AMD Releases AOMP 17.0-0 For Latest Radeon OpenMP Offloading Compiler

On Thursday AMD engineers released AOMP 17.0-0 as the newest version of their LLVM/Clang downstream compiler that carries their latest development patches around Radeon/Instinct OpenMP GPU/accelerator offloading support.

AOMP continues to be where AMD stages their GPU/accelerator OpenMP-related offloading code until it's ready and reviewed for inclusion into the mainline LLVM/Clang compiler. AOMP also works standalone from ROCm aside from the AMDGPU/AMDKFD kernel drivers and libdrm interface.

With the AOMP 17.0-0 there is the version bump to reflect that it's now re-based against the current LLVM 17.0 Git code. AOMP 17.0-0 is built against the ROCm 5.4 code and has the following AMD-specific changes in this OpenMP compiler update:

- Add support for amdclang, amdclang++, and amdflang
- HIP is better than OpenMP for small array size.
- Updated build scripts for Kokkos (and updated to Kokkos v 3.7.00)
- Support for multiple blocksizes in Xteam reduction (1024 limit).
- A new execution mode BigJumpLoop for SPMD non-reduction kernels
- Additional support for OMPT function "translate_time"
- Added Centos-9 and SLES 15 SP4 rpms.
- No longer support SLES15 SP1.

Source downloads along with the Ubuntu, SUSE Linux Enterprise, and CentOS/RHEL binaries via GitHub for AOMP 17.0-0.