Density-equalizing maps, also known as cartograms, are an effective communication tool to represent economic, political or social differences across a geographical region. While most algorithms focus on speed rather than topological accuracy, the diffusion-based method is one of the few ways that guarantees topology preservation and yields a topological exact representation. Nevertheless, to avoid the high computational cost, inexact methods are often used instead, which can result in topological distortions as they don't algorithmically guarantee preservation. Particularly for high-resolution images, the computational cost of the diffusion approach outweighs the benefits in accuracy. In our CS205 project, we address this problem with a software package which is easily deployed from Github on any multicore/multinode cluster and or laptop. From an input image file, the user can obtain a density equalized map with significant speedup. In our tutorial, we illustrate this for the 2016 US election map. On our GitHub website, we show the package scales linearly up to 2 nodes with 8 threads each, which is the size limitation of our cluster.
The libpng package for I/O can be installed on Linux as:
sudo apt-get install libpng-devand installed on Mac OSX as:
brew install libpngThe code can be compiled using the command:
makeThis will generate the serial, the MPI and the hybrid executables. The hybrid executable run for the US county image as:
threads=4
task=2
export OMP_NUM_THREADS=$threads;
mpirun -np $tasks -genv OMP_NUM_THREADS $threads ./exec.openmp "uscounties10.png" "col_counties.txt" "counties.txt" "output.png"The files openmp.sh and mpi.sh generate the MPI/OpenMP speedup diagram on Odyssey with the command:
sbatch openmp.shThe standard output will be mpi.out and openmp.out respectively. If you want to run a different data set beside US counties you need change the SIZE on top of the main file dem_openmp.c.
The GPU-friendly cuda code is located in the GPU. The system requirements are given in the CS205 Team7 PDF report submitted on Tuesday, May 7, 2019. The code can be compiled on a GPU instance using the linux command:
makeAnd can be run with :
./exe_gpu.xThe name of the 3 input files are located in the dem_mpi_gpu.c : uscounties10.png, den_per_county.txt,colchart_counties.txt. For this configuration, the SIZE variable needs to be equal to 3142. If one wants to change the input files, the new files need to be located locally in this folder.
This folder can be cleaned using the following linux command:
make cleanThe code for the two functions write_png_file() and read_png_file() were scavenged from another Github site.