atof-generic.c   [plain text]

/* atof_generic.c - turn a string of digits into a Flonum
   Copyright 1987, 1990, 1991, 1992, 1993, 1994, 1995, 1998, 1999, 2000, 2001
   Free Software Foundation, Inc.

   This file is part of GAS, the GNU Assembler.

   GAS 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.

   GAS 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 GAS; see the file COPYING.  If not, write to the Free
   Software Foundation, 59 Temple Place - Suite 330, Boston, MA
   02111-1307, USA.  */

#include <string.h>

#include "as.h"
#include "safe-ctype.h"

#ifndef FALSE
#define FALSE (0)
#endif
#ifndef TRUE
#define TRUE  (1)
#endif

#ifdef TRACE
static void flonum_print (const FLONUM_TYPE *);
#endif

#define ASSUME_DECIMAL_MARK_IS_DOT

/***********************************************************************\
 *									*
 *	Given a string of decimal digits , with optional decimal	*
 *	mark and optional decimal exponent (place value) of the		*
 *	lowest_order decimal digit: produce a floating point		*
 *	number. The number is 'generic' floating point: our		*
 *	caller will encode it for a specific machine architecture.	*
 *									*
 *	Assumptions							*
 *		uses base (radix) 2					*
 *		this machine uses 2's complement binary integers	*
 *		target flonums use "      "         "       "		*
 *		target flonums exponents fit in a long			*
 *									*
 \***********************************************************************/

/*

  Syntax:

  <flonum> ::= <optional-sign> <decimal-number> <optional-exponent>
  <optional-sign> ::= '+' | '-' | {empty}
  <decimal-number> ::= <integer>
  | <integer> <radix-character>
  | <integer> <radix-character> <integer>
  | <radix-character> <integer>

  <optional-exponent> ::= {empty}
  | <exponent-character> <optional-sign> <integer>

  <integer> ::= <digit> | <digit> <integer>
  <digit> ::= '0' | '1' | '2' | '3' | '4' | '5' | '6' | '7' | '8' | '9'
  <exponent-character> ::= {one character from "string_of_decimal_exponent_marks"}
  <radix-character> ::= {one character from "string_of_decimal_marks"}

  */

int
atof_generic (/* return pointer to just AFTER number we read.  */
	      char **address_of_string_pointer,
	      /* At most one per number.  */
	      const char *string_of_decimal_marks,
	      const char *string_of_decimal_exponent_marks,
	      FLONUM_TYPE *address_of_generic_floating_point_number)
{
  int return_value;		/* 0 means OK.  */
  char *first_digit;
  unsigned int number_of_digits_before_decimal;
  unsigned int number_of_digits_after_decimal;
  long decimal_exponent;
  unsigned int number_of_digits_available;
  char digits_sign_char;

  /*
   * Scan the input string, abstracting (1)digits (2)decimal mark (3) exponent.
   * It would be simpler to modify the string, but we don't; just to be nice
   * to caller.
   * We need to know how many digits we have, so we can allocate space for
   * the digits' value.
   */

  char *p;
  char c;
  int seen_significant_digit;

#ifdef ASSUME_DECIMAL_MARK_IS_DOT
  assert (string_of_decimal_marks[0] == '.'
	  && string_of_decimal_marks[1] == 0);
#define IS_DECIMAL_MARK(c)	((c) == '.')
#else
#define IS_DECIMAL_MARK(c)	(0 != strchr (string_of_decimal_marks, (c)))
#endif

  first_digit = *address_of_string_pointer;
  c = *first_digit;

  if (c == '-' || c == '+')
    {
      digits_sign_char = c;
      first_digit++;
    }
  else
    digits_sign_char = '+';

  switch (first_digit[0])
    {
    case 'n':
    case 'N':
      if (!strncasecmp ("nan", first_digit, 3))
	{
	  address_of_generic_floating_point_number->sign = 0;
	  address_of_generic_floating_point_number->exponent = 0;
	  address_of_generic_floating_point_number->leader =
	    address_of_generic_floating_point_number->low;
	  *address_of_string_pointer = first_digit + 3;
	  return 0;
	}
      break;

    case 'i':
    case 'I':
      if (!strncasecmp ("inf", first_digit, 3))
	{
	  address_of_generic_floating_point_number->sign =
	    digits_sign_char == '+' ? 'P' : 'N';
	  address_of_generic_floating_point_number->exponent = 0;
	  address_of_generic_floating_point_number->leader =
	    address_of_generic_floating_point_number->low;

	  first_digit += 3;
	  if (!strncasecmp ("inity", first_digit, 5))
	    first_digit += 5;

	  *address_of_string_pointer = first_digit;

	  return 0;
	}
      break;
    }

  number_of_digits_before_decimal = 0;
  number_of_digits_after_decimal = 0;
  decimal_exponent = 0;
  seen_significant_digit = 0;
  for (p = first_digit;
       (((c = *p) != '\0')
	&& (!c || !IS_DECIMAL_MARK (c))
	&& (!c || !strchr (string_of_decimal_exponent_marks, c)));
       p++)
    {
      if (ISDIGIT (c))
	{
	  if (seen_significant_digit || c > '0')
	    {
	      ++number_of_digits_before_decimal;
	      seen_significant_digit = 1;
	    }
	  else
	    {
	      first_digit++;
	    }
	}
      else
	{
	  break;		/* p -> char after pre-decimal digits.  */
	}
    }				/* For each digit before decimal mark.  */

#ifndef OLD_FLOAT_READS
  /* Ignore trailing 0's after the decimal point.  The original code here
   * (ifdef'd out) does not do this, and numbers like
   *	4.29496729600000000000e+09	(2**31)
   * come out inexact for some reason related to length of the digit
   * string.
   */
  if (c && IS_DECIMAL_MARK (c))
    {
      unsigned int zeros = 0;	/* Length of current string of zeros */

      for (p++; (c = *p) && ISDIGIT (c); p++)
	{
	  if (c == '0')
	    {
	      zeros++;
	    }
	  else
	    {
	      number_of_digits_after_decimal += 1 + zeros;
	      zeros = 0;
	    }
	}
    }
#else
  if (c && IS_DECIMAL_MARK (c))
    {
      for (p++;
	   (((c = *p) != '\0')
	    && (!c || !strchr (string_of_decimal_exponent_marks, c)));
	   p++)
	{
	  if (ISDIGIT (c))
	    {
	      /* This may be retracted below.  */
	      number_of_digits_after_decimal++;

	      if ( /* seen_significant_digit || */ c > '0')
		{
		  seen_significant_digit = TRUE;
		}
	    }
	  else
	    {
	      if (!seen_significant_digit)
		{
		  number_of_digits_after_decimal = 0;
		}
	      break;
	    }
	}			/* For each digit after decimal mark.  */
    }

  while (number_of_digits_after_decimal
	 && first_digit[number_of_digits_before_decimal
			+ number_of_digits_after_decimal] == '0')
    --number_of_digits_after_decimal;
#endif

  if (flag_m68k_mri)
    {
      while (c == '_')
	c = *++p;
    }
  if (c && strchr (string_of_decimal_exponent_marks, c))
    {
      char digits_exponent_sign_char;

      c = *++p;
      if (flag_m68k_mri)
	{
	  while (c == '_')
	    c = *++p;
	}
      if (c && strchr ("+-", c))
	{
	  digits_exponent_sign_char = c;
	  c = *++p;
	}
      else
	{
	  digits_exponent_sign_char = '+';
	}

      for (; (c); c = *++p)
	{
	  if (ISDIGIT (c))
	    {
	      decimal_exponent = decimal_exponent * 10 + c - '0';
	      /*
	       * BUG! If we overflow here, we lose!
	       */
	    }
	  else
	    {
	      break;
	    }
	}

      if (digits_exponent_sign_char == '-')
	{
	  decimal_exponent = -decimal_exponent;
	}
    }

  *address_of_string_pointer = p;

  number_of_digits_available =
    number_of_digits_before_decimal + number_of_digits_after_decimal;
  return_value = 0;
  if (number_of_digits_available == 0)
    {
      address_of_generic_floating_point_number->exponent = 0;	/* Not strictly necessary */
      address_of_generic_floating_point_number->leader
	= -1 + address_of_generic_floating_point_number->low;
      address_of_generic_floating_point_number->sign = digits_sign_char;
      /* We have just concocted (+/-)0.0E0 */

    }
  else
    {
      int count;		/* Number of useful digits left to scan.  */

      LITTLENUM_TYPE *digits_binary_low;
      unsigned int precision;
      unsigned int maximum_useful_digits;
      unsigned int number_of_digits_to_use;
      unsigned int more_than_enough_bits_for_digits;
      unsigned int more_than_enough_littlenums_for_digits;
      unsigned int size_of_digits_in_littlenums;
      unsigned int size_of_digits_in_chars;
      FLONUM_TYPE power_of_10_flonum;
      FLONUM_TYPE digits_flonum;

      precision = (address_of_generic_floating_point_number->high
		   - address_of_generic_floating_point_number->low
		   + 1);	/* Number of destination littlenums.  */

      /* Includes guard bits (two littlenums worth) */
#if 0 /* The integer version below is very close, and it doesn't
	 require floating point support (which is currently buggy on
	 the Alpha).  */
      maximum_useful_digits = (((double) (precision - 2))
			       * ((double) (LITTLENUM_NUMBER_OF_BITS))
			       / (LOG_TO_BASE_2_OF_10))
	+ 2;			/* 2 :: guard digits.  */
#else
      maximum_useful_digits = (((precision - 2))
			       * ( (LITTLENUM_NUMBER_OF_BITS))
			       * 1000000 / 3321928)
	+ 2;			/* 2 :: guard digits.  */
#endif

      if (number_of_digits_available > maximum_useful_digits)
	{
	  number_of_digits_to_use = maximum_useful_digits;
	}
      else
	{
	  number_of_digits_to_use = number_of_digits_available;
	}

      /* Cast these to SIGNED LONG first, otherwise, on systems with
	 LONG wider than INT (such as Alpha OSF/1), unsignedness may
	 cause unexpected results.  */
      decimal_exponent += ((long) number_of_digits_before_decimal
			   - (long) number_of_digits_to_use);

#if 0
      more_than_enough_bits_for_digits
	= ((((double) number_of_digits_to_use) * LOG_TO_BASE_2_OF_10) + 1);
#else
      more_than_enough_bits_for_digits
	= (number_of_digits_to_use * 3321928 / 1000000 + 1);
#endif

      more_than_enough_littlenums_for_digits
	= (more_than_enough_bits_for_digits
	   / LITTLENUM_NUMBER_OF_BITS)
	+ 2;

      /* Compute (digits) part. In "12.34E56" this is the "1234" part.
	 Arithmetic is exact here. If no digits are supplied then this
	 part is a 0 valued binary integer.  Allocate room to build up
	 the binary number as littlenums.  We want this memory to
	 disappear when we leave this function.  Assume no alignment
	 problems => (room for n objects) == n * (room for 1
	 object).  */

      size_of_digits_in_littlenums = more_than_enough_littlenums_for_digits;
      size_of_digits_in_chars = size_of_digits_in_littlenums
	* sizeof (LITTLENUM_TYPE);

      digits_binary_low = (LITTLENUM_TYPE *)
	alloca (size_of_digits_in_chars);

      memset ((char *) digits_binary_low, '\0', size_of_digits_in_chars);

      /* Digits_binary_low[] is allocated and zeroed.  */

      /*
       * Parse the decimal digits as if * digits_low was in the units position.
       * Emit a binary number into digits_binary_low[].
       *
       * Use a large-precision version of:
       * (((1st-digit) * 10 + 2nd-digit) * 10 + 3rd-digit ...) * 10 + last-digit
       */

      for (p = first_digit, count = number_of_digits_to_use; count; p++, --count)
	{
	  c = *p;
	  if (ISDIGIT (c))
	    {
	      /*
	       * Multiply by 10. Assume can never overflow.
	       * Add this digit to digits_binary_low[].
	       */

	      long carry;
	      LITTLENUM_TYPE *littlenum_pointer;
	      LITTLENUM_TYPE *littlenum_limit;

	      littlenum_limit = digits_binary_low
		+ more_than_enough_littlenums_for_digits
		- 1;

	      carry = c - '0';	/* char -> binary */

	      for (littlenum_pointer = digits_binary_low;
		   littlenum_pointer <= littlenum_limit;
		   littlenum_pointer++)
		{
		  long work;

		  work = carry + 10 * (long) (*littlenum_pointer);
		  *littlenum_pointer = work & LITTLENUM_MASK;
		  carry = work >> LITTLENUM_NUMBER_OF_BITS;
		}

	      if (carry != 0)
		{
		  /*
		   * We have a GROSS internal error.
		   * This should never happen.
		   */
		  as_fatal (_("failed sanity check"));
		}
	    }
	  else
	    {
	      ++count;		/* '.' doesn't alter digits used count.  */
	    }
	}

      /*
       * Digits_binary_low[] properly encodes the value of the digits.
       * Forget about any high-order littlenums that are 0.
       */
      while (digits_binary_low[size_of_digits_in_littlenums - 1] == 0
	     && size_of_digits_in_littlenums >= 2)
	size_of_digits_in_littlenums--;

      digits_flonum.low = digits_binary_low;
      digits_flonum.high = digits_binary_low + size_of_digits_in_littlenums - 1;
      digits_flonum.leader = digits_flonum.high;
      digits_flonum.exponent = 0;
      /*
       * The value of digits_flonum . sign should not be important.
       * We have already decided the output's sign.
       * We trust that the sign won't influence the other parts of the number!
       * So we give it a value for these reasons:
       * (1) courtesy to humans reading/debugging
       *     these numbers so they don't get excited about strange values
       * (2) in future there may be more meaning attached to sign,
       *     and what was
       *     harmless noise may become disruptive, ill-conditioned (or worse)
       *     input.
       */
      digits_flonum.sign = '+';

      {
	/*
	 * Compute the mantssa (& exponent) of the power of 10.
	 * If successful, then multiply the power of 10 by the digits
	 * giving return_binary_mantissa and return_binary_exponent.
	 */

	LITTLENUM_TYPE *power_binary_low;
	int decimal_exponent_is_negative;
	/* This refers to the "-56" in "12.34E-56".  */
	/* FALSE: decimal_exponent is positive (or 0) */
	/* TRUE:  decimal_exponent is negative */
	FLONUM_TYPE temporary_flonum;
	LITTLENUM_TYPE *temporary_binary_low;
	unsigned int size_of_power_in_littlenums;
	unsigned int size_of_power_in_chars;

	size_of_power_in_littlenums = precision;
	/* Precision has a built-in fudge factor so we get a few guard bits.  */

	decimal_exponent_is_negative = decimal_exponent < 0;
	if (decimal_exponent_is_negative)
	  {
	    decimal_exponent = -decimal_exponent;
	  }

	/* From now on: the decimal exponent is > 0. Its sign is separate.  */

	size_of_power_in_chars = size_of_power_in_littlenums
	  * sizeof (LITTLENUM_TYPE) + 2;

	power_binary_low = (LITTLENUM_TYPE *) alloca (size_of_power_in_chars);
	temporary_binary_low = (LITTLENUM_TYPE *) alloca (size_of_power_in_chars);
	memset ((char *) power_binary_low, '\0', size_of_power_in_chars);
	*power_binary_low = 1;
	power_of_10_flonum.exponent = 0;
	power_of_10_flonum.low = power_binary_low;
	power_of_10_flonum.leader = power_binary_low;
	power_of_10_flonum.high = power_binary_low + size_of_power_in_littlenums - 1;
	power_of_10_flonum.sign = '+';
	temporary_flonum.low = temporary_binary_low;
	temporary_flonum.high = temporary_binary_low + size_of_power_in_littlenums - 1;
	/*
	 * (power) == 1.
	 * Space for temporary_flonum allocated.
	 */

	/*
	 * ...
	 *
	 * WHILE	more bits
	 * DO	find next bit (with place value)
	 *	multiply into power mantissa
	 * OD
	 */
	{
	  int place_number_limit;
	  /* Any 10^(2^n) whose "n" exceeds this */
	  /* value will fall off the end of */
	  /* flonum_XXXX_powers_of_ten[].  */
	  int place_number;
	  const FLONUM_TYPE *multiplicand;	/* -> 10^(2^n) */

	  place_number_limit = table_size_of_flonum_powers_of_ten;

	  multiplicand = (decimal_exponent_is_negative
			  ? flonum_negative_powers_of_ten
			  : flonum_positive_powers_of_ten);

	  for (place_number = 1;/* Place value of this bit of exponent.  */
	       decimal_exponent;/* Quit when no more 1 bits in exponent.  */
	       decimal_exponent >>= 1, place_number++)
	    {
	      if (decimal_exponent & 1)
		{
		  if (place_number > place_number_limit)
		    {
		      /* The decimal exponent has a magnitude so great
			 that our tables can't help us fragment it.
			 Although this routine is in error because it
			 can't imagine a number that big, signal an
			 error as if it is the user's fault for
			 presenting such a big number.  */
		      return_value = ERROR_EXPONENT_OVERFLOW;
		      /* quit out of loop gracefully */
		      decimal_exponent = 0;
		    }
		  else
		    {
#ifdef TRACE
		      printf ("before multiply, place_number = %d., power_of_10_flonum:\n",
			      place_number);

		      flonum_print (&power_of_10_flonum);
		      (void) putchar ('\n');
#endif
#ifdef TRACE
		      printf ("multiplier:\n");
		      flonum_print (multiplicand + place_number);
		      (void) putchar ('\n');
#endif
		      flonum_multip (multiplicand + place_number,
				     &power_of_10_flonum, &temporary_flonum);
#ifdef TRACE
		      printf ("after multiply:\n");
		      flonum_print (&temporary_flonum);
		      (void) putchar ('\n');
#endif
		      flonum_copy (&temporary_flonum, &power_of_10_flonum);
#ifdef TRACE
		      printf ("after copy:\n");
		      flonum_print (&power_of_10_flonum);
		      (void) putchar ('\n');
#endif
		    } /* If this bit of decimal_exponent was computable.*/
		} /* If this bit of decimal_exponent was set.  */
	    } /* For each bit of binary representation of exponent */
#ifdef TRACE
	  printf ("after computing power_of_10_flonum:\n");
	  flonum_print (&power_of_10_flonum);
	  (void) putchar ('\n');
#endif
	}

      }

      /*
       * power_of_10_flonum is power of ten in binary (mantissa) , (exponent).
       * It may be the number 1, in which case we don't NEED to multiply.
       *
       * Multiply (decimal digits) by power_of_10_flonum.
       */

      flonum_multip (&power_of_10_flonum, &digits_flonum, address_of_generic_floating_point_number);
      /* Assert sign of the number we made is '+'.  */
      address_of_generic_floating_point_number->sign = digits_sign_char;

    }
  return return_value;
}

#ifdef TRACE
static void
flonum_print (f)
     const FLONUM_TYPE *f;
{
  LITTLENUM_TYPE *lp;
  char littlenum_format[10];
  sprintf (littlenum_format, " %%0%dx", sizeof (LITTLENUM_TYPE) * 2);
#define print_littlenum(LP)	(printf (littlenum_format, LP))
  printf ("flonum @%p %c e%ld", f, f->sign, f->exponent);
  if (f->low < f->high)
    for (lp = f->high; lp >= f->low; lp--)
      print_littlenum (*lp);
  else
    for (lp = f->low; lp <= f->high; lp++)
      print_littlenum (*lp);
  printf ("\n");
  fflush (stdout);
}
#endif

/* end of atof_generic.c */