LLVM Lands Support For OpenMP Offloading To AMD Radeon GPUs
While the AMDGPU back-end has been part of mainline LLVM for years, not until now has it supported an OpenMP toolchain for offloading to Radeon GPUs.
The AMD work for allowing an OpenMP offloading toolchain for the AMDGPU back-end has been merged this week. Ultimately this is for allowing OpenMP code to be offloaded to Radeon GPUs for compute,
AMD has been maintaining AOMP as their downstream of LLVM Clang for supporting OpenMP offloading to Radeon hardware. But exciting is this merge request after one month of review now being mainlined for allowing OpenMP offloading toolchain for the AMDGPU back-end.
At the moment it looks like AOMP still has some patches not yet upstreamed into LLVM but great seeing definite progress in this area. We'll see though if AOMP is still going to be maintained moving forward for new/premature code for AMD support prior to it being mature enough for upstreaming or if AOMP will ultimately be phased out from their ROCm compute stack. Via the AOMP repository is sample code and more.
Similarly, since GCC 10 has been OpenMP and OpenACC offloading to AMD Radeon GPUs. However, in the case of the GNU Compiler Collection its support and compatible GPUs is not nearly as well off as AMDGPU LLVM.