Is this a bug in doloop.c?

[John Lu <jlu at lsil dot com>]

Hi,

I'm working on a port based on gcc-3.4.1, and I'm
trying to use the doloop_end pattern.  In doloop.c:440-469,
the exit test generated by this pattern is analyzed in order to determine 
how to initialize the looping register.  This code sets a variable,
decrement_count, if the looping register needs to be set to N-1,
in order for the loop to iterate N times.  Currently, if the
exit test is:

	(if_then_else (ge (match_operand:QI 0 "register_operand" "")
	                  (const_int 0))

the looping register is set to N-1.  But shouldn't this be done only
if the test is:

	(if_then_else (ge (match_operand:QI 0 "register_operand" "")
	                  (const_int 1))


The relevant code from doloop.c is:

  decrement_count = 0;
  switch (GET_CODE (condition))
    {
    case NE:
      /* Currently only NE tests against zero and one are supported.  */
      if (XEXP (condition, 1) == const0_rtx)
	decrement_count = 1;
      else if (XEXP (condition, 1) != const1_rtx)
	abort ();
      break;

    case GE:
      /* Currently only GE tests against zero are supported.  */
==>   if (XEXP (condition, 1) != const0_rtx)
	abort ();

      /* The iteration count needs decrementing for a GE test.  */
      decrement_count = 1;

    ...

I believe in the marked line, const0_rtx should be replaced const1_rtx.


Thanks,
John Lu