[Bug tree-optimization/49735] New: mips64-elf libgcc build fails with apparently infinite recursion.

[Bug tree-optimization/49735] New: mips64-elf libgcc build fails with apparently infinite recursion.

[bernds at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

           Summary: mips64-elf libgcc build fails with apparently infinite
                    recursion.
           Product: gcc
           Version: 4.7.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: bernds@gcc.gnu.org


Created attachment 24751
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=24751
A testcase

Configure a cross compiler with --target=mips64-elf. When building libgcc, cc1
goes into what seems to be an infinite recursion:

(gdb) bt
#0  0x083b22aa in estimate_calls_size_and_time (node=<value optimized out>,
size=0xf7484030, time=0xf748402c, 
    possible_truths=4294967294) at ../../trunk/gcc/ipa-inline-analysis.c:1748
#1  0x083b229d in estimate_calls_size_and_time (node=<value optimized out>,
size=0xf7484030, time=0xf748402c, 
    possible_truths=4294967294) at ../../trunk/gcc/ipa-inline-analysis.c:1743
#2  0x083b229d in estimate_calls_size_and_time (node=<value optimized out>,
size=0xf7484030, time=0xf748402c, 
    possible_truths=4294967294) at ../../trunk/gcc/ipa-inline-analysis.c:1743

I'm attaching the .i file.

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |build, ice-on-valid-code
             Target|                            |mips64-linux
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2011.07.13 23:12:32
   Target Milestone|---                         |4.7.0
            Summary|mips64-elf libgcc build     |[4.7 Regression] mips64-elf
                   |fails with apparently       |libgcc build fails with
                   |infinite recursion.         |apparently infinite
                   |                            |recursion.
     Ever Confirmed|0                           |1

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-07-13 23:12:32 UTC ---
Confirmed.

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-07-13 23:18:11 UTC ---
Here is a short testcase:
typedef unsigned int UQItype __attribute__((mode (QI)));
static 
UQItype sync_fetch_and_add_1 (UQItype *ptr, UQItype value) 
{
 return __sync_fetch_and_add (ptr, value);
 }; 
typeof (sync_fetch_and_add_1) __sync_fetch_and_add_1 __attribute__((alias
("sync_fetch_and_add" "_" "1")));

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-07-13 23:21:26 UTC ---
I almost want to say it was caused by:
r175169 | hubicka | 2011-06-18 01:33:47 -0700 (Sat, 18 Jun 2011) | 31 lines

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-07-14 00:07:51 UTC ---
This patch fixes the infinite loop:
Index: ipa-inline.c
===================================================================
--- ipa-inline.c    (revision 176253)
+++ ipa-inline.c    (working copy)
@@ -1170,6 +1170,10 @@ recursive_inlining (struct cgraph_edge *
   /* Make sure that function is small enough to be considered for inlining. 
*/
   if (estimate_size_after_inlining (node, edge)  >= limit)
     return false;
+  /* Inlining a builtin recursively into itself should not be done. */
+  if (DECL_BUILT_IN (edge->callee->decl))
+    return false;
+
   heap = fibheap_new ();
   lookup_recursive_calls (node, node, heap);
   if (fibheap_empty (heap))
--- CUT ---

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

--- Comment #5 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-07-14 09:05:32 UTC ---
It probably papers over a different issue though - even recursive inlining
should be limited by a --param, but appearantly that is not working here.

Honza?

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[pinskia at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> 2011-07-14 18:26:54 UTC ---
(In reply to comment #5)
> It probably papers over a different issue though - even recursive inlining
> should be limited by a --param, but appearantly that is not working here.
> 
> Honza?

Oh it is working, just since this is a simple wrapper function (infinite loop
kinda) it is inlinining a lot.

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[rsandifo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

rsandifo@gcc.gnu.org <rsandifo at gcc dot gnu.org> changed:

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

--- Comment #7 from rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> 2011-07-19 18:09:45 UTC ---
(In reply to comment #6)
> (In reply to comment #5)
> > It probably papers over a different issue though - even recursive inlining
> > should be limited by a --param, but appearantly that is not working here.
> > 
> > Honza?
> 
> Oh it is working, just since this is a simple wrapper function (infinite loop
> kinda) it is inlinining a lot.

I think Richard's point was that something like:

void bar (void);
static void foo (void) { bar (); }
void bar (void) __attribute__((alias("foo")));

shouldn't make gcc recurse infinitely (but now does).

Richard

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Target|mips64-linux                |
           Priority|P3                          |P1

--- Comment #8 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-08-02 14:29:11 UTC ---
Indeed.

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

--- Comment #9 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-08-03 20:00:22 UTC ---
Hmm, it looks like recursive_inlining_p predicate broke with presence of
aliases.  I will take a look.

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

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

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

--- Comment #10 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-08-03 20:01:15 UTC ---
Mine, BTW

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[hubicka at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

--- Comment #11 from Jan Hubicka <hubicka at gcc dot gnu.org> 2011-08-05 20:20:11 UTC ---
Author: hubicka
Date: Fri Aug  5 20:20:07 2011
New Revision: 177482

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=177482
Log:
        PR middle-end/49735
        * gcc.c-torture/compile/pr49735.c: New testcase
        * ipa-inline.c (recursive_inlining): Look through aliases.


Added:
    trunk/gcc/testsuite/gcc.c-torture/compile/pr49735.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/ipa-inline.c
    trunk/gcc/testsuite/ChangeLog

[Bug tree-optimization/49735] [4.7 Regression] mips64-elf libgcc build fails with apparently infinite recursion.

[rsandifo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=49735

rsandifo@gcc.gnu.org <rsandifo at gcc dot gnu.org> changed:

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

--- Comment #12 from rsandifo at gcc dot gnu.org <rsandifo at gcc dot gnu.org> 2011-08-06 18:49:17 UTC ---
Just to confirm that the build failure is now fixed.  Thanks Jan!