Abstract
This chapter presents the current best design and implementation practices for the acceleration of dense linear algebra (DLA) on GPUs. Examples are given with fundamental algorithms—from the matrix–matrix multiplication kernel written in CUDA to the higher level algorithms for solving linear systems, eigenvalue and SVD problems. The implementations are available through the MAGMA library—a redesign for GPUs of the popular LAPACK. To generate the extreme level of parallelism needed for the efficient use of GPUs, algorithms of interest are redesigned and then split into well-chosen computational tasks. The tasks execution is scheduled over the computational components of a hybrid system of multicore CPUs with GPU accelerators using either static scheduling or a light-weight runtime system. The use of light-weight runtime systems keeps scheduling overhead low, similar to static scheduling, while enabling the expression of parallelism through sequential-like code. This simplifies the development effort and allows the exploration of the unique strengths of the various hardware components.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
- 2.
The bandwidth for the current generation PCI Express is at most 16 GB/s while the devices achieve over 1,000 Gflop/s performance.
References
Anderson, E., Bai, Z., Bischof, C., Blackford, L.S., Demmel, J.W., Dongarra, J.J. Du Croz, J., Greenbaum, A., Hammarling, S., McKenney, A., Sorensen, D.: LAPACK Users’ Guide. SIAM, Philadelphia (1992). http://www.netlib.org/lapack/lug/
Bientinesi, P., Igual, F.D., Kressner, D., Quintana-Ortí, E.S.: Reduction to condensed forms for symmetric eigenvalue problems on multi-core architectures. In: Proceedings of the 8th International Conference on Parallel Processing and Applied Mathematics: Part I, PPAM’09, pp. 387–395. Springer, Berlin/Heidelberg (2010)
Dongarra, J.J., Sorensen, D.C., Hammarling, S.J.: Block reduction of matrices to condensed forms for eigenvalue computations. J. Comput. Appl. Math. 27(1–2), 215–227 (1989)
Gansterer, W., Kvasnicka, D., Ueberhuber, C.: Multi-sweep algorithms for the symmetric eigenproblem. In: Vector and Parallel Processing - VECPAR’98. Lecture Notes in Computer Science, vol. 1573, pp. 20–28. Springer, Berlin (1999)
Golub, G., Loan, C.V.: Matrix Computations, 3rd edn. Johns Hopkins, Baltimore (1996)
Haidar, A., Ltaief, H., Dongarra, J.: Parallel reduction to condensed forms for symmetric eigenvalue problems using aggregated fine-grained and memory-aware kernels. In: Proceedings of SC ’11, pp. 8:1–8:11. ACM, New York (2011)
Haidar, A., Ltaief, H., Luszczek, P., Dongarra, J.: A comprehensive study of task coalescing for selecting parallelism granularity in a two-stage bidiagonal reduction. In: Proceedings of the IEEE International Parallel and Distributed Processing Symposium, Shanghai, 21–25 May 2012. ISBN 978-1-4673-0975-2
Haidar, A., Tomov, S., Dongarra, J., Solca, R., Schulthess, T.: A novel hybrid CPU-GPU generalized eigensolver for electronic structure calculations based on fine grained memory aware tasks. Int. J. High Perform. Comput. Appl. 28(2), 196–209 (2014)
Haidar, A., Kurzak, J., Luszczek, P.: An improved parallel singular value algorithm and its implementation for multicore hardware. In: SC13, The International Conference for High Performance Computing, Networking, Storage and Analysis, Denver, CO, 17–22 November 2013
Lang, B.: Efficient eigenvalue and singular value computations on shared memory machines. Parallel Comput. 25(7), 845–860 (1999)
Ltaief, H., Luszczek, P., Haidar, A., Dongarra, J.: Enhancing parallelism of tile bidiagonal transformation on multicore architectures using tree reduction. In: Wyrzykowski, R., Dongarra, J., Karczewski, K., Wasniewski, J. (eds.) Proceedings of 9th International Conference, PPAM 2011, Torun, vol. 7203, pp. 661–670 (2012)
MAGMA 1.4.1: http://icl.cs.utk.edu/magma/ (2013)
Nath, R., Tomov, S., Dong, T., Dongarra, J.: Optimizing symmetric dense matrix-vector multiplication on GPUs. In: 2011 International Conference for High Performance Computing, Networking, Storage and Analysis (SC), pp. 1–10. New York, NY, USAm 2011, ACM
Strazdins, P.E.: Lookahead and algorithmic blocking techniques compared for parallel matrix factorization. In: 10th International Conference on Parallel and Distributed Computing and Systems, IASTED, Las Vegas, 1998
Strazdins, P.E.: A comparison of lookahead and algorithmic blocking techniques for parallel matrix factorization. Int. J. Parallel Distrib. Syst. Netw. 4(1), 26–35 (2001)
Tomov, S., Nath, R., Dongarra, J.: Accelerating the reduction to upper Hessenberg, tridiagonal, and bidiagonal forms through hybrid GPU-based computing. Parallel Comput. 36(12), 645–654 (2010)
Yamazaki, I., Dong, T., Solcà, R., Tomov, S., Dongarra, J., Schulthess, T.: Tridiagonalization of a dense symmetric matrix on multiple GPUs and its application to symmetric eigenvalue problems. Concurr. Comput. Pract. Exp. (2013). doi:10.1002/cpe.3152
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2014 Springer International Publishing Switzerland
About this chapter
Cite this chapter
Dongarra, J. et al. (2014). Accelerating Numerical Dense Linear Algebra Calculations with GPUs. In: Kindratenko, V. (eds) Numerical Computations with GPUs. Springer, Cham. https://doi.org/10.1007/978-3-319-06548-9_1
Download citation
DOI: https://doi.org/10.1007/978-3-319-06548-9_1
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-06547-2
Online ISBN: 978-3-319-06548-9
eBook Packages: Computer ScienceComputer Science (R0)