C++ is (nearly) all you need for HPC

NHR PerfLab Seminar

Date: 15 July 2025 at 00:00

Location: Online

To write a HPC code, we use a serial programming language and add in the parallelism with an API, compiler directives, or frameworks, such as MPI, OpenMP, etc. But since the release of C++17, there are built-in parallel patterns that can easily be used for shared-memory and accelerated parallel programming. Since then, C++ has grown to include more features relevant to the types of programs we write for HPC. In this talk, I will highlight the features added in recent C++ standards, including standard parallel algorithms, mdspan, linalg, and execution, and how my group has explored these for HPC applications. Finally, I will describe the limitations of C++ today, and discuss a future for standards-based programming.