Weft depends on gRPC/Protobuf and builds it from source via FetchContent if CMake can't find the config.
The frontend additionally depends on clang (also included via FetchContent - LLVM is not installed on the HPC cluster). The LLVM build is configured to only build clang, but this still takes a while, which you can minimize by only calling the Weft build targets:
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make server weft
Tests are taken from the CUDA samples installation. The CUDA_PATH in Makefiles may need to be updated for each system. The Weft interposer can then be injected via LD_PRELOAD:
LD_PRELOAD=/path/to/libweft.so ./runSample