XiaoMi/mace
MACE is a deep learning inference framework optimized for running neural network models on mobile devices (Android, iOS, Linux, Windows) using heterogeneous compute resources.

Mobile AI Compute Engine is a production inference framework that optimizes deep learning model execution on mobile devices. It leverages hardware accelerators including NEON (ARM SIMD), OpenCL (GPU), and Hexagon DSP (HVX) to achieve high performance while managing power consumption and memory footprint. The framework supports model protection, Winograd convolution acceleration, and automatic kernel partitioning for UI responsiveness.