parallel-interior-point-solver

A parallel implementation for Karmarkar's interior point solver. This work has been presented at the 2024 edition of PDP (Parallel, Distributed and network based Processing)

Interior point solver

A parallel implementation of Karmarkars's interior point method algorithm using OpenMP and CUDA. Developed by Marco Edoardo Santimaria, Samuele Fonio and Giulio Malenza

Usage

To compile the project you will need the following requirements:

• g++ >= 10
• Make
• (optional) CUDA if you plan to use NVIDIA GPU support

Compile target

The Makefile has the following targets:

• cuda: build the project to run on cuda devices
• openmp: build the project to run on CPUs with OpenMP framework. This is also the target to run the single thread version of the software
• makeMatrix: This target uses python3 to generate random matrices to perform benchmarks. To use this target, you need to run the following command: DIAGSIZE=<value> CUT=<value> make makeMatrix, where DIAGSIZE is the number of columns and CUT is the number of rows By default, Makefile will compile for bot cuda and openmp targets.

Main binary usage:

Once compiled, inside build/ there will be the main executables:

• if compiled with cuda target, the executable will be build/Karmarkar.cuda
• if compiled with openmp target, the executable will be build/Karmarkar.openmp

Both the executable have a common interface and the executable shall be run in the following way:

build/Karmarkar.<target> [A matrix filename] [B vector filename] [A matrix Nrow] [A matrix Ncol] [maxIter] [tolerance]