Compiling FFmpeg With CUDA Acceleration NVCC

Problem

When running ./configure for ffmpeg with --enable-nvcc the following error is produced.

My system is a Razor Blade 14 with AMD Ryzen 9 5900HX with Radeon Graphics and NVIDIA GeForce RTX 3070. 32G of RAM and a 1TB NVMe drive.

ERROR: failed checking for nvcc.

Solution

Export the NVCC_APPEND_FLAGS variable before running ./complile to enforce a more modern C standard.

export NVCC_APPEND_FLAGS+='-std=c++14'
./configure --enable-nonfree --enable-cuda-nvcc --enable-libnpp --extra-cflags="-I/opt/cuda/include -march=native" --extra-ldflags=-L/opt/cuda/lib64 --disable-static --enable-shared --enable-gpl --enable-libx264 --enable-libx265 --prefix=/opt/ffmpeg-custom
make -j$(nproc)
sudo make install

Details

I have compiled and installed this custom ffmpeg binary alongside of the one provided by my package manager, hence the --prefix= above. The custom binary can be run with:

LD_LIBRARY_PATH=/opt/ffmpeg-custom/lib /opt/ffmpeg-custom/bin/ffmpeg -hwaccel cuda -hwaccel_output_format cuda -vsync 0 -i in.mp4 -c:v h264_nvenc -preset slow -b:v 10M -y out.mp4

References

• docs.nvidia.com