Hypre Fork/Branch Release for SC21 paper

This release of the fork and corresponding branch has an optimized algorithm for doing assembly on GPUs. This branch was used for a paper submission to the Supercomputing 21 Conference

On ORNL Summit, this can be built with:

module load spectrum-mpi
module load gcc/7.4.0
module load cuda/10.2.89

export CFLAGS="-g -O2"
export CXXFLAGS="-g -O2"
export HYPRE_CUDA_SM=70
export CUDA_HOME=/sw/summit/cuda/10.2.89

export CXX=$(which mpicxx)
export CC=$(which mpicc)
export FC=$(which mpif90)

GPU Build:
./configure --prefix=/PATH/TO/INSTALL/DIR
--without-superlu --disable-bigint --without-openmp --with-cuda --enable-unified-memory --enable-curand --disable-cublas --enable-shared

CPU Build:
./configure --prefix=/PATH/TO/INSTALL/DIR --without-superlu --disable-bigint --without-openmp --enable-shared

make
make install