Efficient firmware operation system for neural network processor

The invention belongs to the technical field of embedded processor firmware, particularly relates to an efficient firmware running system for a neural network processor, and aims to solve the problems that the firmware deployment speed of an existing neural network processor is low, the occupied space is large, and the processing efficiency of files generated after compiling of a neural network compiler is low. The system comprises: a file preprocessing unit configured to obtain a file generated after compiling by a neural network compiler and preprocess the file to obtain a preprocessed file; the embedded hardware platform is configured to store the preprocessed file to a set memory address; and the efficient firmware is configured to read and analyze the preprocessing file, obtain an operator flow file corresponding to the neural network model, and execute each operator in sequence through a pre-constructed operator execution efficient optimization method until one time of complete calculation of the neural