A vector arithmetic co-processor for personal computers, offering extended computational ability, has been developed by the Institute for Microelectronics, Stuttgart, Germany, and the Institute for Applied Mathematics, Karlsruhe. The two academic institutions displayed the CMOS co-processor, the MTM XPA3233, at Comdex/Fall in Las Vegas. The MTM XPA3233 has an Peripheral Component Interconnect interface, and has been developed to provide highly accurate calculations. The Germans say that the quality of elementary floating-point operations and roundings provided by the IEEE standards should be extended to the most frequent numerical data types or mathematical spaces of computation consisting of the four basic data types – real, complex, interval and complex interval – as well as the vectors and matrices over these types. The chip performs dot products of vectors with components of the Institute’s Double Data format to full accuracy or with only one final rounding. The products are accumulated into a very long fixed-point register. A special carry resolution technique is used. To speed the calculations, the Germans have developed extensions to the high level programming languages of Pascal and C: Pascal-XSC and C-XSC – XSC for Extended Scientific Computing, and add-ons to the common compilers that are needed to call the routine of the co-processor. In these languages, the chip’s functionality is directly coupled to the operator symbols in vector and matrix expressions, or to the operators dot product or matmul in Fortran-90. Access to the co-processor from other programming languages is achieved by subroutine calls to a special C subroutine library. The expanded computational capability is gained at modest cost and does not impose a performance penalty, say the designers. As of yet, the academics have not found a commercial manufacturer and it has been the Microelectronics Institute that has been making the chips, consequently there is no price. The makers admit, also, that there are not that many customers for this highly accurate type of computation but say that there are those that really need it. The co-processor has already been used in a power plant to calculate the speed at which the shaft of the turbine has to be accelerated: acceleration has to be done in stages as rotational speeds are critical if dangerous vibrations are to be avoided. The chip was used in these calculations.
