Abstract
This Mini-Symposium consisted of two back to back sessions, each consisting of five presentations, held on the afternoon of Monday, June 21, 2004. A major theme of both sessions was novel data structures for the matrices of dense linear algebra, DLA. Talks one to four of session one all centered on new data layouts of matrices. Cholesky factorization was considered in the first three talks and a contrast was made between a square block hybrid format, a recursive packed format and the two standard data structures of DLA, full and packed format. In both talks one and two, the first two new formats led to level three high performance implementations of Cholesky factorization while using exactly the same amount of storage that standard packed format required. Of course, full format requires twice the amount of storage of the other three formats. In talk one, John Reid presented a definitive study of Cholesky factorization using a standard block based iterative Cholesky factorization, [1]. This factorization is typical of Lapack type factorizations; the major difference of [1] is the type of data structure it uses: talk one uses square blocks of order NB to represent a lower (upper) triangular matrix. In talk two, Jerzy Waśniewski presented the recursive packed format and its related Cholesky factorization algorithm, [2]. This novel format gave especially good Cholesky performance for very large matrices. In talk three, Jerzy Waśniewski demonstrated a detailed tuning strategy for talk one and presented performance results on six important platforms, Alpha, IBM, Intel, Itanium, SGI and Sun. The performance runs covered the algorithms of talks one and two as well as Lapack’s full and packed Cholesky codes, [3]. Overall, the square block hybrid method was best but was not a clear winner. The recursive method suffered because it did not combine blocking with recursion, [4]. Talk four, presented by Fred Gustavson, had a different flavor. Another novel data format was described which showed that two standard full format arrays could represent a triangular or symmetric matrix using only a total storage that was equal to the storage of standard packed storage, [5]. Therefore, this format has the same desirable property of standard full format arrays: one can use standard level 3 BLAS, [6] as well as some 125 or so full format Lapack symmetric / triangular routines on it. Thus new codes written for the new format are trivial to produce as they mostly consist of just calls to already existing codes. The last talk of session one, by James Sexton was on the massively
Similar content being viewed by others
References
Andersen, B.S., Gunnels, J.A., Gustavson, F.G., Reid, J.K., Waśniewski, J.: A Fully Portable High Performance Minimal Storage Hybrid Cholesky Algorithm. ACM TOMS 31(2), 201–227 (2005)
Andersen, B.S., Gustavson, F.G., Waśniewski, J.: A Recursive Formulation of Cholesky Factorization of a Matrix in Packed Storage. ACM TOMS 27(2), 214–244 (2001)
Anderson, E., Bai, Z., Bischof, C., Blackford, L.S., Demmel, J.: LAPACK Users’ Guide, 3rd edn. Society for Industrial and Applied Mathematics, pub.
Gustavson, F.G., Jonsson, I.: Minimal Storage High Performance Cholesky via Blocking and Recursion. IBM Journal of Research and Development 44(6), 823–849 (2000)
Gunnels, J.A., Gustavson, F.G.: A New Array Format for Symmetric and Triangular Matrices. In: Dongarra, J., Madsen, K., Waśniewski, J. (eds.) PARA 2004. LNCS, vol. 3732, pp. 247–255. Springer, Heidelberg (2006)
Dongarra, J.J., Du Croz, J., Hammarling, S., Duff, I.S.: A Set of Level 3 Basic Linear Algebra Subprograms. ACMTransactions on Mathematical Software 16(1), 1–17 (1990)
The BlueGene/L Supercomputer Architecture. J. C. Sexton. Slides on Para04 website
Scholtes, C.: A Method to Derive the Cache Performance of Irregular Applications on Machines with Direct Mapped Caches. Slides on Para04 website
Gunnels, J.A., Gustavson, F.G., Henry, G.M., van de Geijn, R.A.: A Family of High- Performance Matrix Multiplication Algorithms. In: Dongarra, J., Madsen, K., Waśniewski, J. (eds.) PARA 2004. LNCS, vol. 3732, pp. 256–265. Springer, Heidelberg (2006)
Gunnels, J.A., Gustavson, F.G., Pingali, K., Yotov, K.: A General Purpose Compiler that obtains High performance for Some Common Dense Linear Algebra Codes. Slides on Para04 website
Clint Whaley, R., Petitet, A., Dongarra, J.J.: Automated Empirical Optimization of Software and the ATLAS Project. LAWN Report #147, pp. 1-33 (September 2000)
Barnes, D.J., Hopkins, T.R.: Applying Software Testing Metrics to Lapack. In: Dongarra, J., Madsen, K., Waśniewski, J. (eds.) PARA 2004. LNCS, vol. 3732, pp. 228–236. Springer, Heidelberg (2006)
Drackenberg, N.P.: A Matrix-type for Performance Portability. In: Dongarra, J., Madsen, K., Waśniewski, J. (eds.) PARA 2004. LNCS, vol. 3732, pp. 237–246. Springer, Heidelberg (2006)
Author information
Authors and Affiliations
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2006 Springer-Verlag Berlin Heidelberg
About this paper
Cite this paper
Gustavson, F.G., Waśniewski, J. (2006). High Performance Linear Algebra Algorithms: An Introduction. In: Dongarra, J., Madsen, K., Waśniewski, J. (eds) Applied Parallel Computing. State of the Art in Scientific Computing. PARA 2004. Lecture Notes in Computer Science, vol 3732. Springer, Berlin, Heidelberg. https://doi.org/10.1007/11558958_26
Download citation
DOI: https://doi.org/10.1007/11558958_26
Publisher Name: Springer, Berlin, Heidelberg
Print ISBN: 978-3-540-29067-4
Online ISBN: 978-3-540-33498-9
eBook Packages: Computer ScienceComputer Science (R0)