TensorRT plugin and CUDA kernels for whIRLwind inference pipelines. The build produces a shared plugin library:
build/libww_trt_plugins.so- Linux with TensorRT headers/libraries installed under the distro paths (
/usr/include/x86_64-linux-gnuand/usr/lib/x86_64-linux-gnu, or theiraarch64equivalents on Orin) - CUDA toolkit available at
/usr/local/cuda - CMake and a C++17 compiler
pixi run buildThe Pixi task intentionally uses the host compiler and CUDA toolkit (/usr/bin/c++
and /usr/local/cuda/bin/nvcc) so it matches the manual CMake build.