https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121522
--- Comment #3 from Filip Kastl <pheeck at gcc dot gnu.org> ---
>From the transformations that sccopy1 does, this one causes the problem:
;; Function main (main, funcdef_no=3, decl_uid=2989, cgraph_uid=5,
symbol_order=4)
+Replacing SCC of size 3
int main ()
{
int r;
@@ -245,20 +246,19 @@
<bb 3> :
# l_27 = PHI <8(2), o_29(5)>
# o_28 = PHI <0(2), o_29(5)>
- # .MEM_32 = PHI <.MEM_42(D)(2), .MEM_33(5)>
v:
_2 = 1 / l_27;
r_43 = _2 + -805306366;
- # VUSE <.MEM_32>
+ # VUSE <.MEM_42(D)>
_4 = j (-3, 30, -4, -4, -4, -4, -4, 32, -4);
_5 = _4 + -445113689;
_6 = _5 - r_43;
- # VUSE <.MEM_32>
+ # VUSE <.MEM_42(D)>
_7 = j (-2, 23, -3, -3, -3, -3, -3, 22, -3);
_8 = _7 + -1062535215;
_9 = _8 * 2147483647;
_10 = _6 + _9;
- # VUSE <.MEM_32>
+ # VUSE <.MEM_42(D)>
_11 = j (-1, 1, -2, -2, -2, -2, -2, 1, -2);
_12 = _11 + -950633090;
_13 = _12 * -1342177282;
@@ -275,14 +275,12 @@
<bb 5> :
# o_29 = PHI <o_28(3), o_51(7)>
- # .MEM_33 = PHI <.MEM_32(3), .MEM_34(7)>
// predicted unlikely by goto predictor.
goto <bb 3>; [INV]
<bb 6> :
# o_30 = PHI <o_28(4), o_51(7)>
# r_31 = PHI <r_43(4), r_48(7)>
- # .MEM_34 = PHI <.MEM_32(4), .MEM_34(7)>
w:
r_48 = r_31 + 536870911;
_24 = _7 + -1062535222;
@@ -303,7 +301,7 @@
goto <bb 6>; [INV]
<bb 8> :
- # VUSE <.MEM_34>
+ # VUSE <.MEM_42(D)>
return 0;
}
All the removed PHIs (.MEM_32, .MEM_33, .MEM_34) just reference each other.
That's the situation the pass was made for. For normal PHIs, this cleanup is
ok. Is it not ok for memory PHIs?
