Title : 
Mixed abstractions for floating-point arithmetic
         
        
            Author : 
Brillout, Angelo ; Kroening, Daniel ; Wahl, Thomas
         
        
        
        
        
        
            Abstract : 
Floating-point arithmetic is essential for many embedded and safety-critical systems, such as in the avionics industry. Inaccuracies in floating-point calculations can cause subtle changes of the control flow, potentially leading to disastrous errors. In this paper, we present a simple and general, yet powerful framework for building abstractions from formulas, and instantiate this framework to a bit-accurate, sound and complete decision procedure for IEEE-compliant binary floating-point arithmetic. Our procedure benefits in practice from its ability to flexibly harness both over- and underapproximations in the abstraction process. We demonstrate the potency of the procedure for the formal analysis of floating-point software.
         
        
            Keywords : 
avionics; error statistics; floating point arithmetic; real-time systems; safety-critical software; IEEE compliant binary; avionics industry; bit accurate sound; building abstraction formula; complete decision procedure; embedded systems; floating point arithmetic; floating point calculations; floating point software; mixed abstractions; safety critical systems; Aerospace electronics; Computer errors; Computer industry; Digital arithmetic; Electrical equipment industry; Embedded computing; Error correction; Floating-point arithmetic; Laboratories; Logic;
         
        
        
        
            Conference_Titel : 
Formal Methods in Computer-Aided Design, 2009. FMCAD 2009
         
        
            Conference_Location : 
Austin, TX
         
        
            Print_ISBN : 
978-1-4244-4966-8
         
        
            Electronic_ISBN : 
978-1-4244-4966-8
         
        
        
            DOI : 
10.1109/FMCAD.2009.5351141