Re: PRE_DEC, POST_INC

[Florent Defay <spira dot inhabitant at gmail dot com>]

Thank you.

> I am assuming you already have basic generation of auto-incs and you
> have your definitions for legitimate{legitimize}_address all set up
> correctly.

Well, I think they are. But the problem could be this. Here are cuts
from the machine description dealing with auto-inc-dec:

#define HAVE_PRE_INCREMENT  0
#define HAVE_PRE_DECREMENT  1
#define HAVE_POST_INCREMENT 1
#define HAVE_POST_DECREMENT 0

int
tam16_legitimate_address_p (enum machine_mode mode ATTRIBUTE_UNUSED,
rtx x, int strict)
{
  rtx op1,op2;

  /* Post-inc and pre-dec addressing modes allowed. */
  if ((GET_CODE (x) == POST_INC || GET_CODE (x) == PRE_DEC)
      && REG_P (XEXP (x, 0))
      && (strict ? REG_OK_FOR_BASE_STRICT_P (XEXP (x, 0))
          : REG_OK_FOR_BASE_NOSTRICT_P (XEXP (x, 0)))
      /*&& reload_completed*/)
  {
    return 1;
  }
...

The legitimize address function is defined as the legitimate address
function. I took inspiration from msp430's md.
#define LEGITIMIZE_ADDRESS(X, OLDX, MODE, WIN)                          \
  GO_IF_LEGITIMATE_ADDRESS (MODE, X, WIN)

> In this case you could start by tweaking your address costs macros.

I did. Here again I could be wrong in the implementation:

int
tam16_address_costs (rtx x)
{
  switch (GET_CODE (x))
  {
    case REG:
      /* (rn) */
      return COSTS_N_INSNS (2);
      break;

    case PLUS:
      /* X(rn) */
      return COSTS_N_INSNS (4);
      break;

    case POST_INC:
    case PRE_DEC:
      /* -(rn) , (rn)+ */
      return COSTS_N_INSNS (2);
      break;

    default:
      break;
  }

  if (CONSTANT_ADDRESS_P (x))
  {
    return COSTS_N_INSNS (4);
  }

  /* Unknown addressing mode. */
  debug_rtx (x);
  return COSTS_N_INSNS (50);
}

According to you, is this enough to enable basic auto-inc-dec in GCC ?
I tried to compile some test-cases I found in emails about the
subject. The result is there is no pre-dec or post-inc generated at
all.

Regards.

Florent