[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[trippels at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

--- Comment #7 from Markus Trippelsdorf  ---
The issue seems to be fixed. Not sure what revision fixed it.

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

--- Comment #8 from Martin Jambor  ---
Author: jamborm
Date: Fri Dec 11 11:27:11 2015
New Revision: 231559

URL: https://gcc.gnu.org/viewcvs?rev=231559=gcc=rev
Log:
Add an asssert and testcase for PR 68064

2015-12-11  Martin Jambor  

PR ipa/68064
* ipa-prop.c (ipa_compute_jump_functions_for_edge): Add checking
assert that align is nonzero.

testsuite/
* g++.dg/torture/pr68064.C: New test.


Added:
trunk/gcc/testsuite/g++.dg/torture/pr68064.C
Modified:
trunk/gcc/ChangeLog
trunk/gcc/ipa-prop.c
trunk/gcc/testsuite/ChangeLog

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

Martin Jambor  changed:

   What|Removed |Added

 Status|NEW |RESOLVED
 Resolution|--- |DUPLICATE

--- Comment #9 from Martin Jambor  ---
Yes, it has been fixed by Richi's r231246 which is a fix of PR68636
(of which this is basically a duplicate, only manifesting elsewhere).
I am have just committed a testcase and an assert for detecting this
earlier should it happen again, see
https://gcc.gnu.org/ml/gcc-patches/2015-12/msg01246.html

*** This bug has been marked as a duplicate of bug 68636 ***

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[trippels at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

--- Comment #5 from Markus Trippelsdorf  ---
Any update on this issue?

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[hubicka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

--- Comment #6 from Jan Hubicka  ---
Hmm, I can't make much sense of get_pointer_alignment_1 returning 0. I suppose
this happens because here:
  else if (TREE_CODE (exp) == SSA_NAME
   && POINTER_TYPE_P (TREE_TYPE (exp)))
{
  unsigned int ptr_align, ptr_misalign;
  struct ptr_info_def *pi = SSA_NAME_PTR_INFO (exp);

  if (pi && get_ptr_info_alignment (pi, _align, _misalign))
{
  *bitposp = ptr_misalign * BITS_PER_UNIT;
  *alignp = ptr_align * BITS_PER_UNIT;
  /* We cannot really tell whether this result is an approximation.  */
  return true;
}
  else
{
  *bitposp = 0;
  *alignp = BITS_PER_UNIT;
  return false;
}
}
because we get overflow when computing ptr_misalign * BITS_PER_UNIT.
I think get_pointer_alignment should prevent that.  It knows how to deal with
constant:
  else if (TREE_CODE (exp) == INTEGER_CST)
{ 
  *alignp = BIGGEST_ALIGNMENT;
  *bitposp = ((TREE_INT_CST_LOW (exp) * BITS_PER_UNIT)
  & (BIGGEST_ALIGNMENT - 1));
  return true;
}

so I guess the code handling SSA_NAME should also cap to BIGGEST_ALIGNMENT.

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

--- Comment #4 from Martin Jambor  ---
I have proposed committing the fix on the mailing list

https://gcc.gnu.org/ml/gcc-patches/2015-10/msg03383.html

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

Martin Jambor  changed:

   What|Removed |Added

   Assignee|unassigned at gcc dot gnu.org  |jamborm at gcc dot 
gnu.org

--- Comment #2 from Martin Jambor  ---
Mine.

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[jamborm at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

--- Comment #3 from Martin Jambor  ---
The problem is that IPA-CP does not handle zero-aligned pointers (what
would they be?) but get_pointer_alignment_1 actually returns zero for
one and itself returns true, meaning it claims it knows the alignment
is what it says.  The reason for that is that the pointer is actually
either zero or an integer constant, selected by a ternary operator.

Given the circumstances, the behavior of get_pointer_alignment_1 is
actually OK, I suppose, so wee need to check for it like this:

--- a/gcc/ipa-prop.c
+++ b/gcc/ipa-prop.c
@@ -1651,6 +1651,7 @@ ipa_compute_jump_functions_for_edge (struct
ipa_func_body_info *fbi,
  unsigned align;

  if (get_pointer_alignment_1 (arg, , _bitpos)
+ && align != 0
  && align % BITS_PER_UNIT == 0
  && hwi_bitpos % BITS_PER_UNIT == 0)
{

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

Richard Biener  changed:

   What|Removed |Added

   Target Milestone|--- |6.0

[Bug ipa/68064] [6 Regression] ICE: in meet_with, at ipa-cp.c:874

[trippels at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68064

Markus Trippelsdorf  changed:

   What|Removed |Added

 Status|UNCONFIRMED |NEW
   Last reconfirmed||2015-10-23
 CC||hubicka at gcc dot gnu.org
 Ever confirmed|0   |1

--- Comment #1 from Markus Trippelsdorf  ---
markus@x4 Release % cat typer.ii
template  class A {
public:
  class B;
  typedef typename Config::template D::type TypeHandle;
  static A *Tagged() { return B::New(B::kTagged); }
  static TypeHandle Union(TypeHandle);
  static TypeHandle Representation(TypeHandle, typename Config::Region *);
  bool Is();
};

template  class A::B {
  friend A;
  enum { kTaggedPointer = 1 << 31, kTagged = kTaggedPointer };
  static A *New(int p1) { return Config::from_bitset(p1); }
};

struct C {
  typedef int Region;
  template  struct D { typedef A *type; };
  static A *from_bitset(unsigned);
};
A *C::from_bitset(unsigned p1) { return reinterpret_cast(p1); }

namespace {
int *a;
void fn1(A *p1) { A::Union(A::Representation(p1, a)); }
}

void fn2() {
  A b;
  A *c = b.Is() ? 0 : A::Tagged();
  fn1(c);
}

markus@x4 Release % g++ -c -O2 typer.ii
typer.ii:33:1: internal compiler error: in meet_with, at ipa-cp.c:874