https://gcc.gnu.org/bugzilla/show_bug.cgi?id=107117
--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> --- Btw, we only avoid inexact constant foldings for -frounding-math, not with -ftrapping-math (which guards FP exception state). See do_mpfr_ckconv. So we could do better here if we want (but then if at runtime the inexact exception is raised is unspecified).
