AMD AOMP 11.12 Released For OpenMP Offloading To Radeon GPUs

Last week there was the release of AOCC 2.3 as AMD's LLVM Clang downstream focused on Zen-optimized support. Meanwhile on the graphics side of the house, this week ushered in AOMP 11.12 as their LLVM Clang downstream focused on Radeon OpenMP GPU offloading.

AOMP continues maturing as the company's downstream of LLVM Clang that allows OpenMP offloading to Radeon hardware. AMD has been working to upstream their Radeon OMP patches into LLVM, but at least until that's all perfect, AOMP is continuing to advance. Friday's release of AOMP 11.12 follows the recent ROCm 3.10 for which this compiler is now re-based against those sources. We are also still expecting ROCm 4.0 to be released in the coming days after being announced last month at SC20.

With this AOMP 11.12 release there is integration of ROCtracer and ROCprofiler support, changes for supporting the AMD HPC Toolkit, initial support for GPU malloc and free handling, finally there is an official build for Ubuntu 20.04 LTS, and as mentioned updating against the ROCm 3.10 code-base. There are also various fixes and other additions, like supporting fprintf.

Those wanting to try out AOMP 11.12 can find it via GitHub with a generic x86_64 build along with binaries catering to RHEL/CentOS 7, SUSE Linux Enterprise 15 SP1, and Ubuntu 16.04/18.04 along with the newly-added Ubuntu 20.04 LTS.