C3PO’26 Program

How can an ancient language like Fortran drive GPU-based accelerators in a modern supercomputer? The short answer is with a modern compiler (LLVM Flang) and the OpenMP Application Programming Interface! In a GPU-accelerated world, the OpenMP API is one of the key components to support application developers and their need to write portable and high-performance code for heterogeneous systems, especially in the context of (large) Fortran codes. In this talk, we will recap the evolution of the OpenMP API from it's early days in 1997 to the present day and how it has evolved from portable multi-threading for loops (in Fortran) into a directive-based programming model for heterogeneous systems. We will show how LLVM Flang internally handles OpenMP directives for code generation and how OpenMP directives, but also Fortran DO CONCURRENT, are lowered. Finally, we will show the OpenMP API's new features for grid-style programming on GPUs.

-

Michael Klemm is a Principal Member of Technical Staff in the Compilers, Languages, Runtimes & Tools team of the Machine Learning & Software Engineering group at AMD. He is the lead architect of the AMD Fortran Compiler for AMD GPUs. Michael also is the Chief Executive Officer of the OpenMP Architecture Review Board and a lecturer at the Chair of Computer Architecture and Parallel Systems at Technical University Munich. He holds an M.Sc. in Computer Science and a Doctor of Engineering degree (Dr.-Ing.) in Computer Science from the Friedrich-Alexander-University Erlangen-Nuremberg, Germany.