1/* atof_generic.c - turn a string of digits into a Flonum
2   Copyright (C) 1987-2020 Free Software Foundation, Inc.
3
4   This file is part of GAS, the GNU Assembler.
5
6   GAS is free software; you can redistribute it and/or modify
7   it under the terms of the GNU General Public License as published by
8   the Free Software Foundation; either version 3, or (at your option)
9   any later version.
10
11   GAS is distributed in the hope that it will be useful, but WITHOUT
12   ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
13   or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
14   License for more details.
15
16   You should have received a copy of the GNU General Public License
17   along with GAS; see the file COPYING.  If not, write to the Free
18   Software Foundation, 51 Franklin Street - Fifth Floor, Boston, MA
19   02110-1301, USA.  */
20
21#include "as.h"
22#include "safe-ctype.h"
23
24#ifndef FALSE
25#define FALSE (0)
26#endif
27#ifndef TRUE
28#define TRUE  (1)
29#endif
30
31#ifdef TRACE
32static void flonum_print (const FLONUM_TYPE *);
33#endif
34
35#define ASSUME_DECIMAL_MARK_IS_DOT
36
37/***********************************************************************\
38 *									*
39 *	Given a string of decimal digits , with optional decimal	*
40 *	mark and optional decimal exponent (place value) of the		*
41 *	lowest_order decimal digit: produce a floating point		*
42 *	number. The number is 'generic' floating point: our		*
43 *	caller will encode it for a specific machine architecture.	*
44 *									*
45 *	Assumptions							*
46 *		uses base (radix) 2					*
47 *		this machine uses 2's complement binary integers	*
48 *		target flonums use "      "         "       "		*
49 *		target flonums exponents fit in a long			*
50 *									*
51 \***********************************************************************/
52
53/*
54
55  Syntax:
56
57  <flonum> ::= <optional-sign> <decimal-number> <optional-exponent>
58  <optional-sign> ::= '+' | '-' | {empty}
59  <decimal-number> ::= <integer>
60  | <integer> <radix-character>
61  | <integer> <radix-character> <integer>
62  | <radix-character> <integer>
63
64  <optional-exponent> ::= {empty}
65  | <exponent-character> <optional-sign> <integer>
66
67  <integer> ::= <digit> | <digit> <integer>
68  <digit> ::= '0' | '1' | '2' | '3' | '4' | '5' | '6' | '7' | '8' | '9'
69  <exponent-character> ::= {one character from "string_of_decimal_exponent_marks"}
70  <radix-character> ::= {one character from "string_of_decimal_marks"}
71
72  */
73
74int
75atof_generic (/* return pointer to just AFTER number we read.  */
76	      char **address_of_string_pointer,
77	      /* At most one per number.  */
78	      const char *string_of_decimal_marks,
79	      const char *string_of_decimal_exponent_marks,
80	      FLONUM_TYPE *address_of_generic_floating_point_number)
81{
82  int return_value;		/* 0 means OK.  */
83  char *first_digit;
84  unsigned int number_of_digits_before_decimal;
85  unsigned int number_of_digits_after_decimal;
86  long decimal_exponent;
87  unsigned int number_of_digits_available;
88  char digits_sign_char;
89
90  /*
91   * Scan the input string, abstracting (1)digits (2)decimal mark (3) exponent.
92   * It would be simpler to modify the string, but we don't; just to be nice
93   * to caller.
94   * We need to know how many digits we have, so we can allocate space for
95   * the digits' value.
96   */
97
98  char *p;
99  char c;
100  int seen_significant_digit;
101
102#ifdef ASSUME_DECIMAL_MARK_IS_DOT
103  gas_assert (string_of_decimal_marks[0] == '.'
104	  && string_of_decimal_marks[1] == 0);
105#define IS_DECIMAL_MARK(c)	((c) == '.')
106#else
107#define IS_DECIMAL_MARK(c)	(0 != strchr (string_of_decimal_marks, (c)))
108#endif
109
110  first_digit = *address_of_string_pointer;
111  c = *first_digit;
112
113  if (c == '-' || c == '+')
114    {
115      digits_sign_char = c;
116      first_digit++;
117    }
118  else
119    digits_sign_char = '+';
120
121  switch (first_digit[0])
122    {
123    case 'n':
124    case 'N':
125      if (!strncasecmp ("nan", first_digit, 3))
126	{
127	  address_of_generic_floating_point_number->sign = 0;
128	  address_of_generic_floating_point_number->exponent = 0;
129	  address_of_generic_floating_point_number->leader =
130	    address_of_generic_floating_point_number->low;
131	  *address_of_string_pointer = first_digit + 3;
132	  return 0;
133	}
134      break;
135
136    case 'i':
137    case 'I':
138      if (!strncasecmp ("inf", first_digit, 3))
139	{
140	  address_of_generic_floating_point_number->sign =
141	    digits_sign_char == '+' ? 'P' : 'N';
142	  address_of_generic_floating_point_number->exponent = 0;
143	  address_of_generic_floating_point_number->leader =
144	    address_of_generic_floating_point_number->low;
145
146	  first_digit += 3;
147	  if (!strncasecmp ("inity", first_digit, 5))
148	    first_digit += 5;
149
150	  *address_of_string_pointer = first_digit;
151
152	  return 0;
153	}
154      break;
155    }
156
157  number_of_digits_before_decimal = 0;
158  number_of_digits_after_decimal = 0;
159  decimal_exponent = 0;
160  seen_significant_digit = 0;
161  for (p = first_digit;
162       (((c = *p) != '\0')
163	&& (!c || !IS_DECIMAL_MARK (c))
164	&& (!c || !strchr (string_of_decimal_exponent_marks, c)));
165       p++)
166    {
167      if (ISDIGIT (c))
168	{
169	  if (seen_significant_digit || c > '0')
170	    {
171	      ++number_of_digits_before_decimal;
172	      seen_significant_digit = 1;
173	    }
174	  else
175	    {
176	      first_digit++;
177	    }
178	}
179      else
180	{
181	  break;		/* p -> char after pre-decimal digits.  */
182	}
183    }				/* For each digit before decimal mark.  */
184
185#ifndef OLD_FLOAT_READS
186  /* Ignore trailing 0's after the decimal point.  The original code here
187     (ifdef'd out) does not do this, and numbers like
188    	4.29496729600000000000e+09	(2**31)
189     come out inexact for some reason related to length of the digit
190     string.  */
191
192  /* The case number_of_digits_before_decimal = 0 is handled for
193     deleting zeros after decimal.  In this case the decimal mark and
194     the first zero digits after decimal mark are skipped.  */
195  seen_significant_digit = 0;
196  signed long subtract_decimal_exponent = 0;
197
198  if (c && IS_DECIMAL_MARK (c))
199    {
200      unsigned int zeros = 0;	/* Length of current string of zeros.  */
201
202      if (number_of_digits_before_decimal == 0)
203	/* Skip decimal mark.  */
204	first_digit++;
205
206      for (p++; (c = *p) && ISDIGIT (c); p++)
207	{
208	  if (c == '0')
209	    {
210	      if (number_of_digits_before_decimal == 0
211		  && !seen_significant_digit)
212		{
213		  /* Skip '0' and the decimal mark.  */
214		  first_digit++;
215		  subtract_decimal_exponent--;
216		}
217	      else
218		zeros++;
219	    }
220	  else
221	    {
222	      seen_significant_digit = 1;
223	      number_of_digits_after_decimal += 1 + zeros;
224	      zeros = 0;
225	    }
226	}
227    }
228#else
229  if (c && IS_DECIMAL_MARK (c))
230    {
231      for (p++;
232	   (((c = *p) != '\0')
233	    && (!c || !strchr (string_of_decimal_exponent_marks, c)));
234	   p++)
235	{
236	  if (ISDIGIT (c))
237	    {
238	      /* This may be retracted below.  */
239	      number_of_digits_after_decimal++;
240
241	      if ( /* seen_significant_digit || */ c > '0')
242		{
243		  seen_significant_digit = TRUE;
244		}
245	    }
246	  else
247	    {
248	      if (!seen_significant_digit)
249		{
250		  number_of_digits_after_decimal = 0;
251		}
252	      break;
253	    }
254	}			/* For each digit after decimal mark.  */
255    }
256
257  while (number_of_digits_after_decimal
258	 && first_digit[number_of_digits_before_decimal
259			+ number_of_digits_after_decimal] == '0')
260    --number_of_digits_after_decimal;
261#endif
262
263  if (flag_m68k_mri)
264    {
265      while (c == '_')
266	c = *++p;
267    }
268  if (c && strchr (string_of_decimal_exponent_marks, c))
269    {
270      char digits_exponent_sign_char;
271
272      c = *++p;
273      if (flag_m68k_mri)
274	{
275	  while (c == '_')
276	    c = *++p;
277	}
278      if (c && strchr ("+-", c))
279	{
280	  digits_exponent_sign_char = c;
281	  c = *++p;
282	}
283      else
284	{
285	  digits_exponent_sign_char = '+';
286	}
287
288      for (; (c); c = *++p)
289	{
290	  if (ISDIGIT (c))
291	    {
292	      decimal_exponent = decimal_exponent * 10 + c - '0';
293	      /*
294	       * BUG! If we overflow here, we lose!
295	       */
296	    }
297	  else
298	    {
299	      break;
300	    }
301	}
302
303      if (digits_exponent_sign_char == '-')
304	{
305	  decimal_exponent = -decimal_exponent;
306	}
307    }
308
309#ifndef OLD_FLOAT_READS
310  /* Subtract_decimal_exponent != 0 when number_of_digits_before_decimal = 0
311     and first digit after decimal is '0'.  */
312  decimal_exponent += subtract_decimal_exponent;
313#endif
314
315  *address_of_string_pointer = p;
316
317  number_of_digits_available =
318    number_of_digits_before_decimal + number_of_digits_after_decimal;
319  return_value = 0;
320  if (number_of_digits_available == 0)
321    {
322      address_of_generic_floating_point_number->exponent = 0;	/* Not strictly necessary */
323      address_of_generic_floating_point_number->leader
324	= -1 + address_of_generic_floating_point_number->low;
325      address_of_generic_floating_point_number->sign = digits_sign_char;
326      /* We have just concocted (+/-)0.0E0 */
327
328    }
329  else
330    {
331      int count;		/* Number of useful digits left to scan.  */
332
333      LITTLENUM_TYPE *temporary_binary_low = NULL;
334      LITTLENUM_TYPE *power_binary_low = NULL;
335      LITTLENUM_TYPE *digits_binary_low;
336      unsigned int precision;
337      unsigned int maximum_useful_digits;
338      unsigned int number_of_digits_to_use;
339      unsigned int more_than_enough_bits_for_digits;
340      unsigned int more_than_enough_littlenums_for_digits;
341      unsigned int size_of_digits_in_littlenums;
342      unsigned int size_of_digits_in_chars;
343      FLONUM_TYPE power_of_10_flonum;
344      FLONUM_TYPE digits_flonum;
345
346      precision = (address_of_generic_floating_point_number->high
347		   - address_of_generic_floating_point_number->low
348		   + 1);	/* Number of destination littlenums.  */
349
350      /* precision includes two littlenums worth of guard bits,
351	 so this gives us 10 decimal guard digits here.  */
352      maximum_useful_digits = (precision
353			       * LITTLENUM_NUMBER_OF_BITS
354			       * 1000000 / 3321928
355			       + 1);	/* round up.  */
356
357      if (number_of_digits_available > maximum_useful_digits)
358	{
359	  number_of_digits_to_use = maximum_useful_digits;
360	}
361      else
362	{
363	  number_of_digits_to_use = number_of_digits_available;
364	}
365
366      /* Cast these to SIGNED LONG first, otherwise, on systems with
367	 LONG wider than INT (such as Alpha OSF/1), unsignedness may
368	 cause unexpected results.  */
369      decimal_exponent += ((long) number_of_digits_before_decimal
370			   - (long) number_of_digits_to_use);
371
372      more_than_enough_bits_for_digits
373	= (number_of_digits_to_use * 3321928 / 1000000 + 1);
374
375      more_than_enough_littlenums_for_digits
376	= (more_than_enough_bits_for_digits
377	   / LITTLENUM_NUMBER_OF_BITS)
378	+ 2;
379
380      /* Compute (digits) part. In "12.34E56" this is the "1234" part.
381	 Arithmetic is exact here. If no digits are supplied then this
382	 part is a 0 valued binary integer.  Allocate room to build up
383	 the binary number as littlenums.  We want this memory to
384	 disappear when we leave this function.  Assume no alignment
385	 problems => (room for n objects) == n * (room for 1
386	 object).  */
387
388      size_of_digits_in_littlenums = more_than_enough_littlenums_for_digits;
389      size_of_digits_in_chars = size_of_digits_in_littlenums
390	* sizeof (LITTLENUM_TYPE);
391
392      digits_binary_low = (LITTLENUM_TYPE *)
393	xmalloc (size_of_digits_in_chars);
394
395      memset ((char *) digits_binary_low, '\0', size_of_digits_in_chars);
396
397      /* Digits_binary_low[] is allocated and zeroed.  */
398
399      /*
400       * Parse the decimal digits as if * digits_low was in the units position.
401       * Emit a binary number into digits_binary_low[].
402       *
403       * Use a large-precision version of:
404       * (((1st-digit) * 10 + 2nd-digit) * 10 + 3rd-digit ...) * 10 + last-digit
405       */
406
407      for (p = first_digit, count = number_of_digits_to_use; count; p++, --count)
408	{
409	  c = *p;
410	  if (ISDIGIT (c))
411	    {
412	      /*
413	       * Multiply by 10. Assume can never overflow.
414	       * Add this digit to digits_binary_low[].
415	       */
416
417	      long carry;
418	      LITTLENUM_TYPE *littlenum_pointer;
419	      LITTLENUM_TYPE *littlenum_limit;
420
421	      littlenum_limit = digits_binary_low
422		+ more_than_enough_littlenums_for_digits
423		- 1;
424
425	      carry = c - '0';	/* char -> binary */
426
427	      for (littlenum_pointer = digits_binary_low;
428		   littlenum_pointer <= littlenum_limit;
429		   littlenum_pointer++)
430		{
431		  long work;
432
433		  work = carry + 10 * (long) (*littlenum_pointer);
434		  *littlenum_pointer = work & LITTLENUM_MASK;
435		  carry = work >> LITTLENUM_NUMBER_OF_BITS;
436		}
437
438	      if (carry != 0)
439		{
440		  /*
441		   * We have a GROSS internal error.
442		   * This should never happen.
443		   */
444		  as_fatal (_("failed sanity check"));
445		}
446	    }
447	  else
448	    {
449	      ++count;		/* '.' doesn't alter digits used count.  */
450	    }
451	}
452
453      /*
454       * Digits_binary_low[] properly encodes the value of the digits.
455       * Forget about any high-order littlenums that are 0.
456       */
457      while (digits_binary_low[size_of_digits_in_littlenums - 1] == 0
458	     && size_of_digits_in_littlenums >= 2)
459	size_of_digits_in_littlenums--;
460
461      digits_flonum.low = digits_binary_low;
462      digits_flonum.high = digits_binary_low + size_of_digits_in_littlenums - 1;
463      digits_flonum.leader = digits_flonum.high;
464      digits_flonum.exponent = 0;
465      /*
466       * The value of digits_flonum . sign should not be important.
467       * We have already decided the output's sign.
468       * We trust that the sign won't influence the other parts of the number!
469       * So we give it a value for these reasons:
470       * (1) courtesy to humans reading/debugging
471       *     these numbers so they don't get excited about strange values
472       * (2) in future there may be more meaning attached to sign,
473       *     and what was
474       *     harmless noise may become disruptive, ill-conditioned (or worse)
475       *     input.
476       */
477      digits_flonum.sign = '+';
478
479      {
480	/*
481	 * Compute the mantissa (& exponent) of the power of 10.
482	 * If successful, then multiply the power of 10 by the digits
483	 * giving return_binary_mantissa and return_binary_exponent.
484	 */
485
486	int decimal_exponent_is_negative;
487	/* This refers to the "-56" in "12.34E-56".  */
488	/* FALSE: decimal_exponent is positive (or 0) */
489	/* TRUE:  decimal_exponent is negative */
490	FLONUM_TYPE temporary_flonum;
491	unsigned int size_of_power_in_littlenums;
492	unsigned int size_of_power_in_chars;
493
494	size_of_power_in_littlenums = precision;
495	/* Precision has a built-in fudge factor so we get a few guard bits.  */
496
497	decimal_exponent_is_negative = decimal_exponent < 0;
498	if (decimal_exponent_is_negative)
499	  {
500	    decimal_exponent = -decimal_exponent;
501	  }
502
503	/* From now on: the decimal exponent is > 0. Its sign is separate.  */
504
505	size_of_power_in_chars = size_of_power_in_littlenums
506	  * sizeof (LITTLENUM_TYPE) + 2;
507
508	power_binary_low = (LITTLENUM_TYPE *) xmalloc (size_of_power_in_chars);
509	temporary_binary_low = (LITTLENUM_TYPE *) xmalloc (size_of_power_in_chars);
510
511	memset ((char *) power_binary_low, '\0', size_of_power_in_chars);
512	*power_binary_low = 1;
513	power_of_10_flonum.exponent = 0;
514	power_of_10_flonum.low = power_binary_low;
515	power_of_10_flonum.leader = power_binary_low;
516	power_of_10_flonum.high = power_binary_low + size_of_power_in_littlenums - 1;
517	power_of_10_flonum.sign = '+';
518	temporary_flonum.low = temporary_binary_low;
519	temporary_flonum.high = temporary_binary_low + size_of_power_in_littlenums - 1;
520	/*
521	 * (power) == 1.
522	 * Space for temporary_flonum allocated.
523	 */
524
525	/*
526	 * ...
527	 *
528	 * WHILE	more bits
529	 * DO	find next bit (with place value)
530	 *	multiply into power mantissa
531	 * OD
532	 */
533	{
534	  int place_number_limit;
535	  /* Any 10^(2^n) whose "n" exceeds this */
536	  /* value will fall off the end of */
537	  /* flonum_XXXX_powers_of_ten[].  */
538	  int place_number;
539	  const FLONUM_TYPE *multiplicand;	/* -> 10^(2^n) */
540
541	  place_number_limit = table_size_of_flonum_powers_of_ten;
542
543	  multiplicand = (decimal_exponent_is_negative
544			  ? flonum_negative_powers_of_ten
545			  : flonum_positive_powers_of_ten);
546
547	  for (place_number = 1;/* Place value of this bit of exponent.  */
548	       decimal_exponent;/* Quit when no more 1 bits in exponent.  */
549	       decimal_exponent >>= 1, place_number++)
550	    {
551	      if (decimal_exponent & 1)
552		{
553		  if (place_number > place_number_limit)
554		    {
555		      /* The decimal exponent has a magnitude so great
556			 that our tables can't help us fragment it.
557			 Although this routine is in error because it
558			 can't imagine a number that big, signal an
559			 error as if it is the user's fault for
560			 presenting such a big number.  */
561		      return_value = ERROR_EXPONENT_OVERFLOW;
562		      /* quit out of loop gracefully */
563		      decimal_exponent = 0;
564		    }
565		  else
566		    {
567#ifdef TRACE
568		      printf ("before multiply, place_number = %d., power_of_10_flonum:\n",
569			      place_number);
570
571		      flonum_print (&power_of_10_flonum);
572		      (void) putchar ('\n');
573#endif
574#ifdef TRACE
575		      printf ("multiplier:\n");
576		      flonum_print (multiplicand + place_number);
577		      (void) putchar ('\n');
578#endif
579		      flonum_multip (multiplicand + place_number,
580				     &power_of_10_flonum, &temporary_flonum);
581#ifdef TRACE
582		      printf ("after multiply:\n");
583		      flonum_print (&temporary_flonum);
584		      (void) putchar ('\n');
585#endif
586		      flonum_copy (&temporary_flonum, &power_of_10_flonum);
587#ifdef TRACE
588		      printf ("after copy:\n");
589		      flonum_print (&power_of_10_flonum);
590		      (void) putchar ('\n');
591#endif
592		    } /* If this bit of decimal_exponent was computable.*/
593		} /* If this bit of decimal_exponent was set.  */
594	    } /* For each bit of binary representation of exponent */
595#ifdef TRACE
596	  printf ("after computing power_of_10_flonum:\n");
597	  flonum_print (&power_of_10_flonum);
598	  (void) putchar ('\n');
599#endif
600	}
601      }
602
603      /*
604       * power_of_10_flonum is power of ten in binary (mantissa) , (exponent).
605       * It may be the number 1, in which case we don't NEED to multiply.
606       *
607       * Multiply (decimal digits) by power_of_10_flonum.
608       */
609
610      flonum_multip (&power_of_10_flonum, &digits_flonum, address_of_generic_floating_point_number);
611      /* Assert sign of the number we made is '+'.  */
612      address_of_generic_floating_point_number->sign = digits_sign_char;
613
614      free (temporary_binary_low);
615      free (power_binary_low);
616      free (digits_binary_low);
617    }
618  return return_value;
619}
620
621#ifdef TRACE
622static void
623flonum_print (f)
624     const FLONUM_TYPE *f;
625{
626  LITTLENUM_TYPE *lp;
627  char littlenum_format[10];
628  sprintf (littlenum_format, " %%0%dx", sizeof (LITTLENUM_TYPE) * 2);
629#define print_littlenum(LP)	(printf (littlenum_format, LP))
630  printf ("flonum @%p %c e%ld", f, f->sign, f->exponent);
631  if (f->low < f->high)
632    for (lp = f->high; lp >= f->low; lp--)
633      print_littlenum (*lp);
634  else
635    for (lp = f->low; lp <= f->high; lp++)
636      print_littlenum (*lp);
637  printf ("\n");
638  fflush (stdout);
639}
640#endif
641
642/* end of atof_generic.c */
643