Java Numerics Working Group Meeting

West Coast Grand Hotel

Seattle, Washington

November 3, 2002

An open meeting of the Java Numerics Working Group of the Java Grande Forum took place at the ACM Java Grande/ISCOPE Conference held in November in Seattle. This represents the minutes of this meeting.

I. Update on Java 1.5

Joe Darcy provided an update of work going on at Sun of interest to the Working Group. The following points were covered.

• Version 1.4.1 is currently shipping from Sun. Version 1.4.2, which will include routine bug fixes and performance enhancements, is now being prepared.
• Version 1.5 is currently under active development, though its final set of features is as yet undetermined. Among the items being considered are the following.
• Augmenting java.lang.Math (and java.lang.StrictMath) to be on par with the C math library (libm). See Section II below.
• Adoption of JSR 13, which provides better control over rounding precision in java.math.BigDecimal. (This class provides methods for performing extended precision decimal arithmetic.)
• Adoption of JSR 14, generics. This proposal does not include generics for primitive types, which is a severe limitation for the numerics community. Registered Java Developer’s Connection (JDC) members can get a prototype of the JSR14 generics compiler from http://developer.java.sun.com/developer/earlyAccess/adding_generics/
• Inclusion of static imports. As the name implies, static import only imports static methods (such as the methods in java.lang.Math). This is a namespace convenience, allowing static method names from specifically designated imported packages to be abbreviated, e.g. sin(x) rather than Math.sin(x).
• Inclusion of facilities for automatic boxing and unboxing of primitive types. This would allow primitives types to be automatically promoted to objects for passing to methods, which require objects. This might allow the extension of generics to primitives. Some performance issues must be resolved for this, however.
• Inclusion of enhanced for loops (i.e., foreach), for programmer convenience.
• Inclusion of type-safe enums. The type-safe enum pattern is discussed in Josh Bloch's book "Effective Java."

II. Improvements in java.lang.Math

Joe Darcy has proposed the inclusion of additional methods in java.lang.Math with the goal of making this library on par with the C math library (libm). The following functions have been proposed.

§ Hyperbolic functions (e.g., sinh)

§ Inverse hyperbolic functions

§ IEEE 754 recommended functions

§ Fused multiply-add

The Numerics Working Group endorsed this work, registering their support on the Java Developer’s Connection web site. Methods with the same name will also be included in java.lang.StrictMath. These will be based on existing implementations from Sun’s fdlibm.

An ad-hoc subcommittee will be formed to consider the addition of additional functions in the Java math library. One example is the reciprocal square root, which can be computed very quickly on some processors.

III. Multiarray Package

Jose Moreira provided an update on the progress of Java Specification Request (JSR) 83 on multiarrays. A multiarray is a rectangular multidimensional array whose shape and extents are fixed. Such arrays admit compiler optimizations that Java arrays of arrays do not. The current plan is to develop a package of standard classes to provide multiarrays. Support for 0 dimensional arrays (scalars) through 7 dimensional arrays are planned. Multiarrays will be composable from all Java primitive types. The expert group has developed a proposed class hierarchy for multiarrays. A complete specification is expected in a few months. The expert group will provide the implementation of the class, and the University of Manchester will develop conformance tests.

Although not planned as part of JSR-083, a preprocessor illustrating a proposed syntax for multiarrays in Java will also be developed. The preprocessor will map the new syntax into method calls to the multiarray package. This may provide the basis for a future JSR to add multiarray syntax to the Java language itself.

IV. Visual Numeric’s JMSL Library

John Brophy reported on the recently released JMSL library from Visual Numerics, Inc. JMSL contains classes for mathematics, statistics, financial computations, and charting. The math class contains methods for linear algebra, interpolation, approximation, quadrature, ordinary differential equations, transforms, nonlinear equations, optimization, and special functions. The statistics class contains methods for basic statistics, nonparametric tests, regression, ANOVA, time series, goodness of fit tests, distribution functions, and random number generation. While not as comprehensive as the IMSL Fortran and C libraries, JMSL does provide a greater coverage of mathematics and statistics than other math packages for Java. The web page for the library is http://www.vni.com/jmsl/.

V. Additional Issues

Members of the working group re-affirmed their interest in having the following features in Java.

1. True multidimensional arrays indexed using specialized notation, e.g. A[I,j,k].
2. Operator overloading to support the easy expression of alternate arithmetics.
3. Complex numbers that are as efficient as primitive types.
4. A new floating-point mode (i.e., fastfp) that admits the use of fused multiply-add operations in Java, and possibly admits additional compiler optimizations, such as the use of associativity.

Locating champions to carry out the technical work of the group remains an issue.

VI. Future of the Working Group

Attendees agreed that the work of the Working Group remained valuable and should continue. It was suggested that more frequent meetings of the working group (e.g., at six month intervals) would be beneficial. A focused Java Grande workshop on Java numerics issues is planned for 2003.

VII. Attendees

Ron Boisvert, NIST

John Brophy, Visual Numerics

Bryan Carpenter, Indiana University

Joe Darcy, Sun

Han-Ku Lee, Indiana University

Sang Lim, Indiana University

Mark Little, SAS Institute

Mikel Lujan, University of Manchester

Jose Moreira, IBM Research

Roldan Pozo, NIST

Keith Seymour, University of Tennessee

Jaime Spacco, University of Maryland