Open RyanGlScott opened 3 years ago
I think the main issue here is that 80-bit X86 floats don't quite correspond to any IEEE-754 floating-point type. The libBF
library might still be able to faithfully represent them, but it would require some pretty careful management. For symbolic values, they would still work fine for the "real" and "uninterpreted" modes, but I don't know that we could reasonably connect them to a bit-precise solver mode. Maybe we could play some games and fit them into some carefully-chosen IEEE format, but I'm not sure.
The following program:
Will trip up
crux-llvm
:This is the LLVM bitcode produced by Clang 10.0.0:
I originally discovered this by way of SV-COMP's
floats-esbmc-regression/digits_for.c
benchmark program.