sh.c

linux下的gcc编译器

	     requires r0 for the other operand, which is not currently
	     supported, so we can't use the 'o' constraint.
	     Thus we must check for and handle r0+REG addresses here.
	     We punt for now, since this is likely very rare.  */
	  if (GET_CODE (XEXP (inside, 1)) == REG)
	    abort ();
	}
      else if (GET_CODE (inside) == LABEL_REF)
	return "mov.l	%1,%0\n\tmov.l	%1+4,%T0";
      else if (GET_CODE (inside) == POST_INC)
	return "mov.l	%1,%0\n\tmov.l	%1,%T0";
      else
	abort ();

      /* Work out the safe way to copy.  Copy into the second half first.  */
      if (dreg == ptrreg)
	return "mov.l	%T1,%T0\n\tmov.l	%1,%0";
    }

  return "mov.l	%1,%0\n\tmov.l	%T1,%T0";
}

/* Print an instruction which would have gone into a delay slot after
   another instruction, but couldn't because the other instruction expanded
   into a sequence where putting the slot insn at the end wouldn't work.  */

static void
print_slot (insn)
     rtx insn;
{
  final_scan_insn (XVECEXP (insn, 0, 1), asm_out_file, optimize, 0, 1);

  INSN_DELETED_P (XVECEXP (insn, 0, 1)) = 1;
}

const char *
output_far_jump (insn, op)
     rtx insn;
     rtx op;
{
  struct { rtx lab, reg, op; } this;
  rtx braf_base_lab = NULL_RTX;
  const char *jump;
  int far;
  int offset = branch_dest (insn) - INSN_ADDRESSES (INSN_UID (insn));
  rtx prev;

  this.lab = gen_label_rtx ();

  if (TARGET_SH2
      && offset >= -32764
      && offset - get_attr_length (insn) <= 32766)
    {
      far = 0;
      jump = "mov.w	%O0,%1; braf	%1";
    }
  else
    {
      far = 1;
      if (flag_pic)
	{
	  if (TARGET_SH2)
	    jump = "mov.l	%O0,%1; braf	%1";
	  else
	    jump = "mov.l	r0,@-r15; mova	%O0,r0; mov.l	@r0,%1; add	r0,%1; mov.l	@r15+,r0; jmp	@%1";
	}
      else
	jump = "mov.l	%O0,%1; jmp	@%1";
    }
  /* If we have a scratch register available, use it.  */
  if (GET_CODE ((prev = prev_nonnote_insn (insn))) == INSN
      && INSN_CODE (prev) == CODE_FOR_indirect_jump_scratch)
    {
      this.reg = SET_DEST (XVECEXP (PATTERN (prev), 0, 0));
      if (REGNO (this.reg) == R0_REG && flag_pic && ! TARGET_SH2)
	jump = "mov.l	r1,@-r15; mova	%O0,r0; mov.l	@r0,r1; add	r1,r0; mov.l	@r15+,r1; jmp	@%1";
      output_asm_insn (jump, &this.lab);
      if (dbr_sequence_length ())
	print_slot (final_sequence);
      else
	output_asm_insn ("nop", 0);
    }
  else
    {
      /* Output the delay slot insn first if any.  */
      if (dbr_sequence_length ())
	print_slot (final_sequence);

      this.reg = gen_rtx_REG (SImode, 13);
      /* We must keep the stack aligned to 8-byte boundaries on SH5.
	 Fortunately, MACL is fixed and call-clobbered, and we never
	 need its value across jumps, so save r13 in it instead of in
	 the stack.  */
      if (TARGET_SH5)
	output_asm_insn ("lds	r13, macl", 0);
      else
	output_asm_insn ("mov.l	r13,@-r15", 0);
      output_asm_insn (jump, &this.lab);
      if (TARGET_SH5)
	output_asm_insn ("sts	macl, r13", 0);
      else
	output_asm_insn ("mov.l	@r15+,r13", 0);
    }
  if (far && flag_pic && TARGET_SH2)
    {
      braf_base_lab = gen_label_rtx ();
      ASM_OUTPUT_INTERNAL_LABEL (asm_out_file, "L",
				 CODE_LABEL_NUMBER (braf_base_lab));
    }
  if (far)
    output_asm_insn (".align	2", 0);
  ASM_OUTPUT_INTERNAL_LABEL (asm_out_file, "L", CODE_LABEL_NUMBER (this.lab));
  this.op = op;
  if (far && flag_pic)
    {
      if (TARGET_SH2)
	this.lab = braf_base_lab;
      output_asm_insn (".long	%O2-%O0", &this.lab);
    }
  else
    output_asm_insn (far ? ".long	%O2" : ".word %O2-%O0", &this.lab);
  return "";
}

/* Local label counter, used for constants in the pool and inside
   pattern branches.  */

static int lf = 100;

/* Output code for ordinary branches.  */

const char *
output_branch (logic, insn, operands)
     int logic;
     rtx insn;
     rtx *operands;
{
  switch (get_attr_length (insn))
    {
    case 6:
      /* This can happen if filling the delay slot has caused a forward
	 branch to exceed its range (we could reverse it, but only
	 when we know we won't overextend other branches; this should
	 best be handled by relaxation).
	 It can also happen when other condbranches hoist delay slot insn
	 from their destination, thus leading to code size increase.
	 But the branch will still be in the range -4092..+4098 bytes.  */

      if (! TARGET_RELAX)
	{
	  int label = lf++;
	  /* The call to print_slot will clobber the operands.  */
	  rtx op0 = operands[0];
    
	  /* If the instruction in the delay slot is annulled (true), then
	     there is no delay slot where we can put it now.  The only safe
	     place for it is after the label.  final will do that by default.  */
    
	  if (final_sequence
	      && ! INSN_ANNULLED_BRANCH_P (XVECEXP (final_sequence, 0, 0)))
	    {
	      asm_fprintf (asm_out_file, "\tb%s%ss\t%LLF%d\n", logic ? "f" : "t",
	                   ASSEMBLER_DIALECT ? "/" : ".", label);
	      print_slot (final_sequence);
	    }
	  else
	    asm_fprintf (asm_out_file, "\tb%s\t%LLF%d\n", logic ? "f" : "t", label);
    
	  output_asm_insn ("bra\t%l0", &op0);
	  fprintf (asm_out_file, "\tnop\n");
	  ASM_OUTPUT_INTERNAL_LABEL(asm_out_file, "LF", label);
    
	  return "";
	}
      /* When relaxing, handle this like a short branch.  The linker
	 will fix it up if it still doesn't fit after relaxation.  */
    case 2:
      return logic ? "bt%.\t%l0" : "bf%.\t%l0";
    default:
      /* There should be no longer branches now - that would
	 indicate that something has destroyed the branches set
	 up in machine_dependent_reorg.  */
      abort ();
    }
}

const char *
output_branchy_insn (code, template, insn, operands)
     enum rtx_code code;
     const char *template;
     rtx insn;
     rtx *operands;
{
  rtx next_insn = NEXT_INSN (insn);

  if (next_insn && GET_CODE (next_insn) == JUMP_INSN && condjump_p (next_insn))
    {
      rtx src = SET_SRC (PATTERN (next_insn));
      if (GET_CODE (src) == IF_THEN_ELSE && GET_CODE (XEXP (src, 0)) != code)
	{
	  /* Following branch not taken */
	  operands[9] = gen_label_rtx ();
	  emit_label_after (operands[9], next_insn);
	  INSN_ADDRESSES_NEW (operands[9],
			      INSN_ADDRESSES (INSN_UID (next_insn))
			      + get_attr_length (next_insn));
	  return template;
	}
      else
	{
	  int offset = (branch_dest (next_insn)
			- INSN_ADDRESSES (INSN_UID (next_insn)) + 4);
	  if (offset >= -252 && offset <= 258)
	    {
	      if (GET_CODE (src) == IF_THEN_ELSE)
		/* branch_true */
		src = XEXP (src, 1);
	      operands[9] = src;
	      return template;
	    }
	}
    }
  operands[9] = gen_label_rtx ();
  emit_label_after (operands[9], insn);
  INSN_ADDRESSES_NEW (operands[9],
		      INSN_ADDRESSES (INSN_UID (insn))
		      + get_attr_length (insn));
  return template;
}

const char *
output_ieee_ccmpeq (insn, operands)
     rtx insn, *operands;
{
  return output_branchy_insn (NE, "bt\t%l9\\;fcmp/eq\t%1,%0", insn, operands);
}

/* Output to FILE the start of the assembler file.  */

void
output_file_start (file)
     FILE *file;
{
  output_file_directive (file, main_input_filename);

  /* Switch to the data section so that the coffsem symbol
     isn't in the text section.  */
  data_section ();

  if (TARGET_LITTLE_ENDIAN)
    fprintf (file, "\t.little\n");

  if (TARGET_SHCOMPACT)
    fprintf (file, "\t.mode\tSHcompact\n");
  else if (TARGET_SHMEDIA)
    fprintf (file, "\t.mode\tSHmedia\n\t.abi\t%i\n",
	     TARGET_SHMEDIA64 ? 64 : 32);
}

/* Actual number of instructions used to make a shift by N.  */
static const char ashiftrt_insns[] =
  { 0,1,2,3,4,5,8,8,8,8,8,8,8,8,8,8,2,3,4,5,8,8,8,8,8,8,8,8,8,8,8,2};

/* Left shift and logical right shift are the same.  */
static const char shift_insns[]    =
  { 0,1,1,2,2,3,3,4,1,2,2,3,3,4,3,3,1,2,2,3,3,4,3,3,2,3,3,4,4,4,3,3};

/* Individual shift amounts needed to get the above length sequences.
   One bit right shifts clobber the T bit, so when possible, put one bit
   shifts in the middle of the sequence, so the ends are eligible for
   branch delay slots.  */
static const short shift_amounts[32][5] = {
  {0}, {1}, {2}, {2, 1},
  {2, 2}, {2, 1, 2}, {2, 2, 2}, {2, 2, 1, 2},
  {8}, {8, 1}, {8, 2}, {8, 1, 2},
  {8, 2, 2}, {8, 2, 1, 2}, {8, -2, 8}, {8, -1, 8},
  {16}, {16, 1}, {16, 2}, {16, 1, 2},
  {16, 2, 2}, {16, 2, 1, 2}, {16, -2, 8}, {16, -1, 8},
  {16, 8}, {16, 1, 8}, {16, 8, 2}, {16, 8, 1, 2},
  {16, 8, 2, 2}, {16, -1, -2, 16}, {16, -2, 16}, {16, -1, 16}};

/* Likewise, but for shift amounts < 16, up to three highmost bits
   might be clobbered.  This is typically used when combined with some
   kind of sign or zero extension.  */
   
static const char ext_shift_insns[]    =
  { 0,1,1,2,2,3,2,2,1,2,2,3,3,3,2,2,1,2,2,3,3,4,3,3,2,3,3,4,4,4,3,3};

static const short ext_shift_amounts[32][4] = {
  {0}, {1}, {2}, {2, 1},
  {2, 2}, {2, 1, 2}, {8, -2}, {8, -1},
  {8}, {8, 1}, {8, 2}, {8, 1, 2},
  {8, 2, 2}, {16, -2, -1}, {16, -2}, {16, -1},
  {16}, {16, 1}, {16, 2}, {16, 1, 2},
  {16, 2, 2}, {16, 2, 1, 2}, {16, -2, 8}, {16, -1, 8},
  {16, 8}, {16, 1, 8}, {16, 8, 2}, {16, 8, 1, 2},
  {16, 8, 2, 2}, {16, -1, -2, 16}, {16, -2, 16}, {16, -1, 16}};

/* Assuming we have a value that has been sign-extended by at least one bit,
   can we use the ext_shift_amounts with the last shift turned to an arithmetic shift
   to shift it by N without data loss, and quicker than by other means?  */
#define EXT_SHIFT_SIGNED(n) (((n) | 8) == 15)

/* This is used in length attributes in sh.md to help compute the length
   of arbitrary constant shift instructions.  */

int
shift_insns_rtx (insn)
     rtx insn;
{
  rtx set_src = SET_SRC (XVECEXP (PATTERN (insn), 0, 0));
  int shift_count = INTVAL (XEXP (set_src, 1));
  enum rtx_code shift_code = GET_CODE (set_src);

  switch (shift_code)
    {
    case ASHIFTRT:
      return ashiftrt_insns[shift_count];
    case LSHIFTRT:
    case ASHIFT:
      return shift_insns[shift_count];
    default:
      abort();
    }
}

/* Return the cost of a shift.  */

int
shiftcosts (x)
     rtx x;
{
  int value;

  if (TARGET_SHMEDIA)
    return 1;

  if (GET_MODE_SIZE (GET_MODE (x)) > UNITS_PER_WORD)
    {
      if (GET_MODE (x) == DImode
	  && GET_CODE (XEXP (x, 1)) == CONST_INT
	  && INTVAL (XEXP (x, 1)) == 1)
	return 2;

      /* Everything else is invalid, because there is no pattern for it.  */
      return 10000;
    }
  /* If shift by a non constant, then this will be expensive.  */
  if (GET_CODE (XEXP (x, 1)) != CONST_INT)
    return SH_DYNAMIC_SHIFT_COST;

  value = INTVAL (XEXP (x, 1));

  /* Otherwise, return the true cost in instructions.  */
  if (GET_CODE (x) == ASHIFTRT)
    {
      int cost = ashiftrt_insns[value];
      /* If SH3, then we put the constant in a reg and use shad.  */
      if (cost > 1 + SH_DYNAMIC_SHIFT_COST)
	cost = 1 + SH_DYNAMIC_SHIFT_COST;
      return cost;
    }
  else
    return shift_insns[value];
}

/* Return the cost of an AND operation.  */

int
andcosts (x)
     rtx x;
{
  int i;

  /* Anding with a register is a single cycle and instruction.  */
  if (GET_CODE (XEXP (x, 1)) != CONST_INT)
    return 1;

  i = INTVAL (XEXP (x, 1));

  if (TARGET_SHMEDIA)
    {
      if ((GET_CODE (XEXP (x, 1)) == CONST_INT
	   && CONST_OK_FOR_J (INTVAL (XEXP (x, 1))))
	  || EXTRA_CONSTRAINT_S (XEXP (x, 1)))
	return 1;
      else
	return 2;
    }

  /* These constants are single cycle extu.[bw] instructions.  */
  if (i == 0xff || i == 0xffff)
    return 1;
  /* Constants that can be used in an and immediate instruction is a single
     cycle, but this requires r0, so make it a little more expensive.  */
  if (CONST_OK_FOR_L (i))
    return 2;
  /* Constants that can be loaded with a mov immediate and an and.
     This case is probably unnecessary.  */
  if (CONST_OK_FOR_I (i))
    return 2;
  /* Any other constants requires a 2 cycle pc-relative load plus an and.
     This case is probably unnecessary.  */
  return 3;
}

/* Return the cost of an addition or a subtraction.  */

int
addsubcosts (x)
     rtx x;
{
  /* Adding a register is a single cycle insn.  */
  if (GET_CODE (XEXP (x, 1)) == REG
      || GET_CODE (XEXP (x, 1)) == SUBREG)
    return 1;

  /* Likewise for small constants.  */
  if (GET_CODE (XEXP (x, 1)) == CONST_INT
      && CONST_OK_FOR_ADD (INTVAL (XEXP (x, 1))))
    return 1;

  if (TARGET_SHMEDIA)
    switch (GET_CODE (XEXP (x, 1)))
      {
      case CONST:
      case LABEL_REF:
      case SYMBOL_REF:
	return TARGET_SHMEDIA64 ? 5 : 3;

      case CONST_INT:
	if (CONST_OK_FOR_J (INTVAL (XEXP (x, 1))))
          return 2;
	else if (CONST_OK_FOR_J (INTVAL (XEXP (x, 1)) >> 16))
	  return 3;
	else if (CONST_OK_FOR_J ((INTVAL (XEXP (x, 1)) >> 16) >> 16))
	  return 4;

	/* Fall through.  */
      default:
	  return 5;
      }

  /* Any other constant requires a 2 cycle pc-relative load plus an
     addition.  */
  return 3;
}

/* Return the cost of a multiply.  */
int
multcosts (x)
     rtx x ATTRIBUTE_UNUSED;
{
  if (TARGET_SHMEDIA)
    return 3;