Unlock instant, AI-driven research and patent intelligence for your innovation.

Neural network heterogeneous many-core multi-level resource mapping method based on compilation

A neural network and resource mapping technology, which is applied in neural learning methods, biological neural network models, neural architectures, etc., can solve problems such as the inability to perform effective resource mapping and the inability of deep learning loads to fully utilize the performance of heterogeneous many-core platforms. To achieve the effect of improving performance

Pending Publication Date: 2022-03-29
JIANGNAN INST OF COMPUTING TECH
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The purpose of the present invention is to provide a neural network heterogeneous many-core multi-level resource mapping method based on compilation to solve the problem that the current deep learning compiler TVM cannot perform effective resource mapping for domestic heterogeneous many-core processors, resulting in deep The problem that the learning load cannot give full play to the performance of the heterogeneous many-core platform

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
  • Neural network heterogeneous many-core multi-level resource mapping method based on compilation

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0023] Embodiment: The present invention provides a method for mapping heterogeneous many-core multi-level resources based on neural network compilation, which specifically includes the following steps:

[0024] S1. Perform many-core core group resource mapping, as follows:

[0025] S11. Perform loop splitting on the outermost loop x of the neural network operator to obtain the outer loop xo and the inner loop xi, and set the number of cycles of the split outer loop xo to be equal to the number N of many-core core groups;

[0026] S12. For the outer loop xo obtained in S11, bind its calculation process to many-core core group resources;

[0027] S2. Perform slave core thread resource mapping, specifically as follows:

[0028] S21. If the number of cyclic layers of the neural network operator before splitting in step S11 is greater than or equal to 2, execute S23; otherwise, execute S22;

[0029] S22. Perform cyclic splitting on the inner loop xi obtained in step S11 to obtai...

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 compiling-based neural network heterogeneous many-core multi-level resource mapping method. The method comprises the following steps: S1, carrying out many-core group resource mapping; s2, performing slave core thread resource mapping; and S3, carrying out vector component resource mapping. According to the method, the parallel potential of the neural network operator can be fully mined, and the advantage of on-chip multi-level parallel is exerted, so that the performance of the deep learning load on a heterogeneous many-core platform is improved.

Description

technical field [0001] The invention relates to a compilation-based neural network heterogeneous many-core multi-level resource mapping method, which belongs to the technical field of compilation optimization. Background technique [0002] The role of the deep learning compiler is to deploy the deep learning workload on a specific hardware platform to efficiently complete training and reasoning tasks. It can fully mine the algorithm characteristics and pattern characteristics in the field of artificial intelligence, and convert the models of various typical deep learning frameworks into a unified Computing graphs, and then through a series of domain algorithm-guided compilation and optimization technologies and architecture-related underlying optimization technologies, efficient codes for different hardware platforms are generated to accelerate the reasoning process in deep learning. [0003] TVM (Tensor Virtual Machine) is the current mainstream deep learning compiler. It i...

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
IPC IPC(8): G06F8/41G06N3/04G06N3/063G06N3/08
CPCG06F8/41G06N3/063G06N3/08G06N3/045
Inventor 沈莉周文浩王飞肖谦武文浩赵美佳李斌
Owner JIANGNAN INST OF COMPUTING TECH