November 2024 - Cin - Cinelerra-GG Mailing Lists

Tested qsv hwaccel presets
by Terje J. Hanssen 27 Nov '24

27 Nov '24

Attached tested qsv hwaccel presets as gzip'd tar file av1_qsv_10b420.mp4 av1_qsv_10b420.webm av1_qsv_8b420.mp4 av1_qsv_8b420.webm h264_qsv_8b420.mp4 hevc_qsv_10b420.mp4 hevc_qsv_10b422.mp4 hevc_qsv_8b420.mp4 vp9_qsv_10b420.mp4 vp9_qsv_10b420.webm vp9_qsv_8b420.mp4 vp9_qsv_8b420.webm

1 1

do not set ildct flag for av1/h264/vp9 qsv
by Andrew Randrianasulu 26 Nov '24

26 Nov '24

hevc is left out because apparently it works? Terje, can you test this patch?

2 6

Fwd: [FFmpeg-devel] qsvenc.c & AV_CODEC_FLAG_INTERLACED_DCT
by Andrew Randrianasulu 26 Nov '24

26 Nov '24

---------- Forwarded message --------- От: Wang, Fei W <fei.w.wang-at-intel.com(a)ffmpeg.org> Date: пт, 22 нояб. 2024 г., 09:10 Subject: Re: [FFmpeg-devel] qsvenc.c & AV_CODEC_FLAG_INTERLACED_DCT To: ffmpeg-devel(a)ffmpeg.org <ffmpeg-devel(a)ffmpeg.org> On Thu, 2024-11-21 at 11:50 +0300, Andrew Randrianasulu wrote: > Hello. > > > Recently I teached cinelerra-gg to use qsv encoders in ffmpeg > (libavcodec) > and some test were working but some not. > > We nailed it down to top-field first material, like HDV files. bff > files > like consumer version of DV and progressive worked fine. > > I looked at source and ... I thought condition at > > http://git.ffmpeg.org/gitweb/ffmpeg.git/blob/HEAD:/libavcodec/qsvenc.c#l852 > > > must contain case for bff, but I tried this and it was reportedly not > helping? > > > We currently set flags = ildct+ilme on interlaced streams, I hope to > avoid > special-casing qsv encoders, but may be we have no other choice? > > I asked user (Terje) to add +ildct flag to his normal system ffmpeg > line > and it started to fail too: > > https://lists.cinelerra-gg.org/pipermail/cin/2024-November/009123.html AV1 doesn't support interlaced frame. You may add deinterlace filter before encoder after apply this patch: https://patchwork.ffmpeg.org/project/ffmpeg/list/?series=13372 cmd: ffmpeg -hide_banner -qsv_device /dev/dri/renderD128 - hwaccel_output_format qsv -i input.mp4 -pix_fmt nv12 -vf vpp_qsv=deinterlace=advanced -c:v av1_qsv -f null - Thanks Fei > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel(a)ffmpeg.org > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > ffmpeg-devel-request(a)ffmpeg.org with subject "unsubscribe". _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel(a)ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-request(a)ffmpeg.org with subject "unsubscribe".

4 20

test patch for 7.0 ffmpeg
by Andrew Randrianasulu 20 Nov '24

20 Nov '24

It may work or break it completely ... apply by normal patch command from inside thirdparty/ffmpeg (may be "cat pathname | patch -p1", may be "cat patchname | patch -p0") or edit libavcodec/qsvenc.c by hand. rebuild ffmpeg, do not do make clean, cd ../../ so you again in main directory with ./autogen.sh and bin directory with cinelerra. touch cinelerra/ffmpeg.C make Hopefully this rebuild cin so on next start from bin/cin you will have something to test with tff sources. Unfortunately, netBSD does not support hardware virtualization on AMD cpus, so for now I do not have my virtual machines. I also forgot type of cable from PSU to disks, so now I can only power up one of two sata drives. It will be fixed eventually, but for now I hope to live up this netBSD install a bit ...

2 21

10bit av1 and h264 QSV and VAAPI encoding
by Terje J. Hanssen 19 Nov '24

19 Nov '24

Now when 10bit hevc_qsv and hevc_vaapi accelerated encoding looks to be in box, I wonder if something prohibits us to get accelerated encoding for av1 and h264? According to ffmpeg help the following support should be available: av1_qsv Supported pixel formats: nv12 p010le qsv -profile <int> E..V....... (from 0 to INT_MAX) (default unknown) unknown 0 E..V....... main 1 E..V....... av1_vaapi Supported pixel formats: vaapi -profile <int> E..V....... Set profile (seq_profile) (from -99 to 255) (default -99) main 0 E..V....... high 1 E..V....... professional 2 E..V....... h264_qsv Supported pixel formats: nv12 qsv -profile <int> E..V....... (from 0 to INT_MAX) (default unknown) unknown 0 E..V....... baseline 66 E..V....... main 77 E..V....... high 100 E..V....... h264_vaapi Supported pixel formats: vaapi -profile <int> E..V....... Set profile (profile_idc and constraint_set*_flag) (from -99 to 65535) (default -99) constrained_baseline 578 E..V....... main 77 E..V....... high 100 E..V....... high10 110 E..V.......

3 18

more complete 10bit vaapi encode patches
by Andrew Randrianasulu 15 Nov '24

15 Nov '24

Hopefully you can just put "cin_pix_fmt=p010le" in profile and it will be respected Terje, you only need to apply second patch on top of that you already have.

2 7

another 10 bit encoding patch
by Andrew Randrianasulu 15 Nov '24

15 Nov '24

you need to reset your git tree, something like git reset --hard 2037bf63cdf25462ca606f4a28636e12df77c2a0 before applying this patch then run cin like CIN_10BIT_ENC=p010le bin/cin or replace p010le with whatever equivalents y210le (?) sorry, I just hanged my X session so can't copy/paste exact output, but if you launch cingg with error=debug in bin/ffmpeg/encode.opts it will print list of accepted by vaapi library formats at encoding attempt. I think this is still more like prototype to see if we can get it working, I hope to wire it to existing profile reading mechanism.

2 11

Fwd: New Topic
by Phyllis Smith 14 Nov '24

14 Nov '24

Anyone familiar with this package? Cincity uses CinGG but obviously we all use other packages for specific things? ---------- Forwarded message --------- From: Cinelerra GG - Forum <forum(a)cinelerra-gg.org> Date: Wed, Nov 13, 2024 at 5:41 AM Subject: New Topic To: <phylsmith2017(a)gmail.com> Hello phylsmith2017(a)gmail.com! New topic has been created on your subscribed forum - Miscellaneous <https://www.cinelerra-gg.org/forum/forum-miscellaneous/>. *Help on slowmoVideo* I need to slow down a clip which has been recorded at 30 FPS. slowmoVideo is very interesting as it uses the optical flow technology for interpolating frames like respeedr does. And it uses FFMpeg. It can produce super fluid video. It can even be enhanced by a module to use the GPU efficiently. Unfortunately I have problems and at Github no one is answering to this technical issue (maybe because the people spammed around in the past). Does anyone of you have the possibility to take a look at it? My debian package installs perfectly and starts without issues. I can open the clip but on render it crashes. Using the command line I get this error: Have OpenCL: true useOpenCL: true true 1 1 terminate called after throwing an instance of 'cv::Exception' what(): OpenCV(4.6.0) ./modules/video/src/optflowgf.cpp:1120: error: (-215:Assertion failed) _flow0.size() == prev0.size() && _flow0.channels() == 2 && _flow0.depth() == CV_32F in fun... ------------------------------ If you want to unsubscribe from this forum please use the link below.

3 2

10bit VAAPI encoding preset
by Terje J. Hanssen 13 Nov '24

13 Nov '24

After finishing the three hevc_qsv presets for _8b420, _10b420 and _10b422 mp4 https://lists.cinelerra-gg.org/pipermail/cin/2024-November/009039.html I thought to do the corresponding for hevc_vaapi However I am struggling to get 10bit 422 output rendered from Cingg, which so far falls back to 10bit 420 chroma subsampling fprobe -hide_banner hd01_hevc_vaapi_10b422.mp4 Duration: 00:01:11.20, start: 0.000000, bitrate: 11082 kb/s 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], 11080 kb/s, 25 fps, 25 tbr, 12800 tbn (default) In a previous thread I proved that ffmpeg hevc_vaapi encoding has this 10bit-422 capability 3) 10-bit yuv422p10le hevc_vaapi encoding using -vf 'format=y210,hwupload' https://lists.cinelerra-gg.org/pipermail/cin/2024-November/008982.html And yes, there was also issues before hevc_qsv_10b422.mp4 combinations Pixel format/profile finally succeeded. But here are some differences, yet. While hevc_qsv provided a Pixels drop down menu with pixel formats, hevc_vaapi just has a single "vaapi" pixel format. Any idea why, and what can fulfill the vaapi 10bit 422 encoding here?

3 8

10bit VAAPI encoding test thread continued
by Terje J. Hanssen 13 Nov '24

13 Nov '24

Continued and extracted from previous threads_ https://lists.cinelerra-gg.org/pipermail/cin/2024-October/008967.html Den 27.10.2024 21:11, skrev Terje J. Hanssen: > ........snip > > For me it looks like nothing has changed: > > localhost:/Cin # git am 0001-Experimental-try-10bit-vaapi-encoding.patch > fatal: previous rebase directory .git/rebase-apply still exists but > mbox given. > > > remove it again? > > > By the way,testing > > bin/cin > compression: hevc_vaapi.mp4 > Pixels: vaapi (no other options) > cin_hw_dev=vaapi > > Renders > > Color space : YUV > Chroma subsampling : 4:2:0 > Bit depth : 8 bits > > > The same with > CIN_10BIT_ENC=1 bin/cin > > > > yeah, it does not exist in normal cin, only after you add patch. > > > may be just take a break from all this, I do not think we will solve > this problem in few remaining days until end of October. So ... do > not rush > > Ok for me. > Den 04.11.2024 21:49, skrev Andrew Randrianasulu: > > Terje, I think if you are ok with this idea we can return to testing > 10bit-vaapi patch ... > Hopefully in its final form it will just allow same format= string in > encoding profiles as supported by per-file decoding opts files now ,,, > I'm not sure I understood the last line yet, but I should be ready to continue vaapi-testing to-morrow. ==================== Yeah, here we are. The problem as I understood lastly, was to apply the 10bit-vaapi-encoding.patch?

3 35