reorganize files

5 files changed, 3902 insertions, 0 deletions

diff --git a/libgcc/fp-bit.c b/libgcc/fp-bit.c
new file mode 100755
index 0000000..6b8bd70
--- /dev/null
+++ b/libgcc/fp-bit.c
@@ -0,0 +1,1507 @@
+/* This is a software floating point library which can be used instead of
+   the floating point routines in libgcc1.c for targets without hardware
+   floating point. 
+ Copyright (C) 1994, 1995, 1996, 1997, 1998 Free Software Foundation, Inc.
+
+This file 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 2, or (at your option) any
+later version.
+
+In addition to the permissions in the GNU General Public License, the
+Free Software Foundation gives you unlimited permission to link the
+compiled version of this file with other programs, and to distribute
+those programs without any restriction coming from the use of this
+file.  (The General Public License restrictions do apply in other
+respects; for example, they cover modification of the file, and
+distribution when not linked into another program.)
+
+This file 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 this program; see the file COPYING.  If not, write to
+the Free Software Foundation, 59 Temple Place - Suite 330,
+Boston, MA 02111-1307, USA.  */
+
+/* As a special exception, if you link this library with other files,
+   some of which are compiled with GCC, to produce an executable,
+   this library does not by itself cause the resulting executable
+   to be covered by the GNU General Public License.
+   This exception does not however invalidate any other reasons why
+   the executable file might be covered by the GNU General Public License.  */
+
+/* This implements IEEE 754 format arithmetic, but does not provide a
+   mechanism for setting the rounding mode, or for generating or handling
+   exceptions.
+
+   The original code by Steve Chamberlain, hacked by Mark Eichin and Jim
+   Wilson, all of Cygnus Support.  */
+
+/* The intended way to use this file is to make two copies, add `#define FLOAT'
+   to one copy, then compile both copies and add them to libgcc.a.  */
+
+/* Defining FINE_GRAINED_LIBRARIES allows one to select which routines
+   from this file are compiled via additional -D options.
+
+   This avoids the need to pull in the entire fp emulation library
+   when only a small number of functions are needed.
+
+   If FINE_GRAINED_LIBRARIES is not defined, then compile every 
+   suitable routine.  */
+#ifndef FINE_GRAINED_LIBRARIES
+#define L_pack_df
+#define L_unpack_df
+#define L_pack_sf
+#define L_unpack_sf
+#define L_addsub_sf
+#define L_addsub_df
+#define L_mul_sf
+#define L_mul_df
+#define L_div_sf
+#define L_div_df
+#define L_fpcmp_parts_sf
+#define L_fpcmp_parts_df
+#define L_compare_sf
+#define L_compare_df
+#define L_eq_sf
+#define L_eq_df
+#define L_ne_sf
+#define L_ne_df
+#define L_gt_sf
+#define L_gt_df
+#define L_ge_sf
+#define L_ge_df
+#define L_lt_sf
+#define L_lt_df
+#define L_le_sf
+#define L_le_df
+#define L_si_to_sf
+#define L_si_to_df
+#define L_sf_to_si
+#define L_df_to_si
+#define L_f_to_usi
+#define L_df_to_usi
+#define L_negate_sf
+#define L_negate_df
+#define L_make_sf
+#define L_make_df
+#define L_sf_to_df
+#define L_df_to_sf
+#endif
+
+/* The following macros can be defined to change the behaviour of this file:
+   FLOAT: Implement a `float', aka SFmode, fp library.  If this is not
+     defined, then this file implements a `double', aka DFmode, fp library.
+   FLOAT_ONLY: Used with FLOAT, to implement a `float' only library, i.e.
+     don't include float->double conversion which requires the double library.
+     This is useful only for machines which can't support doubles, e.g. some
+     8-bit processors.
+   CMPtype: Specify the type that floating point compares should return.
+     This defaults to SItype, aka int.
+   US_SOFTWARE_GOFAST: This makes all entry points use the same names as the
+     US Software goFast library.  If this is not defined, the entry points use
+     the same names as libgcc1.c.
+   _DEBUG_BITFLOAT: This makes debugging the code a little easier, by adding
+     two integers to the FLO_union_type.  
+   NO_NANS: Disable nan and infinity handling
+   SMALL_MACHINE: Useful when operations on QIs and HIs are faster
+     than on an SI */
+
+/* We don't currently support extended floats (long doubles) on machines
+   without hardware to deal with them.
+
+   These stubs are just to keep the linker from complaining about unresolved
+   references which can be pulled in from libio & libstdc++, even if the
+   user isn't using long doubles.  However, they may generate an unresolved
+   external to abort if abort is not used by the function, and the stubs
+   are referenced from within libc, since libgcc goes before and after the
+   system library.  */
+
+#ifdef EXTENDED_FLOAT_STUBS
+__truncxfsf2 (){ abort(); }
+__extendsfxf2 (){ abort(); }
+__addxf3 (){ abort(); }
+__divxf3 (){ abort(); }
+__eqxf2 (){ abort(); }
+__extenddfxf2 (){ abort(); }
+__gtxf2 (){ abort(); }
+__lexf2 (){ abort(); }
+__ltxf2 (){ abort(); }
+__mulxf3 (){ abort(); }
+__negxf2 (){ abort(); }
+__nexf2 (){ abort(); }
+__subxf3 (){ abort(); }
+__truncxfdf2 (){ abort(); }
+
+__trunctfsf2 (){ abort(); }
+__extendsftf2 (){ abort(); }
+__addtf3 (){ abort(); }
+__divtf3 (){ abort(); }
+__eqtf2 (){ abort(); }
+__extenddftf2 (){ abort(); }
+__gttf2 (){ abort(); }
+__letf2 (){ abort(); }
+__lttf2 (){ abort(); }
+__multf3 (){ abort(); }
+__negtf2 (){ abort(); }
+__netf2 (){ abort(); }
+__subtf3 (){ abort(); }
+__trunctfdf2 (){ abort(); }
+__gexf2 (){ abort(); }
+__fixxfsi (){ abort(); }
+__floatsixf (){ abort(); }
+#else	/* !EXTENDED_FLOAT_STUBS, rest of file */
+
+
+typedef float SFtype __attribute__ ((mode (SF)));
+typedef float DFtype __attribute__ ((mode (DF)));
+
+typedef int HItype __attribute__ ((mode (HI)));
+typedef int SItype __attribute__ ((mode (SI)));
+typedef int DItype __attribute__ ((mode (DI)));
+
+/* The type of the result of a fp compare */
+#ifndef CMPtype
+#define CMPtype SItype
+#endif
+
+typedef unsigned int UHItype __attribute__ ((mode (HI)));
+typedef unsigned int USItype __attribute__ ((mode (SI)));
+typedef unsigned int UDItype __attribute__ ((mode (DI)));
+
+#define MAX_SI_INT   ((SItype) ((unsigned) (~0)>>1))
+#define MAX_USI_INT  ((USItype) ~0)
+
+
+#ifdef FLOAT_ONLY
+#define NO_DI_MODE
+#endif
+
+#ifdef FLOAT
+#	define NGARDS    7L
+#	define GARDROUND 0x3f
+#	define GARDMASK  0x7f
+#	define GARDMSB   0x40
+#	define EXPBITS 8
+#	define EXPBIAS 127
+#	define FRACBITS 23
+#	define EXPMAX (0xff)
+#	define QUIET_NAN 0x100000L
+#	define FRAC_NBITS 32
+#	define FRACHIGH  0x80000000L
+#	define FRACHIGH2 0xc0000000L
+#	define pack_d __pack_f
+#	define unpack_d __unpack_f
+#	define __fpcmp_parts __fpcmp_parts_f
+	typedef USItype fractype;
+	typedef UHItype halffractype;
+	typedef SFtype FLO_type;
+	typedef SItype intfrac;
+
+#else
+#	define PREFIXFPDP dp
+#	define PREFIXSFDF df
+#	define NGARDS 8L
+#	define GARDROUND 0x7f
+#	define GARDMASK  0xff
+#	define GARDMSB   0x80
+#	define EXPBITS 11
+#	define EXPBIAS 1023
+#	define FRACBITS 52
+#	define EXPMAX (0x7ff)
+#	define QUIET_NAN 0x8000000000000LL
+#	define FRAC_NBITS 64
+#	define FRACHIGH  0x8000000000000000LL
+#	define FRACHIGH2 0xc000000000000000LL
+#	define pack_d __pack_d
+#	define unpack_d __unpack_d
+#	define __fpcmp_parts __fpcmp_parts_d
+	typedef UDItype fractype;
+	typedef USItype halffractype;
+	typedef DFtype FLO_type;
+	typedef DItype intfrac;
+#endif
+
+#ifdef US_SOFTWARE_GOFAST
+#	ifdef FLOAT
+#		define add 		fpadd
+#		define sub 		fpsub
+#		define multiply 	fpmul
+#		define divide 		fpdiv
+#		define compare 		fpcmp
+#		define si_to_float 	sitofp
+#		define float_to_si 	fptosi
+#		define float_to_usi 	fptoui
+#		define negate 		__negsf2
+#		define sf_to_df		fptodp
+#		define dptofp 		dptofp
+#else
+#		define add 		dpadd
+#		define sub 		dpsub
+#		define multiply 	dpmul
+#		define divide 		dpdiv
+#		define compare 		dpcmp
+#		define si_to_float 	litodp
+#		define float_to_si 	dptoli
+#		define float_to_usi 	dptoul
+#		define negate 		__negdf2
+#		define df_to_sf 	dptofp
+#endif
+#else
+#	ifdef FLOAT
+#		define add 		__addsf3
+#		define sub 		__subsf3
+#		define multiply 	__mulsf3
+#		define divide 		__divsf3
+#		define compare 		__cmpsf2
+#		define _eq_f2 		__eqsf2
+#		define _ne_f2 		__nesf2
+#		define _gt_f2 		__gtsf2
+#		define _ge_f2 		__gesf2
+#		define _lt_f2 		__ltsf2
+#		define _le_f2 		__lesf2
+#		define si_to_float 	__floatsisf
+#		define float_to_si 	__fixsfsi
+#		define float_to_usi 	__fixunssfsi
+#		define negate 		__negsf2
+#		define sf_to_df		__extendsfdf2
+#else
+#		define add 		__adddf3
+#		define sub 		__subdf3
+#		define multiply 	__muldf3
+#		define divide 		__divdf3
+#		define compare 		__cmpdf2
+#		define _eq_f2 		__eqdf2
+#		define _ne_f2 		__nedf2
+#		define _gt_f2 		__gtdf2
+#		define _ge_f2 		__gedf2
+#		define _lt_f2 		__ltdf2
+#		define _le_f2 		__ledf2
+#		define si_to_float 	__floatsidf
+#		define float_to_si 	__fixdfsi
+#		define float_to_usi 	__fixunsdfsi
+#		define negate 		__negdf2
+#		define df_to_sf		__truncdfsf2
+#	endif
+#endif
+
+
+#ifndef INLINE
+#define INLINE __inline__
+#endif
+
+/* Preserve the sticky-bit when shifting fractions to the right.  */
+#define LSHIFT(a) { a = (a & 1) | (a >> 1); }
+
+/* numeric parameters */
+/* F_D_BITOFF is the number of bits offset between the MSB of the mantissa
+   of a float and of a double. Assumes there are only two float types.
+   (double::FRAC_BITS+double::NGARDS-(float::FRAC_BITS-float::NGARDS))
+ */
+#define F_D_BITOFF (52+8-(23+7))
+
+
+#define NORMAL_EXPMIN (-(EXPBIAS)+1)
+#define IMPLICIT_1 (1LL<<(FRACBITS+NGARDS))
+#define IMPLICIT_2 (1LL<<(FRACBITS+1+NGARDS))
+
+/* common types */
+
+typedef enum
+{
+  CLASS_SNAN,
+  CLASS_QNAN,
+  CLASS_ZERO,
+  CLASS_NUMBER,
+  CLASS_INFINITY
+} fp_class_type;
+
+typedef struct
+{
+#ifdef SMALL_MACHINE
+  char class;
+  unsigned char sign;
+  short normal_exp;
+#else
+  fp_class_type class;
+  unsigned int sign;
+  int normal_exp;
+#endif
+
+  union
+    {
+      fractype ll;
+      halffractype l[2];
+    } fraction;
+} fp_number_type;
+
+typedef union
+{
+  FLO_type value;
+  fractype value_raw;
+
+#ifndef FLOAT
+  halffractype words[2];
+#endif
+
+#ifdef FLOAT_BIT_ORDER_MISMATCH
+  struct
+    {
+      fractype fraction:FRACBITS __attribute__ ((packed));
+      unsigned int exp:EXPBITS __attribute__ ((packed));
+      unsigned int sign:1 __attribute__ ((packed));
+    }
+  bits;
+#endif
+
+#ifdef _DEBUG_BITFLOAT
+  struct
+    {
+      unsigned int sign:1 __attribute__ ((packed));
+      unsigned int exp:EXPBITS __attribute__ ((packed));
+      fractype fraction:FRACBITS __attribute__ ((packed));
+    }
+  bits_big_endian;
+
+  struct
+    {
+      fractype fraction:FRACBITS __attribute__ ((packed));
+      unsigned int exp:EXPBITS __attribute__ ((packed));
+      unsigned int sign:1 __attribute__ ((packed));
+    }
+  bits_little_endian;
+#endif
+}
+FLO_union_type;
+
+
+/* end of header */
+
+/* IEEE "special" number predicates */
+
+#ifdef NO_NANS
+
+#define nan() 0
+#define isnan(x) 0
+#define isinf(x) 0
+#else
+
+INLINE
+static fp_number_type *
+nan ()
+{
+  static fp_number_type thenan;
+
+  return &thenan;
+}
+
+INLINE
+static int
+isnan ( fp_number_type *  x)
+{
+  return x->class == CLASS_SNAN || x->class == CLASS_QNAN;
+}
+
+INLINE
+static int
+isinf ( fp_number_type *  x)
+{
+  return x->class == CLASS_INFINITY;
+}
+
+#endif
+
+INLINE
+static int
+iszero ( fp_number_type *  x)
+{
+  return x->class == CLASS_ZERO;
+}
+
+INLINE 
+static void
+flip_sign ( fp_number_type *  x)
+{
+  x->sign = !x->sign;
+}
+
+extern FLO_type pack_d ( fp_number_type * );
+
+#if defined(L_pack_df) || defined(L_pack_sf)
+FLO_type
+pack_d ( fp_number_type *  src)
+{
+  FLO_union_type dst;
+  fractype fraction = src->fraction.ll;	/* wasn't unsigned before? */
+  int sign = src->sign;
+  int exp = 0;
+
+  if (isnan (src))
+    {
+      exp = EXPMAX;
+      if (src->class == CLASS_QNAN || 1)
+	{
+	  fraction |= QUIET_NAN;
+	}
+    }
+  else if (isinf (src))
+    {
+      exp = EXPMAX;
+      fraction = 0;
+    }
+  else if (iszero (src))
+    {
+      exp = 0;
+      fraction = 0;
+    }
+  else if (fraction == 0)
+    {
+      exp = 0;
+    }
+  else
+    {
+      if (src->normal_exp < NORMAL_EXPMIN)
+	{
+	  /* This number's exponent is too low to fit into the bits
+	     available in the number, so we'll store 0 in the exponent and
+	     shift the fraction to the right to make up for it.  */
+
+	  int shift = NORMAL_EXPMIN - src->normal_exp;
+
+	  exp = 0;
+
+	  if (shift > FRAC_NBITS - NGARDS)
+	    {
+	      /* No point shifting, since it's more that 64 out.  */
+	      fraction = 0;
+	    }
+	  else
+	    {
+	      /* Shift by the value */
+	      fraction >>= shift;
+	    }
+	  fraction >>= NGARDS;
+	}
+      else if (src->normal_exp > EXPBIAS)
+	{
+	  exp = EXPMAX;
+	  fraction = 0;
+	}
+      else
+	{
+	  exp = src->normal_exp + EXPBIAS;
+	  /* IF the gard bits are the all zero, but the first, then we're
+	     half way between two numbers, choose the one which makes the
+	     lsb of the answer 0.  */
+	  if ((fraction & GARDMASK) == GARDMSB)
+	    {
+	      if (fraction & (1 << NGARDS))
+		fraction += GARDROUND + 1;
+	    }
+	  else
+	    {
+	      /* Add a one to the guards to round up */
+	      fraction += GARDROUND;
+	    }
+	  if (fraction >= IMPLICIT_2)
+	    {
+	      fraction >>= 1;
+	      exp += 1;
+	    }
+	  fraction >>= NGARDS;
+	}
+    }
+
+  /* We previously used bitfields to store the number, but this doesn't
+     handle little/big endian systems conveniently, so use shifts and
+     masks */
+#ifdef FLOAT_BIT_ORDER_MISMATCH
+  dst.bits.fraction = fraction;
+  dst.bits.exp = exp;
+  dst.bits.sign = sign;
+#else
+  dst.value_raw = fraction & ((((fractype)1) << FRACBITS) - (fractype)1);
+  dst.value_raw |= ((fractype) (exp & ((1 << EXPBITS) - 1))) << FRACBITS;
+  dst.value_raw |= ((fractype) (sign & 1)) << (FRACBITS | EXPBITS);
+#endif
+
+#if defined(FLOAT_WORD_ORDER_MISMATCH) && !defined(FLOAT)
+  {
+    halffractype tmp = dst.words[0];
+    dst.words[0] = dst.words[1];
+    dst.words[1] = tmp;
+  }
+#endif
+
+  return dst.value;
+}
+#endif
+
+extern void unpack_d (FLO_union_type *, fp_number_type *);
+
+#if defined(L_unpack_df) || defined(L_unpack_sf)
+void
+unpack_d (FLO_union_type * src, fp_number_type * dst)
+{
+  /* We previously used bitfields to store the number, but this doesn't
+     handle little/big endian systems conveniently, so use shifts and
+     masks */
+  fractype fraction;
+  int exp;
+  int sign;
+
+#if defined(FLOAT_WORD_ORDER_MISMATCH) && !defined(FLOAT)
+  FLO_union_type swapped;
+
+  swapped.words[0] = src->words[1];
+  swapped.words[1] = src->words[0];
+  src = &swapped;
+#endif
+  
+#ifdef FLOAT_BIT_ORDER_MISMATCH
+  fraction = src->bits.fraction;
+  exp = src->bits.exp;
+  sign = src->bits.sign;
+#else
+  fraction = src->value_raw & ((((fractype)1) << FRACBITS) - (fractype)1);
+  exp = ((int)(src->value_raw >> FRACBITS)) & ((1 << EXPBITS) - 1);
+  sign = ((int)(src->value_raw >> (FRACBITS + EXPBITS))) & 1;
+#endif
+
+  dst->sign = sign;
+  if (exp == 0)
+    {
+      /* Hmm.  Looks like 0 */
+      if (fraction == 0)
+	{
+	  /* tastes like zero */
+	  dst->class = CLASS_ZERO;
+	}
+      else
+	{
+	  /* Zero exponent with non zero fraction - it's denormalized,
+	     so there isn't a leading implicit one - we'll shift it so
+	     it gets one.  */
+	  dst->normal_exp = exp - EXPBIAS + 1;
+	  fraction <<= NGARDS;
+
+	  dst->class = CLASS_NUMBER;
+#if 1
+	  while (fraction < IMPLICIT_1)
+	    {
+	      fraction <<= 1;
+	      dst->normal_exp--;
+	    }
+#endif
+	  dst->fraction.ll = fraction;
+	}
+    }
+  else if (exp == EXPMAX)
+    {
+      /* Huge exponent*/
+      if (fraction == 0)
+	{
+	  /* Attached to a zero fraction - means infinity */
+	  dst->class = CLASS_INFINITY;
+	}
+      else
+	{
+	  /* Non zero fraction, means nan */
+	  if (fraction & QUIET_NAN)
+	    {
+	      dst->class = CLASS_QNAN;
+	    }
+	  else
+	    {
+	      dst->class = CLASS_SNAN;
+	    }
+	  /* Keep the fraction part as the nan number */
+	  dst->fraction.ll = fraction;
+	}
+    }
+  else
+    {
+      /* Nothing strange about this number */
+      dst->normal_exp = exp - EXPBIAS;
+      dst->class = CLASS_NUMBER;
+      dst->fraction.ll = (fraction << NGARDS) | IMPLICIT_1;
+    }
+}
+#endif
+
+#if defined(L_addsub_sf) || defined(L_addsub_df)
+static fp_number_type *
+_fpadd_parts (fp_number_type * a,
+	      fp_number_type * b,
+	      fp_number_type * tmp)
+{
+  intfrac tfraction;
+
+  /* Put commonly used fields in local variables.  */
+  int a_normal_exp;
+  int b_normal_exp;
+  fractype a_fraction;
+  fractype b_fraction;
+
+  if (isnan (a))
+    {
+      return a;
+    }
+  if (isnan (b))
+    {
+      return b;
+    }
+  if (isinf (a))
+    {
+      /* Adding infinities with opposite signs yields a NaN.  */
+      if (isinf (b) && a->sign != b->sign)
+	return nan ();
+      return a;
+    }
+  if (isinf (b))
+    {
+      return b;
+    }
+  if (iszero (b))
+    {
+      if (iszero (a))
+	{
+	  *tmp = *a;
+	  tmp->sign = a->sign & b->sign;
+	  return tmp;
+	}
+      return a;
+    }
+  if (iszero (a))
+    {
+      return b;
+    }
+
+  /* Got two numbers. shift the smaller and increment the exponent till
+     they're the same */
+  {
+    int diff;
+
+    a_normal_exp = a->normal_exp;
+    b_normal_exp = b->normal_exp;
+    a_fraction = a->fraction.ll;
+    b_fraction = b->fraction.ll;
+
+    diff = a_normal_exp - b_normal_exp;
+
+    if (diff < 0)
+      diff = -diff;
+    if (diff < FRAC_NBITS)
+      {
+	/* ??? This does shifts one bit at a time.  Optimize.  */
+	while (a_normal_exp > b_normal_exp)
+	  {
+	    b_normal_exp++;
+	    LSHIFT (b_fraction);
+	  }
+	while (b_normal_exp > a_normal_exp)
+	  {
+	    a_normal_exp++;
+	    LSHIFT (a_fraction);
+	  }
+      }
+    else
+      {
+	/* Somethings's up.. choose the biggest */
+	if (a_normal_exp > b_normal_exp)
+	  {
+	    b_normal_exp = a_normal_exp;
+	    b_fraction = 0;
+	  }
+	else
+	  {
+	    a_normal_exp = b_normal_exp;
+	    a_fraction = 0;
+	  }
+      }
+  }
+
+  if (a->sign != b->sign)
+    {
+      if (a->sign)
+	{
+	  tfraction = -a_fraction + b_fraction;
+	}
+      else
+	{
+	  tfraction = a_fraction - b_fraction;
+	}
+      if (tfraction >= 0)
+	{
+	  tmp->sign = 0;
+	  tmp->normal_exp = a_normal_exp;
+	  tmp->fraction.ll = tfraction;
+	}
+      else
+	{
+	  tmp->sign = 1;
+	  tmp->normal_exp = a_normal_exp;
+	  tmp->fraction.ll = -tfraction;
+	}
+      /* and renormalize it */
+
+      while (tmp->fraction.ll < IMPLICIT_1 && tmp->fraction.ll)
+	{
+	  tmp->fraction.ll <<= 1;
+	  tmp->normal_exp--;
+	}
+    }
+  else
+    {
+      tmp->sign = a->sign;
+      tmp->normal_exp = a_normal_exp;
+      tmp->fraction.ll = a_fraction + b_fraction;
+    }
+  tmp->class = CLASS_NUMBER;
+  /* Now the fraction is added, we have to shift down to renormalize the
+     number */
+
+  if (tmp->fraction.ll >= IMPLICIT_2)
+    {
+      LSHIFT (tmp->fraction.ll);
+      tmp->normal_exp++;
+    }
+  return tmp;
+
+}
+
+FLO_type
+add (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+  fp_number_type tmp;
+  fp_number_type *res;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  res = _fpadd_parts (&a, &b, &tmp);
+
+  return pack_d (res);
+}
+
+FLO_type
+sub (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+  fp_number_type tmp;
+  fp_number_type *res;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  b.sign ^= 1;
+
+  res = _fpadd_parts (&a, &b, &tmp);
+
+  return pack_d (res);
+}
+#endif
+
+#if defined(L_mul_sf) || defined(L_mul_df)
+static INLINE fp_number_type *
+_fpmul_parts ( fp_number_type *  a,
+	       fp_number_type *  b,
+	       fp_number_type * tmp)
+{
+  fractype low = 0;
+  fractype high = 0;
+
+  if (isnan (a))
+    {
+      a->sign = a->sign != b->sign;
+      return a;
+    }
+  if (isnan (b))
+    {
+      b->sign = a->sign != b->sign;
+      return b;
+    }
+  if (isinf (a))
+    {
+      if (iszero (b))
+	return nan ();
+      a->sign = a->sign != b->sign;
+      return a;
+    }
+  if (isinf (b))
+    {
+      if (iszero (a))
+	{
+	  return nan ();
+	}
+      b->sign = a->sign != b->sign;
+      return b;
+    }
+  if (iszero (a))
+    {
+      a->sign = a->sign != b->sign;
+      return a;
+    }
+  if (iszero (b))
+    {
+      b->sign = a->sign != b->sign;
+      return b;
+    }
+
+  /* Calculate the mantissa by multiplying both 64bit numbers to get a
+     128 bit number */
+  {
+#if defined(NO_DI_MODE)
+    {
+      fractype x = a->fraction.ll;
+      fractype ylow = b->fraction.ll;
+      fractype yhigh = 0;
+      int bit;
+
+      /* ??? This does multiplies one bit at a time.  Optimize.  */
+      for (bit = 0; bit < FRAC_NBITS; bit++)
+	{
+	  int carry;
+
+	  if (x & 1)
+	    {
+	      carry = (low += ylow) < ylow;
+	      high += yhigh + carry;
+	    }
+	  yhigh <<= 1;
+	  if (ylow & FRACHIGH)
+	    {
+	      yhigh |= 1;
+	    }
+	  ylow <<= 1;
+	  x >>= 1;
+	}
+    }
+#elif defined(FLOAT) 
+    {
+      /* Multiplying two 32 bit numbers to get a 64 bit number  on 
+        a machine with DI, so we're safe */
+
+      DItype answer = (DItype)(a->fraction.ll) * (DItype)(b->fraction.ll);
+      
+      high = answer >> 32;
+      low = answer;
+    }
+#else
+    /* Doing a 64*64 to 128 */
+    {
+      UDItype nl = a->fraction.ll & 0xffffffff;
+      UDItype nh = a->fraction.ll >> 32;
+      UDItype ml = b->fraction.ll & 0xffffffff;
+      UDItype mh = b->fraction.ll >>32;
+      UDItype pp_ll = ml * nl;
+      UDItype pp_hl = mh * nl;
+      UDItype pp_lh = ml * nh;
+      UDItype pp_hh = mh * nh;
+      UDItype res2 = 0;
+      UDItype res0 = 0;
+      UDItype ps_hh__ = pp_hl + pp_lh;
+      if (ps_hh__ < pp_hl)
+	res2 += 0x100000000LL;
+      pp_hl = (ps_hh__ << 32) & 0xffffffff00000000LL;
+      res0 = pp_ll + pp_hl;
+      if (res0 < pp_ll)
+	res2++;
+      res2 += ((ps_hh__ >> 32) & 0xffffffffL) + pp_hh;
+      high = res2;
+      low = res0;
+    }
+#endif
+  }
+
+  tmp->normal_exp = a->normal_exp + b->normal_exp;
+  tmp->sign = a->sign != b->sign;
+#ifdef FLOAT
+  tmp->normal_exp += 2;		/* ??????????????? */
+#else
+  tmp->normal_exp += 4;		/* ??????????????? */
+#endif
+  while (high >= IMPLICIT_2)
+    {
+      tmp->normal_exp++;
+      if (high & 1)
+	{
+	  low >>= 1;
+	  low |= FRACHIGH;
+	}
+      high >>= 1;
+    }
+  while (high < IMPLICIT_1)
+    {
+      tmp->normal_exp--;
+
+      high <<= 1;
+      if (low & FRACHIGH)
+	high |= 1;
+      low <<= 1;
+    }
+  /* rounding is tricky. if we only round if it won't make us round later. */
+#if 0
+  if (low & FRACHIGH2)
+    {
+      if (((high & GARDMASK) != GARDMSB)
+	  && (((high + 1) & GARDMASK) == GARDMSB))
+	{
+	  /* don't round, it gets done again later. */
+	}
+      else
+	{
+	  high++;
+	}
+    }
+#endif
+  if ((high & GARDMASK) == GARDMSB)
+    {
+      if (high & (1 << NGARDS))
+	{
+	  /* half way, so round to even */
+	  high += GARDROUND + 1;
+	}
+      else if (low)
+	{
+	  /* but we really weren't half way */
+	  high += GARDROUND + 1;
+	}
+    }
+  tmp->fraction.ll = high;
+  tmp->class = CLASS_NUMBER;
+  return tmp;
+}
+
+FLO_type
+multiply (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+  fp_number_type tmp;
+  fp_number_type *res;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  res = _fpmul_parts (&a, &b, &tmp);
+
+  return pack_d (res);
+}
+#endif
+
+#if defined(L_div_sf) || defined(L_div_df)
+static INLINE fp_number_type *
+_fpdiv_parts (fp_number_type * a,
+	      fp_number_type * b)
+{
+  fractype bit;
+  fractype numerator;
+  fractype denominator;
+  fractype quotient;
+
+  if (isnan (a))
+    {
+      return a;
+    }
+  if (isnan (b))
+    {
+      return b;
+    }
+
+  a->sign = a->sign ^ b->sign;
+
+  if (isinf (a) || iszero (a))
+    {
+      if (a->class == b->class)
+	return nan ();
+      return a;
+    }
+
+  if (isinf (b))
+    {
+      a->fraction.ll = 0;
+      a->normal_exp = 0;
+      return a;
+    }
+  if (iszero (b))
+    {
+      a->class = CLASS_INFINITY;
+      return a;
+    }
+
+  /* Calculate the mantissa by multiplying both 64bit numbers to get a
+     128 bit number */
+  {
+    /* quotient =
+       ( numerator / denominator) * 2^(numerator exponent -  denominator exponent)
+     */
+
+    a->normal_exp = a->normal_exp - b->normal_exp;
+    numerator = a->fraction.ll;
+    denominator = b->fraction.ll;
+
+    if (numerator < denominator)
+      {
+	/* Fraction will be less than 1.0 */
+	numerator *= 2;
+	a->normal_exp--;
+      }
+    bit = IMPLICIT_1;
+    quotient = 0;
+    /* ??? Does divide one bit at a time.  Optimize.  */
+    while (bit)
+      {
+	if (numerator >= denominator)
+	  {
+	    quotient |= bit;
+	    numerator -= denominator;
+	  }
+	bit >>= 1;
+	numerator *= 2;
+      }
+
+    if ((quotient & GARDMASK) == GARDMSB)
+      {
+	if (quotient & (1 << NGARDS))
+	  {
+	    /* half way, so round to even */
+	    quotient += GARDROUND + 1;
+	  }
+	else if (numerator)
+	  {
+	    /* but we really weren't half way, more bits exist */
+	    quotient += GARDROUND + 1;
+	  }
+      }
+
+    a->fraction.ll = quotient;
+    return (a);
+  }
+}
+
+FLO_type
+divide (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+  fp_number_type *res;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  res = _fpdiv_parts (&a, &b);
+
+  return pack_d (res);
+}
+#endif
+
+int __fpcmp_parts (fp_number_type * a, fp_number_type *b);
+
+#if defined(L_fpcmp_parts_sf) || defined(L_fpcmp_parts_df)
+/* according to the demo, fpcmp returns a comparison with 0... thus
+   a<b -> -1
+   a==b -> 0
+   a>b -> +1
+ */
+
+int
+__fpcmp_parts (fp_number_type * a, fp_number_type * b)
+{
+#if 0
+  /* either nan -> unordered. Must be checked outside of this routine. */
+  if (isnan (a) && isnan (b))
+    {
+      return 1;			/* still unordered! */
+    }
+#endif
+
+  if (isnan (a) || isnan (b))
+    {
+      return 1;			/* how to indicate unordered compare? */
+    }
+  if (isinf (a) && isinf (b))
+    {
+      /* +inf > -inf, but +inf != +inf */
+      /* b    \a| +inf(0)| -inf(1)
+       ______\+--------+--------
+       +inf(0)| a==b(0)| a<b(-1)
+       -------+--------+--------
+       -inf(1)| a>b(1) | a==b(0)
+       -------+--------+--------
+       So since unordered must be non zero, just line up the columns...
+       */
+      return b->sign - a->sign;
+    }
+  /* but not both... */
+  if (isinf (a))
+    {
+      return a->sign ? -1 : 1;
+    }
+  if (isinf (b))
+    {
+      return b->sign ? 1 : -1;
+    }
+  if (iszero (a) && iszero (b))
+    {
+      return 0;
+    }
+  if (iszero (a))
+    {
+      return b->sign ? 1 : -1;
+    }
+  if (iszero (b))
+    {
+      return a->sign ? -1 : 1;
+    }
+  /* now both are "normal". */
+  if (a->sign != b->sign)
+    {
+      /* opposite signs */
+      return a->sign ? -1 : 1;
+    }
+  /* same sign; exponents? */
+  if (a->normal_exp > b->normal_exp)
+    {
+      return a->sign ? -1 : 1;
+    }
+  if (a->normal_exp < b->normal_exp)
+    {
+      return a->sign ? 1 : -1;
+    }
+  /* same exponents; check size. */
+  if (a->fraction.ll > b->fraction.ll)
+    {
+      return a->sign ? -1 : 1;
+    }
+  if (a->fraction.ll < b->fraction.ll)
+    {
+      return a->sign ? 1 : -1;
+    }
+  /* after all that, they're equal. */
+  return 0;
+}
+#endif
+
+#if defined(L_compare_sf) || defined(L_compare_df)
+CMPtype
+compare (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  return __fpcmp_parts (&a, &b);
+}
+#endif
+
+#ifndef US_SOFTWARE_GOFAST
+
+/* These should be optimized for their specific tasks someday.  */
+
+#if defined(L_eq_sf) || defined(L_eq_df)
+CMPtype
+_eq_f2 (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  if (isnan (&a) || isnan (&b))
+    return 1;			/* false, truth == 0 */
+
+  return __fpcmp_parts (&a, &b) ;
+}
+#endif
+
+#if defined(L_ne_sf) || defined(L_ne_df)
+CMPtype
+_ne_f2 (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  if (isnan (&a) || isnan (&b))
+    return 1;			/* true, truth != 0 */
+
+  return  __fpcmp_parts (&a, &b) ;
+}
+#endif
+
+#if defined(L_gt_sf) || defined(L_gt_df)
+CMPtype
+_gt_f2 (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  if (isnan (&a) || isnan (&b))
+    return -1;			/* false, truth > 0 */
+
+  return __fpcmp_parts (&a, &b);
+}
+#endif
+
+#if defined(L_ge_sf) || defined(L_ge_df)
+CMPtype
+_ge_f2 (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  if (isnan (&a) || isnan (&b))
+    return -1;			/* false, truth >= 0 */
+  return __fpcmp_parts (&a, &b) ;
+}
+#endif
+
+#if defined(L_lt_sf) || defined(L_lt_df)
+CMPtype
+_lt_f2 (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  if (isnan (&a) || isnan (&b))
+    return 1;			/* false, truth < 0 */
+
+  return __fpcmp_parts (&a, &b);
+}
+#endif
+
+#if defined(L_le_sf) || defined(L_le_df)
+CMPtype
+_le_f2 (FLO_type arg_a, FLO_type arg_b)
+{
+  fp_number_type a;
+  fp_number_type b;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  unpack_d ((FLO_union_type *) & arg_b, &b);
+
+  if (isnan (&a) || isnan (&b))
+    return 1;			/* false, truth <= 0 */
+
+  return __fpcmp_parts (&a, &b) ;
+}
+#endif
+
+#endif /* ! US_SOFTWARE_GOFAST */
+
+#if defined(L_si_to_sf) || defined(L_si_to_df)
+FLO_type
+si_to_float (SItype arg_a)
+{
+  fp_number_type in;
+
+  in.class = CLASS_NUMBER;
+  in.sign = arg_a < 0;
+  if (!arg_a)
+    {
+      in.class = CLASS_ZERO;
+    }
+  else
+    {
+      in.normal_exp = FRACBITS + NGARDS;
+      if (in.sign) 
+	{
+	  /* Special case for minint, since there is no +ve integer
+	     representation for it */
+	  if (arg_a == (SItype) 0x80000000)
+	    {
+	      return -2147483648.0;
+	    }
+	  in.fraction.ll = (-arg_a);
+	}
+      else
+	in.fraction.ll = arg_a;
+
+      while (in.fraction.ll < (1LL << (FRACBITS + NGARDS)))
+	{
+	  in.fraction.ll <<= 1;
+	  in.normal_exp -= 1;
+	}
+    }
+  return pack_d (&in);
+}
+#endif
+
+#if defined(L_sf_to_si) || defined(L_df_to_si)
+SItype
+float_to_si (FLO_type arg_a)
+{
+  fp_number_type a;
+  SItype tmp;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  if (iszero (&a))
+    return 0;
+  if (isnan (&a))
+    return 0;
+  /* get reasonable MAX_SI_INT... */
+  if (isinf (&a))
+    return a.sign ? (-MAX_SI_INT)-1 : MAX_SI_INT;
+  /* it is a number, but a small one */
+  if (a.normal_exp < 0)
+    return 0;
+  if (a.normal_exp > 30)
+    return a.sign ? (-MAX_SI_INT)-1 : MAX_SI_INT;
+  tmp = a.fraction.ll >> ((FRACBITS + NGARDS) - a.normal_exp);
+  return a.sign ? (-tmp) : (tmp);
+}
+#endif
+
+#if defined(L_sf_to_usi) || defined(L_df_to_usi)
+#ifdef US_SOFTWARE_GOFAST
+/* While libgcc2.c defines its own __fixunssfsi and __fixunsdfsi routines,
+   we also define them for GOFAST because the ones in libgcc2.c have the
+   wrong names and I'd rather define these here and keep GOFAST CYG-LOC's
+   out of libgcc2.c.  We can't define these here if not GOFAST because then
+   there'd be duplicate copies.  */
+
+USItype
+float_to_usi (FLO_type arg_a)
+{
+  fp_number_type a;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  if (iszero (&a))
+    return 0;
+  if (isnan (&a))
+    return 0;
+  /* it is a negative number */
+  if (a.sign)
+    return 0;
+  /* get reasonable MAX_USI_INT... */
+  if (isinf (&a))
+    return MAX_USI_INT;
+  /* it is a number, but a small one */
+  if (a.normal_exp < 0)
+    return 0;
+  if (a.normal_exp > 31)
+    return MAX_USI_INT;
+  else if (a.normal_exp > (FRACBITS + NGARDS))
+    return a.fraction.ll << (a.normal_exp - (FRACBITS + NGARDS));
+  else
+    return a.fraction.ll >> ((FRACBITS + NGARDS) - a.normal_exp);
+}
+#endif
+#endif
+
+#if defined(L_negate_sf) || defined(L_negate_df)
+FLO_type
+negate (FLO_type arg_a)
+{
+  fp_number_type a;
+
+  unpack_d ((FLO_union_type *) & arg_a, &a);
+  flip_sign (&a);
+  return pack_d (&a);
+}
+#endif
+
+#ifdef FLOAT
+
+#if defined(L_make_sf)
+SFtype
+__make_fp(fp_class_type class,
+	     unsigned int sign,
+	     int exp, 
+	     USItype frac)
+{
+  fp_number_type in;
+
+  in.class = class;
+  in.sign = sign;
+  in.normal_exp = exp;
+  in.fraction.ll = frac;
+  return pack_d (&in);
+}
+#endif
+
+#ifndef FLOAT_ONLY
+
+/* This enables one to build an fp library that supports float but not double.
+   Otherwise, we would get an undefined reference to __make_dp.
+   This is needed for some 8-bit ports that can't handle well values that
+   are 8-bytes in size, so we just don't support double for them at all.  */
+
+extern DFtype __make_dp (fp_class_type, unsigned int, int, UDItype frac);
+
+#if defined(L_sf_to_df)
+DFtype
+sf_to_df (SFtype arg_a)
+{
+  fp_number_type in;
+
+  unpack_d ((FLO_union_type *) & arg_a, &in);
+  return __make_dp (in.class, in.sign, in.normal_exp,
+		    ((UDItype) in.fraction.ll) << F_D_BITOFF);
+}
+#endif
+
+#endif
+#endif
+
+#ifndef FLOAT
+
+extern SFtype __make_fp (fp_class_type, unsigned int, int, USItype);
+
+#if defined(L_make_df)
+DFtype
+__make_dp (fp_class_type class, unsigned int sign, int exp, UDItype frac)
+{
+  fp_number_type in;
+
+  in.class = class;
+  in.sign = sign;
+  in.normal_exp = exp;
+  in.fraction.ll = frac;
+  return pack_d (&in);
+}
+#endif
+
+#if defined(L_df_to_sf)
+SFtype
+df_to_sf (DFtype arg_a)
+{
+  fp_number_type in;
+  USItype sffrac;
+
+  unpack_d ((FLO_union_type *) & arg_a, &in);
+
+  sffrac = in.fraction.ll >> F_D_BITOFF;
+
+  /* We set the lowest guard bit in SFFRAC if we discarded any non
+     zero bits.  */
+  if ((in.fraction.ll & (((USItype) 1 << F_D_BITOFF) - 1)) != 0)
+    sffrac |= 1;
+
+  return __make_fp (in.class, in.sign, in.normal_exp, sffrac);
+}
+#endif
+
+#endif
+#endif /* !EXTENDED_FLOAT_STUBS */
diff --git a/libgcc/lib1thumb.asm b/libgcc/lib1thumb.asm
new file mode 100755
index 0000000..e0ff746
--- /dev/null
+++ b/libgcc/lib1thumb.asm
@@ -0,0 +1,736 @@
+@ libgcc1 routines for ARM cpu.
+@ Division routines, written by Richard Earnshaw, (rearnsha@armltd.co.uk)
+
+/* Copyright (C) 1995, 1996, 1998 Free Software Foundation, Inc.
+
+This file 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 2, or (at your option) any
+later version.
+
+In addition to the permissions in the GNU General Public License, the
+Free Software Foundation gives you unlimited permission to link the
+compiled version of this file with other programs, and to distribute
+those programs without any restriction coming from the use of this
+file.  (The General Public License restrictions do apply in other
+respects; for example, they cover modification of the file, and
+distribution when not linked into another program.)
+
+This file 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 this program; see the file COPYING.  If not, write to
+the Free Software Foundation, 59 Temple Place - Suite 330,
+Boston, MA 02111-1307, USA.  */
+
+/* As a special exception, if you link this library with other files,
+   some of which are compiled with GCC, to produce an executable,
+   this library does not by itself cause the resulting executable
+   to be covered by the GNU General Public License.
+   This exception does not however invalidate any other reasons why
+   the executable file might be covered by the GNU General Public License.  */
+
+	.code	 16
+
+#ifdef __elf__
+#define __PLT__ (PLT)
+#define TYPE(x) .type SYM(x),function
+#define SIZE(x) .size SYM(x), . - SYM(x)
+#else
+#define __PLT__
+#define TYPE(x)
+#define SIZE(x)
+#endif
+
+#define RET	mov	pc, lr
+
+#define SYM(x) x
+
+work		.req	r4	@ XXXX is this safe ?
+
+#ifdef L_udivsi3
+
+dividend	.req	r0
+divisor		.req	r1
+result		.req	r2
+curbit		.req	r3
+ip		.req	r12
+sp		.req	r13
+lr		.req	r14
+pc		.req	r15
+	
+	.text
+	.globl	SYM (__udivsi3)
+	TYPE 	(__udivsi3)
+	.align	0
+	.thumb_func
+SYM (__udivsi3):
+	cmp	divisor, #0
+	beq	Ldiv0
+	mov	curbit, #1
+	mov	result, #0
+	
+	push	{ work }
+	cmp	dividend, divisor
+	bcc	Lgot_result
+
+	@ Load the constant 0x10000000 into our work register
+	mov	work, #1
+	lsl	work, #28
+Loop1:
+	@ Unless the divisor is very big, shift it up in multiples of
+	@ four bits, since this is the amount of unwinding in the main
+	@ division loop.  Continue shifting until the divisor is 
+	@ larger than the dividend.
+	cmp	divisor, work
+	bcs     Lbignum
+	cmp	divisor, dividend
+	bcs     Lbignum
+	lsl	divisor, #4
+	lsl	curbit,  #4
+	b	Loop1
+
+Lbignum:
+	@ Set work to 0x80000000
+	lsl	work, #3
+Loop2:		
+	@ For very big divisors, we must shift it a bit at a time, or
+	@ we will be in danger of overflowing.
+	cmp	divisor, work
+	bcs	Loop3
+	cmp	divisor, dividend
+	bcs	Loop3
+	lsl	divisor, #1
+	lsl	curbit,  #1
+	b	Loop2
+
+Loop3:
+	@ Test for possible subtractions, and note which bits
+	@ are done in the result.  On the final pass, this may subtract
+	@ too much from the dividend, but the result will be ok, since the
+	@ "bit" will have been shifted out at the bottom.
+	cmp	dividend, divisor
+	bcc     Over1
+	sub	dividend, dividend, divisor
+	orr	result, result, curbit
+Over1:	
+	lsr	work, divisor, #1
+	cmp	dividend, work
+	bcc	Over2
+	sub	dividend, dividend, work
+	lsr	work, curbit, #1
+	orr	result, work
+Over2:	
+	lsr	work, divisor, #2
+	cmp	dividend, work
+	bcc	Over3
+	sub	dividend, dividend, work
+	lsr	work, curbit, #2
+	orr	result, work
+Over3:	
+	lsr	work, divisor, #3
+	cmp	dividend, work
+	bcc	Over4
+	sub	dividend, dividend, work
+	lsr	work, curbit, #3
+	orr	result, work
+Over4:	
+	cmp	dividend, #0			@ Early termination?
+	beq	Lgot_result
+	lsr	curbit,  #4			@ No, any more bits to do?
+	beq	Lgot_result
+	lsr	divisor, #4
+	b	Loop3
+Lgot_result:
+	mov	r0, result
+	pop	{ work }
+	RET
+
+Ldiv0:
+	push	{ lr }
+	bl	SYM (__div0) __PLT__
+	mov	r0, #0			@ about as wrong as it could be
+	pop	{ pc }
+
+	SIZE	(__udivsi3)
+	
+#endif /* L_udivsi3 */
+
+#ifdef L_umodsi3
+
+dividend	.req	r0
+divisor		.req	r1
+overdone	.req	r2
+curbit		.req	r3
+ip		.req	r12
+sp		.req	r13
+lr		.req	r14
+pc		.req	r15
+	
+	.text
+	.globl	SYM (__umodsi3)
+	TYPE	(__umodsi3)
+	.align	0
+	.thumb_func
+SYM (__umodsi3):
+	cmp	divisor, #0
+	beq	Ldiv0
+	mov	curbit, #1
+	cmp	dividend, divisor
+	bcs	Over1
+	RET	
+
+Over1:	
+	@ Load the constant 0x10000000 into our work register
+	push	{ work }
+	mov	work, #1
+	lsl	work, #28
+Loop1:
+	@ Unless the divisor is very big, shift it up in multiples of
+	@ four bits, since this is the amount of unwinding in the main
+	@ division loop.  Continue shifting until the divisor is 
+	@ larger than the dividend.
+	cmp	divisor, work
+	bcs	Lbignum
+	cmp	divisor, dividend
+	bcs	Lbignum
+	lsl	divisor, #4
+	lsl	curbit, #4
+	b	Loop1
+
+Lbignum:
+	@ Set work to 0x80000000
+	lsl	work, #3
+Loop2:
+	@ For very big divisors, we must shift it a bit at a time, or
+	@ we will be in danger of overflowing.
+	cmp	divisor, work
+	bcs	Loop3
+	cmp	divisor, dividend
+	bcs	Loop3
+	lsl	divisor, #1
+	lsl	curbit, #1
+	b	Loop2
+
+Loop3:
+	@ Test for possible subtractions.  On the final pass, this may 
+	@ subtract too much from the dividend, so keep track of which
+	@ subtractions are done, we can fix them up afterwards...
+	mov	overdone, #0
+	cmp	dividend, divisor
+	bcc	Over2
+	sub	dividend, dividend, divisor
+Over2:
+	lsr	work, divisor, #1
+	cmp	dividend, work
+	bcc	Over3
+	sub	dividend, dividend, work
+	mov	ip, curbit
+	mov	work, #1
+	ror	curbit, work
+	orr	overdone, curbit
+	mov	curbit, ip
+Over3:
+	lsr	work, divisor, #2
+	cmp	dividend, work
+	bcc	Over4
+	sub	dividend, dividend, work
+	mov	ip, curbit
+	mov	work, #2
+	ror	curbit, work
+	orr	overdone, curbit
+	mov	curbit, ip
+Over4:
+	lsr	work, divisor, #3
+	cmp	dividend, work
+	bcc	Over5
+	sub	dividend, dividend, work
+	mov	ip, curbit
+	mov	work, #3
+	ror	curbit, work
+	orr	overdone, curbit
+	mov	curbit, ip
+Over5:
+	mov	ip, curbit
+	cmp	dividend, #0			@ Early termination?
+	beq	Over6
+	lsr	curbit, #4			@ No, any more bits to do?
+	beq	Over6
+	lsr	divisor, #4
+	b	Loop3
+
+Over6:	
+	@ Any subtractions that we should not have done will be recorded in
+	@ the top three bits of "overdone".  Exactly which were not needed
+	@ are governed by the position of the bit, stored in ip.
+	@ If we terminated early, because dividend became zero,
+	@ then none of the below will match, since the bit in ip will not be
+	@ in the bottom nibble.
+
+	mov	work, #0xe
+	lsl	work, #28	
+	and	overdone, work
+	bne	Over7
+	pop	{ work }
+	RET					@ No fixups needed
+Over7:
+	mov	curbit, ip
+	mov	work, #3
+	ror	curbit, work
+	tst	overdone, curbit
+	beq	Over8
+	lsr	work, divisor, #3
+	add	dividend, dividend, work
+Over8:
+	mov	curbit, ip
+	mov	work, #2
+	ror	curbit, work
+	tst	overdone, curbit
+	beq	Over9
+	lsr	work, divisor, #2
+	add	dividend, dividend, work
+Over9:
+	mov	curbit, ip
+	mov	work, #1
+	ror	curbit, work
+	tst	overdone, curbit
+	beq	Over10
+	lsr	work, divisor, #1
+	add	dividend, dividend, work
+Over10:
+	pop	{ work }
+	RET	
+
+Ldiv0:
+	push	{ lr }
+	bl	SYM (__div0) __PLT__
+	mov	r0, #0			@ about as wrong as it could be
+	pop	{ pc }
+
+	SIZE	(__umodsi3)
+	
+#endif /* L_umodsi3 */
+
+#ifdef L_divsi3
+
+dividend	.req	r0
+divisor		.req	r1
+result		.req	r2
+curbit		.req	r3
+ip		.req	r12
+sp		.req	r13
+lr		.req	r14
+pc		.req	r15
+	
+	.text
+	.globl	SYM (__divsi3)
+	TYPE	(__divsi3)
+	.align	0
+	.thumb_func
+SYM (__divsi3):
+	cmp	divisor, #0
+	beq	Ldiv0
+	
+	push	{ work }
+	mov	work, dividend
+	eor	work, divisor		@ Save the sign of the result.
+	mov	ip, work
+	mov	curbit, #1
+	mov	result, #0
+	cmp	divisor, #0
+	bpl	Over1
+	neg	divisor, divisor	@ Loops below use unsigned.
+Over1:	
+	cmp	dividend, #0
+	bpl	Over2
+	neg	dividend, dividend
+Over2:	
+	cmp	dividend, divisor
+	bcc	Lgot_result
+
+	mov	work, #1
+	lsl	work, #28
+Loop1:
+	@ Unless the divisor is very big, shift it up in multiples of
+	@ four bits, since this is the amount of unwinding in the main
+	@ division loop.  Continue shifting until the divisor is 
+	@ larger than the dividend.
+	cmp	divisor, work
+	Bcs	Lbignum
+	cmp	divisor, dividend
+	Bcs	Lbignum
+	lsl	divisor, #4
+	lsl	curbit, #4
+	b	Loop1
+
+Lbignum:
+	@ For very big divisors, we must shift it a bit at a time, or
+	@ we will be in danger of overflowing.
+	lsl	work, #3
+Loop2:		
+	cmp	divisor, work
+	Bcs	Loop3
+	cmp	divisor, dividend
+	Bcs	Loop3
+	lsl	divisor, #1
+	lsl	curbit, #1
+	b	Loop2
+
+Loop3:
+	@ Test for possible subtractions, and note which bits
+	@ are done in the result.  On the final pass, this may subtract
+	@ too much from the dividend, but the result will be ok, since the
+	@ "bit" will have been shifted out at the bottom.
+	cmp	dividend, divisor
+	Bcc	Over3
+	sub	dividend, dividend, divisor
+	orr	result, result, curbit
+Over3:
+	lsr	work, divisor, #1
+	cmp	dividend, work
+	Bcc	Over4
+	sub	dividend, dividend, work
+	lsr	work, curbit, #1
+	orr	result, work
+Over4:	
+	lsr	work, divisor, #2
+	cmp	dividend, work
+	Bcc	Over5
+	sub	dividend, dividend, work
+	lsr	work, curbit, #2
+	orr	result, result, work
+Over5:	
+	lsr	work, divisor, #3
+	cmp	dividend, work
+	Bcc	Over6
+	sub	dividend, dividend, work
+	lsr	work, curbit, #3
+	orr	result, result, work
+Over6:	
+	cmp	dividend, #0			@ Early termination?
+	Beq	Lgot_result
+	lsr	curbit, #4			@ No, any more bits to do?
+	Beq	Lgot_result
+	lsr	divisor, #4
+	b	Loop3
+	
+Lgot_result:
+	mov	r0, result
+	mov	work, ip
+	cmp	work, #0
+	Bpl	Over7
+	neg	r0, r0
+Over7:
+	pop	{ work }
+	RET	
+
+Ldiv0:
+	push	{ lr }
+	bl	SYM (__div0) __PLT__
+	mov	r0, #0			@ about as wrong as it could be
+	pop	{ pc }
+
+	SIZE	(__divsi3)
+	
+#endif /* L_divsi3 */
+
+#ifdef L_modsi3
+
+dividend	.req	r0
+divisor		.req	r1
+overdone	.req	r2
+curbit		.req	r3
+ip		.req	r12
+sp		.req	r13
+lr		.req	r14
+pc		.req	r15
+	
+	.text
+	.globl	SYM (__modsi3)
+	TYPE	(__modsi3)
+	.align	0
+	.thumb_func
+SYM (__modsi3):
+	mov	curbit, #1
+	cmp	divisor, #0
+	beq	Ldiv0
+	Bpl	Over1
+	neg	divisor, divisor		@ Loops below use unsigned.
+Over1:	
+	push	{ work }
+	@ Need to save the sign of the dividend, unfortunately, we need
+	@ ip later on.  Must do this after saving the original value of
+	@ the work register, because we will pop this value off first.
+	push	{ dividend }
+	cmp	dividend, #0
+	Bpl	Over2
+	neg	dividend, dividend
+Over2:	
+	cmp	dividend, divisor
+	bcc	Lgot_result
+	mov	work, #1
+	lsl	work, #28
+Loop1:
+	@ Unless the divisor is very big, shift it up in multiples of
+	@ four bits, since this is the amount of unwinding in the main
+	@ division loop.  Continue shifting until the divisor is 
+	@ larger than the dividend.
+	cmp	divisor, work
+	bcs	Lbignum
+	cmp	divisor, dividend
+	bcs	Lbignum
+	lsl	divisor, #4
+	lsl	curbit, #4
+	b	Loop1
+
+Lbignum:
+	@ Set work to 0x80000000
+	lsl	work, #3
+Loop2:
+	@ For very big divisors, we must shift it a bit at a time, or
+	@ we will be in danger of overflowing.
+	cmp	divisor, work
+	bcs	Loop3
+	cmp	divisor, dividend
+	bcs	Loop3
+	lsl	divisor, #1
+	lsl	curbit, #1
+	b	Loop2
+
+Loop3:
+	@ Test for possible subtractions.  On the final pass, this may 
+	@ subtract too much from the dividend, so keep track of which
+	@ subtractions are done, we can fix them up afterwards...
+	mov	overdone, #0
+	cmp	dividend, divisor
+	bcc	Over3
+	sub	dividend, dividend, divisor
+Over3:
+	lsr	work, divisor, #1
+	cmp	dividend, work
+	bcc	Over4
+	sub	dividend, dividend, work
+	mov	ip, curbit
+	mov	work, #1
+	ror	curbit, work
+	orr	overdone, curbit
+	mov	curbit, ip
+Over4:
+	lsr	work, divisor, #2
+	cmp	dividend, work
+	bcc	Over5
+	sub	dividend, dividend, work
+	mov	ip, curbit
+	mov	work, #2
+	ror	curbit, work
+	orr	overdone, curbit
+	mov	curbit, ip
+Over5:
+	lsr	work, divisor, #3
+	cmp	dividend, work
+	bcc	Over6
+	sub	dividend, dividend, work
+	mov	ip, curbit
+	mov	work, #3
+	ror	curbit, work
+	orr	overdone, curbit
+	mov	curbit, ip
+Over6:
+	mov	ip, curbit
+	cmp	dividend, #0			@ Early termination?
+	beq	Over7
+	lsr	curbit, #4			@ No, any more bits to do?
+	beq	Over7
+	lsr	divisor, #4
+	b	Loop3
+
+Over7:	
+	@ Any subtractions that we should not have done will be recorded in
+	@ the top three bits of "overdone".  Exactly which were not needed
+	@ are governed by the position of the bit, stored in ip.
+	@ If we terminated early, because dividend became zero,
+	@ then none of the below will match, since the bit in ip will not be
+	@ in the bottom nibble.
+	mov	work, #0xe
+	lsl	work, #28
+	and	overdone, work
+	beq	Lgot_result
+	
+	mov	curbit, ip
+	mov	work, #3
+	ror	curbit, work
+	tst	overdone, curbit
+	beq	Over8
+	lsr	work, divisor, #3
+	add	dividend, dividend, work
+Over8:
+	mov	curbit, ip
+	mov	work, #2
+	ror	curbit, work
+	tst	overdone, curbit
+	beq	Over9
+	lsr	work, divisor, #2
+	add	dividend, dividend, work
+Over9:
+	mov	curbit, ip
+	mov	work, #1
+	ror	curbit, work
+	tst	overdone, curbit
+	beq	Lgot_result
+	lsr	work, divisor, #1
+	add	dividend, dividend, work
+Lgot_result:
+	pop	{ work }
+	cmp	work, #0
+	bpl	Over10
+	neg	dividend, dividend
+Over10:
+	pop	{ work }
+	RET	
+
+Ldiv0:
+	push    { lr }
+	bl	SYM (__div0) __PLT__
+	mov	r0, #0			@ about as wrong as it could be
+	pop	{ pc }
+	
+	SIZE	(__modsi3)
+		
+#endif /* L_modsi3 */
+
+#ifdef L_dvmd_tls
+
+	.globl	SYM (__div0)
+	TYPE	(__div0)
+	.align	0
+	.thumb_func
+SYM (__div0):
+	RET	
+
+	SIZE	(__div0)
+	
+#endif /* L_divmodsi_tools */
+
+	
+#ifdef L_call_via_rX
+
+/* These labels & instructions are used by the Arm/Thumb interworking code. 
+   The address of function to be called is loaded into a register and then 
+   one of these labels is called via a BL instruction.  This puts the 
+   return address into the link register with the bottom bit set, and the 
+   code here switches to the correct mode before executing the function.  */
+	
+	.text
+	.align 0
+
+.macro call_via register
+	.globl	SYM (_call_via_\register)
+	TYPE	(_call_via_\register)
+	.thumb_func
+SYM (_call_via_\register):
+	bx	\register
+	nop
+	
+	SIZE	(_call_via_\register)
+.endm
+
+	call_via r0
+	call_via r1
+	call_via r2
+	call_via r3
+	call_via r4
+	call_via r5
+	call_via r6
+	call_via r7
+	call_via r8
+	call_via r9
+	call_via sl
+	call_via fp
+	call_via ip
+	call_via sp
+	call_via lr
+
+#endif /* L_call_via_rX */
+
+#ifdef L_interwork_call_via_rX
+
+/* These labels & instructions are used by the Arm/Thumb interworking code,
+   when the target address is in an unknown instruction set.  The address 
+   of function to be called is loaded into a register and then one of these
+   labels is called via a BL instruction.  This puts the return address 
+   into the link register with the bottom bit set, and the code here 
+   switches to the correct mode before executing the function.  Unfortunately
+   the target code cannot be relied upon to return via a BX instruction, so
+   instead we have to store the resturn address on the stack and allow the
+   called function to return here instead.  Upon return we recover the real
+   return address and use a BX to get back to Thumb mode.  */
+	
+	.text
+	.align 0
+
+	.code 32
+	.globl _arm_return
+_arm_return:		
+	ldmia 	r13!, {r12}
+	bx 	r12
+	
+.macro interwork register					
+	.code 16
+	
+	.globl	SYM (_interwork_call_via_\register)
+	TYPE	(_interwork_call_via_\register)
+	.thumb_func
+SYM (_interwork_call_via_\register):
+	bx 	pc
+	nop
+	
+	.code 32
+	.globl .Lchange_\register
+.Lchange_\register:
+	tst	\register, #1
+	stmeqdb	r13!, {lr}
+	adreq	lr, _arm_return
+	bx	\register
+
+	SIZE	(_interwork_call_via_\register)
+.endm
+	
+	interwork r0
+	interwork r1
+	interwork r2
+	interwork r3
+	interwork r4
+	interwork r5
+	interwork r6
+	interwork r7
+	interwork r8
+	interwork r9
+	interwork sl
+	interwork fp
+	interwork ip
+	interwork sp
+
+	/* The lr case has to be handled a little differently...*/
+	.code 16
+	.globl	SYM (_interwork_call_via_lr)
+	TYPE	(_interwork_call_via_lr)
+	.thumb_func
+SYM (_interwork_call_via_lr):
+	bx 	pc
+	nop
+	
+	.code 32
+	.globl .Lchange_lr
+.Lchange_lr:
+	tst	lr, #1
+	stmeqdb	r13!, {lr}
+	mov	ip, lr
+	adreq	lr, _arm_return
+	bx	ip
+
+	SIZE	(_interwork_call_via_lr)
+	
+#endif /* L_interwork_call_via_rX */
diff --git a/libgcc/libgcc1-test.c b/libgcc/libgcc1-test.c
new file mode 100755
index 0000000..0f59cbe
--- /dev/null
+++ b/libgcc/libgcc1-test.c
@@ -0,0 +1,117 @@
+/* This small function uses all the arithmetic operators that
+   libgcc1.c can handle.  If you can link it, then
+   you have provided replacements for all the libgcc1.c functions that
+   your target machine needs.  */
+
+int foo ();
+double dfoo ();
+
+/* We don't want __main here because that can drag in atexit (among other
+   things) which won't necessarily exist yet.  */
+
+main_without__main ()
+{
+  int a = foo (), b = foo ();
+  unsigned int au = foo (), bu = foo ();
+  float af = dfoo (), bf = dfoo ();
+  double ad = dfoo (), bd = dfoo ();
+
+  discard (a * b);
+  discard (a / b);
+  discard (a % b);
+
+  discard (au / bu);
+  discard (au % bu);
+
+  discard (a >> b);
+  discard (a << b);
+
+  discard (au >> bu);
+  discard (au << bu);
+
+  ddiscard (ad + bd);
+  ddiscard (ad - bd);
+  ddiscard (ad * bd);
+  ddiscard (ad / bd);
+  ddiscard (-ad);
+
+  ddiscard (af + bf);
+  ddiscard (af - bf);
+  ddiscard (af * bf);
+  ddiscard (af / bf);
+  ddiscard (-af);
+
+  discard ((int) ad);
+  discard ((int) af);
+
+  ddiscard ((double) a);
+  ddiscard ((float) a);
+  ddiscard ((float) ad);
+
+  discard (ad == bd);
+  discard (ad < bd);
+  discard (ad > bd);
+  discard (ad != bd);
+  discard (ad <= bd);
+  discard (ad >= bd);
+
+  discard (af == bf);
+  discard (af < bf);
+  discard (af > bf);
+  discard (af != bf);
+  discard (af <= bf);
+  discard (af >= bf);
+
+  return 0;
+}
+
+discard (x)
+     int x;
+{}
+
+ddiscard (x)
+     double x;
+{}
+
+foo ()
+{
+  static int table[] = {20, 69, 4, 12};
+  static int idx;
+
+  return table[idx++];
+}
+
+double
+dfoo ()
+{
+  static double table[] = {20.4, 69.96, 4.4, 202.202};
+  static int idx;
+
+  return table[idx++];
+}
+
+/* Provide functions that some versions of the linker use to default
+   the start address if -e symbol is not used, to avoid the warning
+   message saying the start address is defaulted.  */
+extern void start() __asm__("start");
+extern void _start() __asm__("_start");
+extern void __start() __asm__("__start");
+
+/* Provide functions that might be needed by soft-float emulation routines.  */
+void memcpy() {}
+
+void start() {}
+void _start() {}
+void __start() {}
+void mainCRTStartup() {}
+
+/* CYGNUS LOCAL - duplicate definition of memcpy() removed.  */
+
+/* CYGNUS LOCAL v850 */
+#if defined __v850e__ || defined __v850ea__
+/* We need to use the symbol __ctbp in order to force the linker to define it.  */
+extern int _ctbp;
+
+void _func() { _ctbp = 1; }
+#endif
+/* END CYGNUS LOCAL */
diff --git a/libgcc/libgcc1.c b/libgcc/libgcc1.c
new file mode 100755
index 0000000..bece500
--- /dev/null
+++ b/libgcc/libgcc1.c
@@ -0,0 +1,596 @@
+/* Subroutines needed by GCC output code on some machines.  */
+/* Compile this file with the Unix C compiler!  */
+/* Copyright (C) 1987, 1988, 1992, 1994, 1995 Free Software Foundation, Inc.
+
+This file 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 2, or (at your option) any
+later version.
+
+In addition to the permissions in the GNU General Public License, the
+Free Software Foundation gives you unlimited permission to link the
+compiled version of this file with other programs, and to distribute
+those programs without any restriction coming from the use of this
+file.  (The General Public License restrictions do apply in other
+respects; for example, they cover modification of the file, and
+distribution when not linked into another program.)
+
+This file 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 this program; see the file COPYING.  If not, write to
+the Free Software Foundation, 59 Temple Place - Suite 330,
+Boston, MA 02111-1307, USA.  */
+
+/* As a special exception, if you link this library with other files,
+   some of which are compiled with GCC, to produce an executable,
+   this library does not by itself cause the resulting executable
+   to be covered by the GNU General Public License.
+   This exception does not however invalidate any other reasons why
+   the executable file might be covered by the GNU General Public License.  */
+
+#include "config.h"
+
+/* Don't use `fancy_abort' here even if config.h says to use it.  */
+#ifdef abort
+#undef abort
+#endif
+
+/* On some machines, cc is really GCC.  For these machines, we can't
+   expect these functions to be properly compiled unless GCC open codes
+   the operation (which is precisely when the function won't be used).
+   So allow tm.h to specify ways of accomplishing the operations
+   by defining the macros perform_*.
+
+   On a machine where cc is some other compiler, there is usually no
+   reason to define perform_*.  The other compiler normally has other ways
+   of implementing all of these operations.
+
+   In some cases a certain machine may come with GCC installed as cc
+   or may have some other compiler.  Then it may make sense for tm.h
+   to define perform_* only if __GNUC__ is defined.  */
+
+#ifndef perform_mulsi3
+#define perform_mulsi3(a, b) return a * b
+#endif
+
+#ifndef perform_divsi3
+#define perform_divsi3(a, b) return a / b
+#endif
+
+#ifndef perform_udivsi3
+#define perform_udivsi3(a, b) return a / b
+#endif
+
+#ifndef perform_modsi3
+#define perform_modsi3(a, b) return a % b
+#endif
+
+#ifndef perform_umodsi3
+#define perform_umodsi3(a, b) return a % b
+#endif
+
+#ifndef perform_lshrsi3
+#define perform_lshrsi3(a, b) return a >> b
+#endif
+
+#ifndef perform_ashrsi3
+#define perform_ashrsi3(a, b) return a >> b
+#endif
+
+#ifndef perform_ashlsi3
+#define perform_ashlsi3(a, b) return a << b
+#endif
+
+#ifndef perform_adddf3
+#define perform_adddf3(a, b) return a + b
+#endif
+
+#ifndef perform_subdf3
+#define perform_subdf3(a, b) return a - b
+#endif
+
+#ifndef perform_muldf3
+#define perform_muldf3(a, b) return a * b
+#endif
+
+#ifndef perform_divdf3
+#define perform_divdf3(a, b) return a / b
+#endif
+
+#ifndef perform_addsf3
+#define perform_addsf3(a, b) return INTIFY (a + b)
+#endif
+
+#ifndef perform_subsf3
+#define perform_subsf3(a, b) return INTIFY (a - b)
+#endif
+
+#ifndef perform_mulsf3
+#define perform_mulsf3(a, b) return INTIFY (a * b)
+#endif
+
+#ifndef perform_divsf3
+#define perform_divsf3(a, b) return INTIFY (a / b)
+#endif
+
+#ifndef perform_negdf2
+#define perform_negdf2(a) return -a
+#endif
+
+#ifndef perform_negsf2
+#define perform_negsf2(a) return INTIFY (-a)
+#endif
+
+#ifndef perform_fixdfsi
+#define perform_fixdfsi(a) return (nongcc_SI_type) a;
+#endif
+
+#ifndef perform_fixsfsi
+#define perform_fixsfsi(a) return (nongcc_SI_type) a
+#endif
+
+#ifndef perform_floatsidf
+#define perform_floatsidf(a) return (double) a
+#endif
+
+#ifndef perform_floatsisf
+#define perform_floatsisf(a)  return INTIFY ((float) a)
+#endif
+
+#ifndef perform_extendsfdf2
+#define perform_extendsfdf2(a)  return a
+#endif
+
+#ifndef perform_truncdfsf2
+#define perform_truncdfsf2(a)  return INTIFY (a)
+#endif
+
+/* Note that eqdf2 returns a value for "true" that is == 0,
+   nedf2 returns a value for "true" that is != 0,
+   gtdf2 returns a value for "true" that is > 0,
+   and so on.  */
+
+#ifndef perform_eqdf2
+#define perform_eqdf2(a, b) return !(a == b)
+#endif
+
+#ifndef perform_nedf2
+#define perform_nedf2(a, b) return a != b
+#endif
+
+#ifndef perform_gtdf2
+#define perform_gtdf2(a, b) return a > b
+#endif
+
+#ifndef perform_gedf2
+#define perform_gedf2(a, b) return (a >= b) - 1
+#endif
+
+#ifndef perform_ltdf2
+#define perform_ltdf2(a, b) return -(a < b)
+#endif
+
+#ifndef perform_ledf2
+#define perform_ledf2(a, b) return 1 - (a <= b)
+#endif
+
+#ifndef perform_eqsf2
+#define perform_eqsf2(a, b) return !(a == b)
+#endif
+
+#ifndef perform_nesf2
+#define perform_nesf2(a, b) return a != b
+#endif
+
+#ifndef perform_gtsf2
+#define perform_gtsf2(a, b) return a > b
+#endif
+
+#ifndef perform_gesf2
+#define perform_gesf2(a, b) return (a >= b) - 1
+#endif
+
+#ifndef perform_ltsf2
+#define perform_ltsf2(a, b) return -(a < b)
+#endif
+
+#ifndef perform_lesf2
+#define perform_lesf2(a, b) return 1 - (a <= b);
+#endif
+
+/* Define the C data type to use for an SImode value.  */
+
+#ifndef nongcc_SI_type
+#define nongcc_SI_type long int
+#endif
+
+/* Define the C data type to use for a value of word size */
+#ifndef nongcc_word_type
+#define nongcc_word_type nongcc_SI_type
+#endif
+
+/* Define the type to be used for returning an SF mode value
+   and the method for turning a float into that type.
+   These definitions work for machines where an SF value is
+   returned in the same register as an int.  */
+
+#ifndef FLOAT_VALUE_TYPE  
+#define FLOAT_VALUE_TYPE int
+#endif
+
+#ifndef INTIFY
+#define INTIFY(FLOATVAL)  (intify.f = (FLOATVAL), intify.i)
+#endif
+
+#ifndef FLOATIFY
+#define FLOATIFY(INTVAL)  ((INTVAL).f)
+#endif
+
+#ifndef FLOAT_ARG_TYPE
+#define FLOAT_ARG_TYPE union flt_or_int
+#endif
+
+union flt_or_value { FLOAT_VALUE_TYPE i; float f; };
+
+union flt_or_int { int i; float f; };
+
+
+#ifdef L_mulsi3
+nongcc_SI_type
+__mulsi3 (a, b)
+     nongcc_SI_type a, b;
+{
+  perform_mulsi3 (a, b);
+}
+#endif
+
+#ifdef L_udivsi3
+nongcc_SI_type
+__udivsi3 (a, b)
+     unsigned nongcc_SI_type a, b;
+{
+  perform_udivsi3 (a, b);
+}
+#endif
+
+#ifdef L_divsi3
+nongcc_SI_type
+__divsi3 (a, b)
+     nongcc_SI_type a, b;
+{
+  perform_divsi3 (a, b);
+}
+#endif
+
+#ifdef L_umodsi3
+nongcc_SI_type
+__umodsi3 (a, b)
+     unsigned nongcc_SI_type a, b;
+{
+  perform_umodsi3 (a, b);
+}
+#endif
+
+#ifdef L_modsi3
+nongcc_SI_type
+__modsi3 (a, b)
+     nongcc_SI_type a, b;
+{
+  perform_modsi3 (a, b);
+}
+#endif
+
+#ifdef L_lshrsi3
+nongcc_SI_type
+__lshrsi3 (a, b)
+     unsigned nongcc_SI_type a, b;
+{
+  perform_lshrsi3 (a, b);
+}
+#endif
+
+#ifdef L_ashrsi3
+nongcc_SI_type
+__ashrsi3 (a, b)
+     nongcc_SI_type a, b;
+{
+  perform_ashrsi3 (a, b);
+}
+#endif
+
+#ifdef L_ashlsi3
+nongcc_SI_type
+__ashlsi3 (a, b)
+     nongcc_SI_type a, b;
+{
+  perform_ashlsi3 (a, b);
+}
+#endif
+
+#ifdef L_divdf3
+double
+__divdf3 (a, b)
+     double a, b;
+{
+  perform_divdf3 (a, b);
+}
+#endif
+
+#ifdef L_muldf3
+double
+__muldf3 (a, b)
+     double a, b;
+{
+  perform_muldf3 (a, b);
+}
+#endif
+
+#ifdef L_negdf2
+double
+__negdf2 (a)
+     double a;
+{
+  perform_negdf2 (a);
+}
+#endif
+
+#ifdef L_adddf3
+double
+__adddf3 (a, b)
+     double a, b;
+{
+  perform_adddf3 (a, b);
+}
+#endif
+
+#ifdef L_subdf3
+double
+__subdf3 (a, b)
+     double a, b;
+{
+  perform_subdf3 (a, b);
+}
+#endif
+
+/* Note that eqdf2 returns a value for "true" that is == 0,
+   nedf2 returns a value for "true" that is != 0,
+   gtdf2 returns a value for "true" that is > 0,
+   and so on.  */
+
+#ifdef L_eqdf2
+nongcc_word_type
+__eqdf2 (a, b)
+     double a, b;
+{
+  /* Value == 0 iff a == b.  */
+  perform_eqdf2 (a, b);
+}
+#endif
+
+#ifdef L_nedf2
+nongcc_word_type
+__nedf2 (a, b)
+     double a, b;
+{
+  /* Value != 0 iff a != b.  */
+  perform_nedf2 (a, b);
+}
+#endif
+
+#ifdef L_gtdf2
+nongcc_word_type
+__gtdf2 (a, b)
+     double a, b;
+{
+  /* Value > 0 iff a > b.  */
+  perform_gtdf2 (a, b);
+}
+#endif
+
+#ifdef L_gedf2
+nongcc_word_type
+__gedf2 (a, b)
+     double a, b;
+{
+  /* Value >= 0 iff a >= b.  */
+  perform_gedf2 (a, b);
+}
+#endif
+
+#ifdef L_ltdf2
+nongcc_word_type
+__ltdf2 (a, b)
+     double a, b;
+{
+  /* Value < 0 iff a < b.  */
+  perform_ltdf2 (a, b);
+}
+#endif
+
+#ifdef L_ledf2
+nongcc_word_type
+__ledf2 (a, b)
+     double a, b;
+{
+  /* Value <= 0 iff a <= b.  */
+  perform_ledf2 (a, b);
+}
+#endif
+
+#ifdef L_fixdfsi
+nongcc_SI_type
+__fixdfsi (a)
+     double a;
+{
+  perform_fixdfsi (a);
+}
+#endif
+
+#ifdef L_fixsfsi
+nongcc_SI_type
+__fixsfsi (a)
+     FLOAT_ARG_TYPE a;
+{
+  union flt_or_value intify;
+  perform_fixsfsi (FLOATIFY (a));
+}
+#endif
+
+#ifdef L_floatsidf
+double
+__floatsidf (a)
+     nongcc_SI_type a;
+{
+  perform_floatsidf (a);
+}
+#endif
+
+#ifdef L_floatsisf
+FLOAT_VALUE_TYPE
+__floatsisf (a)
+     nongcc_SI_type a;
+{
+  union flt_or_value intify;
+  perform_floatsisf (a);
+}
+#endif
+
+#ifdef L_addsf3
+FLOAT_VALUE_TYPE
+__addsf3 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_value intify;
+  perform_addsf3 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_negsf2
+FLOAT_VALUE_TYPE
+__negsf2 (a)
+     FLOAT_ARG_TYPE a;
+{
+  union flt_or_value intify;
+  perform_negsf2 (FLOATIFY (a));
+}
+#endif
+
+#ifdef L_subsf3
+FLOAT_VALUE_TYPE
+__subsf3 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_value intify;
+  perform_subsf3 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_eqsf2
+nongcc_word_type
+__eqsf2 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_int intify;
+  /* Value == 0 iff a == b.  */
+  perform_eqsf2 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_nesf2
+nongcc_word_type
+__nesf2 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_int intify;
+  /* Value != 0 iff a != b.  */
+  perform_nesf2 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_gtsf2
+nongcc_word_type
+__gtsf2 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_int intify;
+  /* Value > 0 iff a > b.  */
+  perform_gtsf2 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_gesf2
+nongcc_word_type
+__gesf2 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_int intify;
+  /* Value >= 0 iff a >= b.  */
+  perform_gesf2 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_ltsf2
+nongcc_word_type
+__ltsf2 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_int intify;
+  /* Value < 0 iff a < b.  */
+  perform_ltsf2 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_lesf2
+nongcc_word_type
+__lesf2 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_int intify;
+  /* Value <= 0 iff a <= b.  */
+  perform_lesf2 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_mulsf3
+FLOAT_VALUE_TYPE
+__mulsf3 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_value intify;
+  perform_mulsf3 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_divsf3
+FLOAT_VALUE_TYPE
+__divsf3 (a, b)
+     FLOAT_ARG_TYPE a, b;
+{
+  union flt_or_value intify;
+  perform_divsf3 (FLOATIFY (a), FLOATIFY (b));
+}
+#endif
+
+#ifdef L_truncdfsf2
+FLOAT_VALUE_TYPE
+__truncdfsf2 (a)
+     double a;
+{
+  union flt_or_value intify;
+  perform_truncdfsf2 (a);
+}
+#endif
+
+#ifdef L_extendsfdf2
+double
+__extendsfdf2 (a)
+     FLOAT_ARG_TYPE a;
+{
+  union flt_or_value intify;
+  perform_extendsfdf2 (FLOATIFY (a));
+}
+#endif
diff --git a/libgcc/libgcc2.c b/libgcc/libgcc2.c
new file mode 100755
index 0000000..cf7231f
--- /dev/null
+++ b/libgcc/libgcc2.c
@@ -0,0 +1,946 @@
+/* More subroutines needed by GCC output code on some machines.  */
+/* Compile this one with gcc.  */
+/* Copyright (C) 1989, 92-97, 1998 Free Software Foundation, Inc.
+
+This file is part of GNU CC.
+
+GNU CC 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 2, or (at your option)
+any later version.
+
+GNU CC 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 GNU CC; see the file COPYING.  If not, write to
+the Free Software Foundation, 59 Temple Place - Suite 330,
+Boston, MA 02111-1307, USA.  */
+
+/* As a special exception, if you link this library with other files,
+   some of which are compiled with GCC, to produce an executable,
+   this library does not by itself cause the resulting executable
+   to be covered by the GNU General Public License.
+   This exception does not however invalidate any other reasons why
+   the executable file might be covered by the GNU General Public License.  */
+
+#include <stddef.h>
+
+/* Don't use `fancy_abort' here even if config.h says to use it.  */
+#ifdef abort
+#undef abort
+#endif
+
+/* In the first part of this file, we are interfacing to calls generated
+   by the compiler itself.  These calls pass values into these routines
+   which have very specific modes (rather than very specific types), and
+   these compiler-generated calls also expect any return values to have
+   very specific modes (rather than very specific types).  Thus, we need
+   to avoid using regular C language type names in this part of the file
+   because the sizes for those types can be configured to be anything.
+   Instead we use the following special type names.  */
+
+typedef unsigned int UQItype	__attribute__ ((mode (QI)));
+typedef 	 int SItype	__attribute__ ((mode (SI)));
+typedef unsigned int USItype	__attribute__ ((mode (SI)));
+typedef		 int DItype	__attribute__ ((mode (DI)));
+typedef unsigned int UDItype	__attribute__ ((mode (DI)));
+
+typedef 	float SFtype	__attribute__ ((mode (SF)));
+typedef		float DFtype	__attribute__ ((mode (DF)));
+
+typedef int word_type __attribute__ ((mode (__word__)));
+
+/* Make sure that we don't accidentally use any normal C language built-in
+   type names in the first part of this file.  Instead we want to use *only*
+   the type names defined above.  The following macro definitions insure
+   that if we *do* accidentally use some normal C language built-in type name,
+   we will get a syntax error.  */
+
+#define char bogus_type
+#define short bogus_type
+#define int bogus_type
+#define long bogus_type
+#define unsigned bogus_type
+#define float bogus_type
+#define double bogus_type
+
+#define SI_TYPE_SIZE (sizeof (SItype) * 8)
+
+struct DIstruct {SItype low, high;};
+
+/* We need this union to unpack/pack DImode values, since we don't have
+   any arithmetic yet.  Incoming DImode parameters are stored into the
+   `ll' field, and the unpacked result is read from the struct `s'.  */
+
+typedef union
+{
+  struct DIstruct s;
+  DItype ll;
+} DIunion;
+
+#if (defined (L_udivmoddi4) || defined (L_muldi3) || defined (L_udiv_w_sdiv)\
+     || defined (L_divdi3) || defined (L_udivdi3) \
+     || defined (L_moddi3) || defined (L_umoddi3))
+
+#include "longlong.h"
+
+#endif /* udiv or mul */
+
+extern DItype __fixunssfdi (SFtype a);
+extern DItype __fixunsdfdi (DFtype a);
+
+#if defined (L_negdi2) || defined (L_divdi3) || defined (L_moddi3)
+#if defined (L_divdi3) || defined (L_moddi3)
+static inline
+#endif
+DItype
+__negdi2 (DItype u)
+{
+  DIunion w;
+  DIunion uu;
+
+  uu.ll = u;
+
+  w.s.low = -uu.s.low;
+  w.s.high = -uu.s.high - ((USItype) w.s.low > 0);
+
+  return w.ll;
+}
+#endif
+
+/* Unless shift functions are defined whith full ANSI prototypes,
+   parameter b will be promoted to int if word_type is smaller than an int.  */
+#ifdef L_lshrdi3
+DItype
+__lshrdi3 (DItype u, word_type b)
+{
+  DIunion w;
+  word_type bm;
+  DIunion uu;
+
+  if (b == 0)
+    return u;
+
+  uu.ll = u;
+
+  bm = (sizeof (SItype) * 8) - b;
+  if (bm <= 0)
+    {
+      w.s.high = 0;
+      w.s.low = (USItype)uu.s.high >> -bm;
+    }
+  else
+    {
+      USItype carries = (USItype)uu.s.high << bm;
+      w.s.high = (USItype)uu.s.high >> b;
+      w.s.low = ((USItype)uu.s.low >> b) | carries;
+    }
+
+  return w.ll;
+}
+#endif
+
+#ifdef L_ashldi3
+DItype
+__ashldi3 (DItype u, word_type b)
+{
+  DIunion w;
+  word_type bm;
+  DIunion uu;
+
+  if (b == 0)
+    return u;
+
+  uu.ll = u;
+
+  bm = (sizeof (SItype) * 8) - b;
+  if (bm <= 0)
+    {
+      w.s.low = 0;
+      w.s.high = (USItype)uu.s.low << -bm;
+    }
+  else
+    {
+      USItype carries = (USItype)uu.s.low >> bm;
+      w.s.low = (USItype)uu.s.low << b;
+      w.s.high = ((USItype)uu.s.high << b) | carries;
+    }
+
+  return w.ll;
+}
+#endif
+
+#ifdef L_ashrdi3
+DItype
+__ashrdi3 (DItype u, word_type b)
+{
+  DIunion w;
+  word_type bm;
+  DIunion uu;
+
+  if (b == 0)
+    return u;
+
+  uu.ll = u;
+
+  bm = (sizeof (SItype) * 8) - b;
+  if (bm <= 0)
+    {
+      /* w.s.high = 1..1 or 0..0 */
+      w.s.high = uu.s.high >> (sizeof (SItype) * 8 - 1);
+      w.s.low = uu.s.high >> -bm;
+    }
+  else
+    {
+      USItype carries = (USItype)uu.s.high << bm;
+      w.s.high = uu.s.high >> b;
+      w.s.low = ((USItype)uu.s.low >> b) | carries;
+    }
+
+  return w.ll;
+}
+#endif
+
+#ifdef L_ffsdi2
+DItype
+__ffsdi2 (DItype u)
+{
+  DIunion uu, w;
+  uu.ll = u;
+  w.s.high = 0;
+  w.s.low = ffs (uu.s.low);
+  if (w.s.low != 0)
+    return w.ll;
+  w.s.low = ffs (uu.s.high);
+  if (w.s.low != 0)
+    {
+      w.s.low += 8 * sizeof (SItype);
+      return w.ll;
+    }
+  return w.ll;
+}
+#endif
+
+#ifdef L_muldi3
+DItype
+__muldi3 (DItype u, DItype v)
+{
+  DIunion w;
+  DIunion uu, vv;
+
+  uu.ll = u,
+  vv.ll = v;
+
+  w.ll = __umulsidi3 (uu.s.low, vv.s.low);
+  w.s.high += ((USItype) uu.s.low * (USItype) vv.s.high
+	       + (USItype) uu.s.high * (USItype) vv.s.low);
+
+  return w.ll;
+}
+#endif
+
+#ifdef L_udiv_w_sdiv
+#if defined (sdiv_qrnnd)
+USItype
+__udiv_w_sdiv (USItype *rp, USItype a1, USItype a0, USItype d)
+{
+  USItype q, r;
+  USItype c0, c1, b1;
+
+  if ((SItype) d >= 0)
+    {
+      if (a1 < d - a1 - (a0 >> (SI_TYPE_SIZE - 1)))
+	{
+	  /* dividend, divisor, and quotient are nonnegative */
+	  sdiv_qrnnd (q, r, a1, a0, d);
+	}
+      else
+	{
+	  /* Compute c1*2^32 + c0 = a1*2^32 + a0 - 2^31*d */
+	  sub_ddmmss (c1, c0, a1, a0, d >> 1, d << (SI_TYPE_SIZE - 1));
+	  /* Divide (c1*2^32 + c0) by d */
+	  sdiv_qrnnd (q, r, c1, c0, d);
+	  /* Add 2^31 to quotient */
+	  q += (USItype) 1 << (SI_TYPE_SIZE - 1);
+	}
+    }
+  else
+    {
+      b1 = d >> 1;			/* d/2, between 2^30 and 2^31 - 1 */
+      c1 = a1 >> 1;			/* A/2 */
+      c0 = (a1 << (SI_TYPE_SIZE - 1)) + (a0 >> 1);
+
+      if (a1 < b1)			/* A < 2^32*b1, so A/2 < 2^31*b1 */
+	{
+	  sdiv_qrnnd (q, r, c1, c0, b1); /* (A/2) / (d/2) */
+
+	  r = 2*r + (a0 & 1);		/* Remainder from A/(2*b1) */
+	  if ((d & 1) != 0)
+	    {
+	      if (r >= q)
+		r = r - q;
+	      else if (q - r <= d)
+		{
+		  r = r - q + d;
+		  q--;
+		}
+	      else
+		{
+		  r = r - q + 2*d;
+		  q -= 2;
+		}
+	    }
+	}
+      else if (c1 < b1)			/* So 2^31 <= (A/2)/b1 < 2^32 */
+	{
+	  c1 = (b1 - 1) - c1;
+	  c0 = ~c0;			/* logical NOT */
+
+	  sdiv_qrnnd (q, r, c1, c0, b1); /* (A/2) / (d/2) */
+
+	  q = ~q;			/* (A/2)/b1 */
+	  r = (b1 - 1) - r;
+
+	  r = 2*r + (a0 & 1);		/* A/(2*b1) */
+
+	  if ((d & 1) != 0)
+	    {
+	      if (r >= q)
+		r = r - q;
+	      else if (q - r <= d)
+		{
+		  r = r - q + d;
+		  q--;
+		}
+	      else
+		{
+		  r = r - q + 2*d;
+		  q -= 2;
+		}
+	    }
+	}
+      else				/* Implies c1 = b1 */
+	{				/* Hence a1 = d - 1 = 2*b1 - 1 */
+	  if (a0 >= -d)
+	    {
+	      q = -1;
+	      r = a0 + d;
+	    }
+	  else
+	    {
+	      q = -2;
+	      r = a0 + 2*d;
+	    }
+	}
+    }
+
+  *rp = r;
+  return q;
+}
+#else
+/* If sdiv_qrnnd doesn't exist, define dummy __udiv_w_sdiv.  */
+USItype
+__udiv_w_sdiv (USItype *rp __attribute__ ((__unused__)),
+	       USItype a1 __attribute__ ((__unused__)),
+	       USItype a0 __attribute__ ((__unused__)),
+	       USItype d __attribute__ ((__unused__)))
+{
+  return 0;
+}
+#endif
+#endif
+
+#if (defined (L_udivdi3) || defined (L_divdi3) || \
+     defined (L_umoddi3) || defined (L_moddi3))
+#define L_udivmoddi4
+#endif
+
+#ifdef L_udivmoddi4
+static const UQItype __clz_tab[] =
+{
+  0,1,2,2,3,3,3,3,4,4,4,4,4,4,4,4,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,
+  6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,6,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,7,
+  8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+  8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+  8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+  8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,8,
+};
+
+#if (defined (L_udivdi3) || defined (L_divdi3) || \
+     defined (L_umoddi3) || defined (L_moddi3))
+static inline
+#endif
+UDItype
+__udivmoddi4 (UDItype n, UDItype d, UDItype *rp)
+{
+  DIunion ww;
+  DIunion nn, dd;
+  DIunion rr;
+  USItype d0, d1, n0, n1, n2;
+  USItype q0, q1;
+  USItype b, bm;
+
+  nn.ll = n;
+  dd.ll = d;
+
+  d0 = dd.s.low;
+  d1 = dd.s.high;
+  n0 = nn.s.low;
+  n1 = nn.s.high;
+
+#if !UDIV_NEEDS_NORMALIZATION
+  if (d1 == 0)
+    {
+      if (d0 > n1)
+	{
+	  /* 0q = nn / 0D */
+
+	  udiv_qrnnd (q0, n0, n1, n0, d0);
+	  q1 = 0;
+
+	  /* Remainder in n0.  */
+	}
+      else
+	{
+	  /* qq = NN / 0d */
+
+	  if (d0 == 0)
+	    d0 = 1 / d0;	/* Divide intentionally by zero.  */
+
+	  udiv_qrnnd (q1, n1, 0, n1, d0);
+	  udiv_qrnnd (q0, n0, n1, n0, d0);
+
+	  /* Remainder in n0.  */
+	}
+
+      if (rp != 0)
+	{
+	  rr.s.low = n0;
+	  rr.s.high = 0;
+	  *rp = rr.ll;
+	}
+    }
+
+#else /* UDIV_NEEDS_NORMALIZATION */
+
+  if (d1 == 0)
+    {
+      if (d0 > n1)
+	{
+	  /* 0q = nn / 0D */
+
+	  count_leading_zeros (bm, d0);
+
+	  if (bm != 0)
+	    {
+	      /* Normalize, i.e. make the most significant bit of the
+		 denominator set.  */
+
+	      d0 = d0 << bm;
+	      n1 = (n1 << bm) | (n0 >> (SI_TYPE_SIZE - bm));
+	      n0 = n0 << bm;
+	    }
+
+	  udiv_qrnnd (q0, n0, n1, n0, d0);
+	  q1 = 0;
+
+	  /* Remainder in n0 >> bm.  */
+	}
+      else
+	{
+	  /* qq = NN / 0d */
+
+	  if (d0 == 0)
+	    d0 = 1 / d0;	/* Divide intentionally by zero.  */
+
+	  count_leading_zeros (bm, d0);
+
+	  if (bm == 0)
+	    {
+	      /* From (n1 >= d0) /\ (the most significant bit of d0 is set),
+		 conclude (the most significant bit of n1 is set) /\ (the
+		 leading quotient digit q1 = 1).
+
+		 This special case is necessary, not an optimization.
+		 (Shifts counts of SI_TYPE_SIZE are undefined.)  */
+
+	      n1 -= d0;
+	      q1 = 1;
+	    }
+	  else
+	    {
+	      /* Normalize.  */
+
+	      b = SI_TYPE_SIZE - bm;
+
+	      d0 = d0 << bm;
+	      n2 = n1 >> b;
+	      n1 = (n1 << bm) | (n0 >> b);
+	      n0 = n0 << bm;
+
+	      udiv_qrnnd (q1, n1, n2, n1, d0);
+	    }
+
+	  /* n1 != d0...  */
+
+	  udiv_qrnnd (q0, n0, n1, n0, d0);
+
+	  /* Remainder in n0 >> bm.  */
+	}
+
+      if (rp != 0)
+	{
+	  rr.s.low = n0 >> bm;
+	  rr.s.high = 0;
+	  *rp = rr.ll;
+	}
+    }
+#endif /* UDIV_NEEDS_NORMALIZATION */
+
+  else
+    {
+      if (d1 > n1)
+	{
+	  /* 00 = nn / DD */
+
+	  q0 = 0;
+	  q1 = 0;
+
+	  /* Remainder in n1n0.  */
+	  if (rp != 0)
+	    {
+	      rr.s.low = n0;
+	      rr.s.high = n1;
+	      *rp = rr.ll;
+	    }
+	}
+      else
+	{
+	  /* 0q = NN / dd */
+
+	  count_leading_zeros (bm, d1);
+	  if (bm == 0)
+	    {
+	      /* From (n1 >= d1) /\ (the most significant bit of d1 is set),
+		 conclude (the most significant bit of n1 is set) /\ (the
+		 quotient digit q0 = 0 or 1).
+
+		 This special case is necessary, not an optimization.  */
+
+	      /* The condition on the next line takes advantage of that
+		 n1 >= d1 (true due to program flow).  */
+	      if (n1 > d1 || n0 >= d0)
+		{
+		  q0 = 1;
+		  sub_ddmmss (n1, n0, n1, n0, d1, d0);
+		}
+	      else
+		q0 = 0;
+
+	      q1 = 0;
+
+	      if (rp != 0)
+		{
+		  rr.s.low = n0;
+		  rr.s.high = n1;
+		  *rp = rr.ll;
+		}
+	    }
+	  else
+	    {
+	      USItype m1, m0;
+	      /* Normalize.  */
+
+	      b = SI_TYPE_SIZE - bm;
+
+	      d1 = (d1 << bm) | (d0 >> b);
+	      d0 = d0 << bm;
+	      n2 = n1 >> b;
+	      n1 = (n1 << bm) | (n0 >> b);
+	      n0 = n0 << bm;
+
+	      udiv_qrnnd (q0, n1, n2, n1, d1);
+	      umul_ppmm (m1, m0, q0, d0);
+
+	      if (m1 > n1 || (m1 == n1 && m0 > n0))
+		{
+		  q0--;
+		  sub_ddmmss (m1, m0, m1, m0, d1, d0);
+		}
+
+	      q1 = 0;
+
+	      /* Remainder in (n1n0 - m1m0) >> bm.  */
+	      if (rp != 0)
+		{
+		  sub_ddmmss (n1, n0, n1, n0, m1, m0);
+		  rr.s.low = (n1 << b) | (n0 >> bm);
+		  rr.s.high = n1 >> bm;
+		  *rp = rr.ll;
+		}
+	    }
+	}
+    }
+
+  ww.s.low = q0;
+  ww.s.high = q1;
+  return ww.ll;
+}
+#endif
+
+#ifdef L_divdi3
+UDItype __udivmoddi4 ();
+
+DItype
+__divdi3 (DItype u, DItype v)
+{
+  word_type c = 0;
+  DIunion uu, vv;
+  DItype w;
+
+  uu.ll = u;
+  vv.ll = v;
+
+  if (uu.s.high < 0)
+    c = ~c,
+    uu.ll = __negdi2 (uu.ll);
+  if (vv.s.high < 0)
+    c = ~c,
+    vv.ll = __negdi2 (vv.ll);
+
+  w = __udivmoddi4 (uu.ll, vv.ll, (UDItype *) 0);
+  if (c)
+    w = __negdi2 (w);
+
+  return w;
+}
+#endif
+
+#ifdef L_moddi3
+UDItype __udivmoddi4 ();
+DItype
+__moddi3 (DItype u, DItype v)
+{
+  word_type c = 0;
+  DIunion uu, vv;
+  DItype w;
+
+  uu.ll = u;
+  vv.ll = v;
+
+  if (uu.s.high < 0)
+    c = ~c,
+    uu.ll = __negdi2 (uu.ll);
+  if (vv.s.high < 0)
+    vv.ll = __negdi2 (vv.ll);
+
+  (void) __udivmoddi4 (uu.ll, vv.ll, &w);
+  if (c)
+    w = __negdi2 (w);
+
+  return w;
+}
+#endif
+
+#ifdef L_umoddi3
+UDItype __udivmoddi4 ();
+UDItype
+__umoddi3 (UDItype u, UDItype v)
+{
+  UDItype w;
+
+  (void) __udivmoddi4 (u, v, &w);
+
+  return w;
+}
+#endif
+
+#ifdef L_udivdi3
+UDItype __udivmoddi4 ();
+UDItype
+__udivdi3 (UDItype n, UDItype d)
+{
+  return __udivmoddi4 (n, d, (UDItype *) 0);
+}
+#endif
+
+#ifdef L_cmpdi2
+word_type
+__cmpdi2 (DItype a, DItype b)
+{
+  DIunion au, bu;
+
+  au.ll = a, bu.ll = b;
+
+  if (au.s.high < bu.s.high)
+    return 0;
+  else if (au.s.high > bu.s.high)
+    return 2;
+  if ((USItype) au.s.low < (USItype) bu.s.low)
+    return 0;
+  else if ((USItype) au.s.low > (USItype) bu.s.low)
+    return 2;
+  return 1;
+}
+#endif
+
+#ifdef L_ucmpdi2
+word_type
+__ucmpdi2 (DItype a, DItype b)
+{
+  DIunion au, bu;
+
+  au.ll = a, bu.ll = b;
+
+  if ((USItype) au.s.high < (USItype) bu.s.high)
+    return 0;
+  else if ((USItype) au.s.high > (USItype) bu.s.high)
+    return 2;
+  if ((USItype) au.s.low < (USItype) bu.s.low)
+    return 0;
+  else if ((USItype) au.s.low > (USItype) bu.s.low)
+    return 2;
+  return 1;
+}
+#endif
+
+#ifdef L_fixunsdfdi
+#define WORD_SIZE (sizeof (SItype) * 8)
+#define HIGH_WORD_COEFF (((UDItype) 1) << WORD_SIZE)
+
+DItype
+__fixunsdfdi (DFtype a)
+{
+  DFtype b;
+  UDItype v;
+
+  if (a < 0)
+    return 0;
+
+  /* Compute high word of result, as a flonum.  */
+  b = (a / HIGH_WORD_COEFF);
+  /* Convert that to fixed (but not to DItype!),
+     and shift it into the high word.  */
+  v = (USItype) b;
+  v <<= WORD_SIZE;
+  /* Remove high part from the DFtype, leaving the low part as flonum.  */
+  a -= (DFtype)v;
+  /* Convert that to fixed (but not to DItype!) and add it in.
+     Sometimes A comes out negative.  This is significant, since
+     A has more bits than a long int does.  */
+  if (a < 0)
+    v -= (USItype) (- a);
+  else
+    v += (USItype) a;
+  return v;
+}
+#endif
+
+#ifdef L_fixdfdi
+DItype
+__fixdfdi (DFtype a)
+{
+  if (a < 0)
+    return - __fixunsdfdi (-a);
+  return __fixunsdfdi (a);
+}
+#endif
+
+#ifdef L_fixunssfdi
+#define WORD_SIZE (sizeof (SItype) * 8)
+#define HIGH_WORD_COEFF (((UDItype) 1) << WORD_SIZE)
+
+DItype
+__fixunssfdi (SFtype original_a)
+{
+  /* Convert the SFtype to a DFtype, because that is surely not going
+     to lose any bits.  Some day someone else can write a faster version
+     that avoids converting to DFtype, and verify it really works right.  */
+  DFtype a = original_a;
+  DFtype b;
+  UDItype v;
+
+  if (a < 0)
+    return 0;
+
+  /* Compute high word of result, as a flonum.  */
+  b = (a / HIGH_WORD_COEFF);
+  /* Convert that to fixed (but not to DItype!),
+     and shift it into the high word.  */
+  v = (USItype) b;
+  v <<= WORD_SIZE;
+  /* Remove high part from the DFtype, leaving the low part as flonum.  */
+  a -= (DFtype)v;
+  /* Convert that to fixed (but not to DItype!) and add it in.
+     Sometimes A comes out negative.  This is significant, since
+     A has more bits than a long int does.  */
+  if (a < 0)
+    v -= (USItype) (- a);
+  else
+    v += (USItype) a;
+  return v;
+}
+#endif
+
+#ifdef L_fixsfdi
+DItype
+__fixsfdi (SFtype a)
+{
+  if (a < 0)
+    return - __fixunssfdi (-a);
+  return __fixunssfdi (a);
+}
+#endif
+
+#ifdef L_floatdidf
+#define WORD_SIZE (sizeof (SItype) * 8)
+#define HIGH_HALFWORD_COEFF (((UDItype) 1) << (WORD_SIZE / 2))
+#define HIGH_WORD_COEFF (((UDItype) 1) << WORD_SIZE)
+
+DFtype
+__floatdidf (DItype u)
+{
+  DFtype d;
+
+  d = (SItype) (u >> WORD_SIZE);
+  d *= HIGH_HALFWORD_COEFF;
+  d *= HIGH_HALFWORD_COEFF;
+  d += (USItype) (u & (HIGH_WORD_COEFF - 1));
+
+  return d;
+}
+#endif
+
+#ifdef L_floatdisf
+#define WORD_SIZE (sizeof (SItype) * 8)
+#define HIGH_HALFWORD_COEFF (((UDItype) 1) << (WORD_SIZE / 2))
+#define HIGH_WORD_COEFF (((UDItype) 1) << WORD_SIZE)
+#define DI_SIZE (sizeof (DItype) * 8)
+
+/* Define codes for all the float formats that we know of.  Note
+   that this is copied from real.h.  */
+   
+#define UNKNOWN_FLOAT_FORMAT 0
+#define IEEE_FLOAT_FORMAT 1
+#define VAX_FLOAT_FORMAT 2
+#define IBM_FLOAT_FORMAT 3
+
+/* Default to IEEE float if not specified.  Nearly all machines use it.  */
+#ifndef HOST_FLOAT_FORMAT
+#define	HOST_FLOAT_FORMAT	IEEE_FLOAT_FORMAT
+#endif
+
+#if HOST_FLOAT_FORMAT == IEEE_FLOAT_FORMAT
+#define DF_SIZE 53
+#define SF_SIZE 24
+#endif
+
+#if HOST_FLOAT_FORMAT == IBM_FLOAT_FORMAT
+#define DF_SIZE 56
+#define SF_SIZE 24
+#endif
+
+#if HOST_FLOAT_FORMAT == VAX_FLOAT_FORMAT
+#define DF_SIZE 56
+#define SF_SIZE 24
+#endif
+
+SFtype
+__floatdisf (DItype u)
+{
+  /* Do the calculation in DFmode
+     so that we don't lose any of the precision of the high word
+     while multiplying it.  */
+  DFtype f;
+
+  /* Protect against double-rounding error.
+     Represent any low-order bits, that might be truncated in DFmode,
+     by a bit that won't be lost.  The bit can go in anywhere below the
+     rounding position of the SFmode.  A fixed mask and bit position
+     handles all usual configurations.  It doesn't handle the case
+     of 128-bit DImode, however.  */
+  if (DF_SIZE < DI_SIZE
+      && DF_SIZE > (DI_SIZE - DF_SIZE + SF_SIZE))
+    {
+#define REP_BIT ((USItype) 1 << (DI_SIZE - DF_SIZE))
+      if (! (- ((DItype) 1 << DF_SIZE) < u
+	     && u < ((DItype) 1 << DF_SIZE)))
+	{
+	  if ((USItype) u & (REP_BIT - 1))
+	    u |= REP_BIT;
+	}
+    }
+  f = (SItype) (u >> WORD_SIZE);
+  f *= HIGH_HALFWORD_COEFF;
+  f *= HIGH_HALFWORD_COEFF;
+  f += (USItype) (u & (HIGH_WORD_COEFF - 1));
+
+  return (SFtype) f;
+}
+#endif
+
+#ifdef L_fixunsdfsi
+/* Reenable the normal types, in case limits.h needs them.  */
+#undef char
+#undef short
+#undef int
+#undef long
+#undef unsigned
+#undef float
+#undef double
+#undef MIN
+#undef MAX
+#include <limits.h>
+
+USItype
+__fixunsdfsi (DFtype a)
+{
+  if (a >= - (DFtype) LONG_MIN)
+    return (SItype) (a + LONG_MIN) - LONG_MIN;
+  return (SItype) a;
+}
+#endif
+
+#ifdef L_fixunssfsi
+/* Reenable the normal types, in case limits.h needs them.  */
+#undef char
+#undef short
+#undef int
+#undef long
+#undef unsigned
+#undef float
+#undef double
+#undef MIN
+#undef MAX
+#include <limits.h>
+
+USItype
+__fixunssfsi (SFtype a)
+{
+  if (a >= - (SFtype) LONG_MIN)
+    return (SItype) (a + LONG_MIN) - LONG_MIN;
+  return (SItype) a;
+}
+#endif
+
+/* From here on down, the routines use normal data types.  */
+
+#define SItype bogus_type
+#define USItype bogus_type
+#define DItype bogus_type
+#define UDItype bogus_type
+#define SFtype bogus_type
+#define DFtype bogus_type
+
+#undef char
+#undef short
+#undef int
+#undef long
+#undef unsigned
+#undef float
+#undef double