[Cin] 10bit VAAPI encoding test thread
Terje J. Hanssen
terjejhanssen at gmail.com
Sun Oct 27 17:02:46 CET 2024
Den 27.10.2024 16:35, skrev Andrew Randrianasulu:
>
>
> вс, 27 окт. 2024 г., 18:30 Terje J. Hanssen <terjejhanssen at gmail.com>:
>
>
>
>
> Den 27.10.2024 16:21, skrev Andrew Randrianasulu:
>>
>>
>> вс, 27 окт. 2024 г., 13:17 Terje J. Hanssen
>> <terjejhanssen at gmail.com>:
>>
>>
>>
>>
>> Den 26.10.2024 01:37, skrev Andrew Randrianasulu:
>>>
>>>
>>> On Sat, Oct 26, 2024 at 2:14 AM Andrew Randrianasulu
>>> <randrianasulu at gmail.com> wrote:
>>>
>>>
>>>
>>> On Sat, Oct 26, 2024 at 1:29 AM Andrew Randrianasulu
>>> <randrianasulu at gmail.com> wrote:
>>>
>>>
>>>
>>> сб, 26 окт. 2024 г., 00:58 Terje J. Hanssen
>>> <terjejhanssen at gmail.com>:
>>>
>>>
>>>
>>>
>>> Den 25.10.2024 21:48, skrev Andrew Randrianasulu:
>>>> We branching!
>>>>
>>>> I looked at gpu-screen-recorder
>>>>
>>>> https://git.dec05eba.com/gpu-screen-recorder/tree/src/encoder/video/vaapi.c
>>>>
>>>> it sets
>>>>
>>>> |AVHWFramesContext *hw_frame_context =
>>>> (AVHWFramesContext*)frame_context->data;
>>>> hw_frame_context->width =
>>>> video_codec_context->width;
>>>> hw_frame_context->height =
>>>> video_codec_context->height;
>>>> hw_frame_context->sw_format =
>>>> self->params.color_depth ==
>>>> GSR_COLOR_DEPTH_10_BITS ? AV_PIX_FMT_P010LE :
>>>> AV_PIX_FMT_NV12; hw_frame_context->format =
>>>> video_codec_context->pix_fmt;
>>>> hw_frame_context->device_ctx =
>>>> (AVHWDeviceContext*)self->device_ctx->data;|
>>>> ||
>>>> ||
>>>> |so may be we need to set |AV_PIX_FMT_P010LE and not just P010?
>>>
>>> Sorry, It didn't render more than 8bit yuv420p
>>>
>>>
>>>
>>> yeahh ... probably need more debugging (seeing what
>>> pixel format goes into ffmpeg vaapi interface at least).
>>>
>>> I only have decoding on nouveau, and quite flaky at
>>> that - only 2 files so far worked w,/o crashing,
>>> from around 2011/12.
>>>
>>>
>>> I tried to change two instances of NV12 to P010LE
>>> and it shows no visible effect, so I assume it was
>>> supposed to work with encoding ...?
>>>
>>> it does not work even with qsv set to p010 AND patch ?
>>>
>>> if so, I am afraid I only can put this item on shelf
>>> ... for now. Look at mantis bug database, if there
>>> is no entry about 10bit vaapi deciding/encoding -
>>> add it please ....
>>>
>>> you can also test (at some point in time ) this
>>> gpu_screen_recorder, it builds but does not work for
>>> me due to "nouveau" driver.
>>>
>>>
>>>
>>> at least with another patch (attached) I can get this:
>>>
>>> bash-5.1$ bin/cin ~/6029.mp4
>>> Cinelerra Infinity - built: Oct 17 2024 16:18:31
>>> git://git.cinelerra-gg.org/goodguy/cinelerra.git
>>> <http://git.cinelerra-gg.org/goodguy/cinelerra.git>
>>> (c) 2006-2019 Heroine Virtual Ltd. by Adam Williams
>>> 2007-2020 mods for Cinelerra-GG by W.P.Morrow aka goodguy
>>> 2003-2017 mods for Cinelerra-CV by CinelerraCV team
>>> 2015-2024 mods for Cinelerra-GG by Cinelerra-GG team
>>>
>>> Libav version: Lavc61.3.100
>>>
>>> Cinelerra is free software, covered by the GNU General
>>> Public License,
>>> and you are welcome to change it and/or distribute
>>> copies of it under
>>> certain conditions. There is absolutely no warranty for
>>> Cinelerra.
>>>
>>> RenderFarmClient::main_loop: client started
>>> [AVHWDeviceContext @ 0xf925c00] Trying to use DRM render
>>> node for device 0.
>>> [AVHWDeviceContext @ 0xf925c00] libva: VA-API version 1.19.0
>>> [AVHWDeviceContext @ 0xf925c00] libva: Trying to open
>>> /usr/lib/dri/nouveau_drv_video.so
>>> [AVHWDeviceContext @ 0xf925c00] libva: Found init
>>> function __vaDriverInit_1_19
>>> [AVHWDeviceContext @ 0xf925c00] libva: va_openDriver()
>>> returns 0
>>> [AVHWDeviceContext @ 0xf925c00] Initialised VAAPI
>>> connection: version 1.19
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x3231564e -> nv12.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x30313050 -> p010le.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x36313050 ->
>>> unknown.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x30323449 ->
>>> yuv420p.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x32315659 ->
>>> yuv420p.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x56595559 ->
>>> unknown.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x32595559 ->
>>> yuyv422.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x59565955 ->
>>> uyvy422.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x41524742 -> bgra.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x41424752 -> rgba.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x58524742 -> bgr0.
>>> [AVHWDeviceContext @ 0xf925c00] Format 0x58424752 -> rgb0.
>>> [AVHWDeviceContext @ 0xf925c00] VAAPI driver: Mesa
>>> Gallium driver 22.1.7 for NV106.
>>> [AVHWDeviceContext @ 0xf925c00] Driver not found in
>>> known nonstandard list, using standard behaviour.
>>> [AVHWFramesContext @ 0xea43980] Created surface 0x1.
>>> [AVHWFramesContext @ 0xea43980] Direct mapping possible.
>>> [av1_vaapi @ 0xf978680] Input surface format is p010le.
>>> [av1_vaapi @ 0xf978680] Compatible profile
>>> VAProfileAV1Profile0 (32) is not supported by driver.
>>> [av1_vaapi @ 0xf978680] No usable encoding profile found.
>>> FFMPEG::open_encoder err: Function not implemented
>>> int FFMPEG::open_encoder(const char*, const char*):
>>> open failed
>>> av1_vaapi:/dev/shm/yuv-test-reenc-rgba8-mpeg-range.webm
>>> Render::render_single: Session finished.
>>> Total excess of backups: -50
>>> [AVIOContext @ 0xeaf1aa40] Statistics: 4570842 bytes
>>> read, 5 seeks
>>> [AVIOContext @ 0xeaf348c0] Statistics: 2997978 bytes
>>> read, 2 seeks
>>> [AVIOContext @ 0xd7aa8900] Statistics: 4272574 bytes
>>> read, 2 seeks
>>> [AVIOContext @ 0xd7954080] Statistics: 2997978 bytes
>>> read, 2 seeks
>>> [AVIOContext @ 0xf40c2840] Statistics: 9351290 bytes
>>> read, 18 seeks
>>> [AVIOContext @ 0xf4008a00] Statistics: 2997978 bytes
>>> read, 2 seeks
>>> Session time: 0:00:48
>>> Cpu time: user: 0:00:09.659 sys: 0:00:03.077
>>>
>>> so i guess it sets surface format now correctly, just no
>>> encoder entrypoint (obviously on my hw)?
>>> Terje, Andrea - can you test this hack ?
>>>
>>>
>>> one more time, this time with env variable
>>>
>>> CIN_10BIT_ENC=1 bin/cin ~/6029.mp4
>>>
>>> should enable P010LE input surface format ...but you
>>> probably need to set profile etc manually?
>>
>>
>> OK, P010 should be a 10bit 420 format, while Y210 is the
>> 10bit 422 format.
>>
>>
>> I tried your last suggestion and started Cin with
>>
>> CIN_10BIT_ENC=1 bin/cin
>>
>> rendered a 10bit 422 mov input file and got
>>
>> ffprobe -hide_banner hd01_hevc_vaapi.mp4
>> Stream #0:0[0x1](und): Video: hevc (Main) (hev1 /
>> 0x31766568), yuv420p(tv, bt709/unknown/unknown, top coded
>> first (swapped)), 1920x1080 [SAR 1:1 DAR 16:9], 11225 kb/s,
>> 25 fps, 25 tbr, 12800 tbn (default)
>>
>> from Mediainfo
>> Color space : YUV
>> Chroma subsampling : 4:2:0
>> Bit depth : 8 bits
>>
>>
>> But how should I apply your attached "10bit.diff" patch?
>>
>> It didn't work with
>>
>> localhost:/Cin # git am 10bit.diff
>> fatal: previous rebase directory .git/rebase-apply still
>> exists but mbox given.
>>
>>
>>
>> try to
>>
>> rm .git/rebase-apply
>
> localhost:/Cin # rm .git/rebase-apply
> rm: cannot remove '.git/rebase-apply': No such file or directory
>
> localhost:/Cin # git am 10bit.diff
> fatal: previous rebase directory .git/rebase-apply still exists
> but mbox given.
>
> Am I in wrong directory?
>
>
>
>
>
> oh sorry yes
>
>
>
> rm ../.git/rebase-apply (from one level up)
you mean deleting the whole directory?
rm -r ../.git/rebase-apply
>
> localhost:/Cin # ls -la /Cin
> lrwxrwxrwx 1 root root 29 Oct 22 20:23 /Cin ->
> /home/cinelerra/cinelerra-5.1
>
> localhost:/Cin # ls 10bit.diff
> 10bit.diff
>
>>
>>
>>
>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cinelerra-gg.org/pipermail/cin/attachments/20241027/8da8a2f1/attachment-0001.htm>
More information about the Cin
mailing list