An open-source (BSD) library for sparse linear algebra and associated algorithms. It is primarily developed by the Numerical Analysis group at STFC Rutherford Appleton Laboratory ([email protected]).

For detailed information about the SPRAL packages and API see Fortran documentation or C documentation.

• LSMR - Solves sparse least squares problems using LSMR algorithm.
• RANDOM - Pseudo-random number generator.
• RANDOM_MATRIX - Generates random matrices for testing purposes.
• RUTHERFORD_BOEING - Read and write matrices in Rutherford-Boeing format.
• SCALING - Calculates matrix scalings through a variety of algorithms
• SSIDS - Sparse Symmetric Indefinite Direct Solver.
• SSMFE - Sparse Symmetric Matrix-Free Eigensolver. Uses Jacobi-conjugate preconditioned gradients method.

If the functionality you are looking for is not supported, it may be offered by our proprietary licenced HSL Library (free to academics).

Please note that we require METIS and hwloc to be installed (hwloc should be compiled with CUDA support if building for GPU).

We use a standard autotools-based build:

./autogen.sh # If compiling from fresh git checkout
mkdir build
cp nvcc_arch_sm.c build/ # If building for GPU
cd build
../configure --with-metis="-L/path/to/metis -lmetis"
make
make install

When using SSIDS, ensure the following environment variables are set:

export OMP_CANCELLATION=TRUE
export OMP_NESTED=TRUE
export OMP_PROC_BIND=TRUE