OpenMP Parallel Benchmarks

This repository contains a small collection of benchmark programs written in C using OpenMP, focusing on basic parallelization techniques and simple performance evaluation on shared-memory systems.

The project studies how different OpenMP constructs and scheduling policies affect execution time and scalability on a few representative problems.

The implemented benchmarks are based on three main case studies:

• Prime number computation using OpenMP parallel for with reduction, including comparison of different scheduling policies (static, dynamic, guided).
• Gaussian blur image processing, implemented both with loop parallelism and with OpenMP tasks.
• Task dependency example using a producer–consumer style pipeline with OpenMP task dependencies to enforce correct execution order.

Each program includes simple timing measurements in order to compare serial and parallel execution and observe speedup behavior.

---

What is demonstrated

This project demonstrates:

• Basic use of OpenMP directives: parallel, for, reduction, schedule, task, and depend.
• Loop parallelism and runtime scheduling selection.
• Task-based parallelism and explicit task dependency graphs.
• Simple performance measurements and comparison between serial and parallel execution.
• Practical handling of shared data and synchronization in OpenMP programs.

---

Contents

• primes.c
  Parallel computation of prime numbers using parallel for and reduction, with comparison of different scheduling policies.

• gaussian.c
  Gaussian blur image processing using:

  • loop-based parallelization
  • task-based parallelization

• task_dependencies.c
  Example of a pipeline using OpenMP tasks with dependencies (producer–consumer pattern).

• Test images and small input data used by the benchmarks.

---

primes.c (Prime numbers parallelization)

How to run:

Step 1 ->

gcc primes.c -fopenmp

Step 2 ->

./a.out

gaussian.c (Applying gaussian blur to a given image)

How to run:

Step 1 ->

gcc gaussian.c -fopenmp -lm

Step 2 ->

./a.out <blur-radius> <filename>, 

e.g. ./a.out 2 500.bmp

Run with different numbers of threads

• export OMP_NUM_THREADS=1
• export OMP_NUM_THREADS=4

For the benchmarks that support runtime scheduling selection, you can use:

• export OMP_SCHEDULE=static
• export OMP_SCHEDULE=dynamic
• export OMP_SCHEDULE=guided

---

task_dependencies.c (Task dependencies and synchronization problems)

How to run:

Step 1 -> openMP_execution_model.txt needs to be in the same scope as the task_dependencies.c file

Step 2 ->

gcc task_dependencies.c

Step 3 ->

./a.out

👤 Author

GitHub: xrddev

📝 License

Released under the MIT License. Originally built as part of a university project.