apallath / parallel_algos

Implementations of high performance shared and distributed memory parallel algorithms using OpenMP and MPI in C [work in progress].

High performance parallel algorithms

Shared and distributed memory parallel algorithms in C. Custom implementations of functions using OpenMP and MPI, as well as demonstrations of library calls.

Status

Requirements

• Intel oneAPI (for Intel compilers, Intel MKL, and Intel MPI.)

Building from source

Customize environment variables in build.sh. At present, this script is designed for Intel oneAPI (Intel compilers/MPI/MKL).

Generate makefiles and build using CMake by running the script:

./build.sh

Contents

Simple OpenMP and MPI demonstration programs (demo_usage)

• Demonstrations of point-to-point and collective MPI operations (demo_usage/mpi_*.c) ✔️
• Parallel pi computation (via integration) in OpenMP and MPI (demo_usage/{}_pi.c) ✔️

Dense Linear Algebra function implementations (dense_algebra)

Using libraries:

• Matrix multiplication using BLAS (dense_algebra/mm_blas.c) ✔️
• Distributed memory parallel matrix multiplication using PBLAS (dense_algebra/mm_pblas.c) 🔨
• LU factorization using LAPACK (dense_algebra/lu_lapack.c) 🔨
• LU factorization using ScaLAPACK (dense_algebra/lu_scalapack.c) 🔨

Custom implementations:

• Matrix multiplication using OpenMP 🔨
• Matrix multiplication using MPI (SUMMA) 🔨

To run strong scaling and weak scaling benchmarks

• Run

./run_all_benchmarks.sh

in the build directory.

A python script to analyze benchmark outputs is in the sample_benchmarks/ directory.

Sample plots generated by the script: