While there is [1]AOMP for OpenMP device offloading based on the LLVM/Clang compiler , less talked about and not as feature-rich is the AMDGCN back-end within the GNU Compiler Collection (GCC) that is also worked on for OpenMP device offloading capabilities to Radeon GPUs. Squeezing in for the upcoming GCC 14.1 stable release is GFX1103 support for AMD APUs with RDNA3 integrated graphics.

Andrew Stubbs with Bay Libre who has been [2]working on the RDNA2 and RDNA3 support for GCC offloading merged the support today to GCC Git for handling GFX1103, the graphics IP found within the AMD RDNA3 "Phoenix" SoCs.

Stubbs wrote on [3]the commit adding the GFX1103 support:

"Add support for the gfx1103 RDNA3 APU integrated graphics devices. The ROCm documentation warns that these may not be supported, but it seems to work at least partially."

The GCC AMDGCN support isn't as nearly mature as the LLVM AMDGPU back-end and the support found in the likes of AMD's officially supported AOMP compiler, but at least Bay Libre developers and others continue doing a nice job working to enhance this GCC device support.



[1] https://www.phoronix.com/news/AOMP-19.0-0-AMD-OpenMP

[2] https://www.phoronix.com/news/GCC-14-RDNA2-RDNA3-Offload

[3] https://gcc.gnu.org/git/?p=gcc.git;a=commitdiff;h=1bf18629c54adf4893c8db5227a36e1952ee69a3