[Bug bootstrap/51346] [4.7 Regression] LTO bootstrap failed with bootstrap-profiled

[dnovillo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51346

Diego Novillo <dnovillo at gcc dot gnu.org> changed:

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

--- Comment #3 from Diego Novillo <dnovillo at gcc dot gnu.org> 2011-11-30 20:17:08 UTC ---
The reason we get into this ICE is because the call to cgraph_edge() has the
side effect of adding entries to e->caller->call_site_hash:

  if (n > 100)
    { 
      node->call_site_hash = htab_create_ggc (120, edge_hash, edge_eq, NULL);
      for (e2 = node->callees; e2; e2 = e2->next_callee)
        cgraph_add_edge_to_call_site_hash (e2);
      for (e2 = node->indirect_calls; e2; e2 = e2->next_callee)
        cgraph_add_edge_to_call_site_hash (e2);
    }

When cgraph_edge is called with n > 100, the call to
cgraph_add_edge_to_call_site_hash asserts because e->call_stmt is already in
the hash table.

Honza, why is this a problem?  The code is expecting to find the slot empty,
but I'm not sure why it expects that.  Any pointers?


Thanks.  Diego.