[Bug c/63862] New: C frontend converts shift-count to int while standard wants integer promotions

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63862

            Bug ID: 63862
           Summary: C frontend converts shift-count to int while standard
                    wants integer promotions
           Product: gcc
           Version: 5.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: rguenth at gcc dot gnu.org
                CC: jsm28 at gcc dot gnu.org

/* Convert the non vector shift-count to an integer, regardless
             of size of value being shifted.  */
          if (TREE_CODE (TREE_TYPE (op1)) != VECTOR_TYPE
              && TYPE_MAIN_VARIANT (TREE_TYPE (op1)) != integer_type_node)
            op1 = convert (integer_type_node, op1);

while C99 6.5.7 says "The integer promotions are performed on each of the
operands".

The above pessimizes gcc.c-torture/execute/shiftopt-1.c which does

void
utest (unsigned int x)
{
  if (-1 >> x != -1)
    link_error ();
}

and thus expects us to know that the shift amount is nonnegative.  But
with the C frontend generating -1 >> (int)x we don't know that anymore
(and thus if I fix the bogus fold-const.c code which strips sign-changing
conversions on x before asking tree_expr_nonnegative_p it fails to optimize).