[Bug tree-optimization/55616] New: bogus warning about undefined overflow after overflow check

[Bug tree-optimization/55616] New: bogus warning about undefined overflow after overflow check

[fweimer at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

             Bug #: 55616
           Summary: bogus warning about undefined overflow after overflow
                    check
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: fweimer@redhat.com
                CC: nils@redhat.com


Created attachment 28893
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28893
overflow-check.c

Code which performs proper overflow checking results in a compiler warning (at
-O2, without -fwrapv):

overflow-check.c:9:6: warning: assuming signed overflow does not occur when
assuming that (X + c) < X is always false [-Wstrict-overflow]
   if (a > b)
      ^

I'm surprised that the call to something from f is not optimized away with
-fwrapv.  Shouldn't VRP catch this?  Addressing this missed optimization
opportunity might also fix the warning.

The g function in the test case intends to make the if statement in r actually
relevant.

This is not a regression, it happens with 4.7.2 as well.

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|                            |diagnostic
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-12-07
     Ever Confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> 2012-12-07 14:38:48 UTC ---
It's not VRP that warns here but we perform the simplification before reaching
VRP, from forward-propagation which just sees

  long b = a + C;
  if (a > b)
    something ();

and uses fold to simplify a > a + C.  There it assumes that overflow
does not occur.

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[fweimer at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

Florian Weimer <fweimer at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |than at redhat dot com

--- Comment #2 from Florian Weimer <fweimer at redhat dot com> 2013-01-31 09:50:26 UTC ---
Okular triggers this warning as well, in
dviRenderer::TPIC_addPath_special(const QString& cp), due to:

  if (TPIC_path.size() == number_of_elements_in_path)
    TPIC_path.resize(number_of_elements_in_path+100);

QVector::resize(int) is defined as:

template <typename T>
void QVector<T>::resize(int asize)
{ realloc(asize, (asize > d->alloc || (!d->capacity && asize < d->size && asize
< (d->alloc >> 1))) ?
  QVectorData::grow(sizeOfTypedData(), asize, sizeof(T),
QTypeInfo<T>::isStatic)
          : d->alloc); }

asize < d->size turns into d->size < d->size + 100 and triggers the warning.

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-31 10:08:21 UTC ---
I don't see anything bogus on the warning, it is useful to inform the developer
about potentially unintended optimization removing some conditional.

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[fweimer at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

--- Comment #4 from Florian Weimer <fweimer at redhat dot com> 2013-01-31 10:26:12 UTC ---
(In reply to comment #3)
> I don't see anything bogus on the warning, it is useful to inform the developer
> about potentially unintended optimization removing some conditional.

Neither programmer wrote (X + c) < X, this pattern does not occur in Okular nor
Qt.  I'm open to labeling the warning as "very difficult to diagnose" or
"unhelpful" in these cases, and not "bogus".  But I think it's a problem
because it distracts from the real problems this warning intends to catch.

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

--- Comment #5 from Richard Biener <rguenth at gcc dot gnu.org> 2013-01-31 10:47:55 UTC ---
(In reply to comment #4)
> (In reply to comment #3)
> > I don't see anything bogus on the warning, it is useful to inform the developer
> > about potentially unintended optimization removing some conditional.
> 
> Neither programmer wrote (X + c) < X, this pattern does not occur in Okular nor
> Qt.  I'm open to labeling the warning as "very difficult to diagnose" or
> "unhelpful" in these cases, and not "bogus".  But I think it's a problem
> because it distracts from the real problems this warning intends to catch.

Well, but the possible overflow is present in the literal
'number_of_elements_in_path+100'.  Yes, hard to track the warning down to
that possible issue, but I don't think we can improve much on that front.

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

--- Comment #6 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-31 10:52:15 UTC ---
Just write
number_of_elements_in_path+100U
or use unsigned type for
number_of_elements_in_path

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[fweimer at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

--- Comment #7 from Florian Weimer <fweimer at redhat dot com> 2013-01-31 11:05:48 UTC ---
(In reply to comment #6)
> Just write
> number_of_elements_in_path+100U
> or use unsigned type for
> number_of_elements_in_path

Thanks, this is helpful.  It seems you need both (number_of_elements_in_path
already is of type quint16), otherwise the expression is promoted to int.  I
think it might be a permanent fix (not dependent on current optimizer behavior)
because of the GCC extension ensuring that unsigned -> int conversion is always
defined.  Interesting.

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[than at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

--- Comment #8 from Than Ngo <than at redhat dot com> 2013-01-31 11:14:25 UTC ---
number_of_elements_in_path is already defined as unsigned type, so we need both
in this case to get rid of this warning

thanks

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

--- Comment #9 from Jakub Jelinek <jakub at gcc dot gnu.org> 2013-01-31 11:20:41 UTC ---
You haven't provided preprocessed testcase, if
number_of_elements_in_path is unsigned short, then supposedly the compiler
could figure out that number_of_elements_in_path + 100 never overflows (on
targets where sizeof (short) < sizeof (int)), but supposedly the optimization
in that case might not see that one of the + operands has been zero extended
and the other is either constant, or zero extended too.

[Bug tree-optimization/55616] bogus warning about undefined overflow after overflow check

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55616

--- Comment #10 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
GCC 8+ does not warn any more.  I Have not checked why yet.