It isn't common to find articles on IEEE 754 arithmetic outside the
IEEE Computer Arithmetic conferences, but below are two new ones that
may be of interest to list members. The lead author of the second one
is a well-known GNU software developer and expert.
Online PDF versions are available at institutions with journal
subscriptions; although our library no longer gets paper copies of
this journal, I was able to get electronic copies this morning.
Otherwise, interlibrary loan is available even through local and
community libraries.
@String{j-SPE = "Soft\-ware\emdash Prac\-tice and Experience"}
@Article{Ogasawara:2004:OPO,
author = "Takeshi Ogasawara and Hideaki Komatsu and Toshio
Nakatani",
title = "Optimizing precision overhead for x86 processors",
journal = j-SPE,
volume = "34",
number = "9",
pages = "875--893",
day = "25",
month = jul,
year = "2004",
CODEN = "SPEXBL",
ISSN = "0038-0644",
bibdate = "Sat Apr 16 07:26:31 MDT 2005",
bibsource = "http://www.interscience.wiley.com/jpages/0038-0644;
http://www3.interscience.wiley.com/journalfinder.html";,
abstract = "It is a major challenge for a Java JIT compiler to
perform single-precision floating-point operations
efficiently for the x86 processors. In previous
research, the double-precision mode is set as the
default precision mode when methods are invoked.
Sophisticated approaches then use heuristic approaches
to optimization by considering the trade-offs between
roundings and mode switches. However, this convention
introduces redundant mode switches across method
boundaries. Furthermore, methods that include both
single- and double-precision operations cannot switch
the mode, even if single-precision operations are
dominant. We propose a new approach to these problems.
We eliminate redundant mode switches by ignoring the
default precision mode and calling a method in the same
precision mode as the caller. For methods that include
both single- and double-precision methods, we reduce
the overhead of rounding by isolating code segments of
a given method that should be executed in the
single-precision mode. We implemented our approach in
IBM's Just-in-Time compiler, and obtained experimental
results demonstrating that, in SPECjvm98, it
consistently shows the best performance in any
configuration of benchmark programs, inline policies,
and processor architectures compared with previous
research approaches.",
acknowledgement = ack-nhfb,
DOI = "10.1002/spe.596",
keywords = "floating-point operations; Java; precision control;
x86 processors",
onlinedate = "27 Apr 2004",
}
@Article{Eggert:2005:PEN,
author = "P. R. Eggert and D. S. Parker",
title = "Perturbing and evaluating numerical programs without
recompilation --- the wonglediff way",
journal = j-SPE,
volume = "35",
number = "4",
pages = "313--322",
day = "10",
month = apr,
year = "2005",
CODEN = "SPEXBL",
ISSN = "0038-0644",
bibdate = "Sat Apr 16 07:26:37 MDT 2005",
bibsource = "http://www.interscience.wiley.com/jpages/0038-0644;
http://www3.interscience.wiley.com/journalfinder.html";,
abstract = "wonglediff is a program that tests the sensitivity of
arbitrary program executables or processes to changes
that are introduced by a process that runs in parallel.
On Unix and Linux kernels, wonglediff creates a
supervisor process that runs applications and, on the
fly, introduces desired changes to their process state.
When execution terminates, it then summarizes the
resulting changes in the output files. The technique
employed has a variety of uses. This paper describes an
implementation of wonglediff that checks the
sensitivity of programs to random changes in the
floating-point rounding modes. It runs a program
several times, wongling it each time: randomly toggling
the IEEE-754 rounding mode of the program as it
executes. By comparing the resulting output, one gets a
poor man's numerical stability analysis for the
program. Although the analysis does not give any kind
of guarantee about a program's stability, it can reveal
genuine instability, and it does serve as a
particularly useful and revealing idiot light. In our
implementation, differences among the output files from
the program's multiple runs are summarized in a report.
This report is in fact an HTML version of the output
file, with inline mark-up summarizing individual
differences among the multiple instances. When viewed
with a browser, the differences can be highlighted or
rendered in many different ways.",
acknowledgement = ack-nhfb,
DOI = "10.1002/spe.637",
keywords = "diff; IEEE-754 floating point arithmetic; numerical
instability checking; random rounding; rounding modes;
sensitivity analysis",
onlinedate = "21 Dec 2004",
}
-------------------------------------------------------------------------------
- Nelson H. F. Beebe Tel: +1 801 581 5254 -
- University of Utah FAX: +1 801 581 4148 -
- Department of Mathematics, 110 LCB Internet e-mail: beebe@xxxxxxxxxxxxx -
- 155 S 1400 E RM 233 beebe@xxxxxxx beebe@xxxxxxxxxxxx -
- Salt Lake City, UT 84112-0090, USA URL: http://www.math.utah.edu/~beebe -
-------------------------------------------------------------------------------