cuda-lecture4reduction treecuda-lecture4-reduction treecuda-lecture4-reduction treecuda-lecture4-reduction tree.ppt

下载文档 降价啦

6
0
约9.94千字
约 26页
2017-03-18 发布于贵州
举报
版权申诉
保障服务

cuda-lecture4reduction treecuda-lecture4-reduction treecuda-lecture4-reduction treecuda-lecture4-reduction tree.ppt

1、本文档共26页，可阅读全部内容。
2、原创力文档（book118）网站文档一经付费（服务费），不意味着购买了该文档的版权，仅供个人/单位学习、研究之用，不得用于商业用途，未经授权，严禁复制、发行、汇编、翻译或者网络传播等，侵权必究。
3、本站所有内容均由合作方或网友上传，本站不对文档的完整性、权威性及其观点立场正确性做任何保证或承诺！文档内容仅供研究参考，付费前请自行鉴别。如您付费，意味着您自己接受本站规则且自行承担风险，本站不退款、不进行额外附加服务；查看《如何避免下载的几个坑》。如果您已付费下载过本站文档，您可以点击这里二次下载。
4、如文档侵犯商业秘密、侵犯著作权、侵犯人身权等，请点击“版权申诉”（推荐），也可以打举报电话：400-050-0827(电话支持时间：9:00-18:30)。

? David Kirk/NVIDIA and Wen-mei W. Hwu ECE408/CS483/ECE498al, University of Illinois, 2007-2012 ? David Kirk/NVIDIA and Wen-mei W. Hwu ECE408/CS483/ECE498al, University of Illinois, 2007-2012 * Lecture 4Parallel Computation Patterns – Reduction Trees ? David Kirk/NVIDIA and Wen-mei W. Hwu ECE408/CS483/ECE498al, University of Illinois, 2007-2012 * Objective To master Reduction Trees, arguably the most widely used parallel computation pattern Basic concept A class of computation Memory coalescing Control divergence Thread utilization Partition and Summarize A commonly used strategy for processing large input data sets There is no required order of processing elements in a data set (associative and commutative) Partition the data set into smaller chunks Have each thread to process a chunk Use a reduction tree to summarize the results from each chunk into the final answer We will focus on the reduction tree step for now. ? David Kirk/NVIDIA and Wen-mei W. Hwu ECE408/CS483/ECE498al, University of Illinois, 2007-2012 * What is a reduction computation Summarize a set of input values into one value using a “reduction operation” Max Min Sum Product Often with user defined reduction operation function as long as the operation Is associative and commutative Has a well-defined identity value (e.g., 0 for sum) ? David Kirk/NVIDIA and Wen-mei W. Hwu ECE408/CS483/ECE498al, University of Illinois, 2007-2012 * An efficient sequential reduction algorithm performs N operations - O(N) Initialize the result as an identity value for the reduction operation Smallest possible value for max reduction Largest possible value for min reduction 0 for sum reduction 1 for product reduction Scan through the input and perform the reduction operation between the result value and the current input value ? David Kirk/NVIDIA and Wen-mei W. Hwu ECE408/CS483/ECE498al, University of Illinois, 2007-2012 * A parallel reduction tree algorithm performs N-1 Operations in log(N) steps ? David Ki