[Cin] adding onevpl support for buildsystem

Terje J. Hanssen terjejhanssen at gmail.com
Fri Oct 25 18:56:50 CEST 2024




Den 25.10.2024 17:31, skrev Andrew Randrianasulu:
>
>
> On Fri, Oct 25, 2024 at 5:05 PM Terje J. Hanssen 
> <terjejhanssen at gmail.com> wrote:
>
>
>
>
>     Den 25.10.2024 01:55, skrev Terje J. Hanssen:
>>
>>
>>
>>     Den 25.10.2024 00:43, skrev Andrew Randrianasulu:
>
>     snip ...................
>>>
>>>>
>>>>             -----------------------
>>>>
>>>>             localhost:/Cin # bin/cin
>>>>             Cinelerra Infinity - built: Oct 24 2024 16:26:16
>>>>
>>>>             Tested with similar results as my previous ffmpeg_71 build:
>>>>
>>>>             SD-DV to av1_qsv, hevc_qsv and h264_qsv nv12 works
>>>>             HDV to hevc_qsv works with nv12, p010le and yuv422 works
>>>>             HDV to av1_qsv and to h264_qsv don't work
>>>>
>>>>
>>>>         does av1_vaapi work for HDV case?
>>>
>>>
>>>         Yes, I test-rendered HDV (hdv09_04) to hevc_vaapi,
>>>         av1_vaapi, h264_vaapi, all i .mp4 below
>>>         Also FHD (hd01) to av1_vaapi.mp4 worked, see ffprobe below
>>>
>>>         Testing .webm type breaks with the following output:
>>>
>>>             [av1_vaapi @ 0x7f5f74122a80] Driver does not support
>>>             QVBR RC mode (supported modes: CQP, CBR, VBR, ICQ).
>>>             FFMPEG::open_encoder err: Invalid argument
>>>             int FFMPEG::open_encoder(const char*, const char*):
>>>             open failed
>>>             av1_vaapi:/Videoklipp/VAAPI/hdv09_04_av1_vaapi.webm
>>>
>>>
>>>     hm, you probably need to tweak this profile for your driver then ...
>>
>>     I got this tweak to work for hdv09_04.m2t rendering to
>>     av1_vaapi.webm
>>
>>         cin_hw_dev=vaapi
>>         g=30
>>         profile=main
>>         rc_mode=CQP
>>
>>     Render::render_single: Session finished.
>>     ** rendered 5972 frames in 27.408 secs, 217.893 fps
>>     FFMPEG::open_decoder: some stream times estimated:
>>     /Videoklipp/VAAPI/hdv09_04_av1_vaapi.webm
>>     FFMPEG::open_decoder: some stream times estimated:
>>     /Videoklipp/VAAPI/hdv09_04_av1_vaapi.webm
>>     audio0 pad 64 -143 (207)
>>     audio0 pad 0 -143 (143)
>>     audio0 pad 0 -15 (15)
>>
>>
>>     while
>>
>>     [av1_vaapi @ 0x7f8000317a40] Bitrate must be set for VBR RC mode.
>>     FFMPEG::open_encoder  err: Invalid argument
>>     int FFMPEG::open_encoder(const char*, const char*):
>>     open failed av1_vaapi:/Videoklipp/VAAPI/hdv09_04_av1_vaapi.webm
>>
>>
>
>     I think this build uses ffmpeg 7.0 embedded, but is it a way to
>     very which ffmpeg version in use via Cingg?
>
>
> Yes, look at line
>
> Libav version: Lavc61.3.100
>
> 61.19 -> 7.1
> 61.3 -> 7.0

So it is not possible to query somewhere with

ffmpeg --version (or just ffmpeg)

to get it's version on the first line like on the system ffmpeg

# ffmpeg
ffmpeg version 7.1 Copyright (c) 2000-2024 the FFmpeg developers
   built with gcc 14 (SUSE Linux)

or optional via verbose/debug output?

>
>
>
>     As already tested as working on my own build Cingg based on system
>     ffmpeg 7.1,
>     https://lists.cinelerra-gg.org/pipermail/cin/2024-October/008932.html
>
>     this also shows to work on this 7.0(?) build:
>     hdv.m2t and hd.mov (prores 422 hq) rendering with to
>     av1_vaapi.webm on Arc A750 works with
>
>     vprofile=high or vprofile=professional
>
>     but 10bit yuv422 are downscaled to 8bit yuv420 in the output, at
>     least with this profile and/or without other parameters set
>
>
> yeah, seems to be limitation of current code in cingg.
>
> in cinelerra/ffmpeg.C there is function
>
> AVHWDeviceType FFVideoStream::encode_hw_activate()
>
> frames_ctx->sw_format = AV_PIX_FMT_NV12;
>
> if you change AV_PIX_FMT_NV12 with AV_PIX_FMT_P010
>
> does it  encode 10 bit ?

Sorry, no it doesn't, just the same.

And as seen in the ffprobe output below:
     Stream #0:0: Video: av1 (libdav1d) (Main), yuv420p

That is Main profile, even if I set and used vprofile=professional for 
the rendering !?

>
> You said av1_qsv  at least can be set to p012 ? Then, this is 
> advantage (when it works)

I remember I rendered hevc_qsv with p010le and yuv422, but I'm not sure 
I did verify it in the output.
At least I'm not able verify it now, as they also seems to rescale to 
8bit yuv420p.

>
>
>     Render::render_single: Session finished.
>     ** rendered 5972 frames in 28.582 secs, 208.943 fps
>     FFMPEG::open_decoder: some stream times estimated:
>     /Videoklipp/VAAPI/hdv09_04_av1_vaapi.webm
>
>     Render::render_single: Session finished.
>     ** rendered 1781 frames in 13.666 secs, 130.323 fps
>     FFMPEG::open_decoder: some stream times estimated:
>     /Videoklipp/VAAPI/hd01_av1_vaapi.webm
>
>     /Videoklipp/VAAPI # ls -lht hdv* hd*
>     -rw-r--r-- 1 root  root  567M Oct 25 15:30 hd01_av1_vaapi.webm
>     -rw-r--r-- 1 root  root  1.5G Oct 25 15:11 hdv09_04_av1_vaapi.webm
>
>
>     ffprobe -hide_banner hd01_av1_vaapi.webm
>     Input #0, matroska,webm, from 'hd01_av1_vaapi.webm':
>       Metadata:
>         ENCODER         : Lavf61.1.100
>       Duration: 00:01:11.28, start: 0.000000, bitrate: 67255 kb/s
>       Stream #0:0: Video: av1 (libdav1d) (Main), yuv420p(tv,
>     bt709/unknown/unknown), 1920x1080, SAR 1:1 DAR 16:9, 25 fps, 25
>     tbr, 1k tbn
>           Metadata:
>             DURATION        : 00:01:11.243000000
>       Stream #0:1: Audio: vorbis, 48000 Hz, 16 channels, fltp
>           Metadata:
>             DURATION        : 00:01:11.283000000
>
>
>>
>>
>>>
>>>     so, vaapi works in more cases than qsv, it seems?
>
>     So far, yes
>
>>>
>>>
>>>
>>>
>>>         terje at localhost:/Videoklipp/VAAPI> ls -lt hd*
>>>         -rw-r--r-- 1 root  root  349812400 okt. 24 20:54
>>>         hdv09_04_hevc_vaapi.mp4
>>>         -rw-r--r-- 1 root  root  112467398 okt. 24 20:42
>>>         hd01_av1_vaapi.mp4
>>>         -rw-r--r-- 1 root  root  606029742 okt. 24 20:35
>>>         hdv09_04_h264_vaapi.mp4
>>>         -rw-r--r-- 1 root  root  378664255 okt. 24 20:28
>>>         hdv09_04_av1_vaapi.mp4
>>>
>>>
>>>         ffprobe -hide_banner hdv09_04_hevc_vaapi.mp4
>>>         Input #0, mov,mp4,m4a,3gp,3g2,mj2, from
>>>         'hdv09_04_hevc_vaapi.mp4':
>>>           Metadata:
>>>             major_brand     : isom
>>>             minor_version   : 512
>>>             compatible_brands: isomiso2mp41
>>>             encoder         : Lavf61.7.100
>>>           Duration: 00:03:58.88, start: 0.000000, bitrate: 11715 kb/s
>>>           Stream #0:0[0x1](und): Video: hevc (Main) (hev1 /
>>>         0x31766568), yuv420p(tv, bt470bg/unknown/unknown, top coded
>>>         first (swapped)), 1440x1080 [SAR 4:3 DAR 16:9], 11580 kb/s,
>>>         25 fps, 25 tbr, 12800 tbn (default)
>>>               Metadata:
>>>                 handler_name    : VideoHandler
>>>                 vendor_id       : [0][0][0][0]
>>>           Stream #0:1[0x2](und): Audio: aac (LC) (mp4a /
>>>         0x6134706D), 48000 Hz, stereo, fltp, 129 kb/s (default)
>>>               Metadata:
>>>                 handler_name    : SoundHandler
>>>                 vendor_id       : [0][0][0][0]
>>>
>>>
>>>         ffprobe -hide_banner hd01_av1_vaapi.mp4
>>>         Input #0, mov,mp4,m4a,3gp,3g2,mj2, from 'hd01_av1_vaapi.mp4':
>>>           Metadata:
>>>             major_brand     : isom
>>>             minor_version   : 512
>>>             compatible_brands: isomav01iso2mp41
>>>             encoder         : Lavf61.1.100
>>>           Duration: 00:01:11.28, start: 0.000000, bitrate: 12622 kb/s
>>>           Stream #0:0[0x1](und): Video: av1 (libdav1d) (Main) (av01
>>>         / 0x31307661), yuv420p(tv, bt470bg/unknown/unknown, top
>>>         coded first (swapped)), 1920x1080, 12246 kb/s, SAR 1:1 DAR
>>>         16:9, 25 fps, 25 tbr, 12800 tbn (default)
>>>               Metadata:
>>>                 handler_name    : VideoHandler
>>>                 vendor_id       : [0][0][0][0]
>>>           Stream #0:1[0x2](und): Audio: aac (LC) (mp4a /
>>>         0x6134706D), 48000 Hz, 16 channels, fltp, 378 kb/s (default)
>>>               Metadata:
>>>                 handler_name    : SoundHandler
>>>                 vendor_id       : [0][0][0][0]
>>>
>>>
>>>>         can you add auto-scale filter to HDV cases so it will be
>>>>         full HD and not 1440*1080 ? or another similar rescaling ....
>>>>
>>>>         again, thanks for testing.
>>>>
>>>>         But as with any experimentation we get some new questions
>>>>         instead of pure answers ....
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>             ===================================
>>>>
>>>>             In separate posts:
>>>>             Can you possibly setup guideline detail procedure on
>>>>             how to build an appimage and a rpm package of this build,
>>>>             to possible install and test it on my legacy Skylake
>>>>             and Kabylake platforms?
>>>>
>>>>             ===================================
>>>>
>>>>
>>>>         I think basic procedure at
>>>>
>>>>         https://cinelerra-gg.org/download/CinelerraGG_Manual/Build_CinGG_AppImage_from_s.html
>>>>
>>>>         starting from
>>>>
>>>>         2- The script bld_appimage.sh uses a platform specific
>>>>         version of appimagetool so that it can create appimages for
>>>>         x86_64, i686, aarch64, or armv7l architecture. We need to
>>>>         add appimagetool-(platform).AppImage to the /{path to
>>>>         cinelerra- 5.1}/tools directory, or somewhere in your path.
>>>>         You can download the tool for your system (e.g.
>>>>         appimagetool-x86_64.AppImage) from git:
>>>>         https://github.com/AppImage/AppImageKit/releases
>>>>
>>>>
>>>>         for rpm build I think we need to wait for this patch to
>>>>         land in git and then become part of monthly src tarball.
>>>>
>>>>         Then you can edit .spec file with soecifix date-based
>>>>         filename to fetch and run
>>>>
>>>>         rpmbuild -bb ("build binary") our_spec_file.spec
>>>>
>>>>
>     snip ...........
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20241025/deaccf7f/attachment-0001.htm>


More information about the Cin mailing list