[Bug target/65263] New: [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[Bug target/65263] New: [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[doko at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

            Bug ID: 65263
           Summary: [5 Regression] ICE (error: unrecognizable insn / in
                    insn_min_length, at config/rs6000/rs6000.md) on
                    powerpc64le-linux-gnu
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: doko at gcc dot gnu.org

seen with r221076 on powerpc64le-linux-gnu

$ g++ -c -g -O3 -fPIC eigenvalue.ii 
eigenvalue.ii:26:19: warning: direct base 'VirtualMatrice<int>' inaccessible in
'C<int>' due to ambiguity
 template <> class C<int> : H<Complex>, VirtualMatrice<Complex> {
                   ^
eigenvalue.ii: In member function 'virtual bool
C<int>::_ZThn8_NK1CIiE5m_fn1Ei(int) const':
eigenvalue.ii:47:1: error: unrecognizable insn:
 }
 ^
(call_insn/j 3 2 4 (parallel [
            (call (mem:SI (symbol_ref/i:DI ("_ZNK14VirtualMatriceIiE5m_fn1Ei")
[flags 0x1] <function_decl 0x3fff9696fdf8 m_fn1>) [0  S4 A8])
                (const_int 0 [0]))
            (use (const_int 0 [0]))
            (use (reg:SI 65 lr))
            (simple_return)
        ]) eigenvalue.ii:27 -1
     (nil)
    (nil))
eigenvalue.ii:47:1: internal compiler error: in insn_min_length, at
config/rs6000/rs6000.md:7813
Please submit a full bug report,
with preprocessed source if appropriate.

$cat eigenvalue.ii
template <class> class A;
template <class R> struct VirtualMatrice {
  virtual bool m_fn1(int) const { return true; }
  struct B {
    A<R> x;
    B(VirtualMatrice *p1, A<R> p2) : x(p2) { p1->m_fn1(0) ?: throw; }
  };
  void operator*(A<R> p1) { B(this, p1); }
  ~VirtualMatrice();
}
;
template <class> class A {
public:
  operator int *();
  A(int *, long);
};

class G : public A<int> {
public:
  G(long);
};
int typedef Complex;
template <class> class H : VirtualMatrice<int> {};
template <class> class C;
template <> class C<int> : H<Complex>, VirtualMatrice<Complex> {
  bool m_fn1(int) const { return true; }
};
template <class K, class Mat>
void DoIdoAction(int, int, A<K> p3, A<K>, A<K>, A<K>, Mat, Mat &p8) {
  p8 *p3;
}

class D {
  typedef int K;
  class F {
    int operator()() const;
  };
};
int D::F::operator()() const {
  VirtualMatrice<K> *a;
  VirtualMatrice<K> b, &B = *a;
  G c(0), g(1);
  int d, e, f;
  A<K> h(&g[f], 0), i(&g[e], 0), j(&g[d], 0);
  DoIdoAction(0, 3, h, i, j, c, b, B);
}

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[amodra at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

Alan Modra <amodra at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-03-02
                 CC|                            |amodra at gmail dot com,
                   |                            |hubicka at gcc dot gnu.org,
                   |                            |mliska at suse dot cz
     Ever confirmed|0                           |1

--- Comment #1 from Alan Modra <amodra at gmail dot com> ---
This is caused by revision 221040
+2015-02-26  Jan Hubicka  <hubicka@ucw.cz>
+           Martin Liska  <mliska@suse.cz>
+
+       PR bootstrap/65150
[snip]

#2  0x0000000000bca3fa in _fatal_insn_not_found
(insn=insn@entry=0x7ffff6ae4ca8, file=file@entry=0x13852e0
"/src/gcc-current/gcc/config/rs6000/rs6000.md", line=line@entry=7813,
function=function@entry=0x139c330 <insn_min_length(rtx_insn*)::__FUNCTION__>
"insn_min_length") at /src/gcc-current/gcc/rtl-error.c:118
#3  0x0000000000f1e9f2 in insn_min_length (insn=insn@entry=0x7ffff6ae4ca8) at
/src/gcc-current/gcc/config/rs6000/rs6000.md:7813
#4  0x000000000095c8e2 in shorten_branches (first=0x7ffff6ad8038) at
/src/gcc-current/gcc/final.c:1221
#5  0x0000000000eea239 in rs6000_output_mi_thunk (file=0x1c37950,
thunk_fndecl=<optimised out>, delta=<optimised out>, vcall_offset=<optimised
out>, function=<optimised out>) at
/src/gcc-current/gcc/config/rs6000/rs6000.c:25835

That's here:
  /* Run just enough of rest_of_compilation to get the insns emitted.
     There's not really enough bulk here to make other passes such as
     instruction scheduling worth while.  Note that use_thunk calls
     assemble_start_function and assemble_end_function.  */
  insn = get_insns ();
  shorten_branches (insn);

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #2 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Hmm, this seems interesting.  The revision enabled more merging in ICF. The
expansion dies in output MI thunk that is not ICF produced thunk. So only
conclussion I can think of is that ICF redirected call in thunk that made the
target dependent code go wrong.

What is wrong about the insn? Does the declaration of functions differ
significantly?
Let me see if I can reproduce that in a cross.

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #3 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Semantic equality hit:bool VirtualMatrice<R>::m_fn1(int) const [with R =
int]->virtual bool C<int>::m_fn1(int) const
Assembler symbol names:_ZNK14VirtualMatriceIiE5m_fn1Ei->_ZNK1CIiE5m_fn1Ei       
Wrapper cannot be created because of COMDAT                                     
1 local calls have been redirected.                                             
Unified; Function body was removed.                                             

I assume the problem is that thunk can not be generated to non-local symbol. I
specifically prevent comdat local symbols to be produced because these makes
inlining harder.

There are couple options.  First would be to punt redirection on thunks calls,
second would be to produce local alias for thunk calls. I.e. check
TARGET_USE_LOCAL_THUNK_ALIAS_P and produce the local alias specifically for
thunks.  Third would be to produce the local alias always.

I guess first is safest - while I plan to turn thunks to use noninterposable
alias path, I did not because I fear undocumented target issues.  Bettter to
not start with these in stage4.

I will prepare patch after dinner

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[amodra at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #4 from Alan Modra <amodra at gmail dot com> ---
> I assume the problem is that thunk can not be generated to non-local symbol.

Yes, that is at the root of this problem.  The thunk can't be allowed to jump
to anything outside the current object file.

The pattern is a sibcall_local64 except that the symbol_ref doesn't match
rs6000/predicates.md: current_file_function_operand.

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[hubicka at ucw dot cz]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #5 from Jan Hubicka <hubicka at ucw dot cz> ---
Hi,
the following (untested) patch should prevent redirection to happen.  
Index: ipa-icf.c
===================================================================
--- ipa-icf.c    (revision 221096)
+++ ipa-icf.c    (working copy)
@@ -697,12 +697,22 @@ redirect_all_callers (cgraph_node *n, cg
 {
   int nredirected = 0;
   ipa_ref *ref;
+  cgraph_edge *e = n->callers;

-  while (n->callers)
+  while (e)
     {
-      cgraph_edge *e = n->callers;
-      e->redirect_callee (to);
-      nredirected++;
+      /* Redirecting thunks to interposable symbols or symbols in other
sections
+     may not be supported by target output code.  Play safe for now and
+     punt on redirection.  */
+      if (!e->caller->thunk.thunk_p)
+    {
+      struct cgraph_edge *nexte = e->next_callee;
+          e->redirect_callee (to);
+      e = nexte;
+          nredirected++;
+    }
+      else
+    e = e->next_callee;
     }
   for (unsigned i = 0; n->iterate_direct_aliases (i, ref);)
     {
@@ -726,6 +736,17 @@ redirect_all_callers (cgraph_node *n, cg
   return nredirected;
 }

+/* Return true if NODE has thunk.  */
+
+bool
+has_thunk_p (cgraph_node *node, void *)
+{
+  for (cgraph_edge *e = node->callers; e; e = e->next_caller)
+    if (e->caller->thunk.thunk_p)
+      return true;
+  return false;
+}
+
 /* Merges instance with an ALIAS_ITEM, where alias, thunk or redirection can
    be applied.  */

@@ -907,6 +928,7 @@ sem_function::merge (sem_item *alias_ite
       return false;
     }
       if (!create_wrapper
+      && !alias->call_for_symbol_and_aliases (has_thunk_p, NULL, true)
       && !alias->can_remove_if_no_direct_calls_p ())
     {
       if (dump_file)

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[amodra at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #6 from Alan Modra <amodra at gmail dot com> ---
rev 221098 plus the patch segfaults on the testcase in
cgraph_node::function_symbol.  node->callees is NULL.

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P1
   Target Milestone|---                         |5.0

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

Martin Liška <marxin at gcc dot gnu.org> changed:

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

--- Comment #7 from Martin Liška <marxin at gcc dot gnu.org> ---
Created attachment 34922
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=34922&action=edit
Suggested patch

There's fixed version of the patch:

1) If I see correctly, there was error in: struct cgraph_edge *nexte =
e->next_callee
2) We should not remove alias if there's a thunk that wasn't redirected
3) gcc_assert must be changed for these cases

Patch can run the testcase and survives FF with -O3 and -flto. Regression tests
have been running.

Martin
>From gcc-bugs-return-478990-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org Mon Mar 02 13:44:34 2015
Return-Path: <gcc-bugs-return-478990-listarch-gcc-bugs=gcc.gnu.org@gcc.gnu.org>
Delivered-To: listarch-gcc-bugs@gcc.gnu.org
Received: (qmail 27557 invoked by alias); 2 Mar 2015 13:44:34 -0000
Mailing-List: contact gcc-bugs-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Id: <gcc-bugs.gcc.gnu.org>
List-Archive: <http://gcc.gnu.org/ml/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-help@gcc.gnu.org>
Sender: gcc-bugs-owner@gcc.gnu.org
Delivered-To: mailing list gcc-bugs@gcc.gnu.org
Received: (qmail 27488 invoked by uid 48); 2 Mar 2015 13:44:30 -0000
From: "hjl.tools at gmail dot com" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug ipa/65006] [5 Regression] 252.eon in SPEC CPU 2000 miscompiled with LTO
Date: Mon, 02 Mar 2015 13:44:00 -0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: ipa
X-Bugzilla-Version: 5.0
X-Bugzilla-Keywords: wrong-code
X-Bugzilla-Severity: normal
X-Bugzilla-Who: hjl.tools at gmail dot com
X-Bugzilla-Status: RESOLVED
X-Bugzilla-Priority: P1
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: 5.0
X-Bugzilla-Flags:
X-Bugzilla-Changed-Fields: bug_status resolution
Message-ID: <bug-65006-4-u3K2pL6yWO@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-65006-4@http.gcc.gnu.org/bugzilla/>
References: <bug-65006-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-SW-Source: 2015-03/txt/msg00134.txt.bz2
Content-length: 429

https://gcc.gnu.org/bugzilla/show_bug.cgi?ide006

H.J. Lu <hjl.tools at gmail dot com> changed:

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

--- Comment #14 from H.J. Lu <hjl.tools at gmail dot com> ---
Fixed as of r221077.

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[ppluzhnikov at google dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

Paul Pluzhnikov <ppluzhnikov at google dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ppluzhnikov at google dot com

--- Comment #8 from Paul Pluzhnikov <ppluzhnikov at google dot com> ---
r221040 is also possibly causing:

  ../sysdeps/gnu/siglist.c:77:1: internal compiler error: in address_matters_p,
at symtab.c:1908

when building trunk GLIBC (seen with GCC @r221125; same source builds with GCC
@r220312).

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #9 from Jan Hubicka <hubicka at gcc dot gnu.org> ---
Paul, I fixed similar bug yesterday, so please check if it works now for you
and if not, please make new PR with a preprocessed source.

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[ppluzhnikov at google dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #10 from Paul Pluzhnikov <ppluzhnikov at google dot com> ---
(In reply to Jan Hubicka from comment #9)
> Paul, I fixed similar bug yesterday, so please check if it works now

I just built at current SVN trunk (r221126).
Filed PR 65287

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[amodra at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #11 from Alan Modra <amodra at gmail dot com> ---
Martin, your patch passes bootstrap and regression testing on powerpc64-linux.

[Bug target/65263] [5 Regression] ICE (error: unrecognizable insn / in insn_min_length, at config/rs6000/rs6000.md) on powerpc64le-linux-gnu

[aldyh at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=65263

--- Comment #14 from Aldy Hernandez <aldyh at gcc dot gnu.org> ---
*** Bug 65278 has been marked as a duplicate of this bug. ***