Building from Source¶

macOS (Apple Silicon — recommended for development)¶

cmake -B build \
    -DMETAL_COMPUTE=ON \
    -DBUILD_FORGEVIEW=ON \
    -DMPISupport=OFF
cmake --build build -j$(sysctl -n hw.logicalcpu)

Linux (NVIDIA GPU via NVIDIA HPC SDK)¶

cmake -B build \
    -DCMAKE_CXX_COMPILER=nvc++ \
    -DOPENACC_GPU=ON \
    -DMPISupport=ON
cmake --build build -j$(nproc)

Linux (CPU only)¶

cmake -B build -DOPENACC_GPU=OFF -DMPISupport=OFF
cmake --build build -j$(nproc)

CMake Options¶

Option	Default	Effect
`-DMETAL_COMPUTE=ON`	ON (macOS only)	Apple Metal GPU backend
`-DOPENACC_GPU=ON`	ON	NVIDIA CUDA/OpenACC GPU backend
`-DOPENACC_MP=ON`	OFF	OpenACC multi-threading on CPU (PGI only)
`-DMPISupport=ON`	ON	MPI-distributed executables
`-DBUILD_FORGEVIEW=ON`	ON	forgeview TUI monitor
`-DENABLE_DOUBLE_PRECISION=ON`	OFF	Double precision throughout

Docker (Linux GPU)¶

docker pull mechneurolab/forge
docker run --gpus all -it mechneurolab/forge

Build Your Own Image¶

# Base image (NVIDIA HPC SDK + dependencies)
docker build -t forge-hpcsdk docker/forge-hpcsdk/

# Runtime image (clones and builds forge)
docker build -t forge docker/forge/

Testing¶

# CPU tests
./build/cpu_tests

# Metal GPU tests (macOS)
./build/metal_tests