Parallel Implementation of Gray Level Co-occurrence Matrix Construction using MPI

ICS632 Final Project Submission

Aditi Jaiswal, Arianna Bunnell, and Sorapong Khongnawang

Docker Container

SMPI Simulations were run via a Docker container. Instructions on how to install and run the SMPI via Docker can be found at https://simgrid.github.io/SMPI_CourseWare/topic_getting_started/installing_simgrid/

Instructions to Run

The program can be run either on a physical cluster with MPI or using SMPI. To use MPI, replace the SMPI commands below with the MPI equivalent. To run with SMPI, first verify that you are in the Docker container which contains your installation of SMPI.

1. smpicc main.c haralick_imp.c haralick_imp.h stb_image.h misc.c misc.h -o <name> -lm -Ofast
2. smpirun -np <X> -hostfile ./config/hostfile_64.txt -platform ./config/cluster_crossbar_64.xml ./<name> ./data/<image file> <angle> -d <distance> -i <implementation>

-d and -i are optional parameters. The default value for distance is 1, and any integer up to the size of the image can be provided. The default value for implementation is row-based, but sequential and tiling are also valid parameter values.

File Structure

• config contains files needed to run the code using SMPI
  • calibrate_flops.py contains code needed to calibrate the SMPI simulation, taken from https://simgrid.github.io/SMPI_CourseWare/topic_understanding_performance/matrixmultiplication/
  • cluster_crossbar_64.txt contains code to simulate a 100-processor cluster
  • hostfile_64.txt contains code to simulate 100 hosts
• data contains examples of random png image files which were used to obtain results in the writeup. Images were generated with a simple Python script using Pillow and NumPy
• docscontains relevant documentation for the project
  • reference contains papers which served as the inspiration for the project
  • The base folder contains the LaTeX code to create the final writeup