[Bug middle-end/66021] GCC miscompiles Z3

[Bug middle-end/66021] GCC miscompiles Z3

[nunoplopes at sapo dot pt]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66021

--- Comment #1 from Nuno Lopes <nunoplopes at sapo dot pt> ---
Sorry, a bit more information the problem:

On function void
reduce_args_tactic::imp::populate_decl2args_proc::operator()(app * n), when
compiled with -O0 no call to memory::deallocate(void* p) is made, while with
-O2 memory::deallocate is called with p == 0, which cannot happen (since it's
called through dealloc_svect, which bails out if the pointer is null).

Apologies for not being able to reduce the test case; I don't have much
experience with the gcc internals.

[Bug middle-end/66021] GCC miscompiles Z3

[nunoplopes at sapo dot pt]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66021

--- Comment #2 from Nuno Lopes <nunoplopes at sapo dot pt> ---
Created attachment 35465
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35465&action=edit
test case

[Bug middle-end/66021] GCC miscompiles Z3

[nunoplopes at sapo dot pt]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66021

--- Comment #3 from Nuno Lopes <nunoplopes at sapo dot pt> ---
Created attachment 35467
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35467&action=edit
reduced test case

[Bug middle-end/66021] GCC miscompiles Z3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66021

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Severity|critical                    |normal

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Nuno Lopes from comment #3)
> Created attachment 35467 [details]
> reduced test case

The reduced testcase is compiled correctly in that the deallocate is called
unconditional with zero argument:
  bit_vector() : m_num_bits(0), m_capacity(0), m_data(0) {
    memory::deallocate(m_data);
  }

m_data will be NULL always when deallocate is called.

[Bug middle-end/66021] GCC miscompiles Z3

[nunoplopes at sapo dot pt]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66021

Nuno Lopes <nunoplopes at sapo dot pt> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #35467|0                           |1
        is obsolete|                            |

--- Comment #5 from Nuno Lopes <nunoplopes at sapo dot pt> ---
Created attachment 35514
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35514&action=edit
Reduced test case

Previous reduced test case was bogus. This one should be ok.

[Bug middle-end/66021] GCC miscompiles Z3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66021

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|---                         |INVALID

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Note even though the length that is passed to memcpy is zero, the arguments
have to be non-null.  So adding the check around the call to memcpy should fix
the issue.

[Bug middle-end/66021] GCC miscompiles Z3

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66021

--- Comment #7 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Basically GCC understands that memcpy takes non-null arguments and optimizes
based on that.