AMD Unveils “Peano”: Open-Source Compiler for Ryzen AI Development

Recently, AMD announced a new project named “Peano,” an open-source LLVM backend for AMD/Xilinx AI Engine processor development, focusing on the existing Phoenix and Hawk Point Ryzen AI, as well as the forthcoming Ryzen AI 300 series incorporating XDNA2.

According to Phoronix, Peano currently prioritizes supporting the AIE2 architecture implemented by the XDNA accelerators in Phoenix and Hawk Point. It will later extend support to the XDNA2 NPU in Strix Point, though this is still under development.

Typically, AI Engines are ordered, openly pipelined VLIW processors. Each VLIW instruction bundle specifies the behavior of one or more functional units, which commence executing new instructions simultaneously. To accommodate the unique architectural features of the AI Engine, Peano has added LLVM support for several specific characteristics. Additionally, the repository includes support for Clang, LLD, binutils, Compiler-RT, and LLVM-LIBC.

Earlier this year, in January, AMD released the first XDNA driver for the Linux operating system, providing support for the Ryzen AI engine. To run this driver, users need to install Linux kernel 6.7 or higher, which has “IOMMU SVA” support enabled. However, AMD has since made little progress and has not provided a clear roadmap, lagging significantly behind Intel in NPU support.

After months of anticipation, the arrival of Peano finally initiates the necessary support, enabling developers to more effectively leverage Ryzen AI.