csc-training/advancedopenmp

material for the advanced topics in openMP and OpenMP offloading course

Advanced OpenMP - tasks and GPU offloading

Material for the CSC course "Advanced OpenMP - tasks and GPU offloading". The most recent version of the course is found in the main branch, previous versions can be found in Releases/Tags.

Learning outcome

• Awareness of task based parallelism
• Ability to implement task parallel algorithms with OpenMP
• Ability to utilize GPUs efficiently with OpenMP offloading

Prerequisites

• Good knowledge of C/C++ or Fortran
• Basic knowledge of OpenMP for CPUs

GPU programming I-III: exercises

Intro to GPUs in HPC

• Device query

Basics of OpenMP offloading

• Hello world
• Vector addition
• Simple offloading for Laplace equation

Controlling data movement

• Dot product
• Sum and dot product
• Offloading the heat equation

Unstructured data

• Unstructured data in heat equation

Device functions

• Device functions

Interoperability with libraries

• Using HIP library

Asynchronous operations

• Asynchronous operations

Using multiple GPUs

• Multi-GPU parallelization for heat equation