- Cin - Cinelerra-GG Mailing Lists

HW Accel
by Andrea paz 26 Apr '25

26 Apr '25

A few silly questions on topics already addressed several times. I was reading about hardware acceleration in ffmpeg. In my system ffmpeg I have: $ ffmpeg -hwaccels ffmpeg version n7.1 Copyright (c) 2000-2024 the FFmpeg developers built with gcc 14.2.1 (GCC) 20250207 configuration: --prefix=/usr --disable-debug --disable-static --disable-stripping --enable-amf --enable-avisynth --enable-cuda-llvm --enable-lto --enable-fontconfig --enable-frei0r --enable-gmp --enable-gnutls --enable-gpl --enable-ladspa --enable-libaom --enable-libass --enable-libbluray --enable-libbs2b --enable-libdav1d --enable-libdrm --enable-libdvdnav --enable-libdvdread --enable-libfreetype --enable-libfribidi --enable-libglslang --enable-libgsm --enable-libharfbuzz --enable-libiec61883 --enable-libjack --enable-libjxl --enable-libmodplug --enable-libmp3lame --enable-libopencore_amrnb --enable-libopencore_amrwb --enable-libopenjpeg --enable-libopenmpt --enable-libopus --enable-libplacebo --enable-libpulse --enable-librav1e --enable-librsvg --enable-librubberband --enable-libsnappy --enable-libsoxr --enable-libspeex --enable-libsrt --enable-libssh --enable-libsvtav1 --enable-libtheora --enable-libv4l2 --enable-libvidstab --enable-libvmaf --enable-libvorbis --enable-libvpl --enable-libvpx --enable-libwebp --enable-libx264 --enable-libx265 --enable-libxcb --enable-libxml2 --enable-libxvid --enable-libzimg --enable-libzmq --enable-nvdec --enable-nvenc --enable-opencl --enable-opengl --enable-shared --enable-vapoursynth --enable-version3 --enable-vulkan libavutil 59. 39.100 / 59. 39.100 libavcodec 61. 19.100 / 61. 19.100 libavformat 61. 7.100 / 61. 7.100 libavdevice 61. 3.100 / 61. 3.100 libavfilter 10. 4.100 / 10. 4.100 libswscale 8. 3.100 / 8. 3.100 libswresample 5. 3.100 / 5. 3.100 libpostproc 58. 3.100 / 58. 3.100 Hardware acceleration methods: vdpau cuda vaapi qsv drm opencl vulkan If I try the same command with ffmpeg contained in .../thirdparty/ffmpeg-7.0 I get: ./ffmpeg -hwaccels Hardware acceleration methods: vdpau cuda vaapi vulkan So I guess OpenCL; Qsv and drm can only be obtained by manually compiling them. With what exact options in ./configure? (I never know when to use “--with-” and when to use “--enable-”). “drm” can serve in CinGG? Another thing: vulkan is already included in CinGG, but I have never been able to activate the device and get it working. How to do that? With system ffmpeg I can activate it without any problems. I am attaching the part of ffmpeg's documentation regarding hardware acceleration. The following part (regarding decoding) seems important to me: "Note that most acceleration methods are intended for playback and will not be faster than software decoding on modern CPUs. Additionally, ffmpeg will usually need to copy the decoded frames from the GPU memory into the system memory, resulting in further performance loss. This option is thus mainly useful for testing." As usual, I thought it would be a good thing to have playback in hardware but it is not. (However, there is a video of Adam where he manages to get better decoding with his Nvidia: https://www.youtube.com/watch?v=IZdJqrObnik)

2 1

HDR10 on radeonsi/EGL
by Andrew Randrianasulu 26 Apr '25

26 Apr '25

interesting bug report I found. https://gitlab.freedesktop.org/mesa/mesa/-/issues/12669 "radeonsi: Cannot deliver HDR10 0.0-125.0 values to 16 bit float GBM surface, only 0.0-1.0 seems to work. " So, it sort of force-normalized at least in egl ?

4 5

OpenCL on CPU
by Andrew Randrianasulu 25 Apr '25

25 Apr '25

note, openCL is different to openGL, mostly being more about more accurate computations. On AMD FX4300, 32bit userspace but llvm probably uses avx? guest@slax:/dev/shm/mesa/BUILD$ RUSTICL_ENABLE=llvmpipe clpeak Platform: rusticl Device: llvmpipe (LLVM 20.1.3, 256 bits) Driver version : 25.2.0-devel (git-845611bb43) (Linux x86) Compute units : 8 Clock frequency : 300 MHz Global memory bandwidth (GBPS) float : 3.72 float2 : 4.08 float4 : 3.59 float8 : 2.81 float16 : 2.09 Single-precision compute (GFLOPS) float : 14.67 float2 : 17.86 float4 : 15.99 float8 : 14.72 float16 : 14.63 No half precision support! Skipped No double precision support! Skipped Integer compute (GIOPS) int : 13.89 int2 : 13.25 int4 : 12.85 int8 : 13.04 int16 : 11.51 Integer compute Fast 24bit (GIOPS) int : 13.65 int2 : 13.29 int4 : 13.23 int8 : 12.90 int16 : 11.08 Transfer bandwidth (GBPS) enqueueWriteBuffer : 2.82 enqueueReadBuffer : 1.08 enqueueWriteBuffer non-blocking : 2.89 enqueueReadBuffer non-blocking : 1.02 enqueueMapBuffer(for read) : 1.15 memcpy from mapped ptr : 3.02 enqueueUnmap(after write) : 2.22 memcpy to mapped ptr : 3.01 Kernel launch latency : 21.55 us guest@slax:/dev/shm/mesa/BUILD$ command to build somewhat minimal mesa (llvmpipe + amd): meson ../ --prefix=/usr/X11R7 --libdir=lib --strip --buildtype debugoptimized -Degl=enabled -Dosmesa=true -Dplatforms=x11 -Dgallium-drivers=r600,radeonsi,llvmpipe -Dvulkan-drivers=amd,swrast -Dgallium-nine=true -Dgallium-va=enabled -Dgallium-xa=disabled -Dgallium-rusticl=true -Dllvm=enabled -Drust_std=2021 -Dvideo-codecs="all" of course you can set your own prefix ( I have X installed into non-default location). Biggest obstacle for me was that mesa git require some new llvm, and just released two days ago SPIRV-Tools-2024.4 ! And github "release" is of course broken, in sense you need to manually fetch headers at specific commit. Of course "real gpu" will get like >200 GFLOPS , even my puny GF710 was that fast, but possibility of lock up makes this option less attractive ;)

1 3

Found hw transcoding line for ffmpeg!
by Andrew Randrianasulu 25 Apr '25

25 Apr '25

./ffmpeg -hwaccel_output_format vaapi -hwaccel vaapi -i ~/New_hdd/Swimming\ with\ orca\ in\ New\ Zealand-JQ3mDXF3bcE.mkv -c:a copy -c:s copy -c:v h264_vaapi -f mp4 /dev/null -loglevel verbose main detail missing was "-hwaccel_output_format vaapi". with it I can transcode h264->h264 FHD at 42 fps and h264->hevc at 90 fps (know feature of Polaris based encoder at least under Linux).

2 2

Got RX550 working on Linux!
by Andrew Randrianasulu 25 Apr '25

25 Apr '25

With some kind of HDMI (1.4) to VGA adapter bash-5.1$ inxi -Gxi Graphics: Device-1: AMD Lexa PRO [Radeon 540/540X/550/550X / RX 540X/550/550X] vendor: ASRock driver: amdgpu v: kernel bus-ID: 01:00.0 Display: server: X.Org 1.21.1.99 driver: loaded: amdgpu resolution: 1440x900~60Hz OpenGL: renderer: AMD Radeon RX 550 / 550 Series (radeonsi polaris12 ACO DRM 3.49 6.1.44-x64) v: 4.6 Mesa 25.2.0-devel (git-cd8d8ca795) direct render: Yes vainfo Trying display: wayland Trying display: x11 vainfo: VA-API version: 1.19 (libva 2.19.0) vainfo: Driver version: Mesa Gallium driver 25.2.0-devel for AMD Radeon RX 550 / 550 Series (radeonsi, polaris12, ACO, DRM 3.49, 6.1.44-x64) vainfo: Supported profile and entrypoints VAProfileMPEG2Simple : VAEntrypointVLD VAProfileMPEG2Main : VAEntrypointVLD VAProfileVC1Simple : VAEntrypointVLD VAProfileVC1Main : VAEntrypointVLD VAProfileVC1Advanced : VAEntrypointVLD VAProfileH264ConstrainedBaseline: VAEntrypointVLD VAProfileH264ConstrainedBaseline: VAEntrypointEncSlice VAProfileH264Main : VAEntrypointVLD VAProfileH264Main : VAEntrypointEncSlice VAProfileH264High : VAEntrypointVLD VAProfileH264High : VAEntrypointEncSlice VAProfileHEVCMain : VAEntrypointVLD VAProfileHEVCMain : VAEntrypointEncSlice VAProfileHEVCMain10 : VAEntrypointVLD VAProfileJPEGBaseline : VAEntrypointVLD VAProfileNone : VAEntrypointVideoProc bash-5.1$ clpeak Platform: rusticl Device: AMD Radeon RX 550 / 550 Series (radeonsi, polaris12, ACO, DRM 3.49, 6.1.44-x64) Driver version : 25.2.0-devel (git-cd8d8ca795) (Linux x86) Compute units : 8 Clock frequency : 1183 MHz Global memory bandwidth (GBPS) float : 75.30 float2 : 76.62 float4 : 80.28 float8 : 81.12 float16 : 78.93 Single-precision compute (GFLOPS) float : 1188.83 float2 : 1187.27 float4 : 1182.98 float8 : 1174.67 float16 : 1157.09 No half precision support! Skipped No double precision support! Skipped Integer compute (GIOPS) int : 240.17 int2 : 240.10 int4 : 240.01 int8 : 239.80 int16 : 239.45 Integer compute Fast 24bit (GIOPS) int : 595.16 int2 : 594.82 int4 : 594.37 int8 : 593.12 int16 : 590.84 Transfer bandwidth (GBPS) enqueueWriteBuffer : 3.38 enqueueReadBuffer : 1.44 enqueueWriteBuffer non-blocking : 3.39 enqueueReadBuffer non-blocking : 1.45 enqueueMapBuffer(for read) : 2.67 memcpy from mapped ptr : 3.53 enqueueUnmap(after write) : 3.29 memcpy to mapped ptr : 3.48 Kernel launch latency : 90.27 us bash-5.1$ So, mpv plays 4k HDR file from iPhone11 smoothly (finally!) CinGG only can do 15 fps on this file :( (8bit RGBA project, x11 output) Encoding into 8bit HEVC also work: bash-5.1$ mediainfo /dev/shm/vaapi-test.mp4 General Complete name : /dev/shm/vaapi-test.mp4 Format : MPEG-4 Format profile : Base Media Codec ID : isom (isom/iso2/mp41) File size : 132 MiB Duration : 19 s 158 ms Overall bit rate : 57.7 Mb/s Frame rate : 59.940 FPS Writing application : Lavf61.1.100 Video ID : 1 Format : HEVC Format/Info : High Efficiency Video Coding Format profile : Main@L5@Main Codec ID : hev1 Codec ID/Info : High Efficiency Video Coding Duration : 19 s 153 ms Bit rate : 57.6 Mb/s Width : 3 840 pixels Height : 2 160 pixels Display aspect ratio : 16:9 Frame rate mode : Constant Frame rate : 59.940 (60000/1001) FPS Color space : YUV Chroma subsampling : 4:2:0 Bit depth : 8 bits Bits/(Pixel*Frame) : 0.116 Stream size : 132 MiB (100%) Color range : Full Matrix coefficients : BT.709 Codec configuration box : hvcC Audio ID : 2 Format : AAC LC Format/Info : Advanced Audio Codec Low Complexity Codec ID : mp4a-40-2 Duration : 19 s 158 ms Source duration : 19 s 179 ms Bit rate mode : Constant Bit rate : 128 kb/s Channel(s) : 2 channels Channel layout : L R Sampling rate : 48.0 kHz Frame rate : 46.875 FPS (1024 SPF) Compression mode : Lossy Stream size : 300 KiB (0%) Source stream size : 300 KiB (0%) Language : Russian Default : Yes Alternate group : 1 but encoding still was done in seven minutes for 20 seconds test clip (no tonemap either) So I guess my machine seriosly bootlenecked by PCIe speeds/memory speeds? bash-5.1$ dmesg | grep amdgpu [ 9.887359] [drm] amdgpu kernel modesetting enabled. [ 9.887485] amdgpu: CRAT table not found [ 9.887527] amdgpu: Virtual CRAT table created for CPU [ 9.887583] amdgpu: Topology: Add CPU node [ 9.888699] amdgpu 0000:01:00.0: No more image in the PCI ROM [ 9.889843] amdgpu 0000:01:00.0: amdgpu: Fetched VBIOS from ROM BAR [ 9.889886] amdgpu: ATOM BIOS: 113-EXT-700004_L01 [ 9.890070] amdgpu 0000:01:00.0: vgaarb: deactivate vga console [ 9.893944] amdgpu 0000:01:00.0: amdgpu: Trusted Memory Zone (TMZ) feature not supported [ 9.893950] amdgpu 0000:01:00.0: amdgpu: PCIE atomic ops is not supported [ 10.064640] amdgpu 0000:01:00.0: amdgpu: VRAM: 4096M 0x000000F400000000 - 0x000000F4FFFFFFFF (4096M used) [ 10.064655] amdgpu 0000:01:00.0: amdgpu: GART: 256M 0x000000FF00000000 - 0x000000FF0FFFFFFF [ 10.064725] [drm] amdgpu: 4096M of VRAM memory ready [ 10.064729] [drm] amdgpu: 7998M of GTT memory ready. [ 10.143106] amdgpu: hwmgr_sw_init smu backed is polaris10_smu [ 10.572632] snd_hda_intel 0000:01:00.1: bound 0000:01:00.0 (ops amdgpu_dm_audio_component_bind_ops [amdgpu]) [ 10.754610] kfd kfd: amdgpu: skipped device 1002:699f, PCI rejects atomics 730<0 [ 10.754630] amdgpu 0000:01:00.0: amdgpu: SE 2, SH per SE 1, CU per SH 5, active_cu_number 8 [ 10.758061] amdgpu 0000:01:00.0: amdgpu: Using BACO for runtime pm [ 10.758456] [drm] Initialized amdgpu 3.49.0 20150101 for 0000:01:00.0 on minor 0 [ 10.765104] fbcon: amdgpudrmfb (fb0) is primary device [ 10.838142] amdgpu 0000:01:00.0: [drm] fb0: amdgpudrmfb frame buffer device bash-5.1$ dmesg | grep PCIe [ 0.364217] acpi PNP0A03:00: _OSC: platform does not support [PCIeHotplug SHPCHotplug PME LTR] [ 0.364477] acpi PNP0A03:00: _OSC: OS now controls [AER PCIeCapability] [ 0.376895] pci 0000:01:00.0: 32.000 Gb/s available PCIe bandwidth, limited by 5.0 GT/s PCIe x8 link at 0000:00:02.0 (capable of 63.008 Gb/s with 8.0 GT/s PCIe x8 link) Hm, 32 Gb/s should be enough for many 4k streams? 4k ~= 8Kpix, 8*60 = 480Mpix/sec 4 byte/per pixel (30 bit for 10 bpc colors, 2 padding) still just 2Gb/s!

2 3

vf_libplacebo
by Andrew Randrianasulu 22 Apr '25

22 Apr '25

https://forum.doom9.org/showpost.php?p=1983532&postcount=248 Your machine should work with hw vulkan driver for AMD GPU, but there is method to run same filter with software "lavapipe" (much slower, of course) ===== Environment vars: LIBGL_ALWAYS_SOFTWARE=1 __GLX_VENDOR_LIBRARY_NAME=mesa VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/lvp_icd.i686.json:/usr/share/vulkan/icd.d/lvp_icd.x86_64.json Command: ffmpeg -i video.mkv -init_hw_device vulkan -vf hwupload,libplacebo=format=yuv420p,hwdownload,format=yuv420p test.mkv ==== problem with self-building libplacebo it build-depends on some quite heavy Vulkan shader compiler (?) thing, termux does have libplacebo packaged for mpv but binary package for ffmpeg (7.1.1) was built without vulkan support ...

1 0

Tonemapper algorithms (EXR input, require OpenGL)
by Andrew Randrianasulu 22 Apr '25

22 Apr '25

https://github.com/tizian/tonemapper found while looking at why F_tonemap visibly misbehaves for me ...

2 6

Even more on HDR in video
by Andrew Randrianasulu 22 Apr '25

22 Apr '25

https://lightillusion.com/what_is_hdr.html This page points out something interesting: you can have only *some* components of "HDR" in any particular display, so for example display can be Wide Gamut bt2020 BUT SDR! And/or smaller than 4k ;) === Additionally, the individual aspects of the HDR specifications can, and often are, used in isolation. For example, there is nothing to stop a standard gamut display (Rec709), with standard HD or even SD resolution, working with an HDR EOTF with high contrast range. ====== It also points out that viewing environment even more important for both HDR schemas - PQ and HLG. Interestingly it includes table where specific (integer!) signal levels correspond to specific luminance in nits. I still can't fully wrap my mind around what exactly "nit" represent, but apparently in HDR you really grade relative to specific max/min luminance of your mastering display AND suitable HDR display/TV must do some tone mapping magic if your defined luminance (?) goes above level that monitor/tv can show. There was equation on page about tonemapping https://64.github.io/tonemapping/ === Converting a linear RGB triple to a luminance value is easy: L=0.2126R+0.7152G+0.0722B ==== but I am not sure if it works for *any* useful RGB values? Also, issue with viewing distance wil play differently for those VR setups?

3 3

HDR -> sdr tonemap on cpu only
by Andrew Randrianasulu 22 Apr '25

22 Apr '25

Many (if not all) ffmpeg guides insist on using either libplacebo (vulkan API on GPU) or zscale (not available in some ffmpeg builds). Here I found collection of .cube files for doing REC2020 TO rec709. https://www.reddit.com/r/premiere/comments/12t9mh8/convert_hdr_to_sdr/ https://drive.google.com/file/d/1HQGKE5Fzg_W5vsP8wJcmzRIAtCRKevsV/view I extracted one .cube file and run this ffmpeg command (at 1.4 fps on Termux) ffmpeg -i storage/downloads/iPhone11_4K-recorder_59.940HDR10.mov -vf lut3d=file=REC2020PQto709v1.cube:interp=tetrahedral,scale=1280:-2 ~/1.mp4 and it sort of worked, while I think mpv/libplacebo did better job on my desktop. But even just in mpv I had tons of dropped frames, and decoding in cingg without any tonemapping was at may be 9 fps max, out of 60. Ow :( For some reason just throwing F_tonemap to timeline resulted in extremely dark image no matter the settings. I pushed it up by brightness/contrast native plugin but I do not think you supposed to do that? sw tonemapper was added to ffmpeg in 2017 https://github.com/FFmpeg/FFmpeg/blob/master/libavfilter/vf_tonemap.c

3 7

EOTF, PQ, HLG ....
by Andrew Randrianasulu 22 Apr '25

22 Apr '25

It seems that at least for "PQ" HDR systems their EOTF can be used to give some reading into waveform scopes etc? As they set to "Nits" and not -10-110% as usual for SDR (and may be Hybrid Log Gamma?) systems. https://github.com/colour-science/colour/discussions/712 ===== A true PQ curve is an absolute cv-nit scale. The normalized CVs of a few different nit levels are: 100 nit = 0.5081 600 nit = 0.6963 1000 nit = 0.7518 4000 = 0.9026 Here's Steve Shaw's write-up on the matter which further backs up my concerns: https://www.lightillusion.com/uhdtv.html [image: hdr_luminance_levels] <https://user-images.githubusercontent.com/30381480/38954716-4812c0ba-4307-1…> This behavior is how reference displays should work (notwithstanding HDR metadata and consumer-grade secret-sauce), and I can confirm is the way the Dolby Pulsar, Sony x300, and x550 all work. When fed RGB(.5081, .5081, .5081) on all three displays set to D65 ST2084, you should measure 100 nits. If you feed a linear ramp to all the displays, you will see a hard clip of all values above the associated code value for that display's peak luminance. ===== quote end ==== Apple's document HDR and Wide Color Gamut in Final Cut Pro White Paper November 2020 says: ==== Wide Color Gamut Displays that support wider color gamuts with a traditional luminance range are sometimes referred to as wide-gamut, standard-dynamicrange (SDR) displays. Current models of Mac, iPhone, and iPad support a wider-gamut SDR color space called Display P3. Display P3 uses the same color gamut as DCI P3, the standard for movie theater digital projectors, combined with a white point and gamma appropriate for typical viewing environments. The figure below shows that the P3 color gamut contains more colors than the Rec. 709 gamut, with notable improvements in the range of red and green areas of the color spectrum. [..] High Dynamic Range Displays that support both wider color gamuts and higher luminance ranges are referred to as high-dynamic-range (HDR) displays. HDR displays can produce not only high peak luminance values but also very low black levels, revealing more detail in both shadows and highlights. HDR displays typically reach peak brightness levels above 1000 nits while maintaining very high contrast ratios, much higher than the brightness levels and contrast ratios of SDR displays. These levels of brightness mean that images appear more true to life, with a dynamic range that more closely reflects what the eye sees in the real world. The figure below shows how increased brightness levels (measured in nits) combined with a wider color gamut greatly expand the range of reproducible color. HDR displays typically process video at 10 bits per color component rather than 8 bits. The additional color data lets HDR displays render more discrete steps from the minimum to maximum brightness value in each color. In order to encode and transmit HDR images more efficiently, new mathematical transfer functions are used instead of gamma. Two popular standards are the Perceptual Quantizer (PQ) function, defined in both SMPTE ST 2084 and ITU-R Recommendation BT.2100, and Hybrid Log-Gamma (HLG), also defined in Rec. 2100. About Changing a Project’s Color Space After you set a library to wide-gamut HDR, any project with HD resolution or higher created in that library can be set to one of the following color spaces using the Modify command in the Project Properties inspector (described in step 3 above): • Standard - Rec. 709 • Wide Gamut - Rec. 2020 • Wide Gamut HDR - Rec. 2020 PQ • Wide Gamut HDR - Rec. 2020 HLG [..] When performing these adjustments, it may be helpful to show the Waveform video scope and set it to either RGB Parade or RGB Overlay view. In these views, the vertical axis of the waveform is labeled in units of cd/m2 (nits), which is a measure of the display’s output brightness level corresponding to the PQ signal level. Note that for the Luma and Chroma views, and for any non-PQ content, the vertical axis reverts to IRE (where 100 IRE represents 100% signal level). This is because the waveform signal in these cases does not correspond directly to absolute cd/m2 values. === quote ends here finally ===== So, it seems that we currently lack easy user choice between PQ/HLG variants of Wide Gamut bt 2020 color space, I think in ffmpeg/swscale they hide as "trc" (transfer curve) ... but I can be wrong. Also tonemapper on cpu, at least on my potato, tend to be slow, unless you downscale to like 1/4 out of 4k source? So it will be helpful to look into vf_libplacebo, opencl_tonemap etc on machines where they work. Sorry if it feels like too much of confusing info, I think while technically cingg currently capable of *encoding* at least some variants of HDR - more accurate displaying, easy user settings, waveform scope display "morphing" to HDR mode, HDR10+ dynamic metadata calculations all missed, relative to FCP/Resolve.

1 0