可重配置图形处理器的照明算法的性能表征资源-CSDN文库

研究论文

194 浏览量 2021-03-09 01:13:56 上传评论收藏 1.25MB PDF 举报

资源详情

资源评论

Vol. 26 | Issue 5 | Oct. 2019

CONTENTS

Artificial Intelligence

Adaptive transfer learning framework for dense prediction of human activity recognition

摇 Zhang Zhao, Zhang Yong, Teng Yinglei, Guo Da, Deng Haiqin (1)…………………………………………

Self鄄adaptive mechanism based genetic algorithms for combinatorial optimization problems

摇 Qu Zhijian, Wang Shasha, Xu Hongbo, Li Panjing, Li Caihong (11)………………………………………

VM migration algorithm for the balance of energy resource across data centers in cloud computing

摇 Song Da, Fu Xiong, Zhou Jingjing, Wang Junchang, Zhang Lin, Deng Song, Qiao Lei (22)………………

Adv Paxos: making classical Paxos more efficient Li Bin, Jiang Jianguo, Yuan Kaiguo (33)…………………

Image Processing

Low鄄light color image enhancement based on NSST

摇 Wu Xiaochu, Tang Guijin, Liu Xiaohua, Cui Ziguan, Luo Suhuai (41)………………………………………

Competition system for automatically evaluating edge detection algorithms based on the evolutionary game

摇 theory Peng Hong, Ding Guangtai, Zhang Huiran, Hu Dongli (49)…………………………………………

Performance characterization of illumination algorithms for reconfigurable graphics processor

摇 Deng Junyong, Liu Yang, Xie Xiaoyan (60)……………………………………………………………………

Others

Early termination strategy of HEVC based on statistical models

摇 Liu Yitong, Tian Wang, Li Yuchen, Yang Hongwen (72)………………………………………………………

Adaptive access points deployment for indoor bluetooth positioning accuracy enhancement

摇 Tian Zengshan, Ren Haoliang, Zhou Mu (82)…………………………………………………………………

Single鄄cavity triangular substrate integrated waveguide bandpass filter with complementary triangular split

摇 ring resonator Yang Tao, Yan Yonghong, Zhang Yue, Guo Huijuan, Zhu Yongyuan, Huang Wei (94)……

Serial parameters: CN 11鄄3486 / TN*1994*b*16*98*en*P* ￥ 20. 00*1000*10*2019鄄10

Available online at http: / / www. cnki. net

October 2019, 26(5):

試試試試試試試試試試試

60 - 71

The Journal of China

Universities of Posts and

詨詨詨詨詨詨詨詨詨詨詨

Telecommunications

http:椅jcupt. bupt. edu. cn

Performance characterization of illumination algorithms for

reconfigurable graphics processor

Deng Junyong

(苣), Liu Yang

, Xie Xiaoyan

1郾 School of Electronic Engineering, Xi蒺an University of Posts and Telecommunications, Xi蒺an 710121, China

2郾 School of Computer, Xi蒺an University of Posts and Telecommunications, Xi蒺an 710121, China

Abstract

摇 Graphics processing is an increasing important application domain with the demand of real鄄time rendering, video

streaming, virtual reality, and so on. Illumination is a critical module in graphics rendering and is typically

compute鄄bound, memory鄄bound, and power鄄bound in different application cases. It is crucial to decide how to

schedule different illumination algorithms with different features according to the practical requirements in

reconfigurable graphics hardware. This paper analyze the performance characteristics of four main鄄stream lighting

algorithms, Lambert illumination algorithm, Phong illumination algorithm, Blinn鄄Phong illumination algorithm, and

Cook鄄Torrance illumination algorithm, using hardware performance counters on x86 processor platform KabyLake

(KBL). The data movement, computation, power consumption, and memory accessing are evaluated over a range

of application scenarios. Further, by analyzing the system鄄level behavior of these illumination algorithms, obtains

the cons and pros of these specific algorithms were obtained. The associated relationship between performance /

energy and the evaluated metrics was analyzed through Pearson correlation coefficient (PCC) analysis. According to

these performance characterization data, this paper presents some reconfiguration suggestions in reconfigurable

graphics processor.

Keywords摇 performance characterization, illumination algorithms, reconfigurable graphics processor, correlation analysis, computer architecture

Received date: 03鄄01鄄2019

Corresponding author: Deng Junyong, E - mail: djy@ xupt. edu. cn

DOI: 10. 19682 / j. cnki. 1005鄄8885. 2019. 0031

1摇 Introduction

Graphics processors was an important domain with

the growing demand for applications such as 3鄄

dimensional ( 3D) gaming, video streaming, virtual

reality, cryptocurrency mining, and machine learning

[1 - 2 ]. The graphics processors evolved from

configurable implementations of a complex fixed

function pipeline to high performance, highly

programmable, post鄄silicon reconfigurable hardware

where developers could implement their own rendering

algorithms [3 - 5].

The Makimoto wave which proposed by Makimoto

[6] and modified by Xu et al. [ 7 ] indicates that

there is an alternative phenomena of semiconductor

products in between standardization and customization,

as shown in Fig. 1. Reconfigurable computing can

break through the limitations of Von Neumann

architecture with time鄄spatial multi鄄dimensional

computing approach, which combines the

standardization and customization, computing

Issue 5 Deng Junyong, et al. / Performance characterization of illumination algorithms for reconfigurable graphics processor

efficiency and programmable flexibility, and will be a

development direction of future architecture ( http:椅

www. eeworld. com. cn / qrs / 2018 / ic鄄news101651808.

html / . ) [ 8 - 9 ]. Graphics rendering possesses the

characteristics of high flexibility. From the view of

rendering algorithms, there exists many different

algorithms with different efficiency for different

applications, even for one single functionality, such as

illumination, rasterization, clipping, and texture

mapping, etc. , also multiple algorithms [10 - 13 ].

From the view of graphics application programming

interface (API), there also exist many different API,

摇摇摇摇摇摇摇摇

such as OpenGL, standard portable intermediate

representation (SPIR), and Vulkan of Khronos Group

[14 ], Mantle of advanced micro devices ( AMD )

( https: 椅 www. extremetech. com / extreme / 200286鄄

not鄄dead鄄yet鄄amds鄄mantle鄄powers鄄new鄄vulkan鄄api鄄vr鄄

efforts), and so on. For accommodating the algorithm

switching and API extension for different requirement of

graphics processing ranging from many application

fields, the reconfigurable graphics processor is

obtained increasing attention in both academia and

industry.

Fig. 1摇 Makimoto wave

摇

However, under different application requirements,

there is not applicable criteria to make the decision of

when the graphics processor should start to

reconfigure, which algorithm should be chosen. So,

for the design of reconfigurable graphics processor, the

first important step should be establishing a

performance evaluation model which can provide the

associated data to make the decision of when and how

to reconfigure [ 15 ]. This paper focuses on the

illumination processing of graphics rendering, analyzes

four commonly used lighting algorithms, Phong

illumination algorithm [ 10 ], Lambert illumination

algorithm [ 16 ], Blinn鄄Phong illumination algorithm

[ 12 ], and Cook鄄Torrance illumination algorithm

[11], on KBL under different rendering scenarios,

presents the characteristics comparison of data

movement, computation, power consumption,

instruction per clock (IPC), and cache misses per kilo

instructions ( MPKI ), which can be used as the

reconfiguration suggestions.

The contributions of this paper can be summarized as

follows:

1) This paper evaluated four illumination

algorithms, including Lambert illumination algorithm,

Phong illumination algorithm, Blinn鄄Phong illumination

algorithm, and Cook鄄Torrance illumination algorithm.

The performance analysis includes the data movement

per pixel, computations per pixel, power consumption

per pixel, and MPKI of each level caches.

2) In different graphics rendering scenarios, the

algorithms scheduling options are suggested according

to the performance metrics.

The rest of this paper is organized as follows.

Sect. 2 summarizes the trend of reconfigurable graphics

rendering, brief introduction of four popular

illumination algorithms, the analysis of the hardware

platform used in experiment. Sect. 3 describes the

graphics scene datasets, profiling tools, and metrics.

In Sect. 4, this paper presents the performance

comparison of four algorithms in details and provides

剩余12页未读，继续阅读

评论收藏

内容反馈

可重配置图形处理器的照明算法的性能表征

评论0

最新资源

可重配置图形处理器的照明算法的性能表征

评论0

最新资源

相关推荐

基于可配置处理器的AES算法设计

基于可配置处理器的AES算法设计.pdf

3D图形学算法demo

基于FPGA的可重配置分数阶信号变换处理器设计.pdf

开源图形处理器（GPU）Nyuzi.zip

基于FPGA的可配置FFT处理器.pdf

高性能异构多处理器平台及其应用

具有低功耗和高性能的可配置处理器内核.pdf

可配置多核处理器的调试器模块化分层设计.pdf

可配置流处理器核心级指令设计及相关编译技术研究.pdf

高效可配置浮点FFT处理器设计.pdf

高吞吐率可配置FFT处理器IP核的设计与VLSI实现.pdf

C# 配置文件存储 各种序列化算法性能比较 程序1.1

基于3D引擎的图形学算法实现

星载嵌入式处理器软件在轨重配置技术研究.pdf

粗粒度可重配置处理器中基于配置上下文重用和差分重配置的延迟隐藏技术

论文研究-反馈移位寄存器在通用可重构处理器上的配置生成与优化设计.pdf

Spring MVC 配置请求的默认处理器.rar

可配置多核处理器的调试器模块化分层设计

行人惯性导航零速检测算法

基于CORDIC的反正弦和反余弦计算的FPGA实现

混合动力汽车基于规则的控制和ECMS与优化等效因子的实时能源管理策略

BA无标度网络中的SIR模型

使用3DCNN和卷积LSTM进行手势识别学习时空特征

基于三次贝塞尔曲线的类汽车曲率连续路径平滑

基于机器学习的设备剩余寿命预测方法综述

基于无差拍预测控制和扰动观测器的永磁同步电机电流控制

基于FPGA的奇异值和特征值分解的快速实现。

基于BP神经网络的人口预测

C# 配置文件存储各种序列化算法性能比较程序1.1