------- Comment #32 from rguenth at gcc dot gnu dot org 2007-10-24 15:02
-------
I have a patch to "fix" the heuristics, but it doesn't have effect as nobody
hoists PHI nodes that have become invariant apperantly. After PRE I see
<bb 100>:
# HEAP.202_828 = PHI <HEAP.202_747(150), HEAP.202_743(148)>
<bb 101>:
# SMT.222_639 = PHI <SMT.222_638(100), SMT.222_599(152)>
# dtvol_656 = PHI <dtvol_620(100), dtvol_655(152)>
# MPT.242_668 = PHI <MPT.242_744(100), MPT.242_668(152)>
# l_1 = PHI <1(100), l_428(152)>
# VUSE <MPT.242_668>
D.1241_376 = du.data;
... (more VUSEs of MPT.242_668, the only VDEFs in the loop follow)
# dtvol_655 = VDEF <dtvol_656>
# SMT.222_599 = VDEF <SMT.222_639>
(*D.1242_377)[D.1259_397] = D.1277_426;
l_428 = l_1 + 1;
if (l_1 == 5)
goto <bb 102>;
else
goto <bb 152>;
<bb 152>:
goto <bb 101>;
<bb 102>:
the
# SMT.222_639 = PHI <SMT.222_638(100), SMT.222_599(152)>
could be hoisted to BB100 (even to BB92). Who is supposed to do this?
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32921
