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

Programming model oriented to neural network heterogeneous computing platform

A neural network and computing platform technology, applied in the field of programming models oriented to neural network heterogeneous computing platforms, can solve problems such as not being able to meet performance requirements well, and achieve high flexibility, good computing performance, and strong scalability. Effect

Active Publication Date: 2017-10-10
XILINX INC
View PDF5 Cites 38 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Computing platforms based on traditional general-purpose processors (CPUs) cannot meet performance requirements well

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
  • Programming model oriented to neural network heterogeneous computing platform
  • Programming model oriented to neural network heterogeneous computing platform
  • Programming model oriented to neural network heterogeneous computing platform

Examples

Experimental program
Comparison scheme
Effect test

no. 1 example

[0030] Below, we will first combine figure 1 to describe the functions of the constituent parts in the first embodiment.

[0031] figure 1 It is a hybrid compilation model for heterogeneous computing platforms of CPU + neural network special processor.

[0032] exist figure 1 In the neural network (Neural Network, can be referred to as NN) optimization compiler (or "NN optimization compiler") with the neural network model as input, by analyzing the network topology to obtain the control flow and data flow information in the model, Based on this, various optimization transformation techniques are applied to the model. Specifically, the compiler merges the computational operations between different network layers in the neural network model, reducing the computational intensity. For structured and unstructured sparse networks, the compiler will eliminate unnecessary computation and data movement caused by sparse values. In addition, the compiler will fully reuse the network...

no. 2 example

[0056] The following will combine Figure 4 To describe the programming environment of the present invention, that is, the program execution support model.

[0057] Figure 4 It is a program running support model of the heterogeneous computing platform according to the second embodiment of the present invention.

[0058] exist Figure 4 In , the neural network dedicated processor is abbreviated as DPU to distinguish it from the host CPU. It should be understood by those skilled in the art that such nomenclature does not affect the generality of a neural network special-purpose processor. That is to say, in this specification and the accompanying drawings, "neural network dedicated processor" and "DPU" are terms that can be used interchangeably, and are used to represent another processor that is different from the CPU on a heterogeneous computing platform.

[0059] like Figure 4 As shown in the neural network specific processor driver ( Figure 4 (shown as "DPU driver")...

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 provides a programming model oriented to a neural network heterogeneous computing platform. Specifically, the invention provides a compiling method and system of a heterogeneous computing platform and a program running support method and system thereof. A trained neural network model is input to a neural network (NN) optimization complier to generate a NN assembling file corresponding to the NN. The NN assembling file is input to a NN assembler to generate a NN binary file corresponding to the neural network; a host complier tool chain is used for compiling and assembling a neural network application program developed by a user by using the high-level language, and orderly generates a corresponding host assembling file and a host binary file. The host linker is used for linking the NN binary file and the host binary file to generate a single mixed link executable file. The technical scheme provided by the invention has the features of being good in calculation performance, strong in expandability, strong in compatibility and high in flexibility.

Description

technical field [0001] The invention relates to a heterogeneous computing platform and a neural network, and more particularly to a programming model oriented to a neural network heterogeneous computing platform. Background technique [0002] Artificial intelligence has developed rapidly in recent years, which has greatly affected people's lives. All countries in the world have attached great importance to it and made large-scale R&D investment. Artificial neural networks are the core of artificial intelligence applications. The deep learning neural network algorithm is the most common artificial neural network model. Its workload characteristics are computationally intensive (multiply-add operations on the order of G) and data-intensive (parameters ranging from M to hundreds of Mbytes). The computing platform based on the traditional general-purpose processor CPU cannot meet the performance requirements well. In recent years, heterogeneous platforms for accelerating neur...

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): G06F9/45G06N3/02
CPCG06F8/41G06N3/02G06F8/31G06F8/54G06F8/71G06N3/105G06F8/315G06N3/08
Inventor 孙晓明隋凌志罗洪单羿姚颂
Owner XILINX INC
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