AMD Releases ROCm AOMP 14.0 Compiler - Switches To New "amd-stg-open" Branch

AMD released AOMP 14.0 during SC21 week as the newest version of their LLVM/Clang-based compiler providing OpenMP GPU offload support for Radeon graphics processors.

AOMP 14.0 was released this morning as the newest version of this patched-up version of LLVM/Clang that gets OpenMP GPU offload into good shape with Radeon GPUs and AMD Instinct accelerators.

With AOMP 14.0, they have switched to using a new code branch for their LLVM/Clang base. They have switched over to using "amd-stg-open" as a new staging area for code from AMD that they are making publicly available but not yet ready (or otherwise not appropriate/applicable) for upstreaming into the LLVM monorepo. They explain of amd-stg-open in today's announcement, "The amd-stg-open branch is constantly changing as AMD merges upstream development trunk with its internal open development efforts. Some AMD modifications are experimental and/or under review for the LLVM mono-repo. The AOMP release is a snapshot of amd-stg-open and supporting repositories to build various components."

The amd-stg-open branch itself for those curious about AMD's LLVM modifications can be found via this GitHub repository separate from the AOMP repository.

In addition to AOMP 14.0 shifting to use this branch and incorporating code from upstream LLVM/Clang 14.0 Git, the new release pulls in Radeon ROCm 4.5 components, a new HIP-based build method is used, there is support for unified shared memory with Vega GFX90A hardware, support for the atomic hint clause added to support fast floating-point atomics, and a variety of other improvements.

Those wanting to try out this AOMP 14.0 Radeon OpenMP offloading compiler can find the sources as well as CentOS/RHEL, Ubuntu, and SLES 15 binaries via GitHub.