Floating point number error detection method and device based on atomic state function

Abstract

The invention relates to a floating-point number error detection method and device based on an atomic state function. The invention inserts the source code or program to obtain the internal state of the program when it is running; calculates the atomic state function of each floating-point number operation through the internal state of the program when it is running, and excavates the internal relationship between the atomic state function and the floating-point number error, Infer the strong correlation between the atomic state function and the error; build a search framework, and use the search algorithm to obtain a test case set that will trigger a high-risk atomic state function; this test case set is the input that has a high error rate and triggers a significant floating-point error . Since the calculation of the atomic state function does not need to introduce high-precision floating-point numbers and does not depend on test oracle, the present invention does not depend on test oracle, has high error detection rate, low running overhead, does not rely on high-precision floating-point numbers, and has fast running time.

Description

technical field [0001] The invention belongs to the technical field of software testing, and relates to a numerical operation program and an error detection method of a floating-point number program, in particular to a floating-point number error detection method and device based on an atomic state function. The method significantly improves the error detection probability and operating efficiency of the floating-point error detection tool, enabling testers to detect more floating-point errors (defects) with larger errors in a shorter time during the testing process. Background technique [0002] Floating-point error detection is the most effective means to ensure the quality of numerical calculation programs and floating-point programs. Static analysis, dynamic testing and other methods are widely used in floating-point error detection technology. Numerical calculation programs and floating-point program testers use static analysis tools for error detection, use interval c...

AI Technical Summary

Problems solved by technology

In the absence of a test oracle, floating-point errors are hidden (no error message) and thus cannot be detected

[0007] (2) The existing method is based on black-box testing technology, and the error rate is not high

Due to the complexity of floating-point arithmetic, the numerical value of the error has no obvious correlation with the numerical value of the test case

The technology based on black-box testing does not attempt to obtain the runtime state of the program, so that the search algorithm can obtain very little information, and cannot effectively guide the direction of the search, resulting in poor search results and the inability to find test cases that trigger significant errors

[0008] (3) Existing technology relies on high-precision test oracle, and the running time is slow

A large amount of computational overhead makes the existing technology unable to be applied to large-scale projects at the runtime level, and is not practically feasible

Images