2D Laplace Equation Solver using MPI and CUDA Hybrid Programming

Overview

This project implements an iterative solver for the two-dimensional Laplace equation using a hybrid MPI and CUDA programming approach. It supports multi-GPU computations, allowing for efficient parallel processing across multiple GPUs.

Features

• Iterative solution of the 2D Laplace equation.
• Hybrid MPI and CUDA programming for both intra-node GPU acceleration and inter-node communication.
• Support for multi-GPU configurations.
• Choice between Jacobi and Gauss-Seidel iterative methods.

Compilation

To compile the program, follow these steps:

1. Ensure you have CMake installed.
2. Run the following commands in your terminal:

cmake -B build
cmake --build build

Running the Program

The program requires six input parameters to run. Execute the program as follows:

mpirun -np <num_processes> ./your_executable <param1> <param2> <param3> <param4> <param5> <param6>

parameters

1. Grid Points in X Direction (<param1>): Number of grid points along the x-axis.
2. Grid Points in Y Direction (<param2>): Number of grid points along the y-axis.
3. Partitioning in X Direction (<param3>): Number of partitions along the x-axis.
4. Partitioning in Y Direction (<param4>): Number of partitions along the y-axis.
5. Maximum Iterations (<param5>): Maximum number of iterations allowed for the solver.
6. Iterative Method (<param6>): -Input 0 to select the Jacobi iterative method. -Input 1 to select the Gauss-Seidel iterative method. Note: Ensure that the product of <param3> (x-direction partitioning) and <param4> (y-direction partitioning) is less than the actual number of processes (<num_processes>).

Example usage

Assuming you want to run the program with 4 processes, 100 grid points in each direction, 2 partitions in the x-direction, 2 partitions in the y-direction, a maximum of 1000 iterations, and using the Jacobi method:

mpirun -np 4 ./your_executable 100 100 2 2 1000 0

Reference

https://github.com/jcorbino/laplace