在本論文中，我們提出一套針對圖形處理器 (Graphics Processing Unit) 之記憶體壓縮架構。 在傳輸大量資料時，利用所提出的壓縮技術節省匯流排頻寬與記憶體空間。 壓縮與解壓縮單元 (Compression and Decompression Unit) 加入在系統的記憶體階層中之快取記憶體與主記憶體間，使資料經過壓縮後降低資料量，達成降低匯流排負擔與主記憶體空間使用的目的。 本論文提出的壓縮演算法為一針對浮點數表示法開發的演算法，並針對 GPU 同時支援浮點數與整數運算的特性，將壓縮架構拓展為可以一次壓縮多個字元，並進一步觀察即利用資料的特性，藉此達到最有效率的壓縮。 經過實驗，我們提出的單字元與四字元壓縮架構在我們的測試環境下，可以提供平均 51.72% 與 46.81% 的壓縮率。 同時我們的壓縮架構估計可以將記憶體存取的能耗降低到 58.77% ，有效降低記憶體存取的功耗。

In this paper, we present a memory-compression architecture for GPU (Graphic Processing Unit). This architecture saves bus bandwidth and memory space for huge data transfer. Our system improves the compression performance by reducing the redundancies in floating-point data. We propose both Single-word and Quad-word memory compression architectures. Comparing with previous works, we get a good improvement, 16.94% of compression ratio on average, 4.3% of area, 9.3% of power and 14% of operation frequency.

[1] X. Chen; L. Yang; R.P. Dick; L. Shang; H. Lekatsas, “C-Pack: A High-Performance Microprocessor Cache Compression Algorithm,” in IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol.18, no.8, pp.1196-1208, Aug. 2010
[2] E.G. Hallnor; S.K. Reinhardt, “A unified compressed memory hierarchy,” in 2005. HPCA-11. 11th International Symposium on High-Performance Computer Architecture, vol., no., pp.201-212, 12-16 Feb. 2005
[3] S. Roy; R. Kumar; M. Prvulovic, “Improving system performance with compressed memory,” in Proceedings 15th International Parallel and Distributed Processing Symposium., Apr. 2001
[4] I. C. Tuduce; T. Gross, “Adaptive main memory compression,” in Proceedings of the annual conference on USENIX Annual Technical Conference (ATEC '05). USENIX Association, Berkeley, CA, USA, pp.29-29, 2005.
[5] R. B. Tremaine; P. A. Franaszek; J. T. Robinson; C. O. Schulz; T. B. Smith; M. E. Wazlowski; P. M. Bland, “IBM Memory Expansion Technology (MXT),” in IBM Journal of Research and Development , vol.45, no.2, pp.271-285, March 2001
[6] J. L. Nunez; S. Jones, “Gbit/s lossless data compression hardware,” in IEEE Transactions on Very Large Scale Integration (VLSI) Systems, vol.11, no.3, pp.499-510, June 2003
[7] A. Alameldeen; D. A. Wood, “Frequent pattern compression: A significance-based compression scheme for 12 caches,” in Dept. Comp. Scie. , Univ. Wisconsin-Madison, Tech. Rep. 1500, Apr. 2004.
[8] J. Yang; R. Gupta; C. Zhang, “Frequent value encoding for low power data Buses,” in ACM Transactions on Design Automation of Electronic Systems, 9(3), July 2004.
[9] A. Macii et. al., “A New algorithms for energy driven data compression in VLIW embedded processors,” in Proc. IEEE Design Automation and Test Europe (DATE), Munich, Germany, pp. 24-29, Mar. 2003,
[10] H. Lekatsas et. al., “CRAMES: Compressed RAM for Embedded Systems,” in 2005. CODES+ISSS '05. Third IEEE/ACM/IFIP International Conference on Hardware/Software Co-design and System Synthesis, Jersey City, NJ, USA ,Sept. 2005, pp. 93-98.
[11] V. Bui, M.A. Kim, “The Cache and Codec Model for Storing and Manipulating Data ,”in IEEE Micro, vol. 34, no. 4, pp, 28-35, May 2014
[12] M. Thuresson, P. Stentrom, “Accommodation of the Bandwidth of Large Cache Blocks Using Cache/Memory Link Compression,“ in 2008. 37th International Conference on Parallel Processing, ICPP '08, Portland, Sep. 2008, pp. 478-486
[13] D. J. Cao; Y. Y. Lin, “即時性無失真壓縮編碼之研究,” 國立中央大學通訊工程學系博碩士論文, 桃園.
[14] S. Sardashti; D.A. Wood, “Decoupled Compressed Cache: Exploiting Spatial Locality for Energy Optimization,” in IEEE Micro, Vol. 34, no. 3, pp. 91-99, May-June 2014