Building ffmpeg with cuda in 2021

----------------------------------------------------------------------
UPDATE: Just use media-autobuild_suite, it's a billion times easier.
----------------------------------------------------------------------

Prerequisites:
- CUDA Toolkit:			https://developer.nvidia.com/Cuda-toolkit
- Media-autobuild_suite:	https://github.com/m-ab-s/media-autobuild_suite

- Have at least 12 Gb available, the final build will be around 700 Mb. Building itself will take several hours.
- With CUDA installed, run media-autobuild_suite.bat
- Follow instructions, generally self explanatory. Important choices: select "non-free", select "yes, custom ffmpeg options".
- If you get errors during "/making" of ffmpeg, visit ...\build\ffmpeg_options.txt and comment out sketchy modules.
My options, for example: https://pastebin.com/zGhA8CJG

Once building has finished (suite should autoclose itself), copy following files into your directory of choice:
- All files from ...\media-autobuild_suite\local64\bin-video\
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\nppc64_11.dll
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\nppicc64_11.dll
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\nppidei64_11.dll
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\nppig64_11.dll

Now that ffmpeg is ready, you can uninstall CUDA and remove all MAS files. The copied over files is what you'll use.
You're good to go here, and can ignore the rest of this text.



----------------------------------------------------------------------
----------------------------------------------------------------------
This guide should aid with building Ffmpeg with Cuda SDK.
It's heavily based on following answer: https://stackoverflow.com/a/56875864

System:
- OS: Windows 10, version 2004
- CUDA version: 11.2
- GPU: GTX 1660, Turing

Prerequisites:
- Visual Studio 2019 (or newer), with C++ packages
- MSYS2: https://www.msys2.org/
- CUDA Toolkit: https://developer.nvidia.com/Cuda-toolkit

1. Run Msys2.exe, execute following lines:
pacman -Su
pacman -S make
pacman -S diffutils
pacman -S yasm
pacman -S mingw-w64-x86_64-gcc
pacman -S mingw-w64-x86_64-toolchain
PKG_CONFIG_PATH=$PKG_CONFIG_PATH:/usr/loca/lib/pkgconfig
PATH=$PATH:/opt/bin
git clone https://git.ffmpeg.org/ffmpeg.git
git clone https://github.com/libav/libav
git clone https://git.videolan.org/git/ffmpeg/nv-codec-headers.git

2. Launch "x64 Native Tools Command Prompt for VS2019". Run C:\msys32\mingw64.exe using Native Tools command prompt.

3. (Building avconv.exe and avprobe.exe) Using Mingw64.exe, enter following lines:
cd libav
./configure
make
make install
cd ..

4. (Building ffnvcodec.pc) Using Mingw64.exe, enter following lines:
cd nv-codec-headers
make
make install
cd ..

5. Copy file C:\msys32\home\<user>\nv-codec-headers\ffnvcodec.pc to following directory: C:\msys32\usr\local\lib\pkgconfig\

6. Open CUDA Toolkit's installer (e.g. cuda_11.2.0_460.89_win10.exe) using 7-Zip. Find and extract following folders:
- ...\nvcc\include\ to MSYS2\tmp\nvidia_include\
- ...\npp_dev\lib\x64\ to MSYS2\tmp\nvidia_lib\x64\
- ...\npp_dev\include\ to MSYS2\tmp\nvidia_npp_include\

7. (Specifying GPU microarchitecture.)
Open file "C:\msys32\home\<user>\ffmpeg\configure" (it doesn't have extension.)
Find block that starts with "if enabled cuda_nvcc; then" (first occurrence, roughly at line #4345.)
Replace the number in "compute_30" and "sm_30" according to your GPU architecture, use this webpage as a cheat-sheet: https://arnon.dk/matching-sm-architectures-arch-and-gencode-for-various-nvidia-cards/
For example, Turing will have compute_75, sm_75.

8. (Building ffmpeg.exe) Using Mingw64.exe, enter following lines:
export PKG_CONFIG_PATH="/usr/local/lib/pkgconfig"
cd ffmpeg
./configure --enable-nonfree --disable-shared --enable-nvenc --enable-cuda --enable-cuvid --enable-libnpp --extra-cflags=-I/tmp/nvidia_npp_include --extra-cflags=-I/tmp/nvidia_include --extra-ldflags=-L/tmp/nvidia_lib/x64
make
make install

In case of errors during "./configure" procedure, be sure to investigate C:\msys32\home\<user>\ffmpeg\ffbuild\config.log

9. Create a new directory for your compiled ffmpeg. Copy following files into it:
- C:\msys32\home\<user>\ffmpeg\ffmpeg.exe
- C:\msys32\home\<user>\ffmpeg\ffprobe.exe
- C:\msys32\home\<user>\libav\avconv.exe
- C:\msys32\home\<user>\libav\avprobe.exe
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\nppc64_11.dll
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\nppicc64_11.dll
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\nppidei64_11.dll
- C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.2\bin\nppig64_11.dll
Copy all DLL files from C:\msys32\mingw64\bin\ directory as well.

10. Test the program. In case of errors, find and copy remaining missing DLLs.

Documentation: https://docs.nvidia.com/video-technologies/video-codec-sdk/ffmpeg-with-nvidia-gpu/