Floating point architecture with tagged operands

Abstract

A method for representing arithmetic values on which arithmetic operations can be performed uses operands having a fixed number of bits. In a first step, a plurality of operands are stored in a memory, wherein each operand has a bit pattern representing a particular value. In a second step, a tag associated with each of the operands is also stored in the same or a different memory. Each of the tags has a tag value that indicates whether or not its associated operand represents an ordinary operand value or a special operand value. If the operand represents a special operand value, the tag value also indicates which of a predefined set of special operand values is represented by the associated operand. A result of an arithmetic operation can be generated by, in a first step, inputting at least a first operand and a first tag to an arithmetic section. In a second step, the arithmetic section determines a value of the result of the arithmetic operation based on the special operand indicated by the first tag and independently of a value of the first operand. A tag associated with a result of an arithmetic calculation can be generated by, in a first step, generating in an arithmetic section a result of an arithmetic operation on at least one operand. If the generation of the result produces one of a predetermined set of special operands, a tag generator generates a tag having a predetermined tag value corresponding to the produced special operand. Illustratively the tag generator alternatively generates a predetermined tag value indicating that the result is not a special operand if the result is not a special operand.