AMD AOMP 0.7-7 Released For Radeon OpenMP Offloading

Announced at the end of last year was Radeon Open eCosystem 3.0 with the new "AOMP" compiler. Today a new version of AOMP has been released for OpenMP offloading support to AMD Radeon GPUs.

AOMP is the newest of several downstreams of LLVM/Clang maintained by AMD. AOMP tracks upstream LLVM / Clang but with changes for supporting OpenMP API offloading support to Radeon GPUs as part of the ROCm driver stack. While focused on Radeon OpenMP support, AMD does leave the HIP / CUDA / OpenCL support within the AOMP Clang build.

With today's AOMP 0.7-7 Release, LLVM Clang 9.0 stable is being tracked. New with this release are updates for dealing with Fortran OpenMP offloading via the FLANG front-end, ATMI hostcall optimization, support for omp_get_max_threads(), and updated against the ROCm 3.1 sources.

More details on the AOMP 0.7-7 changes via GitHub. Meanwhile AMD says AOMP 11.0 is under development to succeed the v0.7 series. Presumably AOMP 11.0 will switch to tracking LLVM/Clang 11 development otherwise the versioning scheme is more bizarre, though LLVM Clang 11.0 won't be out as stable until around September.