Installation#

The installation is performed in two essentially independent steps. The SAFIRE executable must be compiled, and the safiretools python package can optionally be installed.

Dependencies#

Ensure the following dependencies are installed

  • CMake ≥3.25

  • C++20 compiler

  • MPI

  • HDF5

  • Boost ≥1.70.0

  • BLAS, LAPACK

  • Intel oneAPI MKL (for sparse matrices on CPU)

The build system will fetch the following dependencies automatically if they are not installed

  • nda

  • cxxopts

  • spdlog

  • cpptrace

  • nlohmann_json

  • Catch2 (for tests)

For NVIDIA GPU support, you additionally need

  • CUDA ≥12, compute capability ≥8.0

  • cuTENSOR

  • CCCL (fetched automatically)

Getting SAFIRE#

You can download SAFIRE from GitHub, either through the latest release tarball or

git clone https://github.com/SFQMC/SAFIRE.git

Compiling SAFIRE#

Assuming that all dependencies are installed and available, go to the SAFIRE source directory and run

$ cmake -S . -B build \
        -DCMAKE_INSTALL_PREFIX=$(pwd)/build/install
$ cmake --build build -j 10
$ cmake --install build

This installs the safire executable to build/install/bin. To compile with NVIDIA GPU support use the following:

$ cmake -S . -B build \
        -DCMAKE_INSTALL_PREFIX=$(pwd)/build/install \
        -DENABLE_CUDA=ON
$ cmake --build build -j 10
$ cmake --install build

If CMake does not find cuTENSOR, point it to the unpacked archive with -DCutensor_ROOT=/path/to/libcutensor.

The GPU build can run both CPU and GPU calculations using the --compute CLI argument.

For Flatiron Users

The following snippet will compile SAFIRE in the Rusty environment. Ensure these modules are also loaded whenever you are running SAFIRE.

module purge
module load slurm cmake gcc hdf5 boost intel-oneapi-mkl
module load openmpi            # CPU-only build, or
module load cuda openmpi/cuda  # GPU build

cmake -S . -B build \
      -DCMAKE_INSTALL_PREFIX=$(pwd)/build/install \
      -DENABLE_CUDA=ON -DCutensor_ROOT=... # GPU build only
cmake --build build -j 10
cmake --install build

Installing Companion Software#

It is highly recommended you install safiretools, the companion Python package that simplifies the generation of input files and the processing of output data.

cd python
pip install ".[vis]" # [vis]ualization support

For lattice model calculations, the AutoHF Hartree-Fock code is very useful.

pip install sfqmc-autohf

For ab-initio solids calculations, see CoQuí for installation instructions.

For chemistry, SAFIRE can interface with various quantum chemistry codes. If you want to use the PySCF backend, install

pip install pyscf

Next Steps#

You should now be ready for your first calculation. Continue to the Tutorials.