Re: [Cin] another 10 bit encoding patch

Fri, 15 Nov 2024 05:13:42 -0800 



Den 15.11.2024 13:31, skrev Terje J. Hanssen:




Den 15.11.2024 09:10, skrev Andrew Randrianasulu:




On Fri, Nov 15, 2024 at 9:31 AM Andrew Randrianasulu 
<[email protected]> wrote:




    пт, 15 нояб. 2024 г., 03:21 Terje J. Hanssen
    <[email protected]>:




        Den 14.11.2024 22:21, skrev Andrew Randrianasulu:



        чт, 14 нояб. 2024 г., 23:41 Terje J. Hanssen
        <[email protected]>:




            Den 14.11.2024 18:38, skrev Andrew Randrianasulu:



            чт, 14 нояб. 2024 г., 20:27 Terje J. Hanssen
            <[email protected]>:

                Thank you for the patch.

                A few clarification questions in between before I
                start:

                Den 14.11.2024 14:21, skrev Andrew Randrianasulu:

                > you need to reset your git tree,
                >
                > something like
                >
                > git reset
                --hard 2037bf63cdf25462ca606f4a28636e12df77c2a0

                I did use "git reset --hard" without number at the
                end for the previous
                patch

                What number here, should I find mine somewhere in
                my git tree?



            you can fig them in 'git log' screen, you should have
            two more patches on top of "ORIGIN" line


            Sorry, I'm still a bit confused by the latter "you
            should have two more patches on top of "ORIGIN" line"
            Here is the top from my git log" output:

                commit 68eb98be2183738684f1f4da2729d1ff9989ad16
                Author: Andrew Randrianasulu
                <[email protected]>
                <mailto:[email protected]>
                Date:   Sun Oct 27 19:43:06 2024 +0300

                    Experimental: try 10bit vaapi encoding

                commit e5a5a6da907dbcd40d8612bdbfeea1a2a0ae6cc8
                Author: Andrew Randrianasulu
                <[email protected]>
                <mailto:[email protected]>
                Date:   Fri Oct 18 10:07:33 2024 +0300

                    Add onevpl support to build system

                commit 8681d13675f32e870ab3632eaf89105415fb3961
                Author: Andrew Randrianasulu
                <[email protected]>
                <mailto:[email protected]>
                Date:   Wed Oct 23 16:36:19 2024 +0300

                    Add DESCRIPTION

                commit 90138debee46e0b91adeb5d8a400158b131b0d61
                Author: Andrew Randrianasulu
                <[email protected]>
                <mailto:[email protected]>
                Date:   Wed Oct 23 16:29:58 2024 +0300

                    Add BUGS

                commit 318c884532617e32904d6f4ec05a2b73832f418e
                Author: Andrew Randrianasulu
                <[email protected]>
                <mailto:[email protected]>
                Date:   Wed Oct 23 16:22:58 2024 +0300

                    Update README build instructions


        ah.

        you can just git reset --hard
        e5a5a6da907dbcd40d8612bdbfeea1a2a0ae6cc8


        inother words just reset tree to state without old 10bit patch.

        then apply new.

        if you can't see line with

        (origin/master, origin/HEAD)

        you can scroll down (by cursor arrows)

        or search by pressing "/" and typing your search word and
        pressing enter.

        but I think you can just reset tree to "add onevpl" commit.


        Sorry, if I lost something - because it does not change
        something so far:
        All three attempts result all in yuv422p10le as the previous
        10bit hack did.

        My summary follows:

        /Cin

        # git reset --hard e5a5a6da907dbcd40d8612bdbfeea1a2a0ae6cc8
        HEAD is now at e5a5a6da Add onevpl support to build system

        # git am 0001-yet-another-10bit-attempt.patch
        Applying: yet another 10bit attempt


        I didn't undestand Cingg should be recompiled additional
        after this patch(?)


    yes.





Thanks, after recompiling Cingg with the latest patch, hevc_vaapi 
encoding works individually for each pixel format.

You should try to wire it?


Attach my customized presets for hevc_vaapi_8b420.m4, 
hevc_vaapi_10b420 and hevc_vaapi_10b422.mp4

with output from my latest test runs as follows:

===========================

/Cin # CIN_10BIT_ENC=nv12 bin/cin
Cinelerra Infinity - built: Nov 15 2024 10:13:33

hevc_vaapi_8b420.mp4
    cin_hw_dev=vaapi
    format=nv12,hwupload
    profile=main
    # global_quality=22

** rendered 1780 frames in 42.010 secs, 42.371 fps


ffprobe -hide_banner hd01_hevc_vaapi_8b420.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], 11377 kb/s, 25 fps, 25 tbr, 12800 tbn 
(default)


------------------------

global_quality=22
** rendered 1780 frames in 38.741 secs, 45.946 fps

ffprobe -hide_banner hd01_hevc_vaapi_8b420_gq22.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], 16643 kb/s, 25 fps, 25 tbr, 12800 tbn 
(default)



================================

Cin # CIN_10BIT_ENC=p010le bin/cin


hevc_vaapi_10b420.mp4
    cin_hw_dev=vaapi
    format=p010le,hwupload
    profile=main 10
    # global_quality=22

** rendered 1780 frames in 42.292 secs, 42.088 fps


ffprobe -hide_banner hd01_hevc_vaapi_10b420.mp4

  Stream #0:0[0x1](und): Video: hevc (Main 10) (hev1 / 0x31766568), 
yuv420p10le(tv, bt709/unknown/unknown, top coded first (swapped)), 
1920x1080 [SAR 1:1 DAR 16:9], 11752 kb/s, 25 fps, 25 tbr, 12800 tbn 
(default)


---------------

global_quality=22
** rendered 1780 frames in 39.585 secs, 44.967 fps

ffprobe -hide_banner hd01_hevc_vaapi_10b420_gq22.mp4

  Stream #0:0[0x1](und): Video: hevc (Main 10) (hev1 / 0x31766568), 
yuv420p10le(tv, bt709/unknown/unknown, top coded first (swapped)), 
1920x1080 [SAR 1:1 DAR 16:9], 17038 kb/s, 25 fps, 25 tbr, 12800 tbn 
(default)


===================================

/Cin # CIN_10BIT_ENC=y210le bin/cin

hevc_vaapi_10b422.mp4
    cin_hw_dev=vaapi
    format=y210le,hwupload
    profile=4
    # global_quality=22

** rendered 1780 frames in 41.599 secs, 42.789 fps


ffprobe -hide_banner hd01_hevc_vaapi_10b422.mp4

  Stream #0:0[0x1](und): Video: hevc (Rext) (hev1 / 0x31766568), 
yuv422p10le(tv, bt709/unknown/unknown, top coded first (swapped)), 
1920x1080 [SAR 1:1 DAR 16:9], 13432 kb/s, 25 fps, 25 tbr, 12800 tbn 
(default)


----------------

global_quality=22
** rendered 1780 frames in 41.923 secs, 42.459 fps

ffprobe -hide_banner hd01_hevc_vaapi_10b422_gq22.mp4

  Stream #0:0[0x1](und): Video: hevc (Rext) (hev1 / 0x31766568), 
yuv422p10le(tv, bt709/unknown/unknown, top coded first (swapped)), 
1920x1080 [SAR 1:1 DAR 16:9], 17992 kb/s, 25 fps, 25 tbr, 12800 tbn 
(default)


===================================



As seen "profile 4" in the latter hevc_vaapi_10b422.mp4 preset is 
recognized as "Rext" by ffprobe.

"Rext" was not accepted in the preset.


According to this Phoronix 2020 article, "Intel Adds VA-API Acceleration 
For HEVC REXT To FFmpeg"
.... supporting the y210 pixel format, necessary parsing changes for 
HEVC REXT, and ultimately going through with adding 4:2:2 8-bit/10-bit

https://www.phoronix.com/news/FFmpeg-VA-API-HEVC-REXT









Just recompiled cin  with this patch and (after adding logleve=debug 
to bin/ffmpeg/encode.opts ) I have:


bash-5.1$ bin/cin
Cinelerra Infinity - built: Nov 15 2024 10:55:42

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 @ 0xd7b03740] Trying to use DRM render node for 
device 0.

[AVHWDeviceContext @ 0xd7b03740] libva: VA-API version 1.19.0

[AVHWDeviceContext @ 0xd7b03740] libva: Trying to open 
/usr/lib/dri/nouveau_drv_video.so
[AVHWDeviceContext @ 0xd7b03740] libva: Found init function 
__vaDriverInit_1_19

[AVHWDeviceContext @ 0xd7b03740] libva: va_openDriver() returns 0

[AVHWDeviceContext @ 0xd7b03740] Initialised VAAPI connection: 
version 1.19

[AVHWDeviceContext @ 0xd7b03740] Format 0x3231564e -> nv12.
[AVHWDeviceContext @ 0xd7b03740] Format 0x30313050 -> p010le.
[AVHWDeviceContext @ 0xd7b03740] Format 0x36313050 -> unknown.
[AVHWDeviceContext @ 0xd7b03740] Format 0x30323449 -> yuv420p.
[AVHWDeviceContext @ 0xd7b03740] Format 0x32315659 -> yuv420p.
[AVHWDeviceContext @ 0xd7b03740] Format 0x56595559 -> unknown.
[AVHWDeviceContext @ 0xd7b03740] Format 0x32595559 -> yuyv422.
[AVHWDeviceContext @ 0xd7b03740] Format 0x59565955 -> uyvy422.
[AVHWDeviceContext @ 0xd7b03740] Format 0x41524742 -> bgra.
[AVHWDeviceContext @ 0xd7b03740] Format 0x41424752 -> rgba.
[AVHWDeviceContext @ 0xd7b03740] Format 0x58524742 -> bgr0.
[AVHWDeviceContext @ 0xd7b03740] Format 0x58424752 -> rgb0.

[AVHWDeviceContext @ 0xd7b03740] VAAPI driver: Mesa Gallium driver 
22.1.7 for NV106.
[AVHWDeviceContext @ 0xd7b03740] Driver not found in known 
nonstandard list, using standard behaviour.

[AVHWFramesContext @ 0xd7b30900] Created surface 0x1.

[AVHWFramesContext @ 0xd7b30900] Direct mapping disabled: deriving 
image does not work: 1 (operation failed).

[hevc_vaapi @ 0xd7b0ca40] Input surface format is nv12.

[hevc_vaapi @ 0xd7b0ca40] Compatible profile VAProfileHEVCMain (17) 
is not supported by driver.
[hevc_vaapi @ 0xd7b0ca40] Compatible profile VAProfileHEVCMain (17) 
is not supported by driver.

[hevc_vaapi @ 0xd7b0ca40] No usable encoding profile found.
FFMPEG::open_encoder  err: Function not implemented
int FFMPEG::open_encoder(const char*, const char*):
open failed hevc_vaapi:/dev/shm/vaapi-test.mp4
Render::render_single: Session finished.
Total excess of backups: -50
[AVIOContext @ 0xf4139c40] Statistics: 196608 bytes read, 0 seeks
[AVIOContext @ 0xf413a680] Statistics: 196608 bytes read, 0 seeks
[AVIOContext @ 0xda276c00] Statistics: 294912 bytes read, 0 seeks
[AVIOContext @ 0xda2e6dc0] Statistics: 196608 bytes read, 0 seeks
[AVIOContext @ 0xe4ed3e80] Statistics: 18991217 bytes read, 18 seeks
[AVIOContext @ 0xe4e0a280] Statistics: 196608 bytes read, 0 seeks
Session time: 0:00:29
Cpu time: user: 0:00:14.957 sys: 0:00:01.831

note that it says HevcMain

bash-5.1$ CIN_10BIT_ENC=p010le bin/cin
Cinelerra Infinity - built: Nov 15 2024 10:55:42

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 @ 0xd4bfd880] Trying to use DRM render node for 
device 0.

[AVHWDeviceContext @ 0xd4bfd880] libva: VA-API version 1.19.0

[AVHWDeviceContext @ 0xd4bfd880] libva: Trying to open 
/usr/lib/dri/nouveau_drv_video.so
[AVHWDeviceContext @ 0xd4bfd880] libva: Found init function 
__vaDriverInit_1_19

[AVHWDeviceContext @ 0xd4bfd880] libva: va_openDriver() returns 0

[AVHWDeviceContext @ 0xd4bfd880] Initialised VAAPI connection: 
version 1.19

[AVHWDeviceContext @ 0xd4bfd880] Format 0x3231564e -> nv12.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x30313050 -> p010le.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x36313050 -> unknown.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x30323449 -> yuv420p.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x32315659 -> yuv420p.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x56595559 -> unknown.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x32595559 -> yuyv422.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x59565955 -> uyvy422.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x41524742 -> bgra.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x41424752 -> rgba.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x58524742 -> bgr0.
[AVHWDeviceContext @ 0xd4bfd880] Format 0x58424752 -> rgb0.

[AVHWDeviceContext @ 0xd4bfd880] VAAPI driver: Mesa Gallium driver 
22.1.7 for NV106.
[AVHWDeviceContext @ 0xd4bfd880] Driver not found in known 
nonstandard list, using standard behaviour.

[AVHWFramesContext @ 0xce249500] Created surface 0x1.
[AVHWFramesContext @ 0xce249500] Direct mapping possible.
[hevc_vaapi @ 0xce2a60c0] Input surface format is p010le.

[hevc_vaapi @ 0xce2a60c0] Compatible profile VAProfileHEVCMain10 (18) 
is not supported by driver.
[hevc_vaapi @ 0xce2a60c0] Compatible profile VAProfileHEVCMain10 (18) 
is not supported by driver.

[hevc_vaapi @ 0xce2a60c0] No usable encoding profile found.
FFMPEG::open_encoder  err: Function not implemented
int FFMPEG::open_encoder(const char*, const char*):
open failed hevc_vaapi:/dev/shm/vaapi-test.mp4
Render::render_single: Session finished.
Total excess of backups: -50
[AVIOContext @ 0xd6e988c0] Statistics: 196608 bytes read, 0 seeks
[AVIOContext @ 0xf403aa40] Statistics: 196608 bytes read, 0 seeks
[AVIOContext @ 0xe4e86ec0] Statistics: 294912 bytes read, 0 seeks
[AVIOContext @ 0xe4ed9cc0] Statistics: 196608 bytes read, 0 seeks
[AVIOContext @ 0xef74cfc0] Statistics: 18991217 bytes read, 18 seeks
[AVIOContext @ 0xef709c00] Statistics: 196608 bytes read, 0 seeks
Session time: 0:00:24
Cpu time: user: 0:00:15.011 sys: 0:00:01.294



note what it says HevcMain10 ofc those are 'dry runs' because I do 
not have hardware ...






        --------------------------

        My third Cin test using nv12

        /Cin

        # CIN_10BIT_ENC=nv12 bin/cin
        Cinelerra Infinity - built: Nov 13 2024 20:54:33

        hevc_vaapi_8b420.mp4
            cin_hw_dev=vaapi
            format=nv12,hwupload
            # profile=main
            # global_quality=22

        ** rendered 1780 frames in 41.675 secs, 42.711 fps
        FFMPEG::open_decoder: some stream times estimated:
        /Videoklipp/Cineform/cfhd01.mkv


        ffprobe -hide_banner hd01_hevc_vaapi_8b420.mp4
          Stream #0:0[0x1](und): Video: hevc (Rext) (hev1 /
        0x31766568), yuv422p10le(tv, bt709/unknown/unknown, top coded
        first (swapped)), 1920x1080 [SAR 1:1 DAR 16:9], 24375 kb/s,
        25 fps, 25 tbr, 12800 tbn (default)

        ===========================

        Some output for verification:

        /Cin/cinelerra # egrep -i "10bit|tenbit|nv12|p010|y210" ffmpeg.C
        GET_HW_PIXFMT(nv12,  AV_PIX_FMT_NV12)
                case AV_PIX_FMT_NV12: ctx->get_format = get_hw_nv12; 
        return *p;
            const char *tenbit_enc = getenv("CIN_10BIT_ENC");
                frames_ctx->sw_format = AV_PIX_FMT_NV12;
                if(tenbit_enc)
                frames_ctx->sw_format = av_get_pix_fmt(tenbit_enc);
                picture->format = AV_PIX_FMT_NV12;

        ----------------------

        git log 2>&1 | tee git_log_after_another_10bit_patch

        Top of the output:

        commit 4d42448cc128d7d18e34eb97d4234e73b9acb360
        Author: Andrew Randrianasulu <[email protected]>
        <mailto:[email protected]>
        Date:   Thu Nov 14 16:18:24 2024 +0300

            yet another 10bit attempt

        commit e5a5a6da907dbcd40d8612bdbfeea1a2a0ae6cc8
        Author: Andrew Randrianasulu <[email protected]>
        <mailto:[email protected]>
        Date:   Fri Oct 18 10:07:33 2024 +0300

            Add onevpl support to build system

        =========================









-- 
Cin mailing list
[email protected]
https://lists.cinelerra-gg.org/mailman/listinfo/cin






















					Reply via email to