Processor Descriptor for Neural Network Data Access Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As AI algorithms become more complex, the increasing amount of data and data dimensions processed by processors lead to reduced efficiency due to the need for determining data addresses based on specified parameters in data-read instructions, making data access inefficient.
Innovation Solution
Introducing a descriptor that indicates the shape of tensor data, allowing the processor to determine the corresponding content and execute processing instructions more efficiently by eliminating the need for explicit address input during each data access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If processors determine data address based on parameters specified in data-read instructions, then data access can be performed, but the processing efficiency is reduced due to increasing data amount and dimensions
Solution Approach 1:
The patent applies preliminary action by pre-calculating and storing data addresses in a descriptor before processing occurs. The descriptor contains pre-computed address information that the processor can directly use without real-time calculation, thus improving processing efficiency while reducing the complexity of data access operations during actual processing.
Solution Approach 2:
The patent introduces a descriptor as an intermediary between the processing instruction and the actual data. This descriptor acts as a mediator that contains address information, allowing the processor to access data through the descriptor rather than directly computing addresses from parameters, thereby simplifying the data access process and improving efficiency.
2Ease of operation
If programmers set relevant parameters for data access when designing, then data can be accessed, but the complexity of data access operations increases
Solution Approach 1:
The descriptor is prepared in advance with all necessary address information, eliminating the need for programmers to manually set complex parameters during processing. This preliminary preparation simplifies the operation during actual data access while maintaining the ability to handle complex data structures.
Solution Approach 2:
The patent creates a copy of the address information in the descriptor, which is a simplified representation that the processor can use directly. Instead of requiring the processor to work with complex original parameters, it uses the copied address information from the descriptor, making data access easier while preserving the full capability to access complex data structures.
Data Source
AI summary
The present disclosure provides a data processing method and an apparatus and related products. The products include a control module including an instruction caching unit, an instruction processing unit, and a storage queue unit. The instruction caching unit is configured to store computation instructions associated with an artificial neural network operation; the instruction processing unit is configured to parse the computation instructions to obtain a plurality of operation instructions; and the storage queue unit is configured to store an instruction queue, where the instruction queue includes a plurality of operation instructions or computation instructions to be executed in the sequence of the queue. By adopting the above-mentioned method, the present disclosure can improve the operation efficiency of related products when performing operations of a neural network model.