VCEEnc

Common Features (VCEEnc & VCEEncC)

• Encoding via VCE: Supports H.264/AVC and HEVC (Polaris and later).
• Encoding Modes: CQP (Constant QP), CBR (Constant Bitrate), VBR (Variable Bitrate).
• Parameters: Level/Profile specification, Max Bitrate, Max GOP length, SAR ratio, Color Matrix, Reference distance, and B-frame count.

VCEEnc (AviUtl Plugin) Specifics

• Audio encoding.
• Muxing audio and chapters.
• Automatic field shift support.

VCEEncC (CLI) Specifics

• Input Formats: Supports avs, vpy, y4m, raw, etc.
• Hardware Decoding: H.264, HEVC, MPEG2, VP9, VC-1.
• Hardware Resizing.
• Audio: Extraction from source files and audio encoding.
• Muxing: Supports multiple containers like mp4, mkv, ts.
• Metrics: Calculates SSIM/PSNR of encoding results.
• GPU-Accelerated Filtering (OpenCL):
  • Deinterlacing: afs (Auto Field Shift), nnedi.
  • Subtitle burning.
  • Color space conversion: hdr2sdr.
  • Resizing: spline16, spline36, spline64, lanczos2, lanczos3, lanczos4.
  • Rotation / Inversion.
  • Padding (Black bars).
  • Banding reduction.
  • Denoising: knn (K-nearest neighbor), pmd (Regularized PMD method).
  • Detail enhancement: unsharp, edgelevel (Edge level adjustment), warpsharp.

VPP (Video Post-Processing) options allow you to add filter processing before encoding. The order of application for VPP filters is fixed and does not depend on the order they are specified on the command line. The filters are applied in the following sequence:

1. --vpp-colorspace
2. --vpp-libplacebo-tonemapping
3. --vpp-rff
4. --vpp-delogo
5. --vpp-afs
6. --vpp-nnedi
7. --vpp-bwdif
8. --vpp-yadif
9. --vpp-rtgmc
10. --vpp-rtgmc-bob
11. --vpp-rtgmc-search-prefilter
12. --vpp-rtgmc-edi
13. --vpp-rtgmc-retouch
14. --vpp-rtgmc-shimmer-repair
15. --vpp-rtgmc-primitive
16. --vpp-degrain
17. --vpp-kfm
18. --vpp-decomb
19. --vpp-ivtc
20. --vpp-transform/rotate
21. --vpp-decimate
22. --vpp-mpdecimate
23. --vpp-select-every
24. --vpp-convolution3d
25. --vpp-smooth
26. --vpp-denoise-dct
27. --vpp-bm3d
28. --vpp-fft3d
29. --vpp-msmooth
30. --vpp-knn
31. --vpp-nlmeans
32. --vpp-pmd
33. --vpp-hqdn3d
34. --vpp-anime4k-shader
35. --vpp-onnx
36. --vpp-onnx-deint
37. --vpp-rife-ov
38. --vpp-descale
39. --vpp-subburn
40. --vpp-libplacebo-shader
41. --vpp-resize
42. --vpp-unsharp
43. --vpp-vinverse
44. --vpp-chromashift
45. --vpp-deblock
46. --vpp-deflicker
47. --vpp-stab
48. --vpp-colorfix
49. --vpp-dehalo
50. --vpp-finedehalo
51. --vpp-hqdering
52. --vpp-edgelevel
53. --vpp-msharpen
54. --vpp-cas
55. --vpp-warpsharp
56. --vpp-detailsharpen
57. --vpp-maa
58. --vpp-softlight
59. --vpp-curves
60. --vpp-tweak
61. --vpp-deband
62. --vpp-libplacebo-deband
63. --vpp-padding
64. --vpp-tweak (Note: listed twice in source)

In environments with multiple VCE/VCN compatible GPUs, VCEEncC automatically selects a GPU based on the following logic:

1. Compatibility Check: It selects GPUs that support the requested Codec, Profile, and Level. It also checks for support of 10bit depth and HW decoding if those options are specified.
2. Feature Preference: It prefers GPUs that support B frames.
3. Utilization Optimization: If multiple GPUs meet the above criteria, it prefers the GPU with the lowest Video Engine (VE) utilization and then the lowest GPU core utilization. This is intended to improve throughput by distributing tasks.

Manual Selection: Use the --device option to manually specify which GPU to use.

Note: Utilization checks occur during the initialization phase. If multiple tasks are started simultaneously, they might target the same GPU before the utilization values are updated.

# Example of manual device selection (exact syntax depends on VCEEncC_Options.en.md)
# VCEEncC --device <device_id> ...

In multi-GPU environments, VCEEncC automatically selects the optimal GPU based on your specified options. If you want to bypass automatic selection, use the --device option to specify a GPU manually.

Selection Logic:

1. Capability Check: The tool selects GPUs that support the requested codec, profile, level, bit depth (e.g., 10-bit), and hardware decoding (if avhw is specified).
2. Feature Prioritization: If multiple GPUs are capable, it prioritizes those that support:
   • B-frames (if B-frames are requested).
   • Pre-analysis (if --pa is specified).
3. Efficiency Check: If multiple GPUs meet the above criteria, it selects the one with the lowest Video Engine (VE) utilization and lowest overall GPU utilization to maximize encoding speed and efficiency.

Warning: VE/GPU utilization is measured at the start of encoding. Because there is a slight delay in reporting, starting multiple encoding processes almost simultaneously may cause them to be assigned to the same GPU.

If you need --vpp-onnx support, you must manually prepare the ONNX Runtime DirectML directory. This involves downloading the NuGet package, extracting it, and moving the necessary headers and libraries to a directory defined by the ONNXRUNTIME_DIR environment variable.

For x64 VCEEncC builds, onnxruntime.dll and onnxruntime_providers_shared.dll are automatically copied from %ONNXRUNTIME_DIR%\lib to the output directory.

set ONNXRUNTIME_DIR=F:\VisualStudio2022\Projects\onnxruntime
if not exist "%ONNXRUNTIME_DIR%" mkdir "%ONNXRUNTIME_DIR%"
if not exist "%ONNXRUNTIME_DIR%\include" mkdir "%ONNXRUNTIME_DIR%\include"
if not exist "%ONNXRUNTIME_DIR%\lib" mkdir "%ONNXRUNTIME_DIR%\lib"

curl -L -o onnxruntime-directml.nupkg https://api.nuget.org/v3-flatcontainer/microsoft.ml.onnxruntime.directml/1.23.0/microsoft.ml.onnxruntime.directml.1.23.0.nupkg
7z x -oonnxruntime-directml -y onnxruntime-directml.nupkg

xcopy /e /i /y onnxruntime-directml\build\native\include "%ONNXRUNTIME_DIR%\include"
copy /y onnxruntime-directml\runtimes\win-x64\native\onnxruntime.dll "%ONNXRUNTIME_DIR%\lib\"
copy /y onnxruntime-directml\runtimes\win-x64\native\onnxruntime.lib "%ONNXRUNTIME_DIR%\lib\"
copy /y onnxruntime-directml\runtimes\win-x64\native\onnxruntime_providers_shared.dll "%ONNXRUNTIME_DIR%\lib\"

setx ONNXRUNTIME_DIR "%ONNXRUNTIME_DIR%"