https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120996
--- Comment #11 from Dhruv Chawla <dhruvc at nvidia dot com> ---
Okay, looks like there's a sharing bug with compiler explorer where it doesn't
correctly show the Tree/RTL pane that was opened. Anyways, this is the diff I
see:
=== GCC 15.1.0
<bb 12> [local count: 939524096]:
# iftmp.0_15 = PHI <1.0e+0(10), iftmp.0_61(11)>
if (distbb_59 < elcdst_85)
goto <bb 14>; [50.00%]
else
goto <bb 13>; [50.00%]
<bb 13> [local count: 469762048]:
<bb 14> [local count: 939524096]:
# iftmp.1_16 = PHI <1.0e+0(12), 0.0(13)>
_65 = iftmp.0_15 * iftmp.1_16;
chrg_e_66 = chrg_init_35 * _65;
===
=== GCC trunk
<bb 11> [local count: 554319220]:
_60 = elcdst1_14 * distbb_59;
iftmp.0_61 = 1.0e+0 - _60;
if (distbb_59 < elcdst_86)
goto <bb 13>; [15.25%]
else
goto <bb 12>; [84.75%]
<bb 12> [local count: 469762048]:
<bb 13> [local count: 939524096]:
# iftmp.1_16 = PHI <1.0e+0(11), 0.0(12), 1.0e+0(10)>
# iftmp.0_87 = PHI <iftmp.0_61(11), iftmp.0_61(12), 1.0e+0(10)>
_65 = iftmp.1_16 * iftmp.0_87;
chrg_e_66 = chrg_init_35 * _65;
===
The iftmp PHI has been pushed down to bb 13 in the thread1 dump.
