Parallel Convolution

This project aims to implement and evaluate Gaussian blur, a fundamental image processing operation, using three approaches: serial, MPI, and hybrid MPI + OpenMP. The goal is to analyze the performance of parallel programming techniques in optimizing computationally intensive tasks. This work is part of the ME766 course project. For detailed analysis, refer to Report.pdf, and the implementation codes can be found in Code Files.

---

Implementation Instructions

To run the code, ensure that the image and code files are in the same folder. The following instructions demonstrate how to compile and run the code for 50 iterations on macOS:

General Note

• For grayscale images, use waterfall_grey_1920_2520.raw and specify grey.
• For RGB images, use waterfall_1920_2520.raw and specify rgb.

1. Serial Implementation

Compile:

/opt/homebrew/bin/gcc-14 serial.c -o serial  

Run:

./serial waterfall_grey_1920_2520.raw 1920 2520 50 grey  

2. MPI Implementation

Compile:

mpicc -o mpi_conv mpi_conv.c  

Run (for 2 processes):

mpirun -np 2 ./mpi_conv waterfall_grey_1920_2520.raw 1920 2520 50 grey

3. MPI + OpenMP Hybrid Implementation

Compile:

/opt/homebrew/bin/gcc-14 -fopenmp -o mpi_omp_conv mpi_omp_conv.c -I/opt/homebrew/include -L/opt/homebrew/lib -lmpi  

Run (for 2 processes):

mpirun -np 2 ./mpi_omp_conv waterfall_grey_1920_2520.raw 1920 2520 50 grey  

Results

The table below shows the performance results for various processes, based on the number of iterations and processing approaches for grayscale images:

N	Serial Time (s)	2 Processes MPI Time (s)	2 Processes Hybrid Time (s)	4 Processes MPI Time (s)	4 Processes Hybrid Time (s)	6 Processes MPI Time (s)	6 Processes Hybrid Time (s)	8 Processes MPI Time (s)	8 Processes Hybrid Time (s)
20	3.034	1.549288	0.564127	0.798311	0.506423	0.534291	0.534828	0.534619	0.615634
50	7.935	3.867279	1.396288	1.992304	1.246592	1.334239	1.329070	1.310875	1.526315
100	16.383	7.727510	2.850884	3.986903	2.469982	2.683687	2.741402	2.607915	3.066253
200	32.291	15.493539	5.617990	7.986798	5.052975	5.319520	5.422499	5.076195	6.218083
500	86.180	38.654856	14.047751	19.915830	12.500185	13.675643	13.515103	12.839587	15.371245

References

1. Mangual, Osvaldo, Teixeira, Marvi, Lopez, Reynaldo, Nevarez, Felix. "Hybrid MPI-OpenMP versus MPI Implementations: A Case Study." 2014. DOI: 10.13140/2.1.1219.1364.

2. Lu, Jin, Zhang, Kaisheng, Chen, Ming, Ma, Ke. "Implementation of parallel convolution based on MPI." Proceedings of 2013 3rd International Conference on Computer Science and Network Technology, 2013, pp. 28-31. DOI: 10.1109/ICCSNT.2013.6967057.

3. Jim Ouris. "Parallel Convolution." 2021. Available at: https://github.com/jimouris/parallel-convolution. Accessed: 2024-11-29.

4. Panagiotis Petropoulakis. "Parallel Convolution." 2021. Available at: https://github.com/PetropoulakisPanagiotis/parallel-convolution. Accessed: 2024-11-29.