## Underflow Revisited (1999)

Citations: | 1 - 0 self |

### BibTeX

@MISC{Cuyt99underflowrevisited,

author = {Annie Cuyt and Peter Kuterna and Brigitte Verdonk and Dennis Verschaeren},

title = {Underflow Revisited},

year = {1999}

}

### OpenURL

### Abstract

Underflow is a floating-point phenomenon. Although the use of gradual underflow as defended in [2] and [5] is now widespread, most numerical analysts may not be aware of the fact that several implementations of the same principle are in existence, leading to different behaviour of code on different platforms, mainly with respect to exception signaling. We intend to thoroughly discuss the slight differences among these implementations. Examples will be taken from current hardware and from our own multiprecision software class library. Throughout the discussion the focus is on the analysis of the phenomenon and not on any implementation issues. Many programmers are also unaware of the fact that the IEEE 754 and 854 standards do not guarantee that a program will deliver identical results on all conforming systems. Of all the differences that can occur cross-platform, the underflow exception is just one. ? Research Director FWO-Vlaanderen ffl Supported by an NOI-grant from the Universi...

### Citations

37 |
Underflow and the reliability of numerical software
- Demmel
- 1984
(Show Context)
Citation Context ...erg@uia.ua.ac.be Keywords Computer arithmetic, IEEE floating-point, underflow exception Abstract Underflow is a floating-point phenomenon. Although the use of gradual underflow as defended in [2] and =-=[5]-=- is now widespread, most numerical analysts may not be aware of the fact that several implementations of the same principle are in existence, leading to different behaviour of code on different platfo... |

27 |
A proposed radix and word-length independent standard for floating point arithmetic
- Cody, Karpinski
(Show Context)
Citation Context ...at j 6= 0 has occurred in (2). The IEEE standards [6, 8] relax this condition in the sense that the underflow exception should be signaled "at least" when j 6= 0. A commentary to the standar=-=d however [1]-=- encourages the stricter criterion for setting the underflow flag. That is, it should be set whenever the delivered result is different from what would be delivered in a system with the same precision... |

24 |
Contributions to a proposed standard for binary floating-point arithmetic
- COONEN
- 1984
(Show Context)
Citation Context ...s library developed by one of the authors, which supports a U implementation. The consistency of each implementation was tested using a large set of test vectors which is a generalization of Coonen's =-=[3]-=- and Hough's [7] sets of test vectors and which will be publicly available soon [4]. For each of the precisions (single, double, 64 bit extended on INTEL and 113 bit quadruple on SUN) our test set for... |

6 |
Underflow and the denormalized numbers
- Coonen
- 1981
(Show Context)
Citation Context ...k, vschaerg@uia.ua.ac.be Keywords Computer arithmetic, IEEE floating-point, underflow exception Abstract Underflow is a floating-point phenomenon. Although the use of gradual underflow as defended in =-=[2]-=- and [5] is now widespread, most numerical analysts may not be aware of the fact that several implementations of the same principle are in existence, leading to different behaviour of code on differen... |

2 | A precision independent tool for testing floating-point arithmetic II: conversions
- Cuyt, Verdonk, et al.
- 2000
(Show Context)
Citation Context ...onsistency of each implementation was tested using a large set of test vectors which is a generalization of Coonen's [3] and Hough's [7] sets of test vectors and which will be publicly available soon =-=[4]-=-. For each of the precisions (single, double, 64 bit extended on INTEL and 113 bit quadruple on SUN) our test set for multiplication contained 1152 cases of U underflow, an extra 176 cases of V:U unde... |

1 |
et al. UCBTEST, a suite of programs for testing certain difficult cases of IEEE 754 floating-point arithmetic. Restricted public domain software from http://netlib.belllabs. com/netlib/fp/index.html
- Hough
(Show Context)
Citation Context ...ped by one of the authors, which supports a U implementation. The consistency of each implementation was tested using a large set of test vectors which is a generalization of Coonen's [3] and Hough's =-=[7]-=- sets of test vectors and which will be publicly available soon [4]. For each of the precisions (single, double, 64 bit extended on INTEL and 113 bit quadruple on SUN) our test set for multiplication ... |

1 |
numerical computation guide, Revision A
- Sun
- 1995
(Show Context)
Citation Context ...1) and (2). We also have to point out that at most one of ffl and j is nonzero, in other words fflj = 0, and that moreover always j = 0 when the arithmetic operation is either addition or subtraction =-=[9]-=-. It is easy to prove that the use of denormalized numbers allows to represent all tiny results of the floating-point addition or subtraction operation without rounding error, in other words exactly. ... |

1 |
A class library for multiple precision IEEE floating-point arithmetic
- Verschaeren
(Show Context)
Citation Context ...ext. The IEEE standard however, as it is formulated now, admits this deviation from the principle of exact rounding, for the sake of extended-based hardware platforms. 2.3 The MpIeee class library In =-=[10]-=- a multiprecision, highly performant and yet fully IEEE compliant class library is discussed, with user definable precision t and base 2sfis2 24 . Great care has been taken to support all IEEE feature... |