March 2026 - Cin - Cinelerra-GG Mailing Lists

Vulkan video in qemu
by Andrew Randrianasulu 21 Mar '26

21 Mar '26

https://gist.github.com/peppergrayxyz/fdc9042760273d137dddd3e97034385f Even more bleeding edge (in other words: it very well may cut you quite painfully) than normal Vulkan encode/decode, I guess host Intel (ANV) limitations apply ... Requires patched qemu. It sort of wild people try to game in guest Linux, in Linux-hosted Qemu VM. But well ...

3 3

Offtopic: this is blogpost about "AI" I liked
by Andrew Randrianasulu 20 Mar '26

20 Mar '26

No marketing, just description of work that gone into making domain-specific NN and running it "on potato". https://oldapplestuff.com/blog/MacinAI-Local/

2 1

A bunch of follow-up questions
by Terje J. Hanssen 20 Mar '26

20 Mar '26

This is a bunch of follow-up questions I feel both users and occasionally builders want easy clarified without to much research or experiments. I neither say these cannot be partly answered in nor derived from the manual documentation. Additional questions may appear ;) Q1 What capabilites/features does the pre-build Cingg AppImage and/vs the package build (ex. RPM) have? especially with regards to ffmeg, multimedia codecs, encoders etc. Q2 Is the AppImage static linked and RPM dynamic linked or combined dynamic/staticvwith regards to libs? Q3 Does the AppImage use the thirdparty ffmpeg and the RPM the local system ffmpeg? or are both static builds with third party ffmpeg Q4 Where is possibly the ffmpeg binary for the extracted AppImage and installed RPM? for own build ffmpeg binary is found in cinelerra-5.1/thirdparty/ffmpeg-8.0/ffmpeg Q5 Who make and maintain the thirdparty ffmpeg and what extra codecs are added? Q6 What are the configure scripts for the AppImage vs RPM? Q7 What extra capabilites/features may be achieved and justified with own custom build? what config elements? Q8 How often is it needed or suggested to download (or update if possible) the source repo via git clone?

4 19

I think ffmpeg 8.1 is out
by Andrew Randrianasulu 16 Mar '26

16 Mar '26

https://source.ffmpeg.org/gitweb/ffmpeg-web.git and site indeed now reads: ==== March 16th, 2026, FFmpeg 8.1 "Hoare" A new minor release, FFmpeg 8.1 "Hoare", is now available for download. Here are some of the highlights: * Decoders: xHE-AAC Mps212 (experimental) MPEG-H decoding via libmpeghdec * EXIF Metadata Parsing * LCEVC: support for parsing and forwarding metadata * Vulkan compute-based codecs: ProRes encoding and decoding, DPX decoding * D3D12: D3D12 H.264/AV1 encoding, scale_d3d12, mestimate_d3d12, deinterlace_d3d12 filters * Rockchip H.264/HEVC hardware encoding * IAMF: Projection mode Ambisonic Audio Elements muxing and demuxing * Formats: hxvs demuxer * Filters: drawvg, vpp_amf This release features a lot of internal changes and bugfixes. The groundwork for the upcoming swscale rewrite is progressing. The Vulkan compute-based codecs, and a few filters, no longer depend on runtime GLSL compilation, which speeds up their initialization. A companion post about the Vulkan Compute-based codec implementations has been published on the Khronos blog, featuring technical details on the implementations and future plans. We recommend users, distributors, and system integrators to upgrade unless they use current git master. ==== Hopefully just renaming our ffmpeg 8.0 patches to 8.1 will make it work (but with new libdav1d). I tested ffmpeg.git few days ago (see email archive) and it was working for simple decoding, vaapi/vulkan decoding and vaapi encoding. But there still might be surprises in less used parts like BD creation.

1 0

oops, looks like I broke x11 direct mode while adding workaround for yuva16
by Andrew Randrianasulu 15 Mar '26

15 Mar '26

I use if ( a = b) instead of if (a == b) so it was not working as intended (fps was just 15 instead of 58 for vulkan accelerated h265 4k HDR file) Please test, I tested like this RADV_PERFTEST=video_decode ALLOW_OLD_16BPC=1 bin/cin ~/K38_sdcard1/Documents/iPhone11_4K-recorder_59.940HDR10.mov

2 1

thirdparty ffmpeg-8.0 vs system ffmpeg-8.0.1
by Terje J. Hanssen 14 Mar '26

14 Mar '26

12.03.2026 ---------------- Min siste build med utvidet /home3/cinelerra/cinelerra-5.1/thirdparty konfigurering ./configure --with-single-user --with-booby --enable-libsvtav1 --with-onevpl --with-vulkan --with-libplacebo --with-libzimg fikk lagt til hwaccel kapabiliteter for qsv, drm og vulkan /home3/cinelerra/cinelerra-5.1/thirdparty/ffmpeg-8.0 # ffmpeg -hide_banner -hwaccels Hardware acceleration methods: vdpau cuda vaapi qsv drm vulkan ---------------------------- 02/03/2026, 23:43 I build a standard single-user, static bundled Cingg fra original bld.sh med original config line ./configure --with-single-user --with-booby Looked at the thirdparty ffmpeg-8.0 cd /home3/cinelerra/cinelerra-5.1/thirdparty/ffmpeg-8.0 Trying to verify its available hwaccels methods: ./ffmpeg -hwaccels Hardware acceleration methods: vdpau cuda vaapi Verifying the same with my system ffmpeg-8.0.1 on Tumbleweed-Slowroll I get ffmpeg -hide_banner -hwaccels Hardware acceleration methods: vdpau cuda vaapi qsv drm vulkan I wonder why not also the thirdparty ffmpeg-8.0 does list "qsv" and "vulkan"? Seemingly among encoders "--enable-libvpl" and "--enable-vulkan" are listed.

1 0

Current rebuild for Hw-accelerated QSV GPU decoding and Vulkan
by Terje J. Hanssen 13 Mar '26

13 Mar '26

continuation from a previous thread [Cin]Re: Adding CIN_HW_DEV=qsv for GPU HW decoding? https://lists.cinelerra-gg.org/archives/list/cin@lists.cinelerra-gg.org/mes… The current Slowroll update on Alder Lake /DG2 verifies use of vainfo | grep version libva info: VA-API version 1.23.0 libva info: Trying to open /usr/lib64/dri/iHD_drv_video.so libva info: Found init function __vaDriverInit_1_23 libva info: va_openDriver() returns 0 vainfo: VA-API version: 1.23 (libva 2.23.0) vainfo: Driver version: Intel iHD driver for Intel(R) Gen Graphics - 26.1.1 () Before further testing I had to rebuild my Cingg already patched with 0001-QSV-hw-decoder-test.patch cd /home3/cinelerra/cinelerra-5.1 make clean Edited the build.sh as a "current, all-in-one" ./configure line with libsvtav1, onevpl og vulkan cat bld.sh #!/bin/bash # IMPORTANT comments below to change the configure line # For newer operating system versions, add --enable-libsvtav1 # For really old versions, such as ubuntu 16, add --enable-libaom=no # Faster compile if include --disable-x265_hidepth # To add OpenCV plugins, add --with-opencv=sta,tar=https://download.cinelerra-gg.org/download.php?file=opencv%2Fopencv-20200306.tgz # # original ./configure --with-single-user --with-booby # # 12.03.2026 extended ./configure with Intel QSV api and Vulkan api and filters # --with-onevpl Intel QSV api for unix (no) # --with-vulkan Vulkan api (no) # --with-libplacebo libplacebo ffmpeg Vulkan filter (no) # --with-libzimg libzimg ffmpeg filter (no) # ( ./autogen.sh ./configure --with-single-user --with-booby --enable-libsvtav1 --with-onevpl --with-vulkan --with-libplacebo --with-libzimg make && make install ) 2>&1 | tee log mv Makefile Makefile.cfg cp Makefile.devel Makefile This required two additional dependency packages on Slowroll before the script ran through zypper in libplacebo-devel zimg-devel ./bld.sh -------------------------- cd /home3/cinelerra/cinelerra-5.1 bin/cin Cinelerra Infinity - built: Mar 12 2026 23:52:16 Settings|Preferences|Performance|Use HW Device: none ---- Testing QSV decoding in CPU vs GPU respectively, that is before and after adding remap decoder mpeg2_qsv, shows that decoding in GPU isn't different faster than on CPU in this case, but moving most workload to the GPU should offload the CPU: cat bin/ffmpeg/decode.opts # apply at init decode loglevel=fatal formatprobesize=5000000 scan_all_pmts=1 remap_video_decoder libaom-av1=libdav1d remap_video_decoder mpeg2video=mpeg2_qsv mpeg2video -> h264_qsv ====================== CPU: ** rendered 5972 frames in 24.463 secs, 244.124 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 29% ------------- GPU: ** rendered 5972 frames in 24.562 secs, 243.140 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 42% mpeg2video -> hevc_qsv ====================== CPU: ** rendered 5972 frames in 21.910 secs, 272.570 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 24% -------------- GPU: ** rendered 5972 frames in 22.666 secs, 263.478 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 38% mpeg2video -> av1_qsv ====================== CPU: ** rendered 5972 frames in 24.330 secs, 245.458 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 24% -------- GPU: ** rendered 5972 frames in 24.451 secs, 244.244 fps audio0 pad 64 0 (64) intel_gpu_top Video: < = 39%

2 5

vaapi_h264 encoder on Polaris12
by Andrew Randrianasulu 11 Mar '26

11 Mar '26

Interestingly, if I set async_depth to 64 instead of default 2 speed of encoding full HD 29.97 fps h264 file increases from ** rendered 9133 frames in 442.712 secs, 20.630 fps to ** rendered 9133 frames in 337.237 secs, 27.082 fps so may be try this setting and see if results in higher fps for you?

1 0

Adding CIN_HW_DEV=qsv for GPU HW decoding?
by Terje J. Hanssen 11 Mar '26

11 Mar '26

I have tested transcoding mpeg2video to h264_qsv, hevc_qsv, vp9_qsv and av1_qsv HW GPU accelerated decoding+encoding with ffmpeg. Just by adding "-hwaccel qsv" before the input file, the correct qsv codec for decoding in GPU apparently is auto-detected. ffmpeg -hide_banner -hwaccel qsv -i hdv09_04.m2t -c:v hevc_qsv -y hdv09_04_ffmpeg_hevc_qsv.mp4 With a medium powerful (balanced) Intel Alder Lake (i7-12700KF) /DG2 (A750) the qsv.mp4 video output and playback are clean and faster with GPU decoding. For the legacy, limited SkyLake (h264_qsv) and Coffee Lake (hevc_qsv) igpu's, it is best to keep decoding in CPU, due to igpu buffer errors and some distortions on the playback. So I wonder if it manageable to test-implement an additional environment variable "CIN_HW_DEV=qsv" for GPU HW decoding that adds "-hwaccel qsv" on the input side? If it works, it may be added as a "Use HW Device: qsv" next. Terje J. H

2 15

Cinelerra-gg + ffmpeg.git testing
by Andrew Randrianasulu 11 Mar '26

11 Mar '26

Using this command line LDFLAGS=`pkg-config --libs x11` setarch i686 ./configure --with-single-user --with-vulkan --with-libplacebo --with-git-ffmpeg=https://git.ffmpeg.org/ffmpeg.git --disable-dav1d and moving all thirdparty/src/ffmpeg.git.patch* out of the way I get compile It seems to work with Vulkan decoding of 4K prores test file, at very minimum :) You should obviously omit setarch if compile for x86-64 ffmpeg version git-2026-03-11-e3ee346 Copyright (c) 2000-2026 the FFmpeg developers

1 1