Section: New Results

Floating-point constraint solving

Participants : Matthieu Carlier, Arnaud Gotlieb.

Programs including floating-point computations are known to be hard-to-test. Generating test inputs for those programs requires solving constraints over floating-points computations, which led us to the development of specific constraint filtering techniques. In this work, we extended the Marre and Michel property regarding the use of internal floating-point representation to increase the filtering capabilities of addition to the case of multiplication/division. We came up with an optimized implementation of FPSE (our current FP constraint solver) that was able to deal with large C programs that include (non-linear) floating-point computations. We already got a first publication of this work [17] .