ROCm AOMP 13.0 Released For OpenMP Offloading To Radeon GPUs
AOMP is part of the Radeon Open eCosystem (ROCm) for offloading OpenMP work to AMD GPUs. AOMP is a downstream of the LLVM Clang compiler stack. AMD engineers have been working to upstream their OpenMP offloading patches albeit is a work-in-progress and they churn out new patches faster than what it takes to get reviewed and upstreamed. AOMP is not to be confused with AOCC as AMD's other LLVM Clang downstream focused on providing the latest CPU compiler optimizations/support.
AOMP Release 13.0-2 is their re-base against the current LLVM/Clang 13 upstream development state. LLVM 13.0 stable will be out later in the year while it's been in development for a few months already since the LLVM 12.0 branching.
In addition to moving to the newest LLVM code-base, AOMP 13.0 also has a number of tweaks to its default OpenMP offload behavior, kernel trace improvements, updating the ROCm components against their 4.1.x state, and defaulting to code object v4 format.
The latest AOMP Radeon OpenMP offload compiler release can be downloaded from GitHub with official binaries for Ubuntu / SUSE / RHEL platforms.
5 Comments
