[pf3gnuchains/gcc-fork.git] / gcc / config / avr / avr.c

/* Subroutines for insn-output.c for ATMEL AVR micro controllers
   Copyright (C) 1998, 1999, 2000, 2001, 2002, 2004, 2005, 2006, 2007, 2008,
   2009, 2010, 2011 Free Software Foundation, Inc.
   Contributed by Denis Chertykov (chertykov@gmail.com)

   This file is part of GCC.

   GCC is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 3, or (at your option)
   any later version.
   
   GCC is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.
   
   You should have received a copy of the GNU General Public License
   along with GCC; see the file COPYING3.  If not see
   <http://www.gnu.org/licenses/>.  */

#include "config.h"
#include "system.h"
#include "coretypes.h"
#include "tm.h"
#include "rtl.h"
#include "regs.h"
#include "hard-reg-set.h"
#include "insn-config.h"
#include "conditions.h"
#include "insn-attr.h"
#include "insn-codes.h"
#include "flags.h"
#include "reload.h"
#include "tree.h"
#include "output.h"
#include "expr.h"
#include "c-family/c-common.h"
#include "diagnostic-core.h"
#include "obstack.h"
#include "function.h"
#include "recog.h"
#include "optabs.h"
#include "ggc.h"
#include "langhooks.h"
#include "tm_p.h"
#include "target.h"
#include "target-def.h"
#include "params.h"
#include "df.h"

/* Maximal allowed offset for an address in the LD command */
#define MAX_LD_OFFSET(MODE) (64 - (signed)GET_MODE_SIZE (MODE))

/* Return true if STR starts with PREFIX and false, otherwise.  */
#define STR_PREFIX_P(STR,PREFIX) (0 == strncmp (STR, PREFIX, strlen (PREFIX)))

/* The 4 bits starting at SECTION_MACH_DEP are reserved to store the
   address space where data is to be located.
   As the only non-generic address spaces are all located in Flash,
   this can be used to test if data shall go into some .progmem* section.
   This must be the rightmost field of machine dependent section flags.  */
#define AVR_SECTION_PROGMEM (0xf * SECTION_MACH_DEP)

/* Similar 4-bit region for SYMBOL_REF_FLAGS.  */
#define AVR_SYMBOL_FLAG_PROGMEM (0xf * SYMBOL_FLAG_MACH_DEP)

/* Similar 4-bit region in SYMBOL_REF_FLAGS:
   Set address-space AS in SYMBOL_REF_FLAGS of SYM  */
#define AVR_SYMBOL_SET_ADDR_SPACE(SYM,AS)                       \
  do {                                                          \
    SYMBOL_REF_FLAGS (sym) &= ~AVR_SYMBOL_FLAG_PROGMEM;         \
    SYMBOL_REF_FLAGS (sym) |= (AS) * SYMBOL_FLAG_MACH_DEP;      \
  } while (0)

/* Read address-space from SYMBOL_REF_FLAGS of SYM  */
#define AVR_SYMBOL_GET_ADDR_SPACE(SYM)                          \
  ((SYMBOL_REF_FLAGS (sym) & AVR_SYMBOL_FLAG_PROGMEM)           \
   / SYMBOL_FLAG_MACH_DEP)

/* Known address spaces.  The order must be the same as in the respective
   enum from avr.h (or designated initialized must be used).  */
const avr_addrspace_t avr_addrspace[] =
{
    { ADDR_SPACE_RAM,  0, 2, ""     ,   0 },
    { ADDR_SPACE_FLASH,  1, 2, "__flash",   0 },
    { ADDR_SPACE_FLASH1, 1, 2, "__flash1",  1 },
    { ADDR_SPACE_FLASH2, 1, 2, "__flash2",  2 },
    { ADDR_SPACE_FLASH3, 1, 2, "__flash3",  3 },
    { ADDR_SPACE_FLASH4, 1, 2, "__flash4",  4 },
    { ADDR_SPACE_FLASH5, 1, 2, "__flash5",  5 },
    { ADDR_SPACE_MEMX, 1, 3, "__memx",  0 },
    { 0              , 0, 0, NULL,      0 }
};

/* Map 64-k Flash segment to section prefix.  */
static const char* const progmem_section_prefix[6] =
  {
    ".progmem.data",
    ".progmem1.data",
    ".progmem2.data",
    ".progmem3.data",
    ".progmem4.data",
    ".progmem5.data"
  };

/* Holding RAM addresses of some SFRs used by the compiler and that
   are unique over all devices in an architecture like 'avr4'.  */
  
typedef struct
{
  /* SREG: The pocessor status */
  int sreg;

  /* RAMPX, RAMPY, RAMPD and CCP of XMEGA */
  int ccp;
  int rampd;
  int rampx;
  int rampy;

  /* RAMPZ: The high byte of 24-bit address used with ELPM */ 
  int rampz;

  /* SP: The stack pointer and its low and high byte */
  int sp_l;
  int sp_h;
} avr_addr_t;

static avr_addr_t avr_addr;


/* Prototypes for local helper functions.  */

static const char* out_movqi_r_mr (rtx, rtx[], int*);
static const char* out_movhi_r_mr (rtx, rtx[], int*);
static const char* out_movsi_r_mr (rtx, rtx[], int*);
static const char* out_movqi_mr_r (rtx, rtx[], int*);
static const char* out_movhi_mr_r (rtx, rtx[], int*);
static const char* out_movsi_mr_r (rtx, rtx[], int*);

static int avr_naked_function_p (tree);
static int interrupt_function_p (tree);
static int signal_function_p (tree);
static int avr_OS_task_function_p (tree);
static int avr_OS_main_function_p (tree);
static int avr_regs_to_save (HARD_REG_SET *);
static int get_sequence_length (rtx insns);
static int sequent_regs_live (void);
static const char *ptrreg_to_str (int);
static const char *cond_string (enum rtx_code);
static int avr_num_arg_regs (enum machine_mode, const_tree);
static int avr_operand_rtx_cost (rtx, enum machine_mode, enum rtx_code,
                                 int, bool);
static void output_reload_in_const (rtx*, rtx, int*, bool);
static struct machine_function * avr_init_machine_status (void);


/* Prototypes for hook implementors if needed before their implementation.  */

static bool avr_rtx_costs (rtx, int, int, int, int *, bool);


/* Allocate registers from r25 to r8 for parameters for function calls.  */
#define FIRST_CUM_REG 26

/* Implicit target register of LPM instruction (R0) */
extern GTY(()) rtx lpm_reg_rtx;
rtx lpm_reg_rtx;

/* (Implicit) address register of LPM instruction (R31:R30 = Z) */
extern GTY(()) rtx lpm_addr_reg_rtx;
rtx lpm_addr_reg_rtx;

/* Temporary register RTX (reg:QI TMP_REGNO) */
extern GTY(()) rtx tmp_reg_rtx;
rtx tmp_reg_rtx;

/* Zeroed register RTX (reg:QI ZERO_REGNO) */
extern GTY(()) rtx zero_reg_rtx;
rtx zero_reg_rtx;

/* RTXs for all general purpose registers as QImode */
extern GTY(()) rtx all_regs_rtx[32];
rtx all_regs_rtx[32];

/* SREG, the processor status */
extern GTY(()) rtx sreg_rtx;
rtx sreg_rtx;

/* RAMP* special function registers */
extern GTY(()) rtx rampd_rtx;
extern GTY(()) rtx rampx_rtx;
extern GTY(()) rtx rampy_rtx;
extern GTY(()) rtx rampz_rtx;
rtx rampd_rtx;
rtx rampx_rtx;
rtx rampy_rtx;
rtx rampz_rtx;

/* RTX containing the strings "" and "e", respectively */
static GTY(()) rtx xstring_empty;
static GTY(()) rtx xstring_e;

/* Preprocessor macros to define depending on MCU type.  */
const char *avr_extra_arch_macro;

/* Current architecture.  */
const struct base_arch_s *avr_current_arch;

/* Current device.  */
const struct mcu_type_s *avr_current_device;

/* Section to put switch tables in.  */
static GTY(()) section *progmem_swtable_section;

/* Unnamed sections associated to __attribute__((progmem)) aka. PROGMEM
   or to address space __flash*.  */
static GTY(()) section *progmem_section[6];

/* Condition for insns/expanders from avr-dimode.md.  */
bool avr_have_dimode = true;

/* To track if code will use .bss and/or .data.  */
bool avr_need_clear_bss_p = false;
bool avr_need_copy_data_p = false;

\f

/* Custom function to count number of set bits.  */

static inline int
avr_popcount (unsigned int val)
{
  int pop = 0;

  while (val)
    {
      val &= val-1;
      pop++;
    }

  return pop;
}


/* Constraint helper function.  XVAL is a CONST_INT or a CONST_DOUBLE.
   Return true if the least significant N_BYTES bytes of XVAL all have a
   popcount in POP_MASK and false, otherwise.  POP_MASK represents a subset
   of integers which contains an integer N iff bit N of POP_MASK is set.  */
   
bool
avr_popcount_each_byte (rtx xval, int n_bytes, int pop_mask)
{
  int i;

  enum machine_mode mode = GET_MODE (xval);

  if (VOIDmode == mode)
    mode = SImode;

  for (i = 0; i < n_bytes; i++)
    {
      rtx xval8 = simplify_gen_subreg (QImode, xval, mode, i);
      unsigned int val8 = UINTVAL (xval8) & GET_MODE_MASK (QImode);

      if (0 == (pop_mask & (1 << avr_popcount (val8))))
        return false;
    }

  return true;
}

static void
avr_option_override (void)
{
  flag_delete_null_pointer_checks = 0;

  /* caller-save.c looks for call-clobbered hard registers that are assigned
     to pseudos that cross calls and tries so save-restore them around calls
     in order to reduce the number of stack slots needed.

     This might leads to situations where reload is no more able to cope
     with the challenge of AVR's very few address registers and fails to
     perform the requested spills.  */
  
  if (avr_strict_X)
    flag_caller_saves = 0;

  /* Unwind tables currently require a frame pointer for correctness,
     see toplev.c:process_options().  */

  if ((flag_unwind_tables
       || flag_non_call_exceptions
       || flag_asynchronous_unwind_tables)
      && !ACCUMULATE_OUTGOING_ARGS)
    {
      flag_omit_frame_pointer = 0;
    }

  avr_current_device = &avr_mcu_types[avr_mcu_index];
  avr_current_arch = &avr_arch_types[avr_current_device->arch];
  avr_extra_arch_macro = avr_current_device->macro;
  
  /* RAM addresses of some SFRs common to all Devices in respective Arch. */

  /* SREG: Status Register containing flags like I (global IRQ) */
  avr_addr.sreg = 0x3F + avr_current_arch->sfr_offset;

  /* RAMPZ: Address' high part when loading via ELPM */
  avr_addr.rampz = 0x3B + avr_current_arch->sfr_offset;

  avr_addr.rampy = 0x3A + avr_current_arch->sfr_offset;
  avr_addr.rampx = 0x39 + avr_current_arch->sfr_offset;
  avr_addr.rampd = 0x38 + avr_current_arch->sfr_offset;
  avr_addr.ccp = 0x34 + avr_current_arch->sfr_offset;

  /* SP: Stack Pointer (SP_H:SP_L) */
  avr_addr.sp_l = 0x3D + avr_current_arch->sfr_offset;
  avr_addr.sp_h = avr_addr.sp_l + 1;

  init_machine_status = avr_init_machine_status;

  avr_log_set_avr_log();
}

/* Function to set up the backend function structure.  */

static struct machine_function *
avr_init_machine_status (void)
{
  return ggc_alloc_cleared_machine_function ();
}


/* Implement `INIT_EXPANDERS'.  */
/* The function works like a singleton.  */

void
avr_init_expanders (void)
{
  int regno;

  for (regno = 0; regno < 32; regno ++)
    all_regs_rtx[regno] = gen_rtx_REG (QImode, regno);

  lpm_reg_rtx  = all_regs_rtx[LPM_REGNO];
  tmp_reg_rtx  = all_regs_rtx[TMP_REGNO];
  zero_reg_rtx = all_regs_rtx[ZERO_REGNO];

  lpm_addr_reg_rtx = gen_rtx_REG (HImode, REG_Z);

  sreg_rtx = gen_rtx_MEM (QImode, GEN_INT (avr_addr.sreg));
  rampd_rtx = gen_rtx_MEM (QImode, GEN_INT (avr_addr.rampd));
  rampx_rtx = gen_rtx_MEM (QImode, GEN_INT (avr_addr.rampx));
  rampy_rtx = gen_rtx_MEM (QImode, GEN_INT (avr_addr.rampy));
  rampz_rtx = gen_rtx_MEM (QImode, GEN_INT (avr_addr.rampz));

  xstring_empty = gen_rtx_CONST_STRING (VOIDmode, "");
  xstring_e = gen_rtx_CONST_STRING (VOIDmode, "e");
}


/* Return register class for register R.  */

enum reg_class
avr_regno_reg_class (int r)
{
  static const enum reg_class reg_class_tab[] =
    {
      R0_REG,
      /* r1 - r15 */
      NO_LD_REGS, NO_LD_REGS, NO_LD_REGS,
      NO_LD_REGS, NO_LD_REGS, NO_LD_REGS, NO_LD_REGS,
      NO_LD_REGS, NO_LD_REGS, NO_LD_REGS, NO_LD_REGS,
      NO_LD_REGS, NO_LD_REGS, NO_LD_REGS, NO_LD_REGS,
      /* r16 - r23 */
      SIMPLE_LD_REGS, SIMPLE_LD_REGS, SIMPLE_LD_REGS, SIMPLE_LD_REGS,
      SIMPLE_LD_REGS, SIMPLE_LD_REGS, SIMPLE_LD_REGS, SIMPLE_LD_REGS,
      /* r24, r25 */
      ADDW_REGS, ADDW_REGS,
      /* X: r26, 27 */
      POINTER_X_REGS, POINTER_X_REGS,
      /* Y: r28, r29 */
      POINTER_Y_REGS, POINTER_Y_REGS,
      /* Z: r30, r31 */
      POINTER_Z_REGS, POINTER_Z_REGS,
      /* SP: SPL, SPH */
      STACK_REG, STACK_REG
    };

  if (r <= 33)
    return reg_class_tab[r];
  
  return ALL_REGS;
}


static bool
avr_scalar_mode_supported_p (enum machine_mode mode)
{
  if (PSImode == mode)
    return true;

  return default_scalar_mode_supported_p (mode);
}


/* Return TRUE if DECL is a VAR_DECL located in Flash and FALSE, otherwise.  */

static bool
avr_decl_flash_p (tree decl)
{
  if (TREE_CODE (decl) != VAR_DECL
      || TREE_TYPE (decl) == error_mark_node)
    {
      return false;
    }

  return !ADDR_SPACE_GENERIC_P (TYPE_ADDR_SPACE (TREE_TYPE (decl)));
}


/* Return TRUE if DECL is a VAR_DECL located in the 24-bit Flash
   address space and FALSE, otherwise.  */
 
static bool
avr_decl_memx_p (tree decl)
{
  if (TREE_CODE (decl) != VAR_DECL
      || TREE_TYPE (decl) == error_mark_node)
    {
      return false;
    }

  return (ADDR_SPACE_MEMX == TYPE_ADDR_SPACE (TREE_TYPE (decl)));
}


/* Return TRUE if X is a MEM rtx located in Flash and FALSE, otherwise.  */

bool
avr_mem_flash_p (rtx x)
{
  return (MEM_P (x)
          && !ADDR_SPACE_GENERIC_P (MEM_ADDR_SPACE (x)));
}


/* Return TRUE if X is a MEM rtx located in the 24-bit Flash
   address space and FALSE, otherwise.  */

bool
avr_mem_memx_p (rtx x)
{
  return (MEM_P (x)
          && ADDR_SPACE_MEMX == MEM_ADDR_SPACE (x));
}


/* A helper for the subsequent function attribute used to dig for
   attribute 'name' in a FUNCTION_DECL or FUNCTION_TYPE */

static inline int
avr_lookup_function_attribute1 (const_tree func, const char *name)
{
  if (FUNCTION_DECL == TREE_CODE (func))
    {
      if (NULL_TREE != lookup_attribute (name, DECL_ATTRIBUTES (func)))
        {
          return true;
        }
      
      func = TREE_TYPE (func);
    }

  gcc_assert (TREE_CODE (func) == FUNCTION_TYPE
              || TREE_CODE (func) == METHOD_TYPE);
  
  return NULL_TREE != lookup_attribute (name, TYPE_ATTRIBUTES (func));
}

/* Return nonzero if FUNC is a naked function.  */

static int
avr_naked_function_p (tree func)
{
  return avr_lookup_function_attribute1 (func, "naked");
}

/* Return nonzero if FUNC is an interrupt function as specified
   by the "interrupt" attribute.  */

static int
interrupt_function_p (tree func)
{
  return avr_lookup_function_attribute1 (func, "interrupt");
}

/* Return nonzero if FUNC is a signal function as specified
   by the "signal" attribute.  */

static int
signal_function_p (tree func)
{
  return avr_lookup_function_attribute1 (func, "signal");
}

/* Return nonzero if FUNC is an OS_task function.  */

static int
avr_OS_task_function_p (tree func)
{
  return avr_lookup_function_attribute1 (func, "OS_task");
}

/* Return nonzero if FUNC is an OS_main function.  */

static int
avr_OS_main_function_p (tree func)
{
  return avr_lookup_function_attribute1 (func, "OS_main");
}


/* Implement `ACCUMULATE_OUTGOING_ARGS'.  */

int
avr_accumulate_outgoing_args (void)
{
  if (!cfun)
    return TARGET_ACCUMULATE_OUTGOING_ARGS;

  /* FIXME: For setjmp and in avr_builtin_setjmp_frame_value we don't know
        what offset is correct.  In some cases it is relative to
        virtual_outgoing_args_rtx and in others it is relative to
        virtual_stack_vars_rtx.  For example code see
            gcc.c-torture/execute/built-in-setjmp.c
            gcc.c-torture/execute/builtins/sprintf-chk.c   */
  
  return (TARGET_ACCUMULATE_OUTGOING_ARGS
          && !(cfun->calls_setjmp
               || cfun->has_nonlocal_label));
}


/* Report contribution of accumulated outgoing arguments to stack size.  */

static inline int
avr_outgoing_args_size (void)
{
  return ACCUMULATE_OUTGOING_ARGS ? crtl->outgoing_args_size : 0;
}


/* Implement `STARTING_FRAME_OFFSET'.  */
/* This is the offset from the frame pointer register to the first stack slot
   that contains a variable living in the frame.  */

int
avr_starting_frame_offset (void)
{
  return 1 + avr_outgoing_args_size ();
}


/* Return the number of hard registers to push/pop in the prologue/epilogue
   of the current function, and optionally store these registers in SET.  */

static int
avr_regs_to_save (HARD_REG_SET *set)
{
  int reg, count;
  int int_or_sig_p = (interrupt_function_p (current_function_decl)
                      || signal_function_p (current_function_decl));

  if (set)
    CLEAR_HARD_REG_SET (*set);
  count = 0;

  /* No need to save any registers if the function never returns or 
     has the "OS_task" or "OS_main" attribute.  */
  if (TREE_THIS_VOLATILE (current_function_decl)
      || cfun->machine->is_OS_task
      || cfun->machine->is_OS_main)
    return 0;

  for (reg = 0; reg < 32; reg++)
    {
      /* Do not push/pop __tmp_reg__, __zero_reg__, as well as
         any global register variables.  */
      if (fixed_regs[reg])
        continue;

      if ((int_or_sig_p && !current_function_is_leaf && call_used_regs[reg])
          || (df_regs_ever_live_p (reg)
              && (int_or_sig_p || !call_used_regs[reg])
              /* Don't record frame pointer registers here.  They are treated
                 indivitually in prologue.  */
              && !(frame_pointer_needed
                   && (reg == REG_Y || reg == (REG_Y+1)))))
        {
          if (set)
            SET_HARD_REG_BIT (*set, reg);
          count++;
        }
    }
  return count;
}

/* Return true if register FROM can be eliminated via register TO.  */

static bool
avr_can_eliminate (const int from, const int to)
{
  return ((from == ARG_POINTER_REGNUM && to == FRAME_POINTER_REGNUM)
          || (frame_pointer_needed && to == FRAME_POINTER_REGNUM)
          || ((from == FRAME_POINTER_REGNUM 
               || from == FRAME_POINTER_REGNUM + 1)
              && !frame_pointer_needed));
}

/* Compute offset between arg_pointer and frame_pointer.  */

int
avr_initial_elimination_offset (int from, int to)
{
  if (from == FRAME_POINTER_REGNUM && to == STACK_POINTER_REGNUM)
    return 0;
  else
    {
      int offset = frame_pointer_needed ? 2 : 0;
      int avr_pc_size = AVR_HAVE_EIJMP_EICALL ? 3 : 2;
      
      offset += avr_regs_to_save (NULL);
      return (get_frame_size () + avr_outgoing_args_size()
              + avr_pc_size + 1 + offset);
    }
}

/* Actual start of frame is virtual_stack_vars_rtx this is offset from 
   frame pointer by +STARTING_FRAME_OFFSET.
   Using saved frame = virtual_stack_vars_rtx - STARTING_FRAME_OFFSET
   avoids creating add/sub of offset in nonlocal goto and setjmp.  */

static rtx
avr_builtin_setjmp_frame_value (void)
{
  return gen_rtx_MINUS (Pmode, virtual_stack_vars_rtx, 
                        gen_int_mode (STARTING_FRAME_OFFSET, Pmode));
}

/* Return contents of MEM at frame pointer + stack size + 1 (+2 if 3 byte PC).
   This is return address of function.  */
rtx 
avr_return_addr_rtx (int count, rtx tem)
{
  rtx r;
    
  /* Can only return this function's return address. Others not supported.  */
  if (count)
     return NULL;

  if (AVR_3_BYTE_PC)
    {
      r = gen_rtx_SYMBOL_REF (Pmode, ".L__stack_usage+2");
      warning (0, "'builtin_return_address' contains only 2 bytes of address");
    }
  else
    r = gen_rtx_SYMBOL_REF (Pmode, ".L__stack_usage+1");

  r = gen_rtx_PLUS (Pmode, tem, r);
  r = gen_frame_mem (Pmode, memory_address (Pmode, r));
  r = gen_rtx_ROTATE (HImode, r, GEN_INT (8));
  return  r;
}

/* Return 1 if the function epilogue is just a single "ret".  */

int
avr_simple_epilogue (void)
{
  return (! frame_pointer_needed
          && get_frame_size () == 0
          && avr_outgoing_args_size() == 0
          && avr_regs_to_save (NULL) == 0
          && ! interrupt_function_p (current_function_decl)
          && ! signal_function_p (current_function_decl)
          && ! avr_naked_function_p (current_function_decl)
          && ! TREE_THIS_VOLATILE (current_function_decl));
}

/* This function checks sequence of live registers.  */

static int
sequent_regs_live (void)
{
  int reg;
  int live_seq=0;
  int cur_seq=0;

  for (reg = 0; reg < 18; ++reg)
    {
      if (fixed_regs[reg])
        {
          /* Don't recognize sequences that contain global register
             variables.  */
      
          if (live_seq != 0)
            return 0;
          else
            continue;
        }
      
      if (!call_used_regs[reg])
        {
          if (df_regs_ever_live_p (reg))
            {
              ++live_seq;
              ++cur_seq;
            }
          else
            cur_seq = 0;
        }
    }

  if (!frame_pointer_needed)
    {
      if (df_regs_ever_live_p (REG_Y))
        {
          ++live_seq;
          ++cur_seq;
        }
      else
        cur_seq = 0;

      if (df_regs_ever_live_p (REG_Y+1))
        {
          ++live_seq;
          ++cur_seq;
        }
      else
        cur_seq = 0;
    }
  else
    {
      cur_seq += 2;
      live_seq += 2;
    }
  return (cur_seq == live_seq) ? live_seq : 0;
}

/* Obtain the length sequence of insns.  */

int
get_sequence_length (rtx insns)
{
  rtx insn;
  int length;
  
  for (insn = insns, length = 0; insn; insn = NEXT_INSN (insn))
    length += get_attr_length (insn);
                
  return length;
}

/*  Implement INCOMING_RETURN_ADDR_RTX.  */

rtx
avr_incoming_return_addr_rtx (void)
{
  /* The return address is at the top of the stack.  Note that the push
     was via post-decrement, which means the actual address is off by one.  */
  return gen_frame_mem (HImode, plus_constant (stack_pointer_rtx, 1));
}

/*  Helper for expand_prologue.  Emit a push of a byte register.  */

static void
emit_push_byte (unsigned regno, bool frame_related_p)
{
  rtx mem, reg, insn;

  mem = gen_rtx_POST_DEC (HImode, stack_pointer_rtx);
  mem = gen_frame_mem (QImode, mem);
  reg = gen_rtx_REG (QImode, regno);

  insn = emit_insn (gen_rtx_SET (VOIDmode, mem, reg));
  if (frame_related_p)
    RTX_FRAME_RELATED_P (insn) = 1;

  cfun->machine->stack_usage++;
}


/*  Helper for expand_prologue.  Emit a push of a SFR via tmp_reg.
    SFR is a MEM representing the memory location of the SFR.
    If CLR_P then clear the SFR after the push using zero_reg.  */

static void
emit_push_sfr (rtx sfr, bool frame_related_p, bool clr_p)
{
  rtx insn;
  
  gcc_assert (MEM_P (sfr));

  /* IN __tmp_reg__, IO(SFR) */
  insn = emit_move_insn (tmp_reg_rtx, sfr);
  if (frame_related_p)
    RTX_FRAME_RELATED_P (insn) = 1;
  
  /* PUSH __tmp_reg__ */
  emit_push_byte (TMP_REGNO, frame_related_p);

  if (clr_p)
    {
      /* OUT IO(SFR), __zero_reg__ */
      insn = emit_move_insn (sfr, const0_rtx);
      if (frame_related_p)
        RTX_FRAME_RELATED_P (insn) = 1;
    }
}

static void
avr_prologue_setup_frame (HOST_WIDE_INT size, HARD_REG_SET set)
{
  rtx insn;
  bool isr_p = cfun->machine->is_interrupt || cfun->machine->is_signal;
  int live_seq = sequent_regs_live ();

  HOST_WIDE_INT size_max
    = (HOST_WIDE_INT) GET_MODE_MASK (AVR_HAVE_8BIT_SP ? QImode : Pmode);

  bool minimize = (TARGET_CALL_PROLOGUES
                   && size < size_max
                   && live_seq
                   && !isr_p
                   && !cfun->machine->is_OS_task
                   && !cfun->machine->is_OS_main);
  
  if (minimize
      && (frame_pointer_needed
          || avr_outgoing_args_size() > 8
          || (AVR_2_BYTE_PC && live_seq > 6)
          || live_seq > 7)) 
    {
      rtx pattern;
      int first_reg, reg, offset;

      emit_move_insn (gen_rtx_REG (HImode, REG_X), 
                      gen_int_mode (size, HImode));

      pattern = gen_call_prologue_saves (gen_int_mode (live_seq, HImode),
                                         gen_int_mode (live_seq+size, HImode));
      insn = emit_insn (pattern);
      RTX_FRAME_RELATED_P (insn) = 1;

      /* Describe the effect of the unspec_volatile call to prologue_saves.
         Note that this formulation assumes that add_reg_note pushes the
         notes to the front.  Thus we build them in the reverse order of
         how we want dwarf2out to process them.  */

      /* The function does always set frame_pointer_rtx, but whether that
         is going to be permanent in the function is frame_pointer_needed.  */

      add_reg_note (insn, REG_CFA_ADJUST_CFA,
                    gen_rtx_SET (VOIDmode, (frame_pointer_needed
                                            ? frame_pointer_rtx
                                            : stack_pointer_rtx),
                                 plus_constant (stack_pointer_rtx,
                                                -(size + live_seq))));

      /* Note that live_seq always contains r28+r29, but the other
         registers to be saved are all below 18.  */

      first_reg = 18 - (live_seq - 2);

      for (reg = 29, offset = -live_seq + 1;
           reg >= first_reg;
           reg = (reg == 28 ? 17 : reg - 1), ++offset)
        {
          rtx m, r;

          m = gen_rtx_MEM (QImode, plus_constant (stack_pointer_rtx, offset));
          r = gen_rtx_REG (QImode, reg);
          add_reg_note (insn, REG_CFA_OFFSET, gen_rtx_SET (VOIDmode, m, r));
        }

      cfun->machine->stack_usage += size + live_seq;
    }
  else /* !minimize */
    {
      int reg;
      
      for (reg = 0; reg < 32; ++reg)
        if (TEST_HARD_REG_BIT (set, reg))
          emit_push_byte (reg, true);

      if (frame_pointer_needed
          && (!(cfun->machine->is_OS_task || cfun->machine->is_OS_main)))
        {
          /* Push frame pointer.  Always be consistent about the
             ordering of pushes -- epilogue_restores expects the
             register pair to be pushed low byte first.  */
          
          emit_push_byte (REG_Y, true);
          emit_push_byte (REG_Y + 1, true);
        }
          
      if (frame_pointer_needed
          && size == 0)
        {
          insn = emit_move_insn (frame_pointer_rtx, stack_pointer_rtx);
          RTX_FRAME_RELATED_P (insn) = 1;
        }
      
      if (size != 0)
        {
          /*  Creating a frame can be done by direct manipulation of the
              stack or via the frame pointer. These two methods are:
                  fp =  sp
                  fp -= size
                  sp =  fp
              or
                  sp -= size
                  fp =  sp    (*)
              the optimum method depends on function type, stack and
              frame size.  To avoid a complex logic, both methods are
              tested and shortest is selected.

              There is also the case where SIZE != 0 and no frame pointer is
              needed; this can occur if ACCUMULATE_OUTGOING_ARGS is on.
              In that case, insn (*) is not needed in that case.
              We use the X register as scratch. This is save because in X
              is call-clobbered.
                 In an interrupt routine, the case of SIZE != 0 together with
              !frame_pointer_needed can only occur if the function is not a
              leaf function and thus X has already been saved.  */
              
          int irq_state = -1;
          HOST_WIDE_INT size_cfa = size;
          rtx fp_plus_insns, fp, my_fp;

          gcc_assert (frame_pointer_needed
                      || !isr_p
                      || !current_function_is_leaf);
          
          fp = my_fp = (frame_pointer_needed
                        ? frame_pointer_rtx
                        : gen_rtx_REG (Pmode, REG_X));
          
          if (AVR_HAVE_8BIT_SP)
            {
              /* The high byte (r29) does not change:
                 Prefer SUBI (1 cycle) over SBIW (2 cycles, same size).  */

              my_fp = all_regs_rtx[FRAME_POINTER_REGNUM];
            }

          /* Cut down size and avoid size = 0 so that we don't run
             into ICE like PR52488 in the remainder.  */

          if (size > size_max)
            {
              /* Don't error so that insane code from newlib still compiles
                 and does not break building newlib.  As PR51345 is implemented
                 now, there are multilib variants with -msp8.
                 
                 If user wants sanity checks he can use -Wstack-usage=
                 or similar options.

                 For CFA we emit the original, non-saturated size so that
                 the generic machinery is aware of the real stack usage and
                 will print the above diagnostic as expected.  */
              
              size = size_max;
            }

          size = trunc_int_for_mode (size, GET_MODE (my_fp));
          
          /************  Method 1: Adjust frame pointer  ************/
          
          start_sequence ();

          /* Normally, the dwarf2out frame-related-expr interpreter does
             not expect to have the CFA change once the frame pointer is
             set up.  Thus, we avoid marking the move insn below and
             instead indicate that the entire operation is complete after
             the frame pointer subtraction is done.  */
          
          insn = emit_move_insn (fp, stack_pointer_rtx);
          if (frame_pointer_needed)
            {
              RTX_FRAME_RELATED_P (insn) = 1;
              add_reg_note (insn, REG_CFA_ADJUST_CFA,
                            gen_rtx_SET (VOIDmode, fp, stack_pointer_rtx));
            }

          insn = emit_move_insn (my_fp, plus_constant (my_fp, -size));
          if (frame_pointer_needed)
            {
              RTX_FRAME_RELATED_P (insn) = 1;
              add_reg_note (insn, REG_CFA_ADJUST_CFA,
                            gen_rtx_SET (VOIDmode, fp,
                                         plus_constant (fp, -size_cfa)));
            }
          
          /* Copy to stack pointer.  Note that since we've already
             changed the CFA to the frame pointer this operation
             need not be annotated if frame pointer is needed.
             Always move through unspec, see PR50063.
             For meaning of irq_state see movhi_sp_r insn.  */

          if (cfun->machine->is_interrupt)
            irq_state = 1;

          if (TARGET_NO_INTERRUPTS
              || cfun->machine->is_signal
              || cfun->machine->is_OS_main)
            irq_state = 0;

          if (AVR_HAVE_8BIT_SP)
            irq_state = 2;

          insn = emit_insn (gen_movhi_sp_r (stack_pointer_rtx,
                                            fp, GEN_INT (irq_state)));
          if (!frame_pointer_needed)
            {
              RTX_FRAME_RELATED_P (insn) = 1;
              add_reg_note (insn, REG_CFA_ADJUST_CFA,
                            gen_rtx_SET (VOIDmode, stack_pointer_rtx,
                                         plus_constant (stack_pointer_rtx,
                                                        -size_cfa)));
            }
          
          fp_plus_insns = get_insns ();
          end_sequence ();
          
          /************  Method 2: Adjust Stack pointer  ************/

          /* Stack adjustment by means of RCALL . and/or PUSH __TMP_REG__
             can only handle specific offsets.  */
          
          if (avr_sp_immediate_operand (gen_int_mode (-size, HImode), HImode))
            {
              rtx sp_plus_insns;
              
              start_sequence ();

              insn = emit_move_insn (stack_pointer_rtx,
                                     plus_constant (stack_pointer_rtx, -size));
              RTX_FRAME_RELATED_P (insn) = 1;
              add_reg_note (insn, REG_CFA_ADJUST_CFA,
                            gen_rtx_SET (VOIDmode, stack_pointer_rtx,
                                         plus_constant (stack_pointer_rtx,
                                                        -size_cfa)));
              if (frame_pointer_needed)
                {
                  insn = emit_move_insn (fp, stack_pointer_rtx);
                  RTX_FRAME_RELATED_P (insn) = 1;
                }

              sp_plus_insns = get_insns ();
              end_sequence ();

              /************ Use shortest method  ************/
                  
              emit_insn (get_sequence_length (sp_plus_insns)
                         < get_sequence_length (fp_plus_insns)
                         ? sp_plus_insns
                         : fp_plus_insns);
            }
          else
            {
              emit_insn (fp_plus_insns);
            }

          cfun->machine->stack_usage += size_cfa;
        } /* !minimize && size != 0 */
    } /* !minimize */
}


/*  Output function prologue.  */

void
expand_prologue (void)
{
  HARD_REG_SET set;
  HOST_WIDE_INT size;

  size = get_frame_size() + avr_outgoing_args_size();
  
  /* Init cfun->machine.  */
  cfun->machine->is_naked = avr_naked_function_p (current_function_decl);
  cfun->machine->is_interrupt = interrupt_function_p (current_function_decl);
  cfun->machine->is_signal = signal_function_p (current_function_decl);
  cfun->machine->is_OS_task = avr_OS_task_function_p (current_function_decl);
  cfun->machine->is_OS_main = avr_OS_main_function_p (current_function_decl);
  cfun->machine->stack_usage = 0;
  
  /* Prologue: naked.  */
  if (cfun->machine->is_naked)
    {
      return;
    }

  avr_regs_to_save (&set);

  if (cfun->machine->is_interrupt || cfun->machine->is_signal)
    {
      /* Enable interrupts.  */
      if (cfun->machine->is_interrupt)
        emit_insn (gen_enable_interrupt ());
        
      /* Push zero reg.  */
      emit_push_byte (ZERO_REGNO, true);

      /* Push tmp reg.  */
      emit_push_byte (TMP_REGNO, true);

      /* Push SREG.  */
      /* ??? There's no dwarf2 column reserved for SREG.  */
      emit_push_sfr (sreg_rtx, false, false /* clr */);

      /* Clear zero reg.  */
      emit_move_insn (zero_reg_rtx, const0_rtx);

      /* Prevent any attempt to delete the setting of ZERO_REG!  */
      emit_use (zero_reg_rtx);

      /* Push and clear RAMPD/X/Y/Z if present and low-part register is used.
         ??? There are no dwarf2 columns reserved for RAMPD/X/Y/Z.  */
      
      if (AVR_HAVE_RAMPD)
        emit_push_sfr (rampd_rtx, false /* frame-related */, true /* clr */);

      if (AVR_HAVE_RAMPX
          && TEST_HARD_REG_BIT (set, REG_X)
          && TEST_HARD_REG_BIT (set, REG_X + 1))
        {
          emit_push_sfr (rampx_rtx, false /* frame-related */, true /* clr */);
        }

      if (AVR_HAVE_RAMPY
          && (frame_pointer_needed
              || (TEST_HARD_REG_BIT (set, REG_Y)
                  && TEST_HARD_REG_BIT (set, REG_Y + 1))))
        {
          emit_push_sfr (rampy_rtx, false /* frame-related */, true /* clr */);
        }

      if (AVR_HAVE_RAMPZ
          && TEST_HARD_REG_BIT (set, REG_Z)
          && TEST_HARD_REG_BIT (set, REG_Z + 1))
        {
          emit_push_sfr (rampz_rtx, false /* frame-related */, AVR_HAVE_RAMPD);
        }
    }  /* is_interrupt is_signal */

  avr_prologue_setup_frame (size, set);
  
  if (flag_stack_usage_info)
    current_function_static_stack_size = cfun->machine->stack_usage;
}

/* Output summary at end of function prologue.  */

static void
avr_asm_function_end_prologue (FILE *file)
{
  if (cfun->machine->is_naked)
    {
      fputs ("/* prologue: naked */\n", file);
    }
  else
    {
      if (cfun->machine->is_interrupt)
        {
          fputs ("/* prologue: Interrupt */\n", file);
        }
      else if (cfun->machine->is_signal)
        {
          fputs ("/* prologue: Signal */\n", file);
        }
      else
        fputs ("/* prologue: function */\n", file);
    }

  if (ACCUMULATE_OUTGOING_ARGS)
    fprintf (file, "/* outgoing args size = %d */\n",
             avr_outgoing_args_size());

  fprintf (file, "/* frame size = " HOST_WIDE_INT_PRINT_DEC " */\n",
                 get_frame_size());
  fprintf (file, "/* stack size = %d */\n",
                 cfun->machine->stack_usage);
  /* Create symbol stack offset here so all functions have it. Add 1 to stack
     usage for offset so that SP + .L__stack_offset = return address.  */
  fprintf (file, ".L__stack_usage = %d\n", cfun->machine->stack_usage);
}


/* Implement EPILOGUE_USES.  */

int
avr_epilogue_uses (int regno ATTRIBUTE_UNUSED)
{
  if (reload_completed 
      && cfun->machine
      && (cfun->machine->is_interrupt || cfun->machine->is_signal))
    return 1;
  return 0;
}

/*  Helper for expand_epilogue.  Emit a pop of a byte register.  */

static void
emit_pop_byte (unsigned regno)
{
  rtx mem, reg;

  mem = gen_rtx_PRE_INC (HImode, stack_pointer_rtx);
  mem = gen_frame_mem (QImode, mem);
  reg = gen_rtx_REG (QImode, regno);

  emit_insn (gen_rtx_SET (VOIDmode, reg, mem));
}

/*  Output RTL epilogue.  */

void
expand_epilogue (bool sibcall_p)
{
  int reg;
  int live_seq;
  HARD_REG_SET set;      
  int minimize;
  HOST_WIDE_INT size;
  bool isr_p = cfun->machine->is_interrupt || cfun->machine->is_signal;

  size = get_frame_size() + avr_outgoing_args_size();
  
  /* epilogue: naked  */
  if (cfun->machine->is_naked)
    {
      gcc_assert (!sibcall_p);
      
      emit_jump_insn (gen_return ());
      return;
    }

  avr_regs_to_save (&set);
  live_seq = sequent_regs_live ();
  
  minimize = (TARGET_CALL_PROLOGUES
              && live_seq
              && !isr_p
              && !cfun->machine->is_OS_task
              && !cfun->machine->is_OS_main);
  
  if (minimize
      && (live_seq > 4
          || frame_pointer_needed
          || size))
    {
      /*  Get rid of frame.  */
      
      if (!frame_pointer_needed)
        {
          emit_move_insn (frame_pointer_rtx, stack_pointer_rtx);
        }

      if (size)
        {
          emit_move_insn (frame_pointer_rtx,
                          plus_constant (frame_pointer_rtx, size));
        }
        
      emit_insn (gen_epilogue_restores (gen_int_mode (live_seq, HImode)));
      return;
    }
      
  if (size)
    {
      /* Try two methods to adjust stack and select shortest.  */

      int irq_state = -1;
      rtx fp, my_fp;
      rtx fp_plus_insns;
      HOST_WIDE_INT size_max;

      gcc_assert (frame_pointer_needed
                  || !isr_p
                  || !current_function_is_leaf);
      
      fp = my_fp = (frame_pointer_needed
                    ? frame_pointer_rtx
                    : gen_rtx_REG (Pmode, REG_X));

      if (AVR_HAVE_8BIT_SP)
        {
          /* The high byte (r29) does not change:
             Prefer SUBI (1 cycle) over SBIW (2 cycles).  */
                  
          my_fp = all_regs_rtx[FRAME_POINTER_REGNUM];
        }

      /* For rationale see comment in prologue generation.  */

      size_max = (HOST_WIDE_INT) GET_MODE_MASK (GET_MODE (my_fp));
      if (size > size_max)
        size = size_max;
      size = trunc_int_for_mode (size, GET_MODE (my_fp));
              
      /********** Method 1: Adjust fp register  **********/
              
      start_sequence ();

      if (!frame_pointer_needed)
        emit_move_insn (fp, stack_pointer_rtx);

      emit_move_insn (my_fp, plus_constant (my_fp, size));

      /* Copy to stack pointer.  */

      if (TARGET_NO_INTERRUPTS)
        irq_state = 0;

      if (AVR_HAVE_8BIT_SP)
        irq_state = 2;

      emit_insn (gen_movhi_sp_r (stack_pointer_rtx, fp,
                                 GEN_INT (irq_state)));

      fp_plus_insns = get_insns ();
      end_sequence ();        

      /********** Method 2: Adjust Stack pointer  **********/
      
      if (avr_sp_immediate_operand (gen_int_mode (size, HImode), HImode))
        {
          rtx sp_plus_insns;

          start_sequence ();

          emit_move_insn (stack_pointer_rtx,
                          plus_constant (stack_pointer_rtx, size));

          sp_plus_insns = get_insns ();
          end_sequence ();

          /************ Use shortest method  ************/
          
          emit_insn (get_sequence_length (sp_plus_insns)
                     < get_sequence_length (fp_plus_insns)
                     ? sp_plus_insns
                     : fp_plus_insns);
        }
      else
        emit_insn (fp_plus_insns);
    } /* size != 0 */
          
  if (frame_pointer_needed
      && !(cfun->machine->is_OS_task || cfun->machine->is_OS_main))
    {
      /* Restore previous frame_pointer.  See expand_prologue for
         rationale for not using pophi.  */
              
      emit_pop_byte (REG_Y + 1);
      emit_pop_byte (REG_Y);
    }

  /* Restore used registers.  */
  
  for (reg = 31; reg >= 0; --reg)
    if (TEST_HARD_REG_BIT (set, reg))
      emit_pop_byte (reg);

  if (isr_p)
    {
      /* Restore RAMPZ/Y/X/D using tmp_reg as scratch.
         The conditions to restore them must be tha same as in prologue.  */
      
      if (AVR_HAVE_RAMPZ
          && TEST_HARD_REG_BIT (set, REG_Z)
          && TEST_HARD_REG_BIT (set, REG_Z + 1))
        {
          emit_pop_byte (TMP_REGNO);
          emit_move_insn (rampz_rtx, tmp_reg_rtx);
        }

      if (AVR_HAVE_RAMPY
          && (frame_pointer_needed
              || (TEST_HARD_REG_BIT (set, REG_Y)
                  && TEST_HARD_REG_BIT (set, REG_Y + 1))))
        {
          emit_pop_byte (TMP_REGNO);
          emit_move_insn (rampy_rtx, tmp_reg_rtx);
        }

      if (AVR_HAVE_RAMPX
          && TEST_HARD_REG_BIT (set, REG_X)
          && TEST_HARD_REG_BIT (set, REG_X + 1))
        {
          emit_pop_byte (TMP_REGNO);
          emit_move_insn (rampx_rtx, tmp_reg_rtx);
        }

      if (AVR_HAVE_RAMPD)
        {
          emit_pop_byte (TMP_REGNO);
          emit_move_insn (rampd_rtx, tmp_reg_rtx);
        }

      /* Restore SREG using tmp_reg as scratch.  */
      
      emit_pop_byte (TMP_REGNO);
      emit_move_insn (sreg_rtx, tmp_reg_rtx);

      /* Restore tmp REG.  */
      emit_pop_byte (TMP_REGNO);

      /* Restore zero REG.  */
      emit_pop_byte (ZERO_REGNO);
    }

  if (!sibcall_p)
    emit_jump_insn (gen_return ());
}

/* Output summary messages at beginning of function epilogue.  */

static void
avr_asm_function_begin_epilogue (FILE *file)
{
  fprintf (file, "/* epilogue start */\n");
}


/* Implement TARGET_CANNOT_MODITY_JUMPS_P */

static bool
avr_cannot_modify_jumps_p (void)
{

  /* Naked Functions must not have any instructions after
     their epilogue, see PR42240 */
     
  if (reload_completed
      && cfun->machine
      && cfun->machine->is_naked)
    {
      return true;
    }

  return false;
}


/* Implement `TARGET_MODE_DEPENDENT_ADDRESS_P'.  */

/* FIXME:  PSImode addresses are not mode-dependent in themselves.
      This hook just serves to hack around PR rtl-optimization/52543 by
      claiming that PSImode addresses (which are used for the 24-bit
      address space __memx) were mode-dependent so that lower-subreg.s
      will skip these addresses.  See also the similar FIXME comment along
      with mov<mode> expanders in avr.md.  */

static bool
avr_mode_dependent_address_p (const_rtx addr)
{
  return GET_MODE (addr) != Pmode;
}


/* Helper function for `avr_legitimate_address_p'.  */

static inline bool
avr_reg_ok_for_addr_p (rtx reg, addr_space_t as,
                       RTX_CODE outer_code, bool strict)
{
  return (REG_P (reg)
          && (avr_regno_mode_code_ok_for_base_p (REGNO (reg), QImode,
                                                 as, outer_code, UNKNOWN)
              || (!strict
                  && REGNO (reg) >= FIRST_PSEUDO_REGISTER)));
}


/* Return nonzero if X (an RTX) is a legitimate memory address on the target
   machine for a memory operand of mode MODE.  */

static bool
avr_legitimate_address_p (enum machine_mode mode, rtx x, bool strict)
{
  bool ok = CONSTANT_ADDRESS_P (x);
  
  switch (GET_CODE (x))
    {
    case REG:
      ok = avr_reg_ok_for_addr_p (x, ADDR_SPACE_GENERIC,
                                  MEM, strict);

      if (strict
          && DImode == mode
          && REG_X == REGNO (x))
        {
          ok = false;
        }
      break;

    case POST_INC:
    case PRE_DEC:
      ok = avr_reg_ok_for_addr_p (XEXP (x, 0), ADDR_SPACE_GENERIC,
                                  GET_CODE (x), strict);
      break;

    case PLUS:
      {
        rtx reg = XEXP (x, 0);
        rtx op1 = XEXP (x, 1);
        
        if (REG_P (reg)
            && CONST_INT_P (op1)
            && INTVAL (op1) >= 0)
          {
            bool fit = IN_RANGE (INTVAL (op1), 0, MAX_LD_OFFSET (mode));

            if (fit)
              {
                ok = (! strict
                      || avr_reg_ok_for_addr_p (reg, ADDR_SPACE_GENERIC,
                                                PLUS, strict));
          
                if (reg == frame_pointer_rtx
                    || reg == arg_pointer_rtx)
                  {
                    ok = true;
                  }
              }
            else if (frame_pointer_needed
                     && reg == frame_pointer_rtx)
              {
                ok = true;
              }
          }
      }
      break;
      
    default:
      break;
    }
  
  if (avr_log.legitimate_address_p)
    {
      avr_edump ("\n%?: ret=%d, mode=%m strict=%d "
                 "reload_completed=%d reload_in_progress=%d %s:",
                 ok, mode, strict, reload_completed, reload_in_progress,
                 reg_renumber ? "(reg_renumber)" : "");
      
      if (GET_CODE (x) == PLUS
          && REG_P (XEXP (x, 0))
          && CONST_INT_P (XEXP (x, 1))
          && IN_RANGE (INTVAL (XEXP (x, 1)), 0, MAX_LD_OFFSET (mode))
          && reg_renumber)
        {
          avr_edump ("(r%d ---> r%d)", REGNO (XEXP (x, 0)),
                     true_regnum (XEXP (x, 0)));
        }
      
      avr_edump ("\n%r\n", x);
    }
  
  return ok;
}


/* Former implementation of TARGET_LEGITIMIZE_ADDRESS,
   now only a helper for avr_addr_space_legitimize_address.  */
/* Attempts to replace X with a valid
   memory address for an operand of mode MODE  */

static rtx
avr_legitimize_address (rtx x, rtx oldx, enum machine_mode mode)
{
  bool big_offset_p = false;
  
  x = oldx;
  
  if (GET_CODE (oldx) == PLUS
      && REG_P (XEXP (oldx, 0)))
    {
      if (REG_P (XEXP (oldx, 1)))
        x = force_reg (GET_MODE (oldx), oldx);
      else if (CONST_INT_P (XEXP (oldx, 1)))
        {
          int offs = INTVAL (XEXP (oldx, 1));
          if (frame_pointer_rtx != XEXP (oldx, 0)
              && offs > MAX_LD_OFFSET (mode))
            {
              big_offset_p = true;
              x = force_reg (GET_MODE (oldx), oldx);
            }
        }
    }
  
  if (avr_log.legitimize_address)
    {
      avr_edump ("\n%?: mode=%m\n %r\n", mode, oldx);

      if (x != oldx)
        avr_edump (" %s --> %r\n", big_offset_p ? "(big offset)" : "", x);
    }

  return x;
}


/* Implement `LEGITIMIZE_RELOAD_ADDRESS'.  */
/* This will allow register R26/27 to be used where it is no worse than normal
   base pointers R28/29 or R30/31.  For example, if base offset is greater
   than 63 bytes or for R++ or --R addressing.  */

rtx
avr_legitimize_reload_address (rtx *px, enum machine_mode mode,
                               int opnum, int type, int addr_type,
                               int ind_levels ATTRIBUTE_UNUSED,
                               rtx (*mk_memloc)(rtx,int))
{
  rtx x = *px;
  
  if (avr_log.legitimize_reload_address)
    avr_edump ("\n%?:%m %r\n", mode, x);
  
  if (1 && (GET_CODE (x) == POST_INC
            || GET_CODE (x) == PRE_DEC))
    {
      push_reload (XEXP (x, 0), XEXP (x, 0), &XEXP (x, 0), &XEXP (x, 0),
                   POINTER_REGS, GET_MODE (x), GET_MODE (x), 0, 0,
                   opnum, RELOAD_OTHER);
      
      if (avr_log.legitimize_reload_address)
        avr_edump (" RCLASS.1 = %R\n IN = %r\n OUT = %r\n",
                   POINTER_REGS, XEXP (x, 0), XEXP (x, 0));
      
      return x;
    }
  
  if (GET_CODE (x) == PLUS
      && REG_P (XEXP (x, 0))
      && 0 == reg_equiv_constant (REGNO (XEXP (x, 0)))
      && CONST_INT_P (XEXP (x, 1))
      && INTVAL (XEXP (x, 1)) >= 1)
    {
      bool fit = INTVAL (XEXP (x, 1)) <= MAX_LD_OFFSET (mode);
      
      if (fit)
        {
          if (reg_equiv_address (REGNO (XEXP (x, 0))) != 0)
            {
              int regno = REGNO (XEXP (x, 0));
              rtx mem = mk_memloc (x, regno);
              
              push_reload (XEXP (mem, 0), NULL_RTX, &XEXP (mem, 0), NULL,
                           POINTER_REGS, Pmode, VOIDmode, 0, 0,
                           1, addr_type);
              
              if (avr_log.legitimize_reload_address)
                avr_edump (" RCLASS.2 = %R\n IN = %r\n OUT = %r\n",
                           POINTER_REGS, XEXP (mem, 0), NULL_RTX);
              
              push_reload (mem, NULL_RTX, &XEXP (x, 0), NULL,
                           BASE_POINTER_REGS, GET_MODE (x), VOIDmode, 0, 0,
                           opnum, type);
              
              if (avr_log.legitimize_reload_address)
                avr_edump (" RCLASS.2 = %R\n IN = %r\n OUT = %r\n",
                           BASE_POINTER_REGS, mem, NULL_RTX);
              
              return x;
            }
        }
      else if (! (frame_pointer_needed
                  && XEXP (x, 0) == frame_pointer_rtx))
        {
          push_reload (x, NULL_RTX, px, NULL,
                       POINTER_REGS, GET_MODE (x), VOIDmode, 0, 0,
                       opnum, type);
          
          if (avr_log.legitimize_reload_address)
            avr_edump (" RCLASS.3 = %R\n IN = %r\n OUT = %r\n",
                       POINTER_REGS, x, NULL_RTX);
          
          return x;
        }
    }
  
  return NULL_RTX;
}


/* Helper function to print assembler resp. track instruction
   sequence lengths.  Always return "".
   
   If PLEN == NULL:
       Output assembler code from template TPL with operands supplied
       by OPERANDS.  This is just forwarding to output_asm_insn.
   
   If PLEN != NULL:
       If N_WORDS >= 0  Add N_WORDS to *PLEN.
       If N_WORDS < 0   Set *PLEN to -N_WORDS.
       Don't output anything.
*/

static const char*
avr_asm_len (const char* tpl, rtx* operands, int* plen, int n_words)
{
  if (NULL == plen)
    {
      output_asm_insn (tpl, operands);
    }
  else
    {
      if (n_words < 0)
        *plen = -n_words;
      else
        *plen += n_words;
    }

  return "";
}


/* Return a pointer register name as a string.  */

static const char *
ptrreg_to_str (int regno)
{
  switch (regno)
    {
    case REG_X: return "X";
    case REG_Y: return "Y";
    case REG_Z: return "Z";
    default:
      output_operand_lossage ("address operand requires constraint for"
                              " X, Y, or Z register");
    }
  return NULL;
}

/* Return the condition name as a string.
   Used in conditional jump constructing  */

static const char *
cond_string (enum rtx_code code)
{
  switch (code)
    {
    case NE:
      return "ne";
    case EQ:
      return "eq";
    case GE:
      if (cc_prev_status.flags & CC_OVERFLOW_UNUSABLE)
        return "pl";
      else
        return "ge";
    case LT:
      if (cc_prev_status.flags & CC_OVERFLOW_UNUSABLE)
        return "mi";
      else
        return "lt";
    case GEU:
      return "sh";
    case LTU:
      return "lo";
    default:
      gcc_unreachable ();
    }

  return "";
}


/* Implement `TARGET_PRINT_OPERAND_ADDRESS'.  */
/* Output ADDR to FILE as address.  */

static void
avr_print_operand_address (FILE *file, rtx addr)
{
  switch (GET_CODE (addr))
    {
    case REG:
      fprintf (file, ptrreg_to_str (REGNO (addr)));
      break;

    case PRE_DEC:
      fprintf (file, "-%s", ptrreg_to_str (REGNO (XEXP (addr, 0))));
      break;

    case POST_INC:
      fprintf (file, "%s+", ptrreg_to_str (REGNO (XEXP (addr, 0))));
      break;

    default:
      if (CONSTANT_ADDRESS_P (addr)
          && text_segment_operand (addr, VOIDmode))
        {
          rtx x = addr;
          if (GET_CODE (x) == CONST)
            x = XEXP (x, 0);
          if (GET_CODE (x) == PLUS && GET_CODE (XEXP (x,1)) == CONST_INT)
            {
              /* Assembler gs() will implant word address. Make offset 
                 a byte offset inside gs() for assembler. This is 
                 needed because the more logical (constant+gs(sym)) is not 
                 accepted by gas. For 128K and lower devices this is ok.
                 For large devices it will create a Trampoline to offset
                 from symbol which may not be what the user really wanted.  */
              fprintf (file, "gs(");
              output_addr_const (file, XEXP (x,0));
              fprintf (file, "+" HOST_WIDE_INT_PRINT_DEC ")",
                       2 * INTVAL (XEXP (x, 1)));
              if (AVR_3_BYTE_PC)
                if (warning (0, "pointer offset from symbol maybe incorrect"))
                  {
                    output_addr_const (stderr, addr);
                    fprintf(stderr,"\n");
                  }
            }
          else
            {
              fprintf (file, "gs(");
              output_addr_const (file, addr);
              fprintf (file, ")");
            }
        }
      else
        output_addr_const (file, addr);
    }
}


/* Implement `TARGET_PRINT_OPERAND_PUNCT_VALID_P'.  */

static bool
avr_print_operand_punct_valid_p (unsigned char code)
{
  return code == '~' || code == '!';
}


/* Implement `TARGET_PRINT_OPERAND'.  */
/* Output X as assembler operand to file FILE.
   For a description of supported %-codes, see top of avr.md.  */

static void
avr_print_operand (FILE *file, rtx x, int code)
{
  int abcd = 0;

  if (code >= 'A' && code <= 'D')
    abcd = code - 'A';

  if (code == '~')
    {
      if (!AVR_HAVE_JMP_CALL)
        fputc ('r', file);
    }
  else if (code == '!')
    {
      if (AVR_HAVE_EIJMP_EICALL)
        fputc ('e', file);
    }
  else if (code == 't'
           || code == 'T')
    {
      static int t_regno = -1;
      static int t_nbits = -1;

      if (REG_P (x) && t_regno < 0 && code == 'T')
        {
          t_regno = REGNO (x);
          t_nbits = GET_MODE_BITSIZE (GET_MODE (x));
        }
      else if (CONST_INT_P (x) && t_regno >= 0
               && IN_RANGE (INTVAL (x), 0, t_nbits - 1))
        {
          int bpos = INTVAL (x);

          fprintf (file, "%s", reg_names[t_regno + bpos / 8]);
          if (code == 'T')
            fprintf (file, ",%d", bpos % 8);

          t_regno = -1;
        }
      else
        fatal_insn ("operands to %T/%t must be reg + const_int:", x);
    }
  else if (REG_P (x))
    {
      if (x == zero_reg_rtx)
        fprintf (file, "__zero_reg__");
      else
        fprintf (file, reg_names[true_regnum (x) + abcd]);
    }
  else if (CONST_INT_P (x))
    {
      HOST_WIDE_INT ival = INTVAL (x);
        
      if ('i' != code)
        fprintf (file, HOST_WIDE_INT_PRINT_DEC, ival + abcd);
      else if (low_io_address_operand (x, VOIDmode)
               || high_io_address_operand (x, VOIDmode))
        {
          if (AVR_HAVE_RAMPZ && ival == avr_addr.rampz)
            fprintf (file, "__RAMPZ__");
          else if (AVR_HAVE_RAMPY && ival == avr_addr.rampy)
            fprintf (file, "__RAMPY__");
          else if (AVR_HAVE_RAMPX && ival == avr_addr.rampx)
            fprintf (file, "__RAMPX__");
          else if (AVR_HAVE_RAMPD && ival == avr_addr.rampd)
            fprintf (file, "__RAMPD__");
          else if (AVR_XMEGA && ival == avr_addr.ccp)
            fprintf (file, "__CCP__");
          else if (ival == avr_addr.sreg)   fprintf (file, "__SREG__");
          else if (ival == avr_addr.sp_l)   fprintf (file, "__SP_L__");
          else if (ival == avr_addr.sp_h)   fprintf (file, "__SP_H__");
          else
            {
              fprintf (file, HOST_WIDE_INT_PRINT_HEX,
                       ival - avr_current_arch->sfr_offset);
            }
        }
      else
        fatal_insn ("bad address, not an I/O address:", x);
    }
  else if (MEM_P (x))
    {
      rtx addr = XEXP (x, 0);
      
      if (code == 'm')
        {
          if (!CONSTANT_P (addr))
            fatal_insn ("bad address, not a constant:", addr);
          /* Assembler template with m-code is data - not progmem section */
          if (text_segment_operand (addr, VOIDmode))
            if (warning (0, "accessing data memory with"
                         " program memory address"))
              {
                output_addr_const (stderr, addr);
                fprintf(stderr,"\n");
              }
          output_addr_const (file, addr);
        }
      else if (code == 'i')
        {
          avr_print_operand (file, addr, 'i');
        }
      else if (code == 'o')
        {
          if (GET_CODE (addr) != PLUS)
            fatal_insn ("bad address, not (reg+disp):", addr);

          avr_print_operand (file, XEXP (addr, 1), 0);
        }
      else if (code == 'p' || code == 'r')
        {
          if (GET_CODE (addr) != POST_INC && GET_CODE (addr) != PRE_DEC)
            fatal_insn ("bad address, not post_inc or pre_dec:", addr);
          
          if (code == 'p')
            avr_print_operand_address (file, XEXP (addr, 0));  /* X, Y, Z */
          else
            avr_print_operand (file, XEXP (addr, 0), 0);  /* r26, r28, r30 */
        }
      else if (GET_CODE (addr) == PLUS)
        {
          avr_print_operand_address (file, XEXP (addr,0));
          if (REGNO (XEXP (addr, 0)) == REG_X)
            fatal_insn ("internal compiler error.  Bad address:"
                        ,addr);
          fputc ('+', file);
          avr_print_operand (file, XEXP (addr,1), code);
        }
      else
        avr_print_operand_address (file, addr);
    }
  else if (code == 'i')
    {
      fatal_insn ("bad address, not an I/O address:", x);
    }
  else if (code == 'x')
    {
      /* Constant progmem address - like used in jmp or call */
      if (0 == text_segment_operand (x, VOIDmode))
        if (warning (0, "accessing program memory"
                     " with data memory address"))
          {
            output_addr_const (stderr, x);
            fprintf(stderr,"\n");
          }
      /* Use normal symbol for direct address no linker trampoline needed */
      output_addr_const (file, x);
    }
  else if (GET_CODE (x) == CONST_DOUBLE)
    {
      long val;
      REAL_VALUE_TYPE rv;
      if (GET_MODE (x) != SFmode)
        fatal_insn ("internal compiler error.  Unknown mode:", x);
      REAL_VALUE_FROM_CONST_DOUBLE (rv, x);
      REAL_VALUE_TO_TARGET_SINGLE (rv, val);
      fprintf (file, "0x%lx", val);
    }
  else if (GET_CODE (x) == CONST_STRING)
    fputs (XSTR (x, 0), file);
  else if (code == 'j')
    fputs (cond_string (GET_CODE (x)), file);
  else if (code == 'k')
    fputs (cond_string (reverse_condition (GET_CODE (x))), file);
  else
    avr_print_operand_address (file, x);
}

/* Update the condition code in the INSN.  */

void
notice_update_cc (rtx body ATTRIBUTE_UNUSED, rtx insn)
{
  rtx set;
  enum attr_cc cc = get_attr_cc (insn);
  
  switch (cc)
    {
    default:
      break;

    case CC_OUT_PLUS:
    case CC_OUT_PLUS_NOCLOBBER:
    case CC_LDI:
      {
        rtx *op = recog_data.operand;
        int len_dummy, icc;
        
        /* Extract insn's operands.  */
        extract_constrain_insn_cached (insn);

        switch (cc)
          {
          default:
            gcc_unreachable();
            
          case CC_OUT_PLUS:
            avr_out_plus (op, &len_dummy, &icc);
            cc = (enum attr_cc) icc;
            break;
            
          case CC_OUT_PLUS_NOCLOBBER:
            avr_out_plus_noclobber (op, &len_dummy, &icc);
            cc = (enum attr_cc) icc;
            break;

          case CC_LDI:

            cc = (op[1] == CONST0_RTX (GET_MODE (op[0]))
                  && reg_overlap_mentioned_p (op[0], zero_reg_rtx))
              /* Loading zero-reg with 0 uses CLI and thus clobbers cc0.  */
              ? CC_CLOBBER
              /* Any other "r,rL" combination does not alter cc0.  */
              : CC_NONE;
            
            break;
          } /* inner switch */

        break;
      }
    } /* outer swicth */

  switch (cc)
    {
    default:
      /* Special values like CC_OUT_PLUS from above have been
         mapped to "standard" CC_* values so we never come here.  */
      
      gcc_unreachable();
      break;
      
    case CC_NONE:
      /* Insn does not affect CC at all.  */
      break;

    case CC_SET_N:
      CC_STATUS_INIT;
      break;

    case CC_SET_ZN:
      set = single_set (insn);
      CC_STATUS_INIT;
      if (set)
        {
          cc_status.flags |= CC_NO_OVERFLOW;
          cc_status.value1 = SET_DEST (set);
        }
      break;

    case CC_SET_CZN:
      /* Insn sets the Z,N,C flags of CC to recog_operand[0].
         The V flag may or may not be known but that's ok because
         alter_cond will change tests to use EQ/NE.  */
      set = single_set (insn);
      CC_STATUS_INIT;
      if (set)
        {
          cc_status.value1 = SET_DEST (set);
          cc_status.flags |= CC_OVERFLOW_UNUSABLE;
        }
      break;

    case CC_COMPARE:
      set = single_set (insn);
      CC_STATUS_INIT;
      if (set)
        cc_status.value1 = SET_SRC (set);
      break;
      
    case CC_CLOBBER:
      /* Insn doesn't leave CC in a usable state.  */
      CC_STATUS_INIT;
      break;
    }
}

/* Choose mode for jump insn:
   1 - relative jump in range -63 <= x <= 62 ;
   2 - relative jump in range -2046 <= x <= 2045 ;
   3 - absolute jump (only for ATmega[16]03).  */

int
avr_jump_mode (rtx x, rtx insn)
{
  int dest_addr = INSN_ADDRESSES (INSN_UID (GET_CODE (x) == LABEL_REF
                                            ? XEXP (x, 0) : x));
  int cur_addr = INSN_ADDRESSES (INSN_UID (insn));
  int jump_distance = cur_addr - dest_addr;
  
  if (-63 <= jump_distance && jump_distance <= 62)
    return 1;
  else if (-2046 <= jump_distance && jump_distance <= 2045)
    return 2;
  else if (AVR_HAVE_JMP_CALL)
    return 3;
  
  return 2;
}

/* return an AVR condition jump commands.
   X is a comparison RTX.
   LEN is a number returned by avr_jump_mode function.
   if REVERSE nonzero then condition code in X must be reversed.  */

const char *
ret_cond_branch (rtx x, int len, int reverse)
{
  RTX_CODE cond = reverse ? reverse_condition (GET_CODE (x)) : GET_CODE (x);
  
  switch (cond)
    {
    case GT:
      if (cc_prev_status.flags & CC_OVERFLOW_UNUSABLE)
        return (len == 1 ? ("breq .+2" CR_TAB
                            "brpl %0") :
                len == 2 ? ("breq .+4" CR_TAB
                            "brmi .+2" CR_TAB
                            "rjmp %0") :
                ("breq .+6" CR_TAB
                 "brmi .+4" CR_TAB
                 "jmp %0"));
          
      else
        return (len == 1 ? ("breq .+2" CR_TAB
                            "brge %0") :
                len == 2 ? ("breq .+4" CR_TAB
                            "brlt .+2" CR_TAB
                            "rjmp %0") :
                ("breq .+6" CR_TAB
                 "brlt .+4" CR_TAB
                 "jmp %0"));
    case GTU:
      return (len == 1 ? ("breq .+2" CR_TAB
                          "brsh %0") :
              len == 2 ? ("breq .+4" CR_TAB
                          "brlo .+2" CR_TAB
                          "rjmp %0") :
              ("breq .+6" CR_TAB
               "brlo .+4" CR_TAB
               "jmp %0"));
    case LE:
      if (cc_prev_status.flags & CC_OVERFLOW_UNUSABLE)
        return (len == 1 ? ("breq %0" CR_TAB
                            "brmi %0") :
                len == 2 ? ("breq .+2" CR_TAB
                            "brpl .+2" CR_TAB
                            "rjmp %0") :
                ("breq .+2" CR_TAB
                 "brpl .+4" CR_TAB
                 "jmp %0"));
      else
        return (len == 1 ? ("breq %0" CR_TAB
                            "brlt %0") :
                len == 2 ? ("breq .+2" CR_TAB
                            "brge .+2" CR_TAB
                            "rjmp %0") :
                ("breq .+2" CR_TAB
                 "brge .+4" CR_TAB
                 "jmp %0"));
    case LEU:
      return (len == 1 ? ("breq %0" CR_TAB
                          "brlo %0") :
              len == 2 ? ("breq .+2" CR_TAB
                          "brsh .+2" CR_TAB
                          "rjmp %0") :
              ("breq .+2" CR_TAB
               "brsh .+4" CR_TAB
               "jmp %0"));
    default:
      if (reverse)
        {
          switch (len)
            {
            case 1:
              return "br%k1 %0";
            case 2:
              return ("br%j1 .+2" CR_TAB
                      "rjmp %0");
            default:
              return ("br%j1 .+4" CR_TAB
                      "jmp %0");
            }
        }
      else
        {
          switch (len)
            {
            case 1:
              return "br%j1 %0";
            case 2:
              return ("br%k1 .+2" CR_TAB
                      "rjmp %0");
            default:
              return ("br%k1 .+4" CR_TAB
                      "jmp %0");
            }
        }
    }
  return "";
}

/* Output insn cost for next insn.  */

void
final_prescan_insn (rtx insn, rtx *operand ATTRIBUTE_UNUSED,
                    int num_operands ATTRIBUTE_UNUSED)
{
  if (avr_log.rtx_costs)
    {
      rtx set = single_set (insn);

      if (set)
        fprintf (asm_out_file, "/* DEBUG: cost = %d.  */\n",
                 set_src_cost (SET_SRC (set), optimize_insn_for_speed_p ()));
      else
        fprintf (asm_out_file, "/* DEBUG: pattern-cost = %d.  */\n",
                 rtx_cost (PATTERN (insn), INSN, 0,
                           optimize_insn_for_speed_p()));
    }
}

/* Return 0 if undefined, 1 if always true or always false.  */

int
avr_simplify_comparison_p (enum machine_mode mode, RTX_CODE op, rtx x)
{
  unsigned int max = (mode == QImode ? 0xff :
                      mode == HImode ? 0xffff :
                      mode == PSImode ? 0xffffff :
                      mode == SImode ? 0xffffffff : 0);
  if (max && op && GET_CODE (x) == CONST_INT)
    {
      if (unsigned_condition (op) != op)
        max >>= 1;

      if (max != (INTVAL (x) & max)
          && INTVAL (x) != 0xff)
        return 1;
    }
  return 0;
}


/* Returns nonzero if REGNO is the number of a hard
   register in which function arguments are sometimes passed.  */

int
function_arg_regno_p(int r)
{
  return (r >= 8 && r <= 25);
}

/* Initializing the variable cum for the state at the beginning
   of the argument list.  */

void
init_cumulative_args (CUMULATIVE_ARGS *cum, tree fntype, rtx libname,
                      tree fndecl ATTRIBUTE_UNUSED)
{
  cum->nregs = 18;
  cum->regno = FIRST_CUM_REG;
  if (!libname && stdarg_p (fntype))
    cum->nregs = 0;

  /* Assume the calle may be tail called */
  
  cfun->machine->sibcall_fails = 0;
}

/* Returns the number of registers to allocate for a function argument.  */

static int
avr_num_arg_regs (enum machine_mode mode, const_tree type)
{
  int size;

  if (mode == BLKmode)
    size = int_size_in_bytes (type);
  else
    size = GET_MODE_SIZE (mode);

  /* Align all function arguments to start in even-numbered registers.
     Odd-sized arguments leave holes above them.  */

  return (size + 1) & ~1;
}

/* Controls whether a function argument is passed
   in a register, and which register.  */

static rtx
avr_function_arg (cumulative_args_t cum_v, enum machine_mode mode,
                  const_tree type, bool named ATTRIBUTE_UNUSED)
{
  CUMULATIVE_ARGS *cum = get_cumulative_args (cum_v);
  int bytes = avr_num_arg_regs (mode, type);

  if (cum->nregs && bytes <= cum->nregs)
    return gen_rtx_REG (mode, cum->regno - bytes);

  return NULL_RTX;
}

/* Update the summarizer variable CUM to advance past an argument
   in the argument list.  */
   
static void
avr_function_arg_advance (cumulative_args_t cum_v, enum machine_mode mode,
                          const_tree type, bool named ATTRIBUTE_UNUSED)
{
  CUMULATIVE_ARGS *cum = get_cumulative_args (cum_v);
  int bytes = avr_num_arg_regs (mode, type);

  cum->nregs -= bytes;
  cum->regno -= bytes;

  /* A parameter is being passed in a call-saved register. As the original
     contents of these regs has to be restored before leaving the function,
     a function must not pass arguments in call-saved regs in order to get
     tail-called. */
  
  if (cum->regno >= 8
      && cum->nregs >= 0
      && !call_used_regs[cum->regno])
    {
      /* FIXME: We ship info on failing tail-call in struct machine_function.
         This uses internals of calls.c:expand_call() and the way args_so_far
         is used. targetm.function_ok_for_sibcall() needs to be extended to
         pass &args_so_far, too. At present, CUMULATIVE_ARGS is target
         dependent so that such an extension is not wanted. */
      
      cfun->machine->sibcall_fails = 1;
    }

  /* Test if all registers needed by the ABI are actually available.  If the
     user has fixed a GPR needed to pass an argument, an (implicit) function
     call will clobber that fixed register.  See PR45099 for an example.  */
  
  if (cum->regno >= 8
      && cum->nregs >= 0)
    {
      int regno;

      for (regno = cum->regno; regno < cum->regno + bytes; regno++)
        if (fixed_regs[regno])
          warning (0, "fixed register %s used to pass parameter to function",
                   reg_names[regno]);
    }
      
  if (cum->nregs <= 0)
    {
      cum->nregs = 0;
      cum->regno = FIRST_CUM_REG;
    }
}

/* Implement `TARGET_FUNCTION_OK_FOR_SIBCALL' */
/* Decide whether we can make a sibling call to a function.  DECL is the
   declaration of the function being targeted by the call and EXP is the
   CALL_EXPR representing the call. */

static bool
avr_function_ok_for_sibcall (tree decl_callee, tree exp_callee)
{
  tree fntype_callee;

  /* Tail-calling must fail if callee-saved regs are used to pass
     function args.  We must not tail-call when `epilogue_restores'
     is used.  Unfortunately, we cannot tell at this point if that
     actually will happen or not, and we cannot step back from
     tail-calling. Thus, we inhibit tail-calling with -mcall-prologues. */
  
  if (cfun->machine->sibcall_fails
      || TARGET_CALL_PROLOGUES)
    {
      return false;
    }
  
  fntype_callee = TREE_TYPE (CALL_EXPR_FN (exp_callee));

  if (decl_callee)
    {
      decl_callee = TREE_TYPE (decl_callee);
    }
  else
    {
      decl_callee = fntype_callee;
      
      while (FUNCTION_TYPE != TREE_CODE (decl_callee)
             && METHOD_TYPE != TREE_CODE (decl_callee))
        {
          decl_callee = TREE_TYPE (decl_callee);
        }
    }

  /* Ensure that caller and callee have compatible epilogues */
  
  if (interrupt_function_p (current_function_decl)
      || signal_function_p (current_function_decl)
      || avr_naked_function_p (decl_callee)
      || avr_naked_function_p (current_function_decl)
      /* FIXME: For OS_task and OS_main, we are over-conservative.
         This is due to missing documentation of these attributes
         and what they actually should do and should not do. */
      || (avr_OS_task_function_p (decl_callee)
          != avr_OS_task_function_p (current_function_decl))
      || (avr_OS_main_function_p (decl_callee)
          != avr_OS_main_function_p (current_function_decl)))
    {
      return false;
    }
 
  return true;
}

/***********************************************************************
  Functions for outputting various mov's for a various modes
************************************************************************/

/* Return true if a value of mode MODE is read from flash by
   __load_* function from libgcc.  */

bool
avr_load_libgcc_p (rtx op)
{
  enum machine_mode mode = GET_MODE (op);
  int n_bytes = GET_MODE_SIZE (mode);
        
  return (n_bytes > 2
          && !AVR_HAVE_LPMX
          && MEM_P (op)
          && MEM_ADDR_SPACE (op) == ADDR_SPACE_FLASH);
}

/* Return true if a value of mode MODE is read by __xload_* function.  */

bool
avr_xload_libgcc_p (enum machine_mode mode)
{
  int n_bytes = GET_MODE_SIZE (mode);
  
  return (n_bytes > 1
          || avr_current_device->n_flash > 1);
}


/* If PLEN == NULL: Ouput instructions to load a value from a memory location
   OP[1] in AS1 to register OP[0].
   If PLEN != 0 set *PLEN to the length in words of the instruction sequence.
   Return "".  */

static const char*
avr_out_lpm (rtx insn, rtx *op, int *plen)
{
  rtx xop[3];
  rtx dest = op[0];
  rtx src = SET_SRC (single_set (insn));
  rtx addr;
  int n_bytes = GET_MODE_SIZE (GET_MODE (dest));
  RTX_CODE code;
  addr_space_t as = MEM_ADDR_SPACE (src);

  if (plen)
    *plen = 0;
  
  if (MEM_P (dest))
    {
      warning (0, "writing to address space %qs not supported",
               avr_addrspace[MEM_ADDR_SPACE (dest)].name);
      
      return "";
    }

  addr = XEXP (src, 0);
  code = GET_CODE (addr);

  gcc_assert (REG_P (dest));
  gcc_assert (REG == code || POST_INC == code);

  /* Only 1-byte moves from __flash are representes as open coded
     mov insns.  All other loads from flash are not handled here but
     by some UNSPEC instead, see respective FIXME in machine description.  */
  
  gcc_assert (as == ADDR_SPACE_FLASH);
  gcc_assert (n_bytes == 1);

  xop[0] = dest;
  xop[1] = lpm_addr_reg_rtx;
  xop[2] = lpm_reg_rtx;

  switch (code)
    {
    default:
      gcc_unreachable();

    case REG:

      gcc_assert (REG_Z == REGNO (addr));
      
      return AVR_HAVE_LPMX
        ? avr_asm_len ("lpm %0,%a1", xop, plen, 1)
        : avr_asm_len ("lpm" CR_TAB
                       "mov %0,%2", xop, plen, 2);
      
    case POST_INC:
      
      gcc_assert (REG_Z == REGNO (XEXP (addr, 0)));

      return AVR_HAVE_LPMX
        ? avr_asm_len ("lpm %0,%a1+", xop, plen, 1)
        : avr_asm_len ("lpm"        CR_TAB
                       "adiw %1, 1" CR_TAB
                       "mov %0,%2", xop, plen, 3);
    }

  return "";
}


/* If PLEN == NULL: Ouput instructions to load $0 with a value from
   flash address $1:Z.  If $1 = 0 we can use LPM to read, otherwise
   use ELPM.
   If PLEN != 0 set *PLEN to the length in words of the instruction sequence.
   Return "".  */

const char*
avr_load_lpm (rtx insn, rtx *op, int *plen)
{
  rtx xop[4];
  int n, n_bytes = GET_MODE_SIZE (GET_MODE (op[0]));
  rtx xsegment = op[1];
  bool clobber_z = PARALLEL == GET_CODE (PATTERN (insn));
  bool r30_in_tmp = false;
  
  if (plen)
    *plen = 0;
  
  xop[1] = lpm_addr_reg_rtx;
  xop[2] = lpm_reg_rtx;
  xop[3] = xstring_empty;
  
  /* Set RAMPZ as needed.  */
  
  if (REG_P (xsegment))
    {
      avr_asm_len ("out __RAMPZ__,%0", &xsegment, plen, 1);
      xop[3] = xstring_e;
    }
  
  /* Load the individual bytes from LSB to MSB.  */
  
  for (n = 0; n < n_bytes; n++)
    {
      xop[0] = all_regs_rtx[REGNO (op[0]) + n];
      
      if ((CONST_INT_P (xsegment) && AVR_HAVE_LPMX)
          || (REG_P (xsegment) && AVR_HAVE_ELPMX))
        {
          if (n == n_bytes-1)
            avr_asm_len ("%3lpm %0,%a1", xop, plen, 1);
          else if (REGNO (xop[0]) == REG_Z)
            {
              avr_asm_len ("%3lpm %2,%a1+", xop, plen, 1);
              r30_in_tmp = true;
            }
          else
            avr_asm_len ("%3lpm %0,%a1+", xop, plen, 1);
        }
      else
        {
          gcc_assert (clobber_z);
          
          avr_asm_len ("%3lpm" CR_TAB
                       "mov %0,%2", xop, plen, 2);

          if (n != n_bytes-1)
            avr_asm_len ("adiw %1,1", xop, plen, 1);
        }
    }
  
  if (r30_in_tmp)
    avr_asm_len ("mov %1,%2", xop, plen, 1);
  
  if (!clobber_z
      && n_bytes > 1
      && !reg_unused_after (insn, lpm_addr_reg_rtx)
      && !reg_overlap_mentioned_p (op[0], lpm_addr_reg_rtx))
    {
      xop[2] = GEN_INT (n_bytes-1);
      avr_asm_len ("sbiw %1,%2", xop, plen, 1);
    }
  
  if (REG_P (xsegment) && AVR_HAVE_RAMPD)
    {
      /* Reset RAMPZ to 0 so that EBI devices don't read garbage from RAM */
      
      avr_asm_len ("out __RAMPZ__,__zero_reg__", xop, plen, 1);
    }

  return "";
}


/* Worker function for xload_8 insn.  */

const char*
avr_out_xload (rtx insn ATTRIBUTE_UNUSED, rtx *op, int *plen)
{
  rtx xop[4];

  xop[0] = op[0];
  xop[1] = op[1];
  xop[2] = lpm_addr_reg_rtx;
  xop[3] = AVR_HAVE_LPMX ? op[0] : lpm_reg_rtx;

  if (plen)
    *plen = 0;

  avr_asm_len ("sbrc %1,7" CR_TAB
               "ld %3,%a2" CR_TAB
               "sbrs %1,7", xop, plen, 3);

  avr_asm_len (AVR_HAVE_LPMX ? "lpm %3,%a2" : "lpm", xop, plen, 1);

  if (REGNO (xop[0]) != REGNO (xop[3]))
    avr_asm_len ("mov %0,%3", xop, plen, 1);
  
  return "";
}


const char*
output_movqi (rtx insn, rtx operands[], int *real_l)
{
  rtx dest = operands[0];
  rtx src = operands[1];
  
  if (avr_mem_flash_p (src)
      || avr_mem_flash_p (dest))
    {
      return avr_out_lpm (insn, operands, real_l);
    }

  if (real_l)
    *real_l = 1;
  
  if (register_operand (dest, QImode))
    {
      if (register_operand (src, QImode)) /* mov r,r */
        {
          if (test_hard_reg_class (STACK_REG, dest))
            return "out %0,%1";
          else if (test_hard_reg_class (STACK_REG, src))
            return "in %0,%1";
          
          return "mov %0,%1";
        }
      else if (CONSTANT_P (src))
        {
          output_reload_in_const (operands, NULL_RTX, real_l, false);
          return "";
        }
      else if (MEM_P (src))
        return out_movqi_r_mr (insn, operands, real_l); /* mov r,m */
    }
  else if (MEM_P (dest))
    {
      rtx xop[2];

      xop[0] = dest;
      xop[1] = src == const0_rtx ? zero_reg_rtx : src;

      return out_movqi_mr_r (insn, xop, real_l);
    }
  return "";
}


const char *
output_movhi (rtx insn, rtx xop[], int *plen)
{
  rtx dest = xop[0];
  rtx src = xop[1];

  gcc_assert (GET_MODE_SIZE (GET_MODE (dest)) == 2);
  
  if (avr_mem_flash_p (src)
      || avr_mem_flash_p (dest))
    {
      return avr_out_lpm (insn, xop, plen);
    }

  if (REG_P (dest))
    {
      if (REG_P (src)) /* mov r,r */
        {
          if (test_hard_reg_class (STACK_REG, dest))
            {
              if (AVR_HAVE_8BIT_SP)
                return avr_asm_len ("out __SP_L__,%A1", xop, plen, -1);

              if (AVR_XMEGA)
                return avr_asm_len ("out __SP_L__,%A1" CR_TAB
                                    "out __SP_H__,%B1", xop, plen, -2);
              
              /* Use simple load of SP if no interrupts are  used.  */
              
              return TARGET_NO_INTERRUPTS
                ? avr_asm_len ("out __SP_H__,%B1" CR_TAB
                               "out __SP_L__,%A1", xop, plen, -2)

                : avr_asm_len ("in __tmp_reg__,__SREG__"  CR_TAB
                               "cli"                      CR_TAB
                               "out __SP_H__,%B1"         CR_TAB
                               "out __SREG__,__tmp_reg__" CR_TAB
                               "out __SP_L__,%A1", xop, plen, -5);
            }
          else if (test_hard_reg_class (STACK_REG, src))
            {
              return !AVR_HAVE_SPH
                ? avr_asm_len ("in %A0,__SP_L__" CR_TAB
                               "clr %B0", xop, plen, -2)
                
                : avr_asm_len ("in %A0,__SP_L__" CR_TAB
                               "in %B0,__SP_H__", xop, plen, -2);
            }

          return AVR_HAVE_MOVW
            ? avr_asm_len ("movw %0,%1", xop, plen, -1)

            : avr_asm_len ("mov %A0,%A1" CR_TAB
                           "mov %B0,%B1", xop, plen, -2);
        } /* REG_P (src) */
      else if (CONSTANT_P (src))
        {
          return output_reload_inhi (xop, NULL, plen);
        }
      else if (MEM_P (src))
        {
          return out_movhi_r_mr (insn, xop, plen); /* mov r,m */
        }
    }
  else if (MEM_P (dest))
    {
      rtx xop[2];

      xop[0] = dest;
      xop[1] = src == const0_rtx ? zero_reg_rtx : src;

      return out_movhi_mr_r (insn, xop, plen);
    }
  
  fatal_insn ("invalid insn:", insn);
  
  return "";
}

static const char*
out_movqi_r_mr (rtx insn, rtx op[], int *plen)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx x = XEXP (src, 0);
  
  if (CONSTANT_ADDRESS_P (x))
    {
      return optimize > 0 && io_address_operand (x, QImode)
        ? avr_asm_len ("in %0,%i1", op, plen, -1)
        : avr_asm_len ("lds %0,%m1", op, plen, -2);
    }
  else if (GET_CODE (x) == PLUS
           && REG_P (XEXP (x, 0))
           && CONST_INT_P (XEXP (x, 1)))
    {
      /* memory access by reg+disp */

      int disp = INTVAL (XEXP (x, 1));
      
      if (disp - GET_MODE_SIZE (GET_MODE (src)) >= 63)
        {
          if (REGNO (XEXP (x, 0)) != REG_Y)
            fatal_insn ("incorrect insn:",insn);

          if (disp <= 63 + MAX_LD_OFFSET (GET_MODE (src)))
            return avr_asm_len ("adiw r28,%o1-63" CR_TAB
                                "ldd %0,Y+63"     CR_TAB
                                "sbiw r28,%o1-63", op, plen, -3);

          return avr_asm_len ("subi r28,lo8(-%o1)" CR_TAB
                              "sbci r29,hi8(-%o1)" CR_TAB
                              "ld %0,Y"            CR_TAB
                              "subi r28,lo8(%o1)"  CR_TAB
                              "sbci r29,hi8(%o1)", op, plen, -5);
        }
      else if (REGNO (XEXP (x, 0)) == REG_X)
        {
          /* This is a paranoid case LEGITIMIZE_RELOAD_ADDRESS must exclude
             it but I have this situation with extremal optimizing options.  */
          
          avr_asm_len ("adiw r26,%o1" CR_TAB
                       "ld %0,X", op, plen, -2);
          
          if (!reg_overlap_mentioned_p (dest, XEXP (x,0))
              && !reg_unused_after (insn, XEXP (x,0)))
            {
              avr_asm_len ("sbiw r26,%o1", op, plen, 1);
            }

          return "";
        }

      return avr_asm_len ("ldd %0,%1", op, plen, -1);
    }
  
  return avr_asm_len ("ld %0,%1", op, plen, -1);
}

static const char*
out_movhi_r_mr (rtx insn, rtx op[], int *plen)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx base = XEXP (src, 0);
  int reg_dest = true_regnum (dest);
  int reg_base = true_regnum (base);
  /* "volatile" forces reading low byte first, even if less efficient,
     for correct operation with 16-bit I/O registers.  */
  int mem_volatile_p = MEM_VOLATILE_P (src);

  if (reg_base > 0)
    {
      if (reg_dest == reg_base)         /* R = (R) */
        return avr_asm_len ("ld __tmp_reg__,%1+" CR_TAB
                            "ld %B0,%1"          CR_TAB
                            "mov %A0,__tmp_reg__", op, plen, -3);

      if (reg_base != REG_X)
        return avr_asm_len ("ld %A0,%1" CR_TAB
                            "ldd %B0,%1+1", op, plen, -2);
      
      avr_asm_len ("ld %A0,X+" CR_TAB
                   "ld %B0,X", op, plen, -2);
          
      if (!reg_unused_after (insn, base))
        avr_asm_len ("sbiw r26,1", op, plen, 1);

      return "";
    }
  else if (GET_CODE (base) == PLUS) /* (R + i) */
    {
      int disp = INTVAL (XEXP (base, 1));
      int reg_base = true_regnum (XEXP (base, 0));
      
      if (disp > MAX_LD_OFFSET (GET_MODE (src)))
        {
          if (REGNO (XEXP (base, 0)) != REG_Y)
            fatal_insn ("incorrect insn:",insn);
          
          return disp <= 63 + MAX_LD_OFFSET (GET_MODE (src))
            ? avr_asm_len ("adiw r28,%o1-62" CR_TAB
                           "ldd %A0,Y+62"    CR_TAB
                           "ldd %B0,Y+63"    CR_TAB
                           "sbiw r28,%o1-62", op, plen, -4)

            : avr_asm_len ("subi r28,lo8(-%o1)" CR_TAB
                           "sbci r29,hi8(-%o1)" CR_TAB
                           "ld %A0,Y"           CR_TAB
                           "ldd %B0,Y+1"        CR_TAB
                           "subi r28,lo8(%o1)"  CR_TAB
                           "sbci r29,hi8(%o1)", op, plen, -6);
        }

      /* This is a paranoid case. LEGITIMIZE_RELOAD_ADDRESS must exclude
         it but I have this situation with extremal
         optimization options.  */

      if (reg_base == REG_X)
        return reg_base == reg_dest
          ? avr_asm_len ("adiw r26,%o1"      CR_TAB
                         "ld __tmp_reg__,X+" CR_TAB
                         "ld %B0,X"          CR_TAB
                         "mov %A0,__tmp_reg__", op, plen, -4)

          : avr_asm_len ("adiw r26,%o1" CR_TAB
                         "ld %A0,X+"    CR_TAB
                         "ld %B0,X"     CR_TAB
                         "sbiw r26,%o1+1", op, plen, -4);

      return reg_base == reg_dest
        ? avr_asm_len ("ldd __tmp_reg__,%A1" CR_TAB
                       "ldd %B0,%B1"         CR_TAB
                       "mov %A0,__tmp_reg__", op, plen, -3)

        : avr_asm_len ("ldd %A0,%A1" CR_TAB
                       "ldd %B0,%B1", op, plen, -2);
    }
  else if (GET_CODE (base) == PRE_DEC) /* (--R) */
    {
      if (reg_overlap_mentioned_p (dest, XEXP (base, 0)))
        fatal_insn ("incorrect insn:", insn);

      if (!mem_volatile_p)
        return avr_asm_len ("ld %B0,%1" CR_TAB
                            "ld %A0,%1", op, plen, -2);
      
      return REGNO (XEXP (base, 0)) == REG_X
        ? avr_asm_len ("sbiw r26,2"  CR_TAB
                       "ld %A0,X+"   CR_TAB
                       "ld %B0,X"    CR_TAB
                       "sbiw r26,1", op, plen, -4)
        
        : avr_asm_len ("sbiw %r1,2"  CR_TAB
                       "ld %A0,%p1"  CR_TAB
                       "ldd %B0,%p1+1", op, plen, -3);
    }
  else if (GET_CODE (base) == POST_INC) /* (R++) */
    {
      if (reg_overlap_mentioned_p (dest, XEXP (base, 0)))
        fatal_insn ("incorrect insn:", insn);

      return avr_asm_len ("ld %A0,%1"  CR_TAB
                          "ld %B0,%1", op, plen, -2);
    }
  else if (CONSTANT_ADDRESS_P (base))
    {
      return optimize > 0 && io_address_operand (base, HImode)
        ? avr_asm_len ("in %A0,%i1" CR_TAB
                       "in %B0,%i1+1", op, plen, -2)

        : avr_asm_len ("lds %A0,%m1" CR_TAB
                       "lds %B0,%m1+1", op, plen, -4);
    }
  
  fatal_insn ("unknown move insn:",insn);
  return "";
}

static const char*
out_movsi_r_mr (rtx insn, rtx op[], int *l)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx base = XEXP (src, 0);
  int reg_dest = true_regnum (dest);
  int reg_base = true_regnum (base);
  int tmp;

  if (!l)
    l = &tmp;
  
  if (reg_base > 0)
    {
      if (reg_base == REG_X)        /* (R26) */
        {
          if (reg_dest == REG_X)
            /* "ld r26,-X" is undefined */
            return *l=7, ("adiw r26,3"        CR_TAB
                          "ld r29,X"          CR_TAB
                          "ld r28,-X"         CR_TAB
                          "ld __tmp_reg__,-X" CR_TAB
                          "sbiw r26,1"        CR_TAB
                          "ld r26,X"          CR_TAB
                          "mov r27,__tmp_reg__");
          else if (reg_dest == REG_X - 2)
            return *l=5, ("ld %A0,X+"          CR_TAB
                          "ld %B0,X+"          CR_TAB
                          "ld __tmp_reg__,X+"  CR_TAB
                          "ld %D0,X"           CR_TAB
                          "mov %C0,__tmp_reg__");
          else if (reg_unused_after (insn, base))
            return  *l=4, ("ld %A0,X+"  CR_TAB
                           "ld %B0,X+" CR_TAB
                           "ld %C0,X+" CR_TAB
                           "ld %D0,X");
          else
            return  *l=5, ("ld %A0,X+"  CR_TAB
                           "ld %B0,X+" CR_TAB
                           "ld %C0,X+" CR_TAB
                           "ld %D0,X"  CR_TAB
                           "sbiw r26,3");
        }
      else
        {
          if (reg_dest == reg_base)
            return *l=5, ("ldd %D0,%1+3" CR_TAB
                          "ldd %C0,%1+2" CR_TAB
                          "ldd __tmp_reg__,%1+1"  CR_TAB
                          "ld %A0,%1"  CR_TAB
                          "mov %B0,__tmp_reg__");
          else if (reg_base == reg_dest + 2)
            return *l=5, ("ld %A0,%1"             CR_TAB
                          "ldd %B0,%1+1"          CR_TAB
                          "ldd __tmp_reg__,%1+2"  CR_TAB
                          "ldd %D0,%1+3"          CR_TAB
                          "mov %C0,__tmp_reg__");
          else
            return *l=4, ("ld %A0,%1"    CR_TAB
                          "ldd %B0,%1+1" CR_TAB
                          "ldd %C0,%1+2" CR_TAB
                          "ldd %D0,%1+3");
        }
    }
  else if (GET_CODE (base) == PLUS) /* (R + i) */
    {
      int disp = INTVAL (XEXP (base, 1));
      
      if (disp > MAX_LD_OFFSET (GET_MODE (src)))
        {
          if (REGNO (XEXP (base, 0)) != REG_Y)
            fatal_insn ("incorrect insn:",insn);

          if (disp <= 63 + MAX_LD_OFFSET (GET_MODE (src)))
            return *l = 6, ("adiw r28,%o1-60" CR_TAB
                            "ldd %A0,Y+60"    CR_TAB
                            "ldd %B0,Y+61"    CR_TAB
                            "ldd %C0,Y+62"    CR_TAB
                            "ldd %D0,Y+63"    CR_TAB
                            "sbiw r28,%o1-60");

          return *l = 8, ("subi r28,lo8(-%o1)" CR_TAB
                          "sbci r29,hi8(-%o1)" CR_TAB
                          "ld %A0,Y"           CR_TAB
                          "ldd %B0,Y+1"        CR_TAB
                          "ldd %C0,Y+2"        CR_TAB
                          "ldd %D0,Y+3"        CR_TAB
                          "subi r28,lo8(%o1)"  CR_TAB
                          "sbci r29,hi8(%o1)");
        }

      reg_base = true_regnum (XEXP (base, 0));
      if (reg_base == REG_X)
        {
          /* R = (X + d) */
          if (reg_dest == REG_X)
            {
              *l = 7;
              /* "ld r26,-X" is undefined */
              return ("adiw r26,%o1+3"    CR_TAB
                      "ld r29,X"          CR_TAB
                      "ld r28,-X"         CR_TAB
                      "ld __tmp_reg__,-X" CR_TAB
                      "sbiw r26,1"        CR_TAB
                      "ld r26,X"          CR_TAB
                      "mov r27,__tmp_reg__");
            }
          *l = 6;
          if (reg_dest == REG_X - 2)
            return ("adiw r26,%o1"      CR_TAB
                    "ld r24,X+"         CR_TAB
                    "ld r25,X+"         CR_TAB
                    "ld __tmp_reg__,X+" CR_TAB
                    "ld r27,X"          CR_TAB
                    "mov r26,__tmp_reg__");

          return ("adiw r26,%o1" CR_TAB
                  "ld %A0,X+"    CR_TAB
                  "ld %B0,X+"    CR_TAB
                  "ld %C0,X+"    CR_TAB
                  "ld %D0,X"     CR_TAB
                  "sbiw r26,%o1+3");
        }
      if (reg_dest == reg_base)
        return *l=5, ("ldd %D0,%D1"          CR_TAB
                      "ldd %C0,%C1"          CR_TAB
                      "ldd __tmp_reg__,%B1"  CR_TAB
                      "ldd %A0,%A1"          CR_TAB
                      "mov %B0,__tmp_reg__");
      else if (reg_dest == reg_base - 2)
        return *l=5, ("ldd %A0,%A1"          CR_TAB
                      "ldd %B0,%B1"          CR_TAB
                      "ldd __tmp_reg__,%C1"  CR_TAB
                      "ldd %D0,%D1"          CR_TAB
                      "mov %C0,__tmp_reg__");
      return *l=4, ("ldd %A0,%A1" CR_TAB
                    "ldd %B0,%B1" CR_TAB
                    "ldd %C0,%C1" CR_TAB
                    "ldd %D0,%D1");
    }
  else if (GET_CODE (base) == PRE_DEC) /* (--R) */
    return *l=4, ("ld %D0,%1" CR_TAB
                  "ld %C0,%1" CR_TAB
                  "ld %B0,%1" CR_TAB
                  "ld %A0,%1");
  else if (GET_CODE (base) == POST_INC) /* (R++) */
    return *l=4, ("ld %A0,%1" CR_TAB
                  "ld %B0,%1" CR_TAB
                  "ld %C0,%1" CR_TAB
                  "ld %D0,%1");
  else if (CONSTANT_ADDRESS_P (base))
    return *l=8, ("lds %A0,%m1"   CR_TAB
                  "lds %B0,%m1+1" CR_TAB
                  "lds %C0,%m1+2" CR_TAB
                  "lds %D0,%m1+3");
    
  fatal_insn ("unknown move insn:",insn);
  return "";
}

static const char*
out_movsi_mr_r (rtx insn, rtx op[], int *l)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx base = XEXP (dest, 0);
  int reg_base = true_regnum (base);
  int reg_src = true_regnum (src);
  int tmp;
  
  if (!l)
    l = &tmp;
  
  if (CONSTANT_ADDRESS_P (base))
    return *l=8,("sts %m0,%A1" CR_TAB
                 "sts %m0+1,%B1" CR_TAB
                 "sts %m0+2,%C1" CR_TAB
                 "sts %m0+3,%D1");
  if (reg_base > 0)                 /* (r) */
    {
      if (reg_base == REG_X)                /* (R26) */
        {
          if (reg_src == REG_X)
            {
              /* "st X+,r26" is undefined */
              if (reg_unused_after (insn, base))
                return *l=6, ("mov __tmp_reg__,r27" CR_TAB
                              "st X,r26"            CR_TAB
                              "adiw r26,1"          CR_TAB
                              "st X+,__tmp_reg__"   CR_TAB
                              "st X+,r28"           CR_TAB
                              "st X,r29");
              else
                return *l=7, ("mov __tmp_reg__,r27" CR_TAB
                              "st X,r26"            CR_TAB
                              "adiw r26,1"          CR_TAB
                              "st X+,__tmp_reg__"   CR_TAB
                              "st X+,r28"           CR_TAB
                              "st X,r29"            CR_TAB
                              "sbiw r26,3");
            }
          else if (reg_base == reg_src + 2)
            {
              if (reg_unused_after (insn, base))
                return *l=7, ("mov __zero_reg__,%C1" CR_TAB
                              "mov __tmp_reg__,%D1"  CR_TAB
                              "st %0+,%A1"           CR_TAB
                              "st %0+,%B1"           CR_TAB
                              "st %0+,__zero_reg__"  CR_TAB
                              "st %0,__tmp_reg__"    CR_TAB
                              "clr __zero_reg__");
              else
                return *l=8, ("mov __zero_reg__,%C1" CR_TAB
                              "mov __tmp_reg__,%D1"  CR_TAB
                              "st %0+,%A1"           CR_TAB
                              "st %0+,%B1"           CR_TAB
                              "st %0+,__zero_reg__"  CR_TAB
                              "st %0,__tmp_reg__"    CR_TAB
                              "clr __zero_reg__"     CR_TAB
                              "sbiw r26,3");
            }
          return *l=5, ("st %0+,%A1" CR_TAB
                        "st %0+,%B1" CR_TAB
                        "st %0+,%C1" CR_TAB
                        "st %0,%D1"  CR_TAB
                        "sbiw r26,3");
        }
      else
        return *l=4, ("st %0,%A1"    CR_TAB
                      "std %0+1,%B1" CR_TAB
                      "std %0+2,%C1" CR_TAB
                      "std %0+3,%D1");
    }
  else if (GET_CODE (base) == PLUS) /* (R + i) */
    {
      int disp = INTVAL (XEXP (base, 1));
      reg_base = REGNO (XEXP (base, 0));
      if (disp > MAX_LD_OFFSET (GET_MODE (dest)))
        {
          if (reg_base != REG_Y)
            fatal_insn ("incorrect insn:",insn);

          if (disp <= 63 + MAX_LD_OFFSET (GET_MODE (dest)))
            return *l = 6, ("adiw r28,%o0-60" CR_TAB
                            "std Y+60,%A1"    CR_TAB
                            "std Y+61,%B1"    CR_TAB
                            "std Y+62,%C1"    CR_TAB
                            "std Y+63,%D1"    CR_TAB
                            "sbiw r28,%o0-60");

          return *l = 8, ("subi r28,lo8(-%o0)" CR_TAB
                          "sbci r29,hi8(-%o0)" CR_TAB
                          "st Y,%A1"           CR_TAB
                          "std Y+1,%B1"        CR_TAB
                          "std Y+2,%C1"        CR_TAB
                          "std Y+3,%D1"        CR_TAB
                          "subi r28,lo8(%o0)"  CR_TAB
                          "sbci r29,hi8(%o0)");
        }
      if (reg_base == REG_X)
        {
          /* (X + d) = R */
          if (reg_src == REG_X)
            {
              *l = 9;
              return ("mov __tmp_reg__,r26"  CR_TAB
                      "mov __zero_reg__,r27" CR_TAB
                      "adiw r26,%o0"         CR_TAB
                      "st X+,__tmp_reg__"    CR_TAB
                      "st X+,__zero_reg__"   CR_TAB
                      "st X+,r28"            CR_TAB
                      "st X,r29"             CR_TAB
                      "clr __zero_reg__"     CR_TAB
                      "sbiw r26,%o0+3");
            }
          else if (reg_src == REG_X - 2)
            {
              *l = 9;
              return ("mov __tmp_reg__,r26"  CR_TAB
                      "mov __zero_reg__,r27" CR_TAB
                      "adiw r26,%o0"         CR_TAB
                      "st X+,r24"            CR_TAB
                      "st X+,r25"            CR_TAB
                      "st X+,__tmp_reg__"    CR_TAB
                      "st X,__zero_reg__"    CR_TAB
                      "clr __zero_reg__"     CR_TAB
                      "sbiw r26,%o0+3");
            }
          *l = 6;
          return ("adiw r26,%o0" CR_TAB
                  "st X+,%A1"    CR_TAB
                  "st X+,%B1"    CR_TAB
                  "st X+,%C1"    CR_TAB
                  "st X,%D1"     CR_TAB
                  "sbiw r26,%o0+3");
        }
      return *l=4, ("std %A0,%A1" CR_TAB
                    "std %B0,%B1" CR_TAB
                    "std %C0,%C1" CR_TAB
                    "std %D0,%D1");
    }
  else if (GET_CODE (base) == PRE_DEC) /* (--R) */
    return *l=4, ("st %0,%D1" CR_TAB
                  "st %0,%C1" CR_TAB
                  "st %0,%B1" CR_TAB
                  "st %0,%A1");
  else if (GET_CODE (base) == POST_INC) /* (R++) */
    return *l=4, ("st %0,%A1" CR_TAB
                  "st %0,%B1" CR_TAB
                  "st %0,%C1" CR_TAB
                  "st %0,%D1");
  fatal_insn ("unknown move insn:",insn);
  return "";
}

const char *
output_movsisf (rtx insn, rtx operands[], int *l)
{
  int dummy;
  rtx dest = operands[0];
  rtx src = operands[1];
  int *real_l = l;
  
  if (avr_mem_flash_p (src)
      || avr_mem_flash_p (dest))
    {
      return avr_out_lpm (insn, operands, real_l);
    }

  if (!l)
    l = &dummy;
  
  if (register_operand (dest, VOIDmode))
    {
      if (register_operand (src, VOIDmode)) /* mov r,r */
        {
          if (true_regnum (dest) > true_regnum (src))
            {
              if (AVR_HAVE_MOVW)
                {
                  *l = 2;
                  return ("movw %C0,%C1" CR_TAB
                          "movw %A0,%A1");
                }
              *l = 4;
              return ("mov %D0,%D1" CR_TAB
                      "mov %C0,%C1" CR_TAB
                      "mov %B0,%B1" CR_TAB
                      "mov %A0,%A1");
            }
          else
            {
              if (AVR_HAVE_MOVW)
                {
                  *l = 2;
                  return ("movw %A0,%A1" CR_TAB
                          "movw %C0,%C1");
                }
              *l = 4;
              return ("mov %A0,%A1" CR_TAB
                      "mov %B0,%B1" CR_TAB
                      "mov %C0,%C1" CR_TAB
                      "mov %D0,%D1");
            }
        }
      else if (CONSTANT_P (src))
        {
          return output_reload_insisf (operands, NULL_RTX, real_l);
        }
      else if (GET_CODE (src) == MEM)
        return out_movsi_r_mr (insn, operands, real_l); /* mov r,m */
    }
  else if (GET_CODE (dest) == MEM)
    {
      const char *templ;

      if (src == CONST0_RTX (GET_MODE (dest)))
          operands[1] = zero_reg_rtx;

      templ = out_movsi_mr_r (insn, operands, real_l);

      if (!real_l)
        output_asm_insn (templ, operands);

      operands[1] = src;
      return "";
    }
  fatal_insn ("invalid insn:", insn);
  return "";
}


/* Handle loads of 24-bit types from memory to register.  */

static const char*
avr_out_load_psi (rtx insn, rtx *op, int *plen)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx base = XEXP (src, 0);
  int reg_dest = true_regnum (dest);
  int reg_base = true_regnum (base);
  
  if (reg_base > 0)
    {
      if (reg_base == REG_X)        /* (R26) */
        {
          if (reg_dest == REG_X)
            /* "ld r26,-X" is undefined */
            return avr_asm_len ("adiw r26,2"        CR_TAB
                                "ld r28,X"          CR_TAB
                                "ld __tmp_reg__,-X" CR_TAB
                                "sbiw r26,1"        CR_TAB
                                "ld r26,X"          CR_TAB
                                "mov r27,__tmp_reg__", op, plen, -6);
          else
            {
              avr_asm_len ("ld %A0,X+" CR_TAB
                           "ld %B0,X+" CR_TAB
                           "ld %C0,X", op, plen, -3);

              if (reg_dest != REG_X - 2
                  && !reg_unused_after (insn, base))
                {
                  avr_asm_len ("sbiw r26,2", op, plen, 1);
                }

              return "";
            }
        }
      else /* reg_base != REG_X */
        {
          if (reg_dest == reg_base)
            return avr_asm_len ("ldd %C0,%1+2"          CR_TAB
                                "ldd __tmp_reg__,%1+1"  CR_TAB
                                "ld  %A0,%1"            CR_TAB
                                "mov %B0,__tmp_reg__", op, plen, -4);
          else
            return avr_asm_len ("ld  %A0,%1"    CR_TAB
                                "ldd %B0,%1+1"  CR_TAB
                                "ldd %C0,%1+2", op, plen, -3);
        }
    }
  else if (GET_CODE (base) == PLUS) /* (R + i) */
    {
      int disp = INTVAL (XEXP (base, 1));
      
      if (disp > MAX_LD_OFFSET (GET_MODE (src)))
        {
          if (REGNO (XEXP (base, 0)) != REG_Y)
            fatal_insn ("incorrect insn:",insn);

          if (disp <= 63 + MAX_LD_OFFSET (GET_MODE (src)))
            return avr_asm_len ("adiw r28,%o1-61" CR_TAB
                                "ldd %A0,Y+61"    CR_TAB
                                "ldd %B0,Y+62"    CR_TAB
                                "ldd %C0,Y+63"    CR_TAB
                                "sbiw r28,%o1-61", op, plen, -5);

          return avr_asm_len ("subi r28,lo8(-%o1)" CR_TAB
                              "sbci r29,hi8(-%o1)" CR_TAB
                              "ld  %A0,Y"           CR_TAB
                              "ldd %B0,Y+1"        CR_TAB
                              "ldd %C0,Y+2"        CR_TAB
                              "subi r28,lo8(%o1)"  CR_TAB
                              "sbci r29,hi8(%o1)", op, plen, -7);
        }

      reg_base = true_regnum (XEXP (base, 0));
      if (reg_base == REG_X)
        {
          /* R = (X + d) */
          if (reg_dest == REG_X)
            {
              /* "ld r26,-X" is undefined */
              return avr_asm_len ("adiw r26,%o1+2"     CR_TAB
                                  "ld  r28,X"          CR_TAB
                                  "ld  __tmp_reg__,-X" CR_TAB
                                  "sbiw r26,1"         CR_TAB
                                  "ld  r26,X"          CR_TAB
                                  "mov r27,__tmp_reg__", op, plen, -6);
            }
          
            avr_asm_len ("adiw r26,%o1"      CR_TAB
                         "ld r24,X+"         CR_TAB
                         "ld r25,X+"         CR_TAB
                         "ld r26,X", op, plen, -4);

            if (reg_dest != REG_X - 2)
              avr_asm_len ("sbiw r26,%o1+2", op, plen, 1);

            return "";
        }
      
      if (reg_dest == reg_base)
        return avr_asm_len ("ldd %C0,%C1" CR_TAB
                            "ldd __tmp_reg__,%B1"  CR_TAB
                            "ldd %A0,%A1" CR_TAB
                            "mov %B0,__tmp_reg__", op, plen, -4);

        return avr_asm_len ("ldd %A0,%A1" CR_TAB
                            "ldd %B0,%B1" CR_TAB
                            "ldd %C0,%C1", op, plen, -3);
    }
  else if (GET_CODE (base) == PRE_DEC) /* (--R) */
    return avr_asm_len ("ld %C0,%1" CR_TAB
                        "ld %B0,%1" CR_TAB
                        "ld %A0,%1", op, plen, -3);
  else if (GET_CODE (base) == POST_INC) /* (R++) */
    return avr_asm_len ("ld %A0,%1" CR_TAB
                        "ld %B0,%1" CR_TAB
                        "ld %C0,%1", op, plen, -3);

  else if (CONSTANT_ADDRESS_P (base))
    return avr_asm_len ("lds %A0,%m1" CR_TAB
                        "lds %B0,%m1+1" CR_TAB
                        "lds %C0,%m1+2", op, plen , -6);
  
  fatal_insn ("unknown move insn:",insn);
  return "";
}

/* Handle store of 24-bit type from register or zero to memory.  */

static const char*
avr_out_store_psi (rtx insn, rtx *op, int *plen)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx base = XEXP (dest, 0);
  int reg_base = true_regnum (base);
  
  if (CONSTANT_ADDRESS_P (base))
    return avr_asm_len ("sts %m0,%A1"   CR_TAB
                        "sts %m0+1,%B1" CR_TAB
                        "sts %m0+2,%C1", op, plen, -6);
  
  if (reg_base > 0)                 /* (r) */
    {
      if (reg_base == REG_X)        /* (R26) */
        {
          gcc_assert (!reg_overlap_mentioned_p (base, src));
          
          avr_asm_len ("st %0+,%A1"  CR_TAB
                       "st %0+,%B1" CR_TAB
                       "st %0,%C1", op, plen, -3);

          if (!reg_unused_after (insn, base))
            avr_asm_len ("sbiw r26,2", op, plen, 1);

          return "";
        }
      else
        return avr_asm_len ("st %0,%A1"    CR_TAB
                            "std %0+1,%B1" CR_TAB
                            "std %0+2,%C1", op, plen, -3);
    }
  else if (GET_CODE (base) == PLUS) /* (R + i) */
    {
      int disp = INTVAL (XEXP (base, 1));
      reg_base = REGNO (XEXP (base, 0));

      if (disp > MAX_LD_OFFSET (GET_MODE (dest)))
        {
          if (reg_base != REG_Y)
            fatal_insn ("incorrect insn:",insn);

          if (disp <= 63 + MAX_LD_OFFSET (GET_MODE (dest)))
            return avr_asm_len ("adiw r28,%o0-61" CR_TAB
                                "std Y+61,%A1"    CR_TAB
                                "std Y+62,%B1"    CR_TAB
                                "std Y+63,%C1"    CR_TAB
                                "sbiw r28,%o0-60", op, plen, -5);

          return avr_asm_len ("subi r28,lo8(-%o0)" CR_TAB
                              "sbci r29,hi8(-%o0)" CR_TAB
                              "st Y,%A1"           CR_TAB
                              "std Y+1,%B1"        CR_TAB
                              "std Y+2,%C1"        CR_TAB
                              "subi r28,lo8(%o0)"  CR_TAB
                              "sbci r29,hi8(%o0)", op, plen, -7);
        }
      if (reg_base == REG_X)
        {
          /* (X + d) = R */
          gcc_assert (!reg_overlap_mentioned_p (XEXP (base, 0), src));
          
          avr_asm_len ("adiw r26,%o0" CR_TAB
                       "st X+,%A1"    CR_TAB
                       "st X+,%B1"    CR_TAB
                       "st X,%C1", op, plen, -4);

          if (!reg_unused_after (insn, XEXP (base, 0)))
            avr_asm_len ("sbiw r26,%o0+2", op, plen, 1);

          return "";
        }
      
      return avr_asm_len ("std %A0,%A1" CR_TAB
                          "std %B0,%B1" CR_TAB
                          "std %C0,%C1", op, plen, -3);
    }
  else if (GET_CODE (base) == PRE_DEC) /* (--R) */
    return avr_asm_len ("st %0,%C1" CR_TAB
                        "st %0,%B1" CR_TAB
                        "st %0,%A1", op, plen, -3);
  else if (GET_CODE (base) == POST_INC) /* (R++) */
    return avr_asm_len ("st %0,%A1" CR_TAB
                        "st %0,%B1" CR_TAB
                        "st %0,%C1", op, plen, -3);

  fatal_insn ("unknown move insn:",insn);
  return "";
}


/* Move around 24-bit stuff.  */

const char *
avr_out_movpsi (rtx insn, rtx *op, int *plen)
{
  rtx dest = op[0];
  rtx src = op[1];
  
  if (avr_mem_flash_p (src)
      || avr_mem_flash_p (dest))
    {
      return avr_out_lpm (insn, op, plen);
    }
  
  if (register_operand (dest, VOIDmode))
    {
      if (register_operand (src, VOIDmode)) /* mov r,r */
        {
          if (true_regnum (dest) > true_regnum (src))
            {
              avr_asm_len ("mov %C0,%C1", op, plen, -1);

              if (AVR_HAVE_MOVW)
                return avr_asm_len ("movw %A0,%A1", op, plen, 1);
              else
                return avr_asm_len ("mov %B0,%B1"  CR_TAB
                                    "mov %A0,%A1", op, plen, 2);
            }
          else
            {
              if (AVR_HAVE_MOVW)
                avr_asm_len ("movw %A0,%A1", op, plen, -1);
              else
                avr_asm_len ("mov %A0,%A1"  CR_TAB
                             "mov %B0,%B1", op, plen, -2);
              
              return avr_asm_len ("mov %C0,%C1", op, plen, 1);
            }
        }
      else if (CONSTANT_P (src))
        {
          return avr_out_reload_inpsi (op, NULL_RTX, plen);
        }
      else if (MEM_P (src))
        return avr_out_load_psi (insn, op, plen); /* mov r,m */
    }
  else if (MEM_P (dest))
    {
      rtx xop[2];

      xop[0] = dest;
      xop[1] = src == CONST0_RTX (GET_MODE (dest)) ? zero_reg_rtx : src;
      
      return avr_out_store_psi (insn, xop, plen);
    }
    
  fatal_insn ("invalid insn:", insn);
  return "";
}


static const char*
out_movqi_mr_r (rtx insn, rtx op[], int *plen)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx x = XEXP (dest, 0);
  
  if (CONSTANT_ADDRESS_P (x))
    {
      return optimize > 0 && io_address_operand (x, QImode)
        ? avr_asm_len ("out %i0,%1", op, plen, -1)
        : avr_asm_len ("sts %m0,%1", op, plen, -2);
    }
  else if (GET_CODE (x) == PLUS
           && REG_P (XEXP (x, 0))
           && CONST_INT_P (XEXP (x, 1)))
    {
      /* memory access by reg+disp */

      int disp = INTVAL (XEXP (x, 1));

      if (disp - GET_MODE_SIZE (GET_MODE (dest)) >= 63)
        {
          if (REGNO (XEXP (x, 0)) != REG_Y)
            fatal_insn ("incorrect insn:",insn);

          if (disp <= 63 + MAX_LD_OFFSET (GET_MODE (dest)))
            return avr_asm_len ("adiw r28,%o0-63" CR_TAB
                                "std Y+63,%1"     CR_TAB
                                "sbiw r28,%o0-63", op, plen, -3);

          return avr_asm_len ("subi r28,lo8(-%o0)" CR_TAB
                              "sbci r29,hi8(-%o0)" CR_TAB
                              "st Y,%1"            CR_TAB
                              "subi r28,lo8(%o0)"  CR_TAB
                              "sbci r29,hi8(%o0)", op, plen, -5);
        }
      else if (REGNO (XEXP (x,0)) == REG_X)
        {
          if (reg_overlap_mentioned_p (src, XEXP (x, 0)))
            {
              avr_asm_len ("mov __tmp_reg__,%1" CR_TAB
                           "adiw r26,%o0"       CR_TAB
                           "st X,__tmp_reg__", op, plen, -3);
            }
          else
            {
              avr_asm_len ("adiw r26,%o0" CR_TAB
                           "st X,%1", op, plen, -2);
            }
          
          if (!reg_unused_after (insn, XEXP (x,0)))
            avr_asm_len ("sbiw r26,%o0", op, plen, 1);

          return "";
        }
      
      return avr_asm_len ("std %0,%1", op, plen, -1);
    }
  
  return avr_asm_len ("st %0,%1", op, plen, -1);
}


/* Helper for the next function for XMEGA.  It does the same
   but with low byte first.  */

static const char*
avr_out_movhi_mr_r_xmega (rtx insn, rtx op[], int *plen)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx base = XEXP (dest, 0);
  int reg_base = true_regnum (base);
  int reg_src = true_regnum (src);

  /* "volatile" forces writing low byte first, even if less efficient,
     for correct operation with 16-bit I/O registers like SP.  */
  int mem_volatile_p = MEM_VOLATILE_P (dest);

  if (CONSTANT_ADDRESS_P (base))
    return optimize > 0 && io_address_operand (base, HImode)
      ? avr_asm_len ("out %i0,%A1" CR_TAB
                     "out %i0+1,%B1", op, plen, -2)

      : avr_asm_len ("sts %m0,%A1" CR_TAB
                     "sts %m0+1,%B1", op, plen, -4);
  
  if (reg_base > 0)
    {
      if (reg_base != REG_X)
        return avr_asm_len ("st %0,%A1" CR_TAB
                            "std %0+1,%B1", op, plen, -2);
      
      if (reg_src == REG_X)
        /* "st X+,r26" and "st -X,r26" are undefined.  */
        avr_asm_len ("mov __tmp_reg__,r27" CR_TAB
                     "st X,r26"            CR_TAB
                     "adiw r26,1"          CR_TAB
                     "st X,__tmp_reg__", op, plen, -4);
      else
        avr_asm_len ("st X+,%A1" CR_TAB
                     "st X,%B1", op, plen, -2);

      return reg_unused_after (insn, base)
        ? ""
        : avr_asm_len ("sbiw r26,1", op, plen, 1);
    }
  else if (GET_CODE (base) == PLUS)
    {
      int disp = INTVAL (XEXP (base, 1));
      reg_base = REGNO (XEXP (base, 0));
      if (disp > MAX_LD_OFFSET (GET_MODE (dest)))
        {
          if (reg_base != REG_Y)
            fatal_insn ("incorrect insn:",insn);
          
          return disp <= 63 + MAX_LD_OFFSET (GET_MODE (dest))
            ? avr_asm_len ("adiw r28,%o0-62" CR_TAB
                           "std Y+62,%A1"    CR_TAB
                           "std Y+63,%B1"    CR_TAB
                           "sbiw r28,%o0-62", op, plen, -4)

            : avr_asm_len ("subi r28,lo8(-%o0)" CR_TAB
                           "sbci r29,hi8(-%o0)" CR_TAB
                           "st Y,%A1"           CR_TAB
                           "std Y+1,%B1"        CR_TAB
                           "subi r28,lo8(%o0)"  CR_TAB
                           "sbci r29,hi8(%o0)", op, plen, -6);
        }
      
      if (reg_base != REG_X)
        return avr_asm_len ("std %A0,%A1" CR_TAB
                            "std %B0,%B1", op, plen, -2);
      /* (X + d) = R */
      return reg_src == REG_X
        ? avr_asm_len ("mov __tmp_reg__,r26"  CR_TAB
                       "mov __zero_reg__,r27" CR_TAB
                       "adiw r26,%o0"         CR_TAB
                       "st X+,__tmp_reg__"    CR_TAB
                       "st X,__zero_reg__"    CR_TAB
                       "clr __zero_reg__"     CR_TAB
                       "sbiw r26,%o0+1", op, plen, -7)

        : avr_asm_len ("adiw r26,%o0" CR_TAB
                       "st X+,%A1"    CR_TAB
                       "st X,%B1"     CR_TAB
                       "sbiw r26,%o0+1", op, plen, -4);
    }
  else if (GET_CODE (base) == PRE_DEC) /* (--R) */
    {
      if (!mem_volatile_p)
        return avr_asm_len ("st %0,%B1" CR_TAB
                            "st %0,%A1", op, plen, -2);

      return REGNO (XEXP (base, 0)) == REG_X
        ? avr_asm_len ("sbiw r26,2"  CR_TAB
                       "st X+,%A1"   CR_TAB
                       "st X,%B1"    CR_TAB
                       "sbiw r26,1", op, plen, -4)

        : avr_asm_len ("sbiw %r0,2"  CR_TAB
                       "st %p0,%A1"  CR_TAB
                       "std %p0+1,%B1", op, plen, -3);
    }
  else if (GET_CODE (base) == POST_INC) /* (R++) */
    {
      return avr_asm_len ("st %0,%A1"  CR_TAB
                          "st %0,%B1", op, plen, -2);
      
    }
  fatal_insn ("unknown move insn:",insn);
  return "";
}


static const char*
out_movhi_mr_r (rtx insn, rtx op[], int *plen)
{
  rtx dest = op[0];
  rtx src = op[1];
  rtx base = XEXP (dest, 0);
  int reg_base = true_regnum (base);
  int reg_src = true_regnum (src);
  int mem_volatile_p;

  /* "volatile" forces writing high-byte first (no-xmega) resp.
     low-byte first (xmega) even if less efficient, for correct
     operation with 16-bit I/O registers like.  */

  if (AVR_XMEGA)
    return avr_out_movhi_mr_r_xmega (insn, op, plen);

  mem_volatile_p = MEM_VOLATILE_P (dest);

  if (CONSTANT_ADDRESS_P (base))
    return optimize > 0 && io_address_operand (base, HImode)
      ? avr_asm_len ("out %i0+1,%B1" CR_TAB
                     "out %i0,%A1", op, plen, -2)

      : avr_asm_len ("sts %m0+1,%B1" CR_TAB
                     "sts %m0,%A1", op, plen, -4);
  
  if (reg_base > 0)
    {
      if (reg_base != REG_X)
        return avr_asm_len ("std %0+1,%B1" CR_TAB
                            "st %0,%A1", op, plen, -2);
      
      if (reg_src == REG_X)
        /* "st X+,r26" and "st -X,r26" are undefined.  */
        return !mem_volatile_p && reg_unused_after (insn, src)
          ? avr_asm_len ("mov __tmp_reg__,r27" CR_TAB
                         "st X,r26"            CR_TAB
                         "adiw r26,1"          CR_TAB
                         "st X,__tmp_reg__", op, plen, -4)

          : avr_asm_len ("mov __tmp_reg__,r27" CR_TAB
                         "adiw r26,1"          CR_TAB
                         "st X,__tmp_reg__"    CR_TAB
                         "sbiw r26,1"          CR_TAB
                         "st X,r26", op, plen, -5);
      
      return !mem_volatile_p && reg_unused_after (insn, base)
        ? avr_asm_len ("st X+,%A1" CR_TAB
                       "st X,%B1", op, plen, -2)
        : avr_asm_len ("adiw r26,1" CR_TAB
                       "st X,%B1"   CR_TAB
                       "st -X,%A1", op, plen, -3);
    }
  else if (GET_CODE (base) == PLUS)
    {
      int disp = INTVAL (XEXP (base, 1));
      reg_base = REGNO (XEXP (base, 0));
      if (disp > MAX_LD_OFFSET (GET_MODE (dest)))
        {
          if (reg_base != REG_Y)
            fatal_insn ("incorrect insn:",insn);
          
          return disp <= 63 + MAX_LD_OFFSET (GET_MODE (dest))
            ? avr_asm_len ("adiw r28,%o0-62" CR_TAB
                           "std Y+63,%B1"    CR_TAB
                           "std Y+62,%A1"    CR_TAB
                           "sbiw r28,%o0-62", op, plen, -4)

            : avr_asm_len ("subi r28,lo8(-%o0)" CR_TAB
                           "sbci r29,hi8(-%o0)" CR_TAB
                           "std Y+1,%B1"        CR_TAB
                           "st Y,%A1"           CR_TAB
                           "subi r28,lo8(%o0)"  CR_TAB
                           "sbci r29,hi8(%o0)", op, plen, -6);
        }
      
      if (reg_base != REG_X)
        return avr_asm_len ("std %B0,%B1" CR_TAB
                            "std %A0,%A1", op, plen, -2);
      /* (X + d) = R */
      return reg_src == REG_X
        ? avr_asm_len ("mov __tmp_reg__,r26"  CR_TAB
                       "mov __zero_reg__,r27" CR_TAB
                       "adiw r26,%o0+1"       CR_TAB
                       "st X,__zero_reg__"    CR_TAB
                       "st -X,__tmp_reg__"    CR_TAB
                       "clr __zero_reg__"     CR_TAB
                       "sbiw r26,%o0", op, plen, -7)

        : avr_asm_len ("adiw r26,%o0+1" CR_TAB
                       "st X,%B1"       CR_TAB
                       "st -X,%A1"      CR_TAB
                       "sbiw r26,%o0", op, plen, -4);
    }
  else if (GET_CODE (base) == PRE_DEC) /* (--R) */
    {
      return avr_asm_len ("st %0,%B1" CR_TAB
                          "st %0,%A1", op, plen, -2);
    }
  else if (GET_CODE (base) == POST_INC) /* (R++) */
    {
      if (!mem_volatile_p)
        return avr_asm_len ("st %0,%A1"  CR_TAB
                            "st %0,%B1", op, plen, -2);
      
      return REGNO (XEXP (base, 0)) == REG_X
        ? avr_asm_len ("adiw r26,1"  CR_TAB
                       "st X,%B1"    CR_TAB