[Bug rtl-optimization/58542] [4.7/4.8/4.9 Regression] Arguments of __atomic_* functions are converted in unsigned mode

[rth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=58542

Richard Henderson <rth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |rth at gcc dot gnu.org

--- Comment #10 from Richard Henderson <rth at gcc dot gnu.org> ---
Thanks for reminding me of this.

The bug isn't in expand_atomic_compare_and_swap, and the parameters
to create_convert_operand_to, because the argument is already TImode:

 <integer_cst 0x7ffff19d75e0
   type <integer_type 0x7ffff19e1a80 __int128 unsigned>
   constant -1>

The bug is either in expand_expr_* in creating a const_int instead of
a const_double, or in convert_to_mode (or its callees) in misinterpreting
the const_int.