sh.c

GCC编译器源代码

    {
      if (REGNO (src) == MACH_REG)
	return "sts	mach,%S0\n\tsts	macl,%R0";

      /* When mov.d r1,r2 do r2->r3 then r1->r2;
         when mov.d r1,r0 do r1->r0 then r2->r1.  */

      if (REGNO (src) + 1 == REGNO (dst))
	return "mov	%T1,%T0\n\tmov	%1,%0";
      else
	return "mov	%1,%0\n\tmov	%T1,%T0";
    }
  else if (GET_CODE (src) == CONST_INT)
    {
      if (INTVAL (src) < 0)
	output_asm_insn ("mov	#-1,%S0", operands);
      else
	output_asm_insn ("mov	#0,%S0", operands);

      return "mov	%1,%R0";
    }
  else if (GET_CODE (src) == MEM)
    {
      int ptrreg = -1;
      int dreg = REGNO (dst);
      rtx inside = XEXP (src, 0);

      if (GET_CODE (inside) == REG)
	ptrreg = REGNO (inside);
      else if (GET_CODE (inside) == SUBREG)
	ptrreg = REGNO (SUBREG_REG (inside)) + SUBREG_WORD (inside);
      else if (GET_CODE (inside) == PLUS)
	{
	  ptrreg = REGNO (XEXP (inside, 0));
	  /* ??? A r0+REG address shouldn't be possible here, because it isn't
	     an offsettable address.  Unfortunately, offsettable addresses use
	     QImode to check the offset, and a QImode offsettable address
	     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;
}

char *
output_far_jump (insn, op)
     rtx insn;
     rtx op;
{
  struct { rtx lab, reg, op; } this;
  char *jump;
  int far;

  this.lab = gen_label_rtx ();

  if (braf_branch_p (insn, 0))
    {
      far = 0;
      jump = "mov.w	%O0,%1;braf	%1";
    }
  else
    {
      far = 1;
      jump = "mov.l	%O0,%1;jmp	@%1";
    }
  /* If we have a scratch register available, use it.  */
  if (GET_CODE (PREV_INSN (insn)) == INSN
      && INSN_CODE (PREV_INSN (insn)) == CODE_FOR_indirect_jump_scratch)
    {
      this.reg = SET_DEST (PATTERN (PREV_INSN (insn)));
      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);
      output_asm_insn ("mov.l	r13,@-r15", 0);
      output_asm_insn (jump, &this.lab);
      output_asm_insn ("mov.l	@r15+,r13", 0);
    }
  if (far)
    output_asm_insn (".align	2", 0);
  ASM_OUTPUT_INTERNAL_LABEL (asm_out_file, "L", CODE_LABEL_NUMBER (this.lab));
  this.op = op;
  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.  */

char *
output_branch (logic, insn, operands)
     int logic;
     rtx insn;
     rtx *operands;
{
  int offset
    = (insn_addresses[INSN_UID (XEXP (XEXP (SET_SRC (PATTERN (insn)), 1), 0))]
       - insn_addresses[INSN_UID (insn)]);

  if (offset == 260
      && final_sequence
      && ! INSN_ANNULLED_BRANCH_P (XVECEXP (final_sequence, 0, 0)))
    {
      /* The filling of the delay slot has caused a forward branch to exceed
	 its range.
         Just emit the insn from the delay slot in front of the branch.  */
      /* The call to print_slot will clobber the operands.  */
      rtx op0 = operands[0];
      print_slot (final_sequence);
      operands[0] = op0;
    }
  else if (offset < -252 || offset > 258)
    {
      /* This can 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.  */

      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 "";
    }
  return logic ? "bt%.\t%l0" : "bf%.\t%l0";
}

int branch_offset ();

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

  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);
	  return template;
	}
      else
	{
	  int offset = branch_offset (next_insn) + 4;
	  if (offset >= -252 && offset <= 256)
	    {
	      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);
  return template;
}

char *
output_ieee_ccmpeq (insn, operands)
     rtx insn, operands;
{
  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;
{
  register int pos;

  output_file_directive (file, main_input_filename);

  /* Switch to the data section so that the coffsem symbol and the
     gcc2_compiled. symbol aren't in the text section.  */
  data_section ();

  if (TARGET_LITTLE_ENDIAN)
    fprintf (file, "\t.little\n");
}

/* Actual number of instructions used to make a shift by N.  */
static 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 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 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 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 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 = INTVAL (XEXP (x, 1));

  /* If shift by a non constant, then this will be expensive.  */
  if (GET_CODE (XEXP (x, 1)) != CONST_INT)
    return SH_DYNAMIC_SHIFT_COST;

  /* 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));
  /* 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 a multiply.  */
int
multcosts (x)
     rtx x;
{
  if (TARGET_SH2)
    {
      /* We have a mul insn, so we can never take more than the mul and the
	 read of the mac reg, but count more because of the latency and extra
	 reg usage.  */
      if (TARGET_SMALLCODE)
	return 2;
      return 3;
    }

  /* If we're aiming at small code, then just count the number of
     insns in a multiply call sequence.  */
  if (TARGET_SMALLCODE)
    return 5;

  /* Otherwise count all the insns in the routine we'd be calling too.  */
  return 20;
}

/* Code to expand a shift.  */

void
gen_ashift (type, n, reg)
     int type;
     int n;
     rtx reg;
{
  /* Negative values here come from the shift_amounts array.  */
  if (n < 0)
    {
      if (type == ASHIFT)
	type = LSHIFTRT;
      else
	type = ASHIFT;
      n = -n;
    }

  switch (type)
    {
    case ASHIFTRT:
      emit_insn (gen_ashrsi3_k (reg, reg, GEN_INT (n)));
      break;
    case LSHIFTRT:
      if (n == 1)
	emit_insn (gen_lshrsi3_m (reg, reg, GEN_INT (n)));
      else
	emit_insn (gen_lshrsi3_k (reg, reg, GEN_INT (n)));
      break;
    case ASHIFT:
      emit_insn (gen_ashlsi3_k (reg, reg, GEN_INT (n)));
      break;
    }
}

/* Same for HImode */

void
gen_ashift_hi (type, n, reg)
     int type;
     int n;
     rtx reg;
{
  /* Negative values here come from the shift_amounts array.  */
  if (n < 0)
    {
      if (type == ASHIFT)
	type = LSHIFTRT;
      else
	type = ASHIFT;
      n = -n;
    }

  switch (type)
    {
    case ASHIFTRT:
      emit_insn (gen_ashrhi3_k (reg, reg, GEN_INT (n)));
      break;
    case LSHIFTRT:
      if (n == 1)
	emit_insn (gen_lshrhi3_m (reg, reg, GEN_INT (n)));
      else
	emit_insn (gen_lshrhi3_k (reg, reg, GEN_INT (n)));
      break;
    case ASHIFT:
      emit_insn (gen_ashlhi3_k (reg, reg, GEN_INT (n)));
      break;
    }
}

/* Output RTL to split a constant shift into its component SH constant
   shift instructions.  */
   
int
gen_shifty_op (code, operands)
     int code;
     rtx *operands;