FASTER NUMERICAL ALGORITHMS VIA EXCEPTION HANDLING

被引：13

作者：

DEMMEL, JW ^{[1
]}

LI, XY ^{[1
]}

机构：

[1] UNIV CALIF BERKELEY,DEPT MATH,BERKELEY,CA 94720

来源：

IEEE TRANSACTIONS ON COMPUTERS | 1994年 / 43卷 / 08期

基金：

美国国家科学基金会;

关键词：

IEEE FLOATING POINT ARITHMETIC; EXCEPTION HANDLING; LINEAR ALGEBRA; LAPACK; SPEEDUP; NANS; BASIC LINEAR ALGEBRA SUBPROGRAMS;

D O I：

10.1109/12.295860

中图分类号：

TP3 [计算技术、计算机技术];

学科分类号：

0812 ;

摘要：

An attractive paradigm for building fast numerical algorithms is the following: 1) try a fast but occasionally unstable algorithm, 2) test the accuracy of the computed answer, and 3) recompute the answer slowly and accurately in the unlikely event it is necessary. This is especially attractive on parallel machines where the fastest algorithms may be less stable than the best serial algorithms. Since unstable algorithms can overflow or cause other exceptions, exception handling is needed to implement this paradigm safely. To implement it efficiently, exception handling cannot be too slow. We illustrate this paradigm with numerical linear algebra algorithms from the LAPACK library.

引用

页码：983 / 992

页数：10

共 19 条

[11] CONDITION ESTIMATES
HAGER, WW
[J]. SIAM JOURNAL ON SCIENTIFIC AND STATISTICAL COMPUTING, 1984, 5 (02): : 311 - 316
[12] FORTRAN CODES FOR ESTIMATING THE ONE-NORM OF A REAL OR COMPLEX MATRIX, WITH APPLICATIONS TO CONDITION ESTIMATION
HIGHAM, NJ
[J]. ACM TRANSACTIONS ON MATHEMATICAL SOFTWARE, 1988, 14 (04): : 381 - 396
[13] Kahan W, 1966, CS41 STANF U COMP SC
[14] Lawson C. L., 1979, ACM Transactions on Mathematical Software, V5, P324, DOI [10.1145/355841.355847, 10.1145/355841.355848]
[15] SITE R, 1992, ALPHA ARCHITECTURE R
[16] 1991, CONNECTION MACHINE C
[17] [No title captured]
[18] 1987, IEEE STANDARD RAIX I
[19] 1985, IEEE STANDARD BINARY

← 1 2 →