A VARIABLE-PRECISION, INTERVAL ARITHMETIC PROCESSOR

This paper presents the software interface and hardware design of a pr ocessor which supports variable-precision, interval arithmetic. The so ftware interface gives the programmer the ability do specify the preci sion of the computation, determine the accuracy of the result, and rec ompute inaccurate results with higher precision. Special instructions and hardware features are provided for operations on matrices, vectors and complex numbers. Providing direct hardware support for variable-p recision, interval arithmetic greatly improves the accuracy and reliab ility of the computation, and is much faster than existing software te chniques for controlling numerical error. Area and delay estimates ind icate that the processor can be implemented on a single chip with a cy cle time that is comparable to existing IEEE double-precision floating point processors.