Neural Network Processor Instruction Generation for Channel Mismatch
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neural network processors experience resource wastage due to the mismatch between the number of channels they support for parallel processing and the number of channels in the input feature data, leading to suboptimal utilization of computational resources.
Innovation Solution
An instruction generating method that determines target splicing information based on hardware parameters and input feature data parameters, allowing for efficient splicing of input feature data to match the number of channels supported by the neural network processor for parallel processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the neural network processor supports a large number of channels for parallel processing, then computational efficiency is improved, but computational resources are wasted when the input feature data has fewer channels
Solution Approach 1:
The input feature data with fewer channels is segmented and spliced with other data to form a complete data set that matches the processor's channel capacity. This allows the processor to operate at full capacity without wasting computational resources on unused channels.
Solution Approach 2:
Multiple input feature data sets are merged or spliced together to create a combined data set whose channel count matches the neural network processor's supported channel capacity. This ensures full utilization of the processor's parallel processing capabilities.
2Adaptability or versatility
If the number of channels in input feature data is increased to match processor capacity, then resource utilization is improved, but data preprocessing complexity increases
Solution Approach 1:
The splicing operation is performed in advance during the instruction generation phase, before the actual neural network processing. This preliminary preparation ensures that the data is properly formatted for full processor utilization without adding complexity during the critical processing phase.
Solution Approach 2:
The instruction generating apparatus acts as an intermediary that automatically performs the splicing operation and generates appropriate processing instructions. This mediator handles the preprocessing complexity, shielding the main processing system from the complexity of data alignment.
Data Source
AI summary
Disclosed are an instruction generating method, an apparatus and an electronic device. The present disclosure relates to the technical field of integrated circuits. The method comprises determining parameter information about input feature data to be processed by a function operator in a neural network model; determining hardware parameters supported by the neural network processor; determining target splicing information corresponding to the parameter information about the input feature data based on the hardware parameters and the parameter information about the input feature data; and generating instructions executable by the neural network processor based on the target splicing information. The technical solution of the present disclosure may avoid the problem of waste of computational resources due to the number of channels supported by the neural network processor for parallel processing is greater than the number of channels of input feature data, thereby fully utilizing the computational resources of the neural network processor.