Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

A data flow programming method and system for cpu/gpu heterogeneous clusters

A heterogeneous cluster and programming method technology, applied in the field of computer compilation, can solve problems such as low program performance, inability to reasonably allocate hardware resources, and insufficient mining of data flow program parallelism, so as to improve program performance and execution performance.

Active Publication Date: 2019-10-08
HUAZHONG UNIV OF SCI & TECH
View PDF2 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] In view of the above defects or improvement needs of the prior art, the present invention provides a data flow programming method and system for CPU / GPU heterogeneous clusters. The existing CPU and GPU work simultaneously and cannot reasonably allocate hardware under the target architecture. resources, did not fully exploit the parallelism existing in data flow programs, and the technical problems of low program performance

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A data flow programming method and system for cpu/gpu heterogeneous clusters
  • A data flow programming method and system for cpu/gpu heterogeneous clusters
  • A data flow programming method and system for cpu/gpu heterogeneous clusters

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0052] Such as figure 2 Shown is the compiling frame diagram in the COStream data stream compiling system of this embodiment 1. The data stream program written in the COStream language is used as the initial input, and the source program is constructed by the top-level grammar through the lexical analysis and semantic analysis at the front end of the COStream compiling system. The abstract syntax tree represented by the node; the abstract syntax tree is converted into a directed acyclic graph (SDF graph) connected by the task unit actor through the data flow stream. The nodes in the figure correspond to the source program task unit to represent calculation, and the edge The data flow corresponding to the source program indicates the dependency relationship between nodes and the direction of data communication; the SDF graph corresponding to the source program is used as the processing object, and the hierarchical secondary task division and scheduling are performed according t...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a method of data flow programming oriented to CPU / GPU heterogeneous clusters. The method comprises the steps of utilizing a data flow program to obtain a SDF diagram, conducting hierarchical secondary task division and scheduling on the SDF diagram, setting a stage number for each task unit to obtain the execution order of the task units; utilizing a CPU kernel or GPU mapping from the task unit to heterogeneous cluster nodes in the SDF diagram and a heterogeneous cluster multitask data communication model to obtain data communications between the task units, and obtaining target codes according to the execution order of the task units and the data communications between the task units. According to the method and system of data flow programming oriented to CPU / GPU heterogeneous clusters, the data communications are optimized with the parallelization of the data flow programs, and the overall execution performance of the programs is improved.

Description

technical field [0001] The invention belongs to the technical field of computer compilation, and more specifically relates to a data flow programming method and system for CPU / GPU heterogeneous clusters. Background technique [0002] Traditional CPU multi-core processors and CPU clusters sometimes cannot meet the needs of some applications for large-scale and ultra-large-scale computing. The high-performance floating-point performance of graphics processing units (Graphic Processing Unit, GPU) on servers with heterogeneous architectures such as CPU / GPU Computing, parallel computing and other aspects have huge advantages, so they are gradually widely used. A heterogeneous computer combining multiple CPUs and multiple GPUs has gradually become a research object. Under this architecture, the traditional core CPU and computing core GPU can work simultaneously, and tasks are executed in a multi-granularity parallel manner. Therefore, the CPU / GPU heterogeneous cluster system has ...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06F9/50G06F9/38
Inventor 于俊清杨瑞瑞唐九飞
Owner HUAZHONG UNIV OF SCI & TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com