[Bug c++/38844] New: [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[Bug c++/38844] New: [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[zsojka at seznam dot cz]

gcc seems to deadlock with the following code at -O1 and above:

=================================
struct A {
        inline __attribute__((always_inline)) A()
        {
                A();
                A();
        }
};

int main()
{
        A();
        return 0;
}
=================================

Command line:
g++ deadlock.cpp -o deadlock.o -O1


-- 
           Summary: [4.3/4.4 Regression] deadlock with
                    __attribute__((always_inline)) at -O1 and above
           Product: gcc
           Version: 4.3.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: zsojka at seznam dot cz
  GCC host triplet: x86_64-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug c++/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[zsojka at seznam dot cz]

------- Comment #1 from zsojka at seznam dot cz  2009-01-14 18:14 -------
Created an attachment (id=17100)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17100&action=view)
preprocessed source


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug c++/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[zsojka at seznam dot cz]

------- Comment #2 from zsojka at seznam dot cz  2009-01-14 19:03 -------
Created an attachment (id=17102)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17102&action=view)
another testcase, this one fails _only_ with -O1


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[rguenth at gcc dot gnu dot org]

-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hubicka at gcc dot gnu dot
                   |                            |org
          Component|c++                         |tree-optimization
           Keywords|                            |ice-on-valid-code
   Target Milestone|---                         |4.3.3


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[pinskia at gcc dot gnu dot org]

------- Comment #3 from pinskia at gcc dot gnu dot org  2009-01-15 01:07 -------
Confirmed.
here is one that fails with both the C and C++ front-ends:
        inline __attribute__((always_inline)) void A(int *this1)
        {
  int this2;
  int this3;
                A(&this2);
                A(&this3);
        }

int main()
{
int this1;
        A(&this1);
        return 0;
}


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pinskia at gcc dot gnu dot
                   |                            |org
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|0                           |1
      Known to fail|4.3.2 4.4.0                 |4.3.2 4.4.0 4.3.0
   Last reconfirmed|0000-00-00 00:00:00         |2009-01-15 01:07:47
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2009-01-15 16:03 -------
Honza, can you please have a look here?


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[hjl dot tools at gmail dot com]

------- Comment #5 from hjl dot tools at gmail dot com  2009-01-15 21:35 -------
It is caused by revision 121570:

http://gcc.gnu.org/ml/gcc-cvs/2007-02/msg00126.html


-- 

hjl dot tools at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hjl dot tools at gmail dot
                   |                            |com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[hjl dot tools at gmail dot com]

------- Comment #6 from hjl dot tools at gmail dot com  2009-01-15 22:38 -------
This patch:

--- ./ipa-inline.c.foo  2009-01-02 11:06:18.000000000 -0800
+++ ./ipa-inline.c      2009-01-15 14:35:28.000000000 -0800
@@ -1412,7 +1412,8 @@ cgraph_decide_inlining_incrementally (st
            }
          continue;
        }
-      inlined |= try_inline (e, mode, depth);
+      if (e->inline_failed)
+       inlined |= try_inline (e, mode, depth);
     }

   /* Now do the automatic inlining.  */

works for me. But I don't know if it is correct.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[hjl dot tools at gmail dot com]

------- Comment #7 from hjl dot tools at gmail dot com  2009-01-15 23:57 -------
Revision 121570 is:

http://gcc.gnu.org/ml/gcc-patches/2007-02/msg00319.html


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[rguenth at gcc dot gnu dot org]

------- Comment #8 from rguenth at gcc dot gnu dot org  2009-01-24 10:21 -------
GCC 4.3.3 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.3                       |4.3.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[jakub at gcc dot gnu dot org]

------- Comment #9 from jakub at gcc dot gnu dot org  2009-01-28 21:08 -------
The #c6 patch doesn't make sense to me.

Two alternative testcases:

inline __attribute__ ((always_inline))
void A (void)
{
  int i;
  asm volatile ("" : "=m" (i));
  A ();
  A ();
}

int
main ()
{
  A ();
  return 0;
}

and

inline void A (void);
inline void B (void);

inline __attribute__ ((always_inline))
void C (void)
{
  B ();
}

inline __attribute__ ((always_inline))
void B (void)
{
  A ();
}

inline __attribute__ ((always_inline))
void A (void)
{
  int i;
  asm volatile ("" : "=m" (i));
  A ();
  C ();
}

int
main ()
{
  A ();
  return 0;
}

As the cgraph_decide_inlining_incrementally + try_inline combo goes
depth-first, it discovers just the first (perhaps indirect) recursive inlining
only, the others mean we keep recursing with deeper and deeper depths.  I
wonder if it couldn't go breadth-first instead.  Honza?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[jakub at gcc dot gnu dot org]

------- Comment #10 from jakub at gcc dot gnu dot org  2009-01-28 21:47 -------
Further testcases:

inline __attribute__((always_inline))
void A (void)
{
  A ();
  A ();
  A ();
}

int
main ()
{
  A ();
  return 0;
}

and:

inline void A (void);
inline void B (void);

inline __attribute__ ((always_inline))
void C (void)
{
  B ();
}

inline __attribute__ ((always_inline))
void B (void)
{
  A ();
}

inline __attribute__ ((always_inline))
void A (void)
{
  A ();
  B ();
  C ();
}

int
main ()
{
  A ();
  return 0;
}

Perhaps just changing node->aux for all nodes for which recursive
cgraph_decide_inlining_incrementally will be called would work.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[rguenth at gcc dot gnu dot org]

------- Comment #11 from rguenth at gcc dot gnu dot org  2009-01-28 22:37 -------
We should still optimize

static void bar(void) { foo (); }

inline void __attribute__((always_inline)) foo(void)
{
  foo ();
  bar ();
}

to

foo() { foo(); foo (); }

or similar cases, that is basically collapse a cycle covering multiple
functions into one.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[hubicka at gcc dot gnu dot org]

------- Comment #12 from hubicka at gcc dot gnu dot org  2009-02-02 16:35 -------
try_inline should not recurse when edge is already inlined.  The following
patch should fix the problem, I will benchmark it tonight
Index: ipa-inline.c
===================================================================
--- ipa-inline.c        (revision 143680)
+++ ipa-inline.c        (working copy)
@@ -1330,16 +1330,18 @@ try_inline (struct cgraph_edge *e, enum 
               cgraph_node_name (e->caller));
     }
   if (e->inline_failed)
-    cgraph_mark_inline (e);
+    {
+      cgraph_mark_inline (e);

-  /* In order to fully inline always_inline functions, we need to
-     recurse here, since the inlined functions might not be processed by
-     incremental inlining at all yet.  
+      /* In order to fully inline always_inline functions, we need to
+        recurse here, since the inlined functions might not be processed by
+        incremental inlining at all yet.  

-     Also flattening needs to be done recursively.  */
+        Also flattening needs to be done recursively.  */

-  if (mode == INLINE_ALL || always_inline)
-    cgraph_decide_inlining_incrementally (e->callee, mode, depth + 1);
+      if (mode == INLINE_ALL || always_inline)
+       cgraph_decide_inlining_incrementally (e->callee, mode, depth + 1);
+    }
   callee->aux = (void *)(size_t) callee_mode;
   return true;
 }


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3/4.4 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[hubicka at gcc dot gnu dot org]

------- Comment #13 from hubicka at gcc dot gnu dot org  2009-02-06 16:47 -------
Subject: Bug 38844

Author: hubicka
Date: Fri Feb  6 16:47:39 2009
New Revision: 143985

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=143985
Log:

        PR tree-optimization/38844
        * ipa-inline.c (try_inline): Stop inlining recursion when edge
        is already inlined.

Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/ipa-inline.c


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[hubicka at gcc dot gnu dot org]

------- Comment #14 from hubicka at gcc dot gnu dot org  2009-02-06 16:49 -------
Fixed on mainline


-- 

hubicka at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[4.3/4.4 Regression]        |[4.3 Regression] deadlock
                   |deadlock with               |with
                   |__attribute__((always_inline|__attribute__((always_inline
                   |)) at -O1 and above         |)) at -O1 and above


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[rguenth at gcc dot gnu dot org]

------- Comment #15 from rguenth at gcc dot gnu dot org  2009-08-04 12:29 -------
GCC 4.3.4 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.4                       |4.3.5


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[rguenth at gcc dot gnu dot org]

------- Comment #16 from rguenth at gcc dot gnu dot org  2010-04-20 14:30 -------
Backporting the fix causes

FAIL: gcc.c-torture/compile/pr32482.c  -O0  (test for excess errors)
Excess errors:
/abuild/rguenther/gcc-4_3-branch/gcc/testsuite/gcc.c-torture/compile/pr32482.c:4:
sorry, unimplemented: inlining failed in call to 'bitmap_empty': function not
considered for inlining
/abuild/rguenther/gcc-4_3-branch/gcc/testsuite/gcc.c-torture/compile/pr32482.c:10:
sorry, unimplemented: called from here


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[rguenth at gcc dot gnu dot org]

------- Comment #17 from rguenth at gcc dot gnu dot org  2010-05-22 18:13 -------
GCC 4.3.5 is being released, adjusting target milestone.


-- 

rguenth at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|4.3.5                       |4.3.6


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

[Bug tree-optimization/38844] [4.3 Regression] deadlock with __attribute__((always_inline)) at -O1 and above

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38844

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
      Known to work|                            |
         Resolution|                            |FIXED
   Target Milestone|4.3.6                       |4.4.0

--- Comment #18 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-06-27 11:43:49 UTC ---
Fixed for 4.4.0.