Advertisement
Published in:
2021 | OriginalPaper | Chapter
18. GPGPUs with OpenMP
Author : Tobias Weinzierl
Published in: Principles of Parallel Scientific Computing
Publisher: Springer International Publishing
Abstract
GPGPUs can be read as multicore chips within the compute node which host their own, separate memory and multiple cores which are slow yet nevertheless extremely powerful. This power results from SIMD on a massive scale. We give a brief, very high level overview of the GPU architecture paradigm, before we introduce OpenMP’s target blocks. The data movement between host and accelerator requires dedicated attention, while a brief discussion of collectives, function offloading and structs (classes) for GPUs close the discussion.
