[Bug ipa/111157] [14 Regression] 416.gamess fails with a run-time abort when compiled with -O2 -flto after r14-3226-gd073e2d75d9ed4

["jamborm at gcc dot gnu.org" <gcc-bugzilla@gcc.gnu.org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111157

Martin Jambor <jamborm at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
           Assignee|unassigned at gcc dot gnu.org      |jamborm at gcc dot gnu.org
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2023-08-25

--- Comment #2 from Martin Jambor <jamborm at gcc dot gnu.org> ---
With the propagation, PRE performs the following:

 void point.constprop.isra (double ISRA.1740, int & restrict ipoint, double &
restrict x, double & restrict y, double & restrict z, int & restrict istat)
 {
   int iy;
   int ix;
   double & restrict prploc;
   int _5;
-  int _9;
   int _11;
   int _12;
   long int _13;
@@ -13284,7 +15377,6 @@
   <bb 2> [local count: 1073741824]:
   # DEBUG D#556 s=> prploc
   # DEBUG prploc => D#556
-  *istat_1(D) = 0;
   if (ISRA.1740_70(D) !=
6.08805302369215843745180305174265706886482420102590225625e-154)
     goto <bb 7>; [50.00%]
   else
@@ -13301,8 +15393,7 @@
   calcom (x_6(D), y_7(D), z_8(D));

   <bb 5> [local count: 536870912]:
-  _9 = *ipoint_4(D);
-  if (_9 > 1)
+  if (_5 > 1)
     goto <bb 6>; [59.00%]
   else
     goto <bb 25>; [41.00%]


The write of zero to where we know there is already zero is however
problematic because all callers expect the pointed to value to be
overwritten and dse2 pass does:

@@ -20229,7 +14076,6 @@
   xpt = 0.0;
   ypt = 0.0;
   zpt = 0.0;
-  istat = 0;
   point.constprop.isra (_73, &ipoint, &xpt, &ypt, &zpt, &istat);
   _77 = istat;
   if (_77 < 0)


So I guess this is a nasty interaction with IPA-modref, and indeed
-fno-ipa-modref avoids the issue.

I'll discuss with Honza whether IPA-modref can be modified to kill the
known values in summaries in such cases or whether it is IPA-CP that
should avoid "knowing" the value.