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

A data structure suitable for hpcg optimization on "Sunway TaihuLight" and its efficient implementation method

A technology of data structure and implementation method, applied in the direction of electrical digital data processing, address formation of the next instruction, complex mathematical operations, etc., can solve the problem that the data structure cannot be directly applied, so as to improve the locality of time and space and avoid data competition conflicts , Improve the effect of parallelism and bandwidth utilization

Active Publication Date: 2020-06-16
INST OF SOFTWARE - CHINESE ACAD OF SCI +1
View PDF1 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Due to the particularity of the domestic Shenwei many-core platform architecture, these existing data structures cannot be directly applied

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 structure suitable for hpcg optimization on "Sunway TaihuLight" and its efficient implementation method
  • A data structure suitable for hpcg optimization on "Sunway TaihuLight" and its efficient implementation method
  • A data structure suitable for hpcg optimization on "Sunway TaihuLight" and its efficient implementation method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0030] The present invention will be described in detail below in conjunction with the accompanying drawings and illustrations.

[0031] The entire matrix is ​​stored in the ELL format, which can allocate continuous memory for the data and ensure the alignment of the data. After using the ELL format, since the number of elements in each row is the same, the start of each row of the matrix can be determined directly by calculating the offset, instead of recording the start of each row by storing an additional array like the CSR format. Thereby reducing the total data access. In the role of block, matrix values ​​and corresponding indexes based on ELL format need to be rearranged according to their corresponding block rows. For the value of the matrix, it is enough to move directly according to the row order after the block. For the index of the matrix, there are generally two different processing schemes according to the processing method of the indexed vector:

[0032] 1) I...

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 data structure suitable for HPCG (High Performance Conjugate Gradients) optimization on "Sunway TaihuLight", and an efficient implementation method thereof. On the basis of the structural characteristics of a Sunway many-core processor system on the "Sunway TaihuLight" and HPCG algorithm optimization requirements, a data structure of an improved ELL (ELLPACK) format is used to replacing an original data structure, wherein the data structure includes the own data structure of matrix data and also supports the data structures related to blocking coloring, index conversion, process communication, vector position mapping and the like; in an intermediate treatment process, and an Athread multi-thread library provided by the Sunway many-core platform is adopted to carry out parallel so as to be finally applied to the kernel function SpMV and SymGS optimization processes of the HPCG algorithm; and corresponding performance and a bandwidth percentage obtain a consistent speed-up ratio, and the highest corresponding performance and the highest bandwidth percentage are independently 18.2 and 17.6.

Description

technical field [0001] The self-defined data structure and its efficient implementation method involved in the present invention support the high-performance implementation of the HighPerformanceConjugate Gradients (HPCG, high-performance conjugate gradient) benchmark test program on the "Sunway TaihuLight" supercomputer. Among them, the intermediate processing process uses the Athread library provided by the domestic Shenwei many-core platform for multi-thread parallelism. Background technique [0002] The High Performance Conjugate Gradients (HPCG, High Performance Conjugate Gradients) benchmark is a new standard for ranking supercomputers around the world. This test benchmark is mainly used to measure the ability of supercomputers to solve large-scale sparse linear systems. Compared with the HighPerformanceLinpack (HPL) benchmark test used by the current TOP500 ranking, its calculation, memory access and communication modes are more representative of the current partial d...

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): G06F9/30G06F9/32G06F9/38G06F17/16
CPCG06F9/30007G06F9/30101G06F9/325G06F9/3836G06F17/16
Inventor 敖玉龙杨超刘芳芳尹万旺魏迪袁欣辉蒋丽娟
Owner INST OF SOFTWARE - CHINESE ACAD OF SCI