GCC 14 Lands Working Support For AMD RDNA2 & RDNA3 GPU Offloading
The GCC 14.1 stable compiler release due out in the coming months will now have initial working support for both AMD RDNA2 (GFX10.3) and RDNA3 (GFX11) graphics processors. This is intended for OpenMP and OpenACC offloading to the GPUs.
This commit brings the RDNA2 and RDNA3 support in order for OpenMP/OpenACC device offloading. Of the consumer discrete RDNA2 graphics cards are the Radeon RX 6000 series while the latest-generation RDNA3 consumer GPUs currently make up the Radeon RX 7000 series.
It's also interesting to note this work is coming out of the consulting firm BayLibre from a recently-formed Compiler Services group. Former Mentor Graphics / Siemens compiler engineers are now working on compiler tech at BayLibre around OpenACC, OpenMP, and device offloading.
AMD for their official OpenMP device offloading is focused on LLVM/Clang where they maintain the "AOMP" downstream compiler. They also officially maintain the upstream AMDGPU LLVM back-end that is not only used for compute but also by the likes of AMDVLK, RadeonSI, and other uses for their shader compiler.