High-level comprehensive method and system for graph calculation

Abstract

The invention discloses a high-level comprehensive method and system for graph calculation, and belongs to the field of big data processing, and the method comprises the steps: (1) generating a graphcalculation program according to a functional programming model taking a point as a center; (2) specifying architecture parameters and micro-architecture parameters by adding an optimization instruction; (3) compiling a graph calculation program into modular data flow intermediate representation according to the data flow graph and the optimization instruction; (4) according to a mapping relationship between the IR module and the hardware template, mapping the intermediate representation of the data stream to a bottom-layer architecture, and instantiating an assembly line and a buffer area inthe hardware template; (5) if the instantiated parameterized hardware templates and the overall architecture meet constraint conditions, turning to the step (6); otherwise, turning to the step (3) after modifying the optimization instruction; and (6) generating a hardware language code which can be synthesized. Effective support can be provided for generating a graph application RTL from an upper-layer language, so that the execution parallelism degree of graph calculation on an FPGA is improved.

Description

technical field [0001] The invention belongs to the field of big data processing, and more specifically relates to a high-level comprehensive method and system for graph-oriented computing. Background technique [0002] In the past ten years, with the emergence of big data analysis problems such as biological information networks, social networks, and web page graphs, graph applications have become more and more important. Graphs are the best way to express the associated attributes of big data, and graph computing is based on graph models. For the mining and analysis process of huge, sparse, and hyperdimensional associations, the current machine learning and deep learning of big data rely on graph computing, and graph computing has become one of the mainstream modes of big data processing. [0003] Graph computing is complex and irregular, which poses new challenges to current hardware. For a general-purpose central processing unit (Central Processing Unit, CPU), even for ...

AI Technical Summary

Problems solved by technology

The HLS system can convert programs written in high-level languages ​​(mostly in C / C++) into register-transfer level (Register-Transfer Level, RTL) codes (such as Verilog, VHDL, etc.), and the HLS system also provides various optimization methods to make Developers can optimize the hardware structure from the high-level language level, and some of them also provide a visual view to facilitate the analysis of the circuit behavior of each clock cycle, which further improves the performance of generating RTL. There is a large gap in many aspects such as graph computing and performance. This phenomenon is especially serious for irregular and complex associated applications such as graph computing.

In general, the existing HLS system cannot provide effective support for the high-parallel execution of graph computing on FPGA

Images