https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115719
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Version|unknown |15.0
Ever confirmed|0 |1
Last reconfirmed| |2024-07-01
Status|UNCONFIRMED |NEW
Target|X86_64 |x86_64-*-*
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
I can confirm that we do not thread the jumps. Note that the optimization
only happens on RTL (by jump-after-combine). On GIMPLE the 2 byte and
4 byte memcmp are inline expanded, but we are not able to thread
<bb 2> :
_1 = __builtin_memcmp (input_5(D), "ab", 2);
if (_1 == 0)
goto <bb 5>; [INV]
else
goto <bb 3>; [INV]
<bb 3> :
_2 = __builtin_memcmp (input_5(D), "abcd", 4);
if (_2 == 0)
goto <bb 5>; [INV]
else
goto <bb 4>; [INV]
<bb 4> :
<bb 5> :
# iftmp.1_3 = PHI <1(3), 0(4), 1(2)>
since we do not have "string ranges". I'm not sure if strlenopt would have
enough infrastructure to thread this or if it is even worth the trouble.
