High-parallelism computing system and instruction scheduling method thereof

Abstract

The invention provides a high-parallelism computing system and an instruction scheduling method thereof. The computing system comprises an instruction reading and distributing module used for readinga plurality of types of instructions according to a specific sequence and distributing the obtained instructions to corresponding function modules according to the types; an internal cache which is used for caching data and instructions required for executing calculation; a plurality of functional modules which are used for sequentially executing the instructions of the category distributed by theinstruction reading and distributing module and reading required data from the internal cache, wherein the specific sequence is obtained by carrying out topological sorting on the instructions according to a directed acyclic graph formed by categories and dependency relationships. Therefore, the directed acyclic graph constructed according to the instruction category and the dependency relationship is subjected to topological sorting to perform value taking, and deadlock caused by instruction dependency can be avoided through relatively simple operation. Preferably, the sorting can be realized in an instruction compiling stage, so that the resource consumption of the computing system is further reduced.

Description

Technical field [0001] The present invention relates to the field of high-parallelism computing, and more specifically, to a high-parallelism computing system and its instruction scheduling method and corresponding compilation method. Background technique [0002] Neural Network has become a research hotspot in the field of image recognition in recent years. The trained neural network model can be used in many fields such as image classification, object recognition and saliency detection. In recent years, the neural network model has shown a trend of increasing computing scale and increasing complexity. The use of traditional CPU platforms can no longer meet its practical requirements. Therefore, the use of FPGA, GPU, ASIC and other heterogeneous computing platforms for neural network accelerator design has become a new research focus. Among them, compared to GPU platforms, FPGAs and ASICs can achieve more flexible hardware architectures and higher computing energy efficiency r...

AI Technical Summary

Problems solved by technology

In recent years, the neural network model has shown a trend of increasing computing scale and increasing complexity. The use of traditional CPU platforms has been unable to meet its practical requirements.

Images