Following news of the embarrassing bug in the floating point unit of Intel Corp’s Pentium chip, Mathworks Inc, Natick, Massachusetts has stepped into the breech and posted to the Internet a description of the way in which it patched its own software. The company also intends to distribute a modification to Matlab, its numeric computation and graphic visualisation software, which overcomes the troublesome floating point computational error. Matlab provides maths and visualisation functionality, specialised applications toolboxes, and a high level language primarily targeted at technical professionals for designing and prototyping. MathWorks promotes Matlab as an open and programmable environment, featuring object-oriented graphics, sparse matrix support and sound output. Simulink extends Matlab to provide a block diagram environment for modelling, analysing and simulating a broad range of dynamic non-linear systems. The modifications to Matlab compensate for the flaw in the Pentium hardware. The probability of generating an erroneous result in a typical Matlab computation is actually quite low, it explains. But the real concern is loss of confidence. We do not want our customers to have to worry that arithmetic operations are not being done correctly. Unlike traditional visual data analysis packages, which tend to be stand-alone programs that post-process data from other sources, Matlab enables the user to analyse, transform and visualise within a single environment. Matlab uses an underlying graphical framework, called Handle Graphics. Mathworks says this enables easy customisation of plots. Users can open multiple graph windows simultaneously, place axes anywhere in a window and control where printed output appears on the page. More importantly, says MathWorks, the pictures are ‘alive’. Matlab’s internal code is optimised in C, with inner loops hand coded in assembly language. Matlab also provides array notations for manipulating data and evaluating mathematical functions. When a Matlab function is applied to a vector, the function operates on each element of the vector. Matlab has application toolboxes for signal processing, automatic control and neural networks. Each is built on top of Matlab’s numerics with access to the toolbox source code, so the user is able to inspect, customise and extend the algorithms and functionality of the toolbox to suit the need. The toolboxes can be used together in a seamless manner, for instance applying optimisation and neural network tools to advance signal processing problems and display the results as colour three dimensional graphs-within a single environment. Information on the approach that was used by MathWorks to compensate for the error in the Pentium microprocessor can be accessed on the Internet through the newsgroups comp.soft-sys.matlab and comp.sys.intel, or via the MathWorks home page on the World Wide Web, http://www.mathworks.com.
