https://gcc.gnu.org/bugzilla/show_bug.cgi?id=99069
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, but phiopt2 turns it into
bool f (bool x, bool y)
{
bool _5;
bool _6;
<bb 2> [local count: 1073741824]:
_5 = x_2(D) & y_3(D);
_6 = -_5;
return _6;
and we end up with
_Z1fbb:
.LFB0:
.cfi_startproc
andl %edi, %esi
movl %esi, %eax
andl $1, %eax
ret
which is optimal. On GIMPLE we fail to elide -bool to bool I guess.
