distances.
    However, I don't want to accept that nothing can be done about them. ;)

    I will try to force 12bit and see if they are still there...

    Mini M8S II 2G/16G (S905X)

  • Once we have nougat-exp working, I can try to address some of this. For now, the changes in the state of "testing1" are sitting as pull requests for kszaq, so hopefully soon we can see official nougat builds.


    This is part of the reason why for some media I prefer 8bit output + dithering, and why I'm interested to play around on the S912 device I ordered. It has dolby registers for controlling stuff like noise, 10->12bit extending, then 12->10bit dithering. Those capabilities are only on the nougat-exp branch with the h264 pts (web-dl) issue, which is a big incentive for me to get it working properly before my S912 device arrives :)

    Edited once, last by johngalt ().

  • What I can tell you johngalt is that today I did a short playback test of my samples on a devices that came with Android 7.1.1 - all playback issues we are experiencing with Nougat kernel were there, sometimes even worse than on LE. We're fixing Amlogic bugs...

  • What I can tell you johngalt is that today I did a short playback test of my samples on a devices that came with Android 7.1.1 - all playback issues we are experiencing with Nougat kernel were there, sometimes even worse than on LE. We're fixing Amlogic bugs...

    That's good to know :) . The PRs I put up for review revert some amports changes to avoid the h264 playback issues entirely. At this point, that seems to be a good base and just needs more testing by users who had the nougat output bug (one user reported it as fixed).


    What was the device? I ask because I'm curious if it used one of the full multi-instance decoders (defined in ucode). It's interesting: with copy/pasted functions, in one instance there were pts issues reintroduced in the "standard" h264 decoder, and fixed in the multi-instance decoder with those massive multi-instance decoding commits (with changes going far beyond multi-instance decoding). At this point, I'm very confused as to the type of development Amlogic review goes through.

  • First of all, thx kszaq and johngalt for your work! :)


    I am on S905X 2GB testing1 nand installed. When I am watching a movie with following parameters:....

    ... the first ~8 seconds of the movie video and audio are async. In this short time the movie plays faster to catch up the audio and sync.


    Since this bug do not exist on builds <8.0.2a and even on MM builds, I think I have to tell you about it if you not even know about it.

  • I tried again Testing 1 version,

    8 bit output all ok.

    10 bit output: 444 no video output.

    10 bit output: 422 only 24p output.

    10 bit output: 420 no video output.

  • That's interesting, especially because by default you shouldn't have hw accel enabled on it (sd). Thank you for the report.


    I tried again Testing 1 version,

    8 bit output all ok.

    10 bit output: 444 no video output.

    10 bit output: 422 only 24p output.

    10 bit output: 420 no video output.

    You have a different issue than the guys with the "standard" nougat output issue. Sorry, I'm not sure if you'll ever get 10bit output working on the hardware since you also had the issue on the early marshmallow test builds. Also, 4:2:0 set through attr won't work currently no matter the setup. For now, I recommend you stick with 8bit.

  • It's interesting: with copy/pasted functions, in one instance there were pts issues reintroduced in the "standard" h264 decoder, and fixed in the multi-instance decoder with those massive multi-instance decoding commits

    I can confirm this as well. Until I re-enabled Multi_dec in the Experimental Kernels, video playback was terrible.

    What I can tell you johngalt is that today I did a short playback test of my samples on a devices that came with Android 7.1.1 - all playback issues we are experiencing with Nougat kernel were there, sometimes even worse than on LE. We're fixing Amlogic bugs...

    Well no surprises there - seems like ops normal for AML - I suspect this is why we have not seen Xiaomi rush out a AML Android 7.1 Nougat update for their S905X Mi Box.

  • I can confirm this as well. Until I re-enabled Multi_dec in the Experimental Kernels, video playback was terrible.

    Thank you, good to know. For nougat-exp, I'll enable for S912 in config and try enabling for S905* in ucode.

  • Sorry for the delay...

    I finally had the time to test 12bit and look for the artifacts.


    Result:

    - 12bit does not change anything... Vertical lines are still there in 10bit videos.

    - I definitely see the weird artifacts in the "Sony Camp" video as well, it is 59.94 fps. At ~1.03min over the mountain (sky). It looks like heavy compression artifacts you get at low bitrates... Macroblocking... The video has a very high bitrate though and the artifacts are not there when I play via the TV's internal media player.

    Mini M8S II 2G/16G (S905X)

  • I tried again Testing 1 version,

    8 bit output all ok.

    10 bit output: 444 no video output.

    10 bit output: 422 only 24p output.

    10 bit output: 420 no video output.

    You mentioned a Marantz AV and a JVC X5000 projector before...
    How long is the cable to the projector?
    Did you try going directly from the box to the projector? Do you have the same limitations then?

    Mini M8S II 2G/16G (S905X)

  • Hmmm this is what I feared...
    15m cables are almost never capable of those bandwidths.

    Have you tried to use a short (3m max), good HDMI 2 cable directly between box and JVC?

    Mini M8S II 2G/16G (S905X)

  • With a 2m cable is ok. So I think tha I have to buy another one 18gb 15m cable.

    May can you fix 420 OR 422 output?

    Edited once, last by saltanar ().

  • Don't do it...
    You might get lucky, but in my experience, finding a 15m cable that can pass the full 18gbps is very difficult to come by.
    Is there a chance for you to use two shorter cables and a signal amplifier in between?
    As far as I know, even 10m cables are not such "divas" as 15m ones... Maybe you can use one 10m and one 5m cable or two 7.5m cables?

    Mini M8S II 2G/16G (S905X)

  • Sorry for the delay...

    I finally had the time to test 12bit and look for the artifacts.


    Result:

    - 12bit does not change anything... Vertical lines are still there in 10bit videos.

    - I definitely see the weird artifacts in the "Sony Camp" video as well, it is 59.94 fps. At ~1.03min over the mountain (sky). It looks like heavy compression artifacts you get at low bitrates... Macroblocking... The video has a very high bitrate though and the artifacts are not there when I play via the TV's internal media player.

    It's still 10bit video at 12bit output, so that makes sense unfortunately. I'll check out that demo, ty.


    With a 2m cable is ok. So I think tha I have to buy another one 18gb 15m cable.

    May can you fix 420 OR 422 output?

    Manually setting 422 works, and fixing manual 420 is planned.

  • Hmmm this is what I feared...
    15m cables are almost never capable of those bandwidths.

    Have you tried to use a short (3m max), good HDMI 2 cable directly between box and JVC?

    FWIW, there are no such things as "HDMI 2.0" cables. The only thing the HDMI alliance does is specify a 'premium' label (with QR code so you can check it yourself), which in practice means it works for 2.0, because it was actually tested.

    So far all three premium cables I bought work great with 4k60/444, the non-premium "2.0 + ethernet cables" are hit and miss. They are sadly 3 times the prices of regular cables, though :(