[Bug ipa/61602] [4.10 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[Bug ipa/61602] [4.10 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |lto
                 CC|                            |hubicka at gcc dot gnu.org
          Component|lto                         |ipa
   Target Milestone|---                         |4.10.0
            Summary|ICE in lto1 on              |[4.10 Regression] ICE in
                   |x86_64-linux-gnu in         |lto1 on x86_64-linux-gnu in
                   |ipa_single_use, at          |ipa_single_use, at
                   |ipa.c:1257                  |ipa.c:1257

[Bug ipa/61602] [4.10 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[su at cs dot ucdavis.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

--- Comment #1 from Zhendong Su <su at cs dot ucdavis.edu> ---
To help diagnose the issue, here is another test case that triggers the same
ICE without using LTO.  

$ gcc-trunk -v
Using built-in specs.
COLLECT_GCC=gcc-trunk
COLLECT_LTO_WRAPPER=/usr/local/gcc-trunk/libexec/gcc/x86_64-unknown-linux-gnu/4.10.0/lto-wrapper
Target: x86_64-unknown-linux-gnu
Configured with: ../gcc-trunk/configure --prefix=/usr/local/gcc-trunk
--enable-languages=c,c++ --disable-werror --enable-multilib
Thread model: posix
gcc version 4.10.0 20140708 (experimental) [trunk revision 212357] (GCC) 
$ 
$ gcc-trunk -O1 -c small.c         
small.c:9:1: internal compiler error: in ipa_single_use, at ipa.c:1255
 }
 ^
0x81fc21 ipa_single_use
    ../../gcc-trunk/gcc/ipa.c:1255
0x81fc21 execute
    ../../gcc-trunk/gcc/ipa.c:1303
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <http://gcc.gnu.org/bugs.html> for instructions.
$ 


-------------------------------------------


static int a, *b = &a, **c = &b, ***d = &c;

int
main ()
{
  d = 0;
  b = 0;
  return 0;
}

[Bug ipa/61602] [4.10 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2014-07-09
                 CC|                            |jakub at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Started with r211925 (at least the #c1 testcase).

[Bug ipa/61602] [4.10 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[su at cs dot ucdavis.edu]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

--- Comment #3 from Zhendong Su <su at cs dot ucdavis.edu> ---
Hi, it would be really nice if this one could be fixed very soon as it keeps
popping up during my testing. Many thanks!

[Bug ipa/61602] [5 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[tbsaunde at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

tbsaunde at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |tbsaunde at gcc dot gnu.org

--- Comment #4 from tbsaunde at gcc dot gnu.org ---
So, during the single use pass we have 3 variables in symtab a, b and d. 
a->ref_list.referring is empty so the map doesn't contain an entry for a.  So 
single_use_map is only { b-> main, d -> main } and we assert because
single_use_map.get(a) returns NULL.

However I'm not sure if the problem is the assert being incorrect or if
a.ref_list.referring should contain main.

[Bug ipa/61602] [5 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> ---
Still broken - Honza?

[Bug ipa/61602] [5 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[trippels at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

Markus Trippelsdorf <trippels at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |trippels at gcc dot gnu.org

--- Comment #6 from Markus Trippelsdorf <trippels at gcc dot gnu.org> ---
An overnight csmith run came up with the same issue:

 ~ % cat crash1.i
int a;
int *b = &a, **c = &b;
int
main ()
{
  int **d = &b;
  *d = 0;
}

 ~ % gcc -O2 -flto crash2.i
lto1: internal compiler error: in ipa_single_use, at ipa.c:1299

[Bug ipa/61602] [5 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |hubicka at gcc dot gnu.org

--- Comment #7 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Mine.

[Bug ipa/61602] [5 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

--- Comment #8 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
OK, the problem is that the variable becomes unreachable after other variable
is declared writeonly and its constructor is killed. I am testing:
Index: ipa.c
===================================================================
--- ipa.c       (revision 218730)
+++ ipa.c       (working copy)
@@ -714,14 +714,18 @@ set_readonly_bit (varpool_node *vnode, v
 /* Set writeonly bit and clear the initalizer, since it will not be needed. 
*/

 bool
-set_writeonly_bit (varpool_node *vnode, void *data ATTRIBUTE_UNUSED)
+set_writeonly_bit (varpool_node *vnode, void *data)
 {
   vnode->writeonly = true;
   if (optimize)
     {
       DECL_INITIAL (vnode->decl) = NULL;
       if (!vnode->alias)
-       vnode->remove_all_references ();
+       {
+         if (vnode->num_references ())
+           *(bool *)data = true;
+         vnode->remove_all_references ();
+       }
     }
   return false;
 }
@@ -739,15 +743,18 @@ clear_addressable_bit (varpool_node *vno
 /* Discover variables that have no longer address taken or that are read only
    and update their flags.

+   Return true when unreachable symbol removan should be done.
+
    FIXME: This can not be done in between gimplify and omp_expand since
    readonly flag plays role on what is shared and what is not.  Currently we
do
    this transformation as part of whole program visibility and re-do at
    ipa-reference pass (to take into account clonning), but it would
    make sense to do it before early optimizations.  */

-void
+bool
 ipa_discover_readonly_nonaddressable_vars (void)
 {
+  bool remove_p = false;
   varpool_node *vnode;
   if (dump_file)
     fprintf (dump_file, "Clearing variable flags:");
@@ -762,14 +769,16 @@ ipa_discover_readonly_nonaddressable_var
        bool read = false;
        bool explicit_refs = true;
-       process_references (vnode, &written, &address_taken, &read,
&explicit_refs);
+       process_references (vnode, &written, &address_taken, &read,
+                           &explicit_refs);
        if (!explicit_refs)
          continue;
        if (!address_taken)
          {
            if (TREE_ADDRESSABLE (vnode->decl) && dump_file)
              fprintf (dump_file, " %s (non-addressable)", vnode->name ());
-           vnode->call_for_node_and_aliases (clear_addressable_bit, NULL,
true);
+           vnode->call_for_node_and_aliases (clear_addressable_bit, NULL,
+                                             true);
          }
        if (!address_taken && !written
            /* Making variable in explicit section readonly can cause section
@@ -785,11 +794,13 @@ ipa_discover_readonly_nonaddressable_var
          {
            if (dump_file)
              fprintf (dump_file, " %s (write-only)", vnode->name ());
-           vnode->call_for_node_and_aliases (set_writeonly_bit, NULL, true);
+           vnode->call_for_node_and_aliases (set_writeonly_bit, &remove_p, 
+                                            true);
          }
       }
   if (dump_file)
     fprintf (dump_file, "\n");
+  return remove_p;
 }

 /* Free inline summary.  */
Index: cgraph.h
===================================================================
--- cgraph.h    (revision 218726)
+++ cgraph.h    (working copy)
@@ -2188,7 +2188,7 @@ void record_references_in_initializer (t

 /* In ipa.c  */
 void cgraph_build_static_cdtor (char which, tree body, int priority);
-void ipa_discover_readonly_nonaddressable_vars (void);
+bool ipa_discover_readonly_nonaddressable_vars (void);

 /* In varpool.c  */
 tree ctor_for_folding (tree);
Index: testsuite/gcc.dg/torture/pr61602.c
===================================================================
--- testsuite/gcc.dg/torture/pr61602.c  (revision 0)
+++ testsuite/gcc.dg/torture/pr61602.c  (revision 0)
@@ -0,0 +1,8 @@
+int a;
+int *b = &a, **c = &b;
+int
+main ()
+{
+  int **d = &b;
+  *d = 0;
+}
Index: ipa-reference.c
===================================================================
--- ipa-reference.c     (revision 218726)
+++ ipa-reference.c     (working copy)
@@ -676,11 +676,12 @@ propagate (void)
     XCNEWVEC (struct cgraph_node *, symtab->cgraph_count);
   int order_pos;
   int i;
+  bool remove_p;

   if (dump_file)
     cgraph_node::dump_cgraph (dump_file);

-  ipa_discover_readonly_nonaddressable_vars ();
+  remove_p = ipa_discover_readonly_nonaddressable_vars ();
   generate_summary ();

   /* Propagate the local information through the call graph to produce
@@ -867,7 +868,7 @@ propagate (void)
   if (dump_file)
     splay_tree_delete (reference_vars_to_consider);
   reference_vars_to_consider = NULL;
-  return 0;
+  return remove_p ? TODO_remove_functions : 0;
 }

 /* Return true if we need to write summary of NODE. */

[Bug ipa/61602] [5 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

--- Comment #9 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Author: hubicka
Date: Mon Dec 15 06:19:51 2014
New Revision: 218731

URL: https://gcc.gnu.org/viewcvs?rev=218731&root=gcc&view=rev
Log:

    PR ipa/61602
    * gcc.dg/torture/pr61602.c: New testcase.

    * cgraph.h (ipa_discover_readonly_nonaddressable_vars): Return bool.
    * ipa.c (set_writeonly_bit): Track if reference was removed.
    (ipa_discover_readonly_nonaddressable_vars): Return true if any
    references was removed.
    * ipa-reference.c (propagate): Return TODO_remove_functions if
    reference was removed.

Added:
    trunk/gcc/testsuite/gcc.dg/torture/pr61602.c
Modified:
    trunk/gcc/cgraph.h
    trunk/gcc/ipa-reference.c
    trunk/gcc/ipa.c
    trunk/gcc/testsuite/ChangeLog

[Bug ipa/61602] [5 Regression] ICE in lto1 on x86_64-linux-gnu in ipa_single_use, at ipa.c:1257

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61602

Jan Hubicka <hubicka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #10 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Fixed.