= ispm-sparse-lib

CUDA sparse matrix-vector multiplication for a custom matrix format based on Sliced ELLPACK and a few associated routines for building a GPU iterative linear solver. The associated foam-extend solver code is currently provided at https://github.com/amonakov/openfoam-extend-foam-extend-3.1, branch feature/ispmCudaSolvers. See the respective README and source.

== Building

make -j4 CUDA_ARCH=20
export ISPM_DIR=`pwd`

The CUDA compiler nvcc must be in PATH.

CUDA_ARCH denotes the target CUDA compute capability. See this link for the list. The above example is for Tesla M20xx, for Tesla K10 specify CUDA_ARCH=30 and for Tesla K20 and K40 — CUDA_ARCH=35 (the default).

After building this library proceed to build the foam-extend solver with ./Allwmake in its source subdirectory. The ISPM_DIR environment variable should point to the checkout of this library, as above.