Highly distributed solution of sparse linear equations

 SuperLU is a general purpose library for the direct solution of large,
 sparse, nonsymmetric systems of linear equations. The library is
 written in C and is callable from either C or Fortran program. It uses
 MPI, OpenMP and CUDA to support various forms of parallelism. It
 supports both real and complex datatypes, both single and double
 precision, and 64-bit integer indexing. The library routines performs
 an LU decomposition with partial pivoting and triangular system solves
 through forward and back substitution. The LU factorization routines
 can handle non-square matrices but the triangular solves are performed
 only for square matrices. The matrix columns may be preordered (before
 factorization) either through library or user supplied routines. This
 preordering for sparsity is completely separate from the
 factorization. Working precision iterative refinement subroutines are
 provided for improved backward stability. Routines are also provided
 to equilibrate the system, estimate the condition number, calculate
 the relative backward error, and estimate error bounds for the refined
 solutions.
 .
 SuperLU_DIST implements the algorithms for distributed memory,
 targetting highly parallel distributed memory hybrid systems. The
 numerical factorization routines are already implemented for hybrid
 systems with multiple GPUs. Further work will be needed to implement
 the other phases of the algorithms on the hybrid systems and to
 enhance strong scaling to extreme scale.
 .
 The main library is libsuperlu_dist.so but a fortran wrapper library
 is also provided as libsuperlu_dist_fortran.so
 .
 This package provides development files for building client
 applications against superlu-dist.