AMD Releases AOMP 16.0-1 With Initial Support For RDNA3 "GFX11" GPUs

AMD today published AOMP 16.0-1 as their newest LLVM/Clang downstream focused on providing the latest Radeon OpenMP GPU offloading support. Notable with this AOMP build is providing initial support for GFX1100 - GFX1103 GPUs. The GFX11 IP block is coming with the soon-to-launch RDNA3 graphics cards and with this AOMP support gives us hope AMD will be providing punctual ROCm support for these next-generation graphics cards.

After ROCm support for RDNA and RDNA2 GPUs was very tardy, it's exciting to see with today's AOMP 16.0-1 release that GFX1100 - GFX1103 support is already being mentioned for the unreleased RDNA3 graphics processors. We'll have to wait and see after the 3 November announcement, but this provides some hope that there will be timely ROCm support for these upcoming consumer GPUs. On the LLVM shader compiler side, for months now there has been GFX11 AMDGPU compiler support in the works and initially introduced in LLVM 15.0.

This updated LLVM/Clang 16 downstream has also enhanced xteam reductions, optimized wait for signals to improve performance, support for order (concurrent), OpenMP warning clean-ups, atomic min/max support for Mi200 GPUs, support for device new and delete, and switching over to the ROCm 5.3 sources.

Downloads and more details on the AOMP 16.0-1 Radeon OpenMP compiler release via GitHub.