[Bug target/54087] __atomic_fetch_add does not use xadd instruction

["ubizjak at gmail dot com" <gcc-bugzilla at gcc dot gnu dot org>]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54087

--- Comment #5 from Uros Bizjak <ubizjak at gmail dot com> 2012-08-02 17:35:30 UTC ---
Created attachment 27927
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=27927
Patch that introduces atomic_fetch_sub<mode> with const_int operands

This patch introduces atomic_fetch_sub<mode>:

--cut here--
(define_expand "atomic_fetch_sub<mode>"
  [(set (match_operand:SWI 0 "register_operand")
    (unspec_volatile:SWI
      [(match_operand:SWI 1 "memory_operand")
       (match_operand:SI 3 "const_int_operand")]        ;; model
      UNSPECV_XCHG))
   (set (match_dup 1)
    (minus:SWI (match_dup 1)
           (match_operand:SWI 2 "const_int_operand")))
   (clobber (reg:CC FLAGS_REG))]
  "TARGET_XADD"
{
  /* Avoid overflows.  */
  if (mode_signbit_p (<MODE>mode, operands[2]))
    FAIL;

  operands[2] = negate_rtx (<MODE>mode, operands[2]);

  emit_insn (gen_atomic_fetch_add<mode> (operands[0], operands[1],
                     operands[2], operands[3]));
  DONE;
})
--cut here--

We have to take care of overflows, so we can handle only const_int operands
with known-to-be overflow free values.