Systems and methods for model checking the precision of programs employing floating-point operations

摘要

Methods and systems for verifying the precision of a program that utilizes floating point operations are disclosed. Interval and affine arithmetic can be employed to build a model of the program including floating point operations and variables that are expressed as reals and integers, thereby permitting accurate determination of precision loss using a model checker. Abstract interpretation can be also employed to simplify the model. In addition, counterexample-guided abstraction refinement can be used to refine the values of parametric error constants introduced in the model.