TinyIREE: An ML Execution Environment for Embedded Systems From Compilation to Deployment

Machine learning model deployment for training and execution has been an important topic industry academic research in the last decade. Much of attention focused on developing specific toolchains to support acceleration hardware. In this article, we present Intermediate Representation Execution Environment (IREE), a unified compiler runtime stack with explicit goal scale down machine programs smallest footprints mobile edge devices, while maintaining ability up larger targets. IREE adopts compiler-based approach optimizes heterogeneous hardware accelerators through use Multi-Level IR (MLIR) infrastructure, which provides means quickly design implement multilevel intermediate representations (IR). More specifically, article is TinyIREE, set options that accommodate limited memory computation resources embedded systems bare-metal platforms, also demonstrating IREE’s intuitive workflow generates workloads different ISA extensions application binary interface (ABIs) LLVM.