Parallel-Computing

Reproducing IntroPARCO_2024_H1

This project implements various optimizations for sequential and parallel algorithms using OpenMP and AVX2.

Prerequisites

Before starting, ensure you have the following tools and libraries installed in your environment:

Required Tools:

• g++ (preferably version 9.3 or later)
• WSL (Windows Subsystem for Linux) if you're working on a Windows system.
• Make (optional, if you want to automate the compilation using a Makefile).

Required Libraries:

• omp.h (included with a compiler that supports OpenMP)
• immintrin.h (included with the compiler for AVX2 instructions)

To verify the availability of the required libraries, compile and run a simple test program:

#include <omp.h>
#include <immintrin.h>
#include <stdio.h>

int main() {
    printf("OpenMP and AVX2 are available!\n");
    return 0;
}

Compile the test program using:

g++ -fopenmp -mavx2 test.c -o test

Compilation Instructions

To compile the project, use the following command:

g++ -O3 -mavx2 -march=native -fopenmp -ffast-math -funroll-loops introPARCO_2024_H1.c all_optimization.h sequential_implementation.c implicit_parallelization.c explicit_parallelization.c -o introPARCO_2024

Explanation of the Flags:

• -O3 : Enables high-level optimizations.

• -mavx2: Enables AVX2 instructions.

• -march=native: Utilizes features specific to your CPU.

• -fopenmp: Enables OpenMP support.

• -ffast-math: Improves the speed of math operations (may slightly reduce precision).

• -funroll-loops: Expands loops for performance improvement.

This will create an executable named introPARCO_2024.

Running the Program

To run the program, use the following command:

Running the Program To run the program, use the following command:

./introPARCO_2024

Code Structure

The project consists of the following files:

• introPARCO_2024_H1.c: The main entry point of the program.
• all_optimization.h: Header file containing common definitions and macros.
• sequential_implementation.c: Implementation of the sequential version of the algorithms.
• implicit_parallelization.c: Parallel implementation using an implicit approach (e.g., OpenMP).
• explicit_parallelization.c: Parallel implementation using explicit AVX2 intrinsics.

Common Issues

1. Error: #include <omp.h> not found

• Ensure you are using a compiler that supports OpenMP (e.g., g++). -On Ubuntu, you can install it with:

sudo apt update
sudo apt install g++

2. Error:#include <immintrin.h>not found

• Ensure your compiler supports AVX2 instructions. Check your CPU support using:

lscpu | grep avx2

If AVX2 is not listed, your hardware does not support these instructions. 3. Slow performance

• Ensure you are using the -march=native flag and that your system supports AVX2.

Example of output

Here is an example of the output generated by the program:

EVALUATING THE PERFORMANCES AND MEASURING TIME TAKEN FOR MATRIX TRANSPOSITION

Theoretical peak performance: 172.031997681 FLOP/s
Theoretical peak memory bandwidth: 102400 MB/s

B = Bandwidth MB
etc...........


COMPUTING SPEEDUP AND EFFICIENCY GAINS FOR THE OPENMP IMPLEMENTATION

S = Speedup    E = Efficiency
etc...........