Radeon ROCm 3.7 Release Enables OpenMP 5.0 By Default In AOMP

AMD's software team has released version 3.7 of ROCm, the Radeon Open eCosystem stack as their alternative to NVIDIA's closed-source CUDA compute environment.

The ROCm 3.7 release comes just hours after they released AOMP 11.8 as AMD's downstream of LLVM/Clang focused on providing Radeon OpenMP offloading support until the work is all upstreamed in LLVM/Clang. With AOMP 11.8 they are using the branched LLVM 11.0 code that is nearing its stable release, OPMD updates for the FLANG Fortran compiler, OpenMP debugging improvements, and other changes.

As part of the ROCm 3.7 release, some of the AOMP enhancements include that OpenMP 5.0 support is now enabled by default while the OpenMP 4.5 compliance path remains available and a variety of other code restructuring and improvements.

With ROCm 3.7 their Communications Collective Library has also been updated for compatibility with the NVIDIA NCCL 2.7 API, and a variety of other lower-level improvements for this AMD Radeon open-source GPU compute stack.

More details on ROCm 3.7 via the project's GitHub repository. ROCm 3.7 still doesn't mention any official support for Navi/GFX10 but I have heard that the code should be working albeit unofficially.