https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121595
--- Comment #4 from Matteo Nicoli <matteo.nicoli001 at gmail dot com> --- (In reply to Andrew Pinski from comment #3) > (In reply to Matteo Nicoli from comment #2) > > This optimization must still be available when using the -ffast-math option, > > right? > > No, just -fno-trapping-math is enough to be able to optimize away the > `+0.0`. Basically `a+0.0` can cause a trap and currently GCC's default is > trapping-math. > > The reason why `fabs(a+0.0)` can be optimized to just `fabs(a)` is because > while `a+0.0` might change the sign; it is ignored with fabs as fabs will > always zero out the sign bit. > > Note also `copy_sign(a+0.0, b)` can also be optimized to just `copy_sign(a, > b)` for similar reasons (that is the sign bit of `a+0.0` is ignored so the > `+0.0` part can be dropped). Thank you Andrew
