PetropoulakisPanagiotis / parallel-convolution

Parallel Convolution using MPI and OpenMP

Parallel Convolution

Parallel convolution performs n convolutions to an image using MPI and OpenMP for parallelism

How It Works

The image is partitioned into square blocks and each process is "mapped" to a specific block. At first, each process convolutes only the inner pixels with a given kernel, and then, each process exchanges with its neighbours its outer pixels to perform a convolution to the remaining pixels. This process is repeated n times.

Notes:

• You can find some plots inside the measurements folder
• Filter size = 3*3
• All processes are called workers. Process 0 sends initially the input image to other workers, and in the end, the same process prints some statistics.
• In plots, the pre and post-processing time are excluded.
• Max neighbours per process = 8
• The input image must be square

Key Features

• Persistent Communication
• Virtual Topology - Cartesian
• Derived Data Types - Vector, Struct, Contiguous
• Non-Blocking Communication
• Dynamic Receiving - MPI_Status, MPI_Waitany, use of MPI_PROC_NULL
• Static Scheduling with OpenMP

Requirements

• MPI-3
• OpenMP

Compilation and Excecution

$ cd parallel-convolution/convolution
$ make
$ mpiexec -n 4 ./convolution

Contributing

• Introduce parallel I/O to read a real input image

Helpful Links:

1. http://matlabtricks.com/post-5/3x3-convolution-kernels-with-online-demo
2. http://setosa.io/ev/image-kernels/
3. https://lodev.org/cgtutor/filtering.html
4. https://www.openmp.org/
5. https://www.mpich.org/static/downloads/3.2.1/mpich-3.2.1-userguide.pdf

Authors

• Petropoulakis Panagiotis [e-mail]
• Andreas Charalambous [e-mail]