Re: [Committed] ifcvt.c solution to PR9814

[Richard Henderson <rth at redhat dot com>]

On Sat, May 28, 2005 at 12:03:11PM -0600, Roger Sayle wrote:
> 	* ifcvt.c (noce_emit_move_insn): Construct a SET pattern directly
> 	instead of calling emit_move_insn with an arbitrary RHS.

Hmm.  How about this rearrangement:

  if (GET_CODE (x) == STRICT_LOW_PART)
    {
      enum machine_mode outmode;
      rtx outer, inner;
      int bitpos;
    
      outer = XEXP (x, 0);
      inner = XEXP (outer, 0);
      outmode = GET_MODE (outer);
      bitpos = SUBREG_BYTE (outer) * BITS_PER_UNIT;
      store_bit_field (inner, GET_MODE_BITSIZE (outmode), bitpos, outmode, y);
    }
  else if (OBJECT_P (y) || CONSTANT_P (y) || GET_CODE (y) == SUBREG)
    emit_move_insn (x, y)
  else
    {
      emit_insn (gen_rtx_SET (VOIDmode, x, y));
      ...
    }

I'm thinking of the situation in which we have a rhs constant that's
too large for a direct set, but for which the mov<mode> pattern would
have expanded to a pair of sets.

Also, you have a typo SET_DEST for SET_SRC.


r~