簡易檢索 / 詳目顯示

回結果列表
研究生： 劉庭瑋
Liu, Ting-Wei
論文名稱： 使用 CUDA 函數庫實現 BiCGstab 與 BiCGstab(L)方法 的 GPU 平行化
GPU parallelization of BiCGstab and BiCGstab(L) methods using CUDA Libraries
指導教授： 陳人豪
Chen, Jen-Hao
口試委員: 劉晉良
Liu, Jinn-Liang
陳仁純
Chen, Ren-Chuen
學位類別： 碩士
Master
系所名稱： 南大校區系所調整院務中心 - 應用數學系所
應用數學系所(English)
論文出版年： 2018
畢業學年度： 106
語文別： 英文
論文頁數： 32
中文關鍵詞： 平行化 、ＣＵＤＡ函式庫 、BiCGstab 演算法 、BiCGstab(L) 演算法 、顯示卡計算 、壓縮矩陣
外文關鍵詞： parallelization, CUDA Libraries, BiCGstab, BiCGstab(L), GPU, interation
相關次數： 點閱：2  下載：0 
分享至:
查詢本校圖書館目錄 查詢臺灣博碩士論文知識加值系統 勘誤回報

    • 這篇論文提出了BiCGstab和BiCGstab($l$)演算法的GPU平行化。並使用兩個易於使用的CUDA函示庫cuBLAS和cuSPARSE來協助撰寫程式碼，這兩個函式庫最常被用來在執行線性代數運算,利用Poisson-Fermi模型構建的TRPV通道矩陣和佛羅里達大學稀疏矩陣集合中選取的矩陣來檢驗這兩種方法的性能。數值結果表明BiCGstab($l$)在有幾乎純虛根特徵值的矩陣中優於BiCGstab。

    This thesis presents the GPU parallelizations of the BiCGstab and BiCGstab($l$) algorithms. Two easy-to-use CUDA libraries, cuBLAS and cuSPARSE, are em- ployed to perform the linear algebra operations in the methods. The matrices constructed by Poisson-Fermi model for TRPV channel and selected from the university of florida sparse matrix collection are used to test the performance of these two methods. The numerical results show that BiCGstab($l$) outperforms BiCGstab in some matrices which have almost pure imaginary eigenvalues.

    Contents Abstract i Acknowledgement iii 1 Introduction 1 2 GPU computing 3 2.1 GPUVSCPU 3 2.2 CUDA 4 2.2.1 cuBLAS(CUDA Basic Linear Algebra Subroutines library) 5 2.2.2 cuSPARSE(CUDA Sparse Matrix library) 5 3 Algorithm 6 3.1 BiCGstab method. 6 3.2 BiCGstab(l) method 7 4 Parallelization 9 4.1 Compressed Sparse matrix(CSR) 9 4.2 Subroutine of Library 10 4.2.1 Subroutine of cuBLAS 11 4.2.2 Subroutine of cuSPARSE 12 4.3 Parallelization of BiCGstab and BiCGstab(l) 14 4.4 Computational cost of BiCGstab and BiCGstab(l) 20 5 Result 21 5.1 Example1. TRPV 22 5.2 Example2. ORSIRR1 24 5.3 Example3. fs1834 26 6 Conclusion 30 Reference 31

    References

    [1] Matrix Market. https://math.nist.gov/MatrixMarket/.
    [2] NVIDIA (2018) NVIDIA CUBLAS LIBRARY user guide 9.1.
    [3] NVIDIA (2018) NVIDIA CUDA C programming guide 9.1.
    [4] NVIDIA (2018) NVIDIA CUSPARSE LIBRARY user guide 9.1.
    [5] The University of Florida Fparse Matrix Collection. https://sparse.tamu.edu .
    [6] Jen-Hao Chen, Ren-Chuen Chen, and Jinn-Liang Liu. A gpu poisson-fermi solver for ion channel simulations. arXiv:1711.04060, Comput. Phys. Com- mun., 229:99–105, 2018.
    [7] Ruipeng Li and Yousef Saad. Gpu-accelerated preconditioned iterative linear solvers. The Journal of Supercomputing, 63(2):443–466, 2013.
    [8] Youcef Saad and Martin H Schultz. Gmres: A generalized minimal resid- ual algorithm for solving nonsymmetric linear systems. SIAM Journal on scientific and statistical computing, 7(3):856–869, 1986.
    [9] Yousef Saad. Krylov subspace methods for solving large unsymmetric linear systems. Mathematics of computation, 37(155):105–126, 1981.
    [10] Gerard LG Sleijpen and Diederik R Fokkema. Bicgstab (l) for linear equations involving unsymmetric matrices with complex spectrum. Electronic Transac- tions on Numerical Analysis, 1(11):2000, 1993.
    [11] Peter Sonneveld. Cgs, a fast lanczos-type solver for nonsymmetric linear systems. SIAM journal on scientific and statistical computing, 10(1):36–52, 1989.
    [12] Henk A Van der Vorst. Bi-CGSTAB: A fast and smoothly converging variant of Bi-CG for the solution of nonsymmetric linear systems. SIAM Journal on scientific and Statistical Computing, 13(2):631–644, 1992.
    [13] Wikipedia contributors. Cuda — Wikipedia, the free encyclopedia, 2018. [Online; accessed 8-May-2018].

    QR CODE