Method for rapidly processing floating-point operations which involve exceptions

摘要

A method for processing exception conditions in a floating-point system (40) begins by determining in hardware that an exception will occur (14) for a given floating-point operation and operand(s). After the hardware determines that an exception is to occur, an exception handler is invoked from software. The exception handler numerically alters the operands to create biased operands which may be processed by the hardware (20) via the given floating-point operation without creating another exception condition. The biased operands are then processed by the hardware via the given floating-point operation to produce a biased floating-point result. The biased floating-point result is unbiased (22) by the exception handler to achieve an unbiased result. The speed of floating-point exception handling is improved by using biasing techniques and both software and hardware to mathematically resolve floating-point exceptions to correct unbiased results.