CPU and GPU build trees

Keep a separate build directory for each backend. CUDA and HIP select different translation units, compilers and dependency libraries; changing flags inside one configured tree can leave incompatible cached paths.

Use scripts/build.sh from the repository root. It loads the matching modules, selects the toolchain, builds and runs tests. The following Tohtori commands use explicit directories so switching backends cannot reuse another backend’s cache:

./scripts/build.sh --machine=tohtori --cpu --build-dir=builds/cpu
./scripts/build.sh --machine=tohtori --with-cuda --build-dir=builds/cuda

On LUMI the build directory belongs on flash. Configure runs on the login node; compilation and tests run in a scheduled GPU allocation:

./scripts/build.sh --machine=lumi --with-rocm --partition=standard-g --wait

Variant	Main options selected by the script	Dependency
CPU spectral	CUDA and HIP off; HeFFTe on	CPU HeFFTe with FFTW
CUDA spectral	OpenPFC_ENABLE_CUDA=ON	CUDA toolkit and CUDA-enabled HeFFTe
HIP spectral	OpenPFC_ENABLE_HIP=ON	ROCm and ROCm-enabled HeFFTe
CPU finite differences only	--cpu --no-heffte	MPI; spectral targets are omitted

HeFFTe is installed outside the OpenPFC clone. To select a site-specific install, set HEFFTE_PREFIX or HEFFTE_MODULE before invoking the script. Run ./scripts/build.sh --help for supported overrides, architecture choices and machine defaults.

Use the installation guide for dependency setup, the LUMI guide for scheduled HIP builds and storage, and the GPU application tutorial for launching the resulting executables. A build directory selects compiled capabilities; a JSON backend value cannot enable a backend absent from that build.