以記憶體追蹤方式在單指令多執行緒架構中資料分享程度之分析研究

在本論文中，我們透過量化單一指令多執行續程式中記憶體存取的資料分享程度進而分析應用程式的區域特性。此外，我們也提供了不同執行環境下之資料分享的視覺化方法。為了量化資料分享程度，本論文使用含有執行階段記憶體位置的記憶體追蹤以完成記憶體存取的資料分享程度分析。在此分析中，我們重新定義重複使用距離的概念以提供分析以及不同執行環境的需要。

關鍵字

SIMT ；重用；距離；程度；記憶；追蹤

並列摘要

In this work, we address the problem of quantifying the data sharing degree of the memory access behavior within specific SIMT applications in order to quantify the locality characteristics of the application’s workload. In addition, we also offer way to visualize the way the sharing patterns of the applications and the way they change under different models of runtime scenarios. For the purposes of quantifying the data sharing degree a memory trace is generated that contains information of the addresses accessed at a specific point of execution. Then, the information contained in the traces is used to perform the data sharing degree analysis of memory accesses. In this analysis, we have redefined the reuse distance concept in order to make it suitable to our analytical requirements, at the same time considering the particulars of the execution model previously mentioned

並列關鍵字

SIMT ； Reuse ； Distance ； Degree ； Memory ； Trace

參考文獻

[1] J. Hennessy; D. Patterson, 5th ed., Computer Architecture. Elsevier, 2012.

[4] A. Kerr; G. Diamos; S. Yalamanchili. “A Characterization and Analysis of PTX Kernels”, in IEEE International Symposium on Workload Characterization, October 2009.

[6] Goswami, N., Shankar, R., Joshi, M., Li, T. “Exploring GPU Workloads: Characterization Methodology, Analysis and Microarchitecture Evaluation implications”. In IEEE International Symposium on Workload Characterization, December 2010.

[8] Jia, W.; Shaw, K.; Martonosi, M. “”Characterizing and Improving the Use of Demand-Fetched Caches in GPU”. In International Conference on Supercomputing, 2012.

[9] Kuo, H.K.; Lai, B.C.C.; Jou, J.Y. “A Cache Hierarchy Aware Thread Mapping Methodology for GPGPUs”.

國際替代計量

以記憶體追蹤方式在單指令多執行緒架構中資料分享程度之分析研究

全文下載

主題瀏覽