Overview of the Architecture, Circuit Design, and Physical建筑电路设计物理概述.pptVIP

下载本文档

1
0
约1.84万字
约 76页
2017-03-08 发布于上海
举报
版权申诉

Overview of the Architecture, Circuit Design, and Physical建筑电路设计物理概述.ppt

1、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。。
2、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。
3、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)。
4、该文档为VIP文档，如果想要下载，成为VIP会员后，下载免费。
5、成为VIP后，下载本文档将扣除1次下载权益。下载后，不支持退款、换文档。如有疑问请联系我们。
6、成为VIP后，您将拥有八大权益，权益包括：VIP文档下载权益、阅读免打扰、文档格式转换、高级专利检索、专属身份标志、高级客服、多端互通、版权登记。
7、VIP文档为合作方或网友上传，每下载1次，网站将根据用户上传文档的质量评分、类型等，对文档贡献者给予高额补贴、流量扶持。如果你也想贡献VIP文档。上传文档

Overview of the Architecture, Circuit Design, and Physical建筑电路设计物理概述

* * * * * * * * * * * * * * * * * * * The Processors Used Benchmark 1 –GEMM Dense Matrix-Matrix Multiplication – High Computational Intensity and regular memory access Expect to reach close to peak on most platforms Explored two blocking formats: Column major and Block data layout Benchmark 1 –GEMM 3.0 7.8 29.5 204.7 — SP 5.4 4.0 16.9 14.6 51.1 DP IA64 AMD64 X1E CellPM Cell+PM Gflop/s CellPM CellPM 23 88 245 5117 — SP 42 45 141 365 1277 DP IA64 AMD64 X1E CellPM Cell+PM Mflop/W BENCHMARK 2 – Sparse Matrix Vector Multiply Seems like a poor choice at first glance due to low computational intensity and irregular data accesses. But less local store latency, task parallelism, 8 SPE load store units and DMA prove otherwise. Most of the matrix entries are zero, thus the nonzeros are sparsely distributed and can be streamed via DMA Like DGEMM, can exploit a FMA well Very low computational intensity (1 FMA for every 12+ bytes)? Non FP instructions can dominate Row lengths can be unique and in multiples of 4 symmetric unsymmetric 0.83 0.80 - 7.68 - - SP 0.67 0.60 2.64 4.00 4.35 3.38* DP - 3.04 CellFSS 0.41 0.53 - 4.08 - SP 0.36 0.36 1.14 2.34 2.46 DP IA64 AMD64 X1E CellPM Cell+PM Gflop/s symmetric unsymmetric 6.38 8.99 - 192 - - SP 5.15 6.74 22.0 100 109 84.5* DP - 76.0 CellFSS 3.15 5.96 - 102 - SP 2.77 4.04 9.50 58.5 61.5 DP IA64 AMD64 X1E CellPM Cell+PM Mflop/W SpMV - Results Stencil Based Computations Stencil computations codes represent wide array of scientific applications Each point in multidimensional grid is updated from subset of neighbours Finite difference operations used to solver complex numerical systems Here simple heat equations and 3D hyperbolic PDE are examined Relatively low computational intensity results in low % of peak on superscalars Memory bandwidth bound – Low computational intensity Stencils - Results Gflop/s 1.97 1.07 3.26 21.2 - 65.8 SP 7.25 CellFSS 1.19 0.57 3.91 8.2 21.1 DP IA64 AMD64 X1E CellPM Cell+PM Mflop/W 15.2 12 27.2 530 - 1645 SP 181