Instruction and logic for the detection of a numerical accumulation error

摘要

A processor includes circuitry for decoding at least one instruction and execution unit. The decoded statement can calculate a floating point result. The execution unit includes circuitry to execute the instruction, determine the floating point result, calculate the amount of lost accuracy in a mantissa of the floating point result, compare the amount of lost accuracy with a precision limit of the numerical accumulation error based determine on the comparison if a numerical accumulation error has occurred and write a value to a flag. The amount of lost precision corresponds to several lost bits in the mantissa of the floating point result. The value written to the flag may be based on the determination that the numerical accumulation error has occurred. The flag may serve to indicate that the numerical accumulation error has occurred.