The Case for OpenMP*: Why ISO Fortran Is Not Enough for Heterogeneous Parallelism

Even though GPUs are the dominant paradigm for data-parallel computations, software outlives hardware. Thus, standard approaches are needed to increase the longevity of heterogeneous software, regardless of hardware target.

OpenMP* is such an approach.

This session focuses on how industry-standard OpenMP provides a rich set of directives to address the limitations of Fortran, a mature and modern programming language that, nonetheless, evolves slowly.

Topics covered:

• Differences between Fortran and OpenMP in regard to supporting heterogeneous parallelism
• Using detailed code examples to illustrate when ISO Fortran is sufficient and when Fortran+OpenMP is a better alternative to achieve heterogeneous acceleration
• Using OpenMP to explicitly control data transfer between disjoint memories to overcome the control issues of host-device data transfer using Fortran DO CONCURRENT loops to offload computations to an accelerator

Skill level: Intermediate

Download the following stand-alone product or as part of the Intel® oneAPI Base Toolkit:

• Intel® Fortran Compiler

• Fortran-OpenMP Matrix Multiply Sample
• ISO3DfD OpenMP Offload
• See All Code Samples