Yesterday along with releasing [1]ROCm 7.2 there was also the release of AOMP 22.0-2 as the newest version of their open-source downstream of LLVM/Clang/Flang that is focused on offering the best OpenMP/OpenACC offloading support to Instinct/Radeon hardware.

AOMP continues to be where AMD compiler engineers stage their latest GPU/accelerator offloading improvements before they are able to get the code upstreamed in LLVM proper. AOMP 22.0-2 is re-based against the latest LLVM trunk code while also introducing some new functionality on AMD's side.

While ROCm 7.2.0 released yesterday, with AOMP 22.0-2 they aren't quite aligned but is their release that re-bases against the ROCm 7.1.1 source code.

Besides re-basing against ROCm 7.1 and pulling in newer LLVM 22 upstream code, the rest of the AMD work in this release is focused on the Flang front-end for Fortran programming language support.

AOMP 22.0-2's Flang compiler front-end adds standalone tile support, enables no-loop kernels, support for -fno-fast-real-mod, improved support for split distribute and parallel, and various other fixes and improvements.

Downloads and more details on the AOMP 22.0-2 compiler release via [2]GitHub .



[1] https://www.phoronix.com/news/AMD-ROCm-7.2-Released

[2] https://github.com/ROCm/aomp/releases/tag/rel_22.0-2