A three dimensional Eikonal equation solver using the parallel Fast Sweeping Method (FSM) that computes the distance field of the given domain. FSM is an iterative algorithm that uses upwind difference scheme for discretization and Gauss-Seidel iterations with alternating sweeping orderings to solve the discretized system.
Here, the two different versions of the parallel FSM are implemented.
-
Shared Memory: Single GPU (CUDA)
- NVIDIA Graphics Processing Unit (GPU)
- GCC (GNU C Compiler)
- NVCC (Nvidia CUDA Compiler)
- Make
-
Hybrid Memory: MPI/OpenACC
- Graphics Processing Unit (GPU)
- C/OpenACC compiler from PGI Group (pgcc)
- MPI library from Open MPI Project
- NetCDF4 | HDF5 | SZIP
- Make
Execute the make command from within the folder. Change the location of the directories of the libraries accordingly in the Makefile to correctly build the program.
make
Change the log level, default log level INFO.
make log=[NO|ERROR|INFO|DEBUG]
Once you compile the code the binary executable is created within a folder called bin
in the same directory. By default the executable is named as PDFS
.
-
Single GPU CUDA
Usage: ./bin/PDFS <filename.vti> <outputPrefix> filename.vti: VTI input file for the initial distance field outputPrefix: Prefix string to be added to the output file
-
MPI/OpenACC
Usage: mpirun -n <nproc> ./bin/PDFS -i <filename.nc> -p <outputPrefix> [--nx <val>] [--ny <val>] [--nz <val>] -i: input: filename.nc: NetCDF4 input file for the initial distance field -p: prefix: outputPrefix: Prefix string to be added to the output file --nx: Decomposition in x (Optional, default 1) --ny: Decomposition in y (Optional, default 1) --nz: Decomposition in z (Optional, default 1) nproc: (nx * ny * nz)