Neural network processor and control method of neural network processor

A control method of a neural network processor applied to a multi-core neural network processor, the multi-core neural network processor comprising N single-core neural network processors, wherein N is an integer greater than 1, the method comprising:

• obtaining a first task descriptor, the first task descriptor configured to describe information of a task distributed to the neural network processor;
• analyzing the first task descriptor, and then determining subtasks distributed to each selected processor according to an analysis result thereof, wherein the selected processor is a processor that is selected from the N single-core neural network processors and configured to perform the task;
• modifying the first task descriptor to obtain a second task descriptor corresponding to each selected processor, the second task descriptor configured to describe information of a subtask distributed to the selected processor;
• and
• configuring each second task descriptor to a corresponding selected processor, and starting each selected processor to process a corresponding subtask;
• and wherein
• the step of modifying the first task descriptor, comprises:
• if a processing mode indicated in the first task descriptor is a multi-core single-batch mode, modifying a first fragment information address of the first task descriptor to a second fragment information address, wherein the first fragment information address is an address of fragment information of the multi-core neural network processor, and the second fragment information address is an address of fragment information of the selected processor.