AMD has recently announced the launch of a new project called Peano, an open LLVM compiler designed specifically for AMD/Xilinx processors. This project is specifically focused on the Ryzen AI SoC, which includes Phoenix and Hawk Point equipment, as well as future XDNA2 solutions for the Ryzen AI 300 series.
Earlier this year, in January, AMD unveiled an open XDNA driver to provide support for Ryzen AI on Linux. Although the driver has not yet been integrated into the main core, rumors suggest that this integration may happen in the near future.
The announcement of PEANO was made by Stephen Nyenderfer from AMD/Xilinx. He elaborated that the compiler is built to support the AIE2 architecture, as implemented in XDNA accelerators on Phoenix and Hawk Point.
The key features of the PEANO compiler include support for llvm code generation, as well as Clang, LLD, Binutils, Compiler-RT, and LLVM-Libc tools. The AI darts targeted by this compiler are VLIW processors with an exposed conveyor, allowing the compiler to plan instructions that can be stacked on top of each other.
Currently, the PEANO project is hosted on GitHub in the branch xilinx/llvm/aie, and the support for XDNA2 in the upcoming Ryzen AI 300 “Strix Point” equipment is still under development.
The introduction of the open PEANO compiler is set to enhance the functionality of Ryzen AI accelerators for Linux and facilitate the integration of the XDNA driver into the Linux kernel. This naming convention of the compiler is believed to be inspired by the Italian mathematician Giuseppe Peano, mirroring the practice of NVIDIA naming its projects after scientists.
Overall, the PEANO project marks a significant milestone for AMD, as it will help expedite the growth of the Linux ecosystem tailored for Ryzen AI processors.