专业英语论文翻译
2020-03-04 09:11:11 来源：范文大全收藏下载本文
A Parallelization Cost Model for GPU
GPU并行成本模型 2009137127 周幼兰
Abstract - Using GPU for general computing has become an important research direction in high performance computing technology.However, this is not a lole optimization method.Due to the impact of device initialization cost, data transmiion delay, specific characteristics of programs, and other factors, the general computing on GPU may not always achieve the desired speedup, and sometimes results in program execution performance degradation.On the basis of in-depth analysis of GPU internal proceing mechanisms, the main factors affecting GPU implementation performance are pointed out, and a parallel cost model for GPU based on static program analysis is proposed to provide judgement basis for using GPU in general computing.摘要：在高性能计算技术领域，使用GPU执行通用计算已成为一个重要的研究方向。但它并不是一种无损优化方法。由于受设备初始化成本、数据传输延迟、程序本身特征和一些其他因素的影响，基于GUP的通用计算不可能总是达到预期的加速，有时还会导致程序执行性能下降。在对GPU内部处理机制进行深度分析的基础上，得到影响GPU执行性能的主要因素，且得出基于静态程序分析的GPU并行成本模型为GPU在通用计算中的使用提供了判断依据的结论。
I.INTRODUCTION Graphics proceing Unit (GPU) has developed at a speed much faster than the Moore’s Law in recent years, not only improving image proceing, virtual reality, computer simulation, and the development of related applications, but also providing a good running platform for general-purpose computing using GPU beyond graphic proceing.
The application of GPU in general-purpose computing makes a series of new challenges faced by the development of the high performance computing technology.Currently, in order to reduce the programming complexity of GPU in general-purpose computing, many GPU manufacturers and research institutions proposed a number of programming languages and programming models close to traditional programming methods, but different styles, such as Brook+[1], CUDA[2] and OpenCL[3], etc.However, as GPU has its own specific characteristic of hardware architecture and development, programmers must have a high level of expertise.Using GPU for general-purpose computing, the increased performance achieved in large part depends on the hardware knowledge and programming skills of programmers. At present, the most studies for the GPU’s parallelism both at home and abroad directly rewrite and transplant programs on the basis of original serial programs.Because software programmers often lack a deep understanding of the hardware platform and have no corresponding capability of programming hardware programs, program transplantation lead to the increased effects achieved of all kinds of applications accelerating general-purpose computing using GPU have obvious difference [4-6] .Program performance analysis technology as a basic method of understanding program behavior, plays an important role for comparing the performance difference between different program implementation, identifying performance bottlenecks of programs, and understanding the hardware resource utilization, and is the important part of development and optimization of high
performance computing programs [7].How to take advantage of program performance analysis technology and combine the architecture characteristics of CPU and GPU to guide the planning and optimization of parallel programs so that a variety of computing resources of CPU and GPU are fully utilized is a problem worthy of study using GPU in general-purpose computing at present.简介：
近几年来，计算机图形处理器（GPU）比摩尔定律发展得更迅猛，这种发展不仅体现在改善图形处理、虚拟现实、计算机模拟以及相关运用方面，还体现在为使用GPU作图形以外处理的通用计算提供了良好的运行平台。
GPU在通用计算方面的运用面临着高性能计算技术发展的一系列新挑战。目前，为了减少通用计算中GPU编程复杂性，许多GPU制造商和研究机构提出了一系列编程语言和编程模式，这些编程模式类似于传统德编程方法，但是具有不同的编程风格，例如Brook++、CUDA以及OpenCL等等。然而，当GPU拥有自己独特的硬件结构和发展特征时，与此同时程序员必须拥有较高水平的专业技能。通用计算中使用GPU来提高性能很大程度上取决于程序员的硬件知识和编程技术。迄今为止，国内外对于GPU的平行性研究大多数是直接写入和将程序直接植入到原始串行程序基础上。由于软件程序员往往对硬件平台缺乏深层次的了解，并且没有相应的编写硬件程序的能力，所以程序植入导致了各种各样的运用程序的增加效果有着明显的差异，这些运用程序加速了使用GPU的通用计算。 作为一种理解程序行为的基本方法，程序性能分析技术在比较不同程序执行的性能差异、找出程序的性能瓶颈和了解硬件资源的利用率方面扮演着重要角色，而且它还是高性能计算程序的重要组成部分。如何利用程序性能分析技术以及如何结合CPU和GPU的体系特征来引导平行程序的规划和优化以至CPU和GPU的大量计算资源得到充分利用，是目前GPU在通用计算方面一个值得探讨的问题。
V.CONCLUSION There are a number of studies on the applications related to using GPU in general-purpose computing.The most researches focus on using GPU to improve the execution performance of applications.However, how to measure the costs of GPU at runtime is le discued.From the perspective of GPU’s internal operation mechanism, analyze the key factors affecting the GPU implementation performance, and propose a cost test algorithm based on static program analysis.The results obtained by experiments show that the algorithm proposed can estimate relatively accurate GPU execution performance, and thus provide a useful reference for transplantation of traditional high performance computing to GPU.结论：
与使用GPU作通用计算相关运用的研究实例比比皆是，且大多数研究将使用GPU来提高运用程序的执行性能为研究中心，但是至于怎样衡量GPU运行时的成本却很少有人提及。本论文从GPU内部运行机制出发，分析影响GPU执行性能的主要因素，提出基于静态程序分析的成本测试算法。实验所得结果说明所提算法能够相对精确地评估GPU执行性能从而为传统的高性能计算的在GPU中的移植提供了一个实用性的参考。
Comment: Cost model has been widely used in the computer field as a way to evaluate whether a program is excellent or not.And quite a number of researchers in computer field has proposed various kinds of cost models for GPU, of which most are under certain conditions.
But this paper give us a new perception of cost model for GPU in general-purpose computing.The parallelization cost model for GPU can be generally applied for many application areas .The authors attempt to estimate the cost comprehensively.Actually they take the GPU initialization cost, transmiion of data cost as well as the program execution cost into consideration.Each aspect use a special algorithm to calculate the cost .And the cost is measured by the time of each aspect.In this paper ,we can get through the authors’ idea very well ,because they are well organized in form and shown clearly in graphs, charts as well as in equations.The parallelization cost model is more accurate, flexible and portable than models of the past.So we should learn the method they study a topic.That is to consider a question in a broad perspective.And if we keep thinking in this way ,our ability of doing scientific research will be greatly improved.At the same time, we should keep close watch on the field of GPU.As it is such an important proceor that it is used widely used on modern computers.It is even considered the core graphics proceor of computers.And it develops faster nowadays as the need rises.
人人范文网 m.inrrp.com.cn 手机版