https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87325
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEW |WAITING
Last reconfirmed|2018-09-17 00:00:00 |2023-12-5
--- Comment #4 from Richard Biener <rguenth at gcc dot gnu.org> ---
It now works "better":
<bb 3> [local count: 201326592]:
ivtmp.7_24 = (<unnamed-unsigned:24>) &ts;
<bb 4> [local count: 805306368]:
# i_20 = PHI <i_15(6), 0(3)>
# ivtmp.7_26 = PHI <ivtmp.7_25(6), ivtmp.7_24(3)>
_29 = (<address-space-7> void *) ivtmp.7_26;
_30 = _29 + 4;
_1 = MEM[(float *)_30];
_2 = MEM[(float *)_29];
_3 = bar (_2, _1);
r = _3;
_17 = ivtmp.7_26 + 8;
_4 = (const <address-space-7> float *) _17;
_5 = (const float *) _4;
_6 = foo (&r, _5);
if (_6 != 0)
note the (implicit) address-space convert for the 2nd argument to foo (),
the testcase may be actually invalid. The linked bug is still there,
but this one might be fixed by r14-6133-ge00c00730912cd?
Please verify and close if fixed. PR60075 tracks a remaining IVOPTs issue.
