1 /* atof_generic.c - turn a string of digits into a Flonum
2    Copyright (C) 1987-2021 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 #ifdef TRACE
25 static void flonum_print (const FLONUM_TYPE *);
26 #endif
27 
28 #define ASSUME_DECIMAL_MARK_IS_DOT
29 
30 /***********************************************************************\
31  *									*
32  *	Given a string of decimal digits , with optional decimal	*
33  *	mark and optional decimal exponent (place value) of the		*
34  *	lowest_order decimal digit: produce a floating point		*
35  *	number. The number is 'generic' floating point: our		*
36  *	caller will encode it for a specific machine architecture.	*
37  *									*
38  *	Assumptions							*
39  *		uses base (radix) 2					*
40  *		this machine uses 2's complement binary integers	*
41  *		target flonums use "      "         "       "		*
42  *		target flonums exponents fit in a long			*
43  *									*
44  \***********************************************************************/
45 
46 /*
47 
48   Syntax:
49 
50   <flonum> ::= <optional-sign> <decimal-number> <optional-exponent>
51   <optional-sign> ::= '+' | '-' | {empty}
52   <decimal-number> ::= <integer>
53   | <integer> <radix-character>
54   | <integer> <radix-character> <integer>
55   | <radix-character> <integer>
56 
57   <optional-exponent> ::= {empty}
58   | <exponent-character> <optional-sign> <integer>
59 
60   <integer> ::= <digit> | <digit> <integer>
61   <digit> ::= '0' | '1' | '2' | '3' | '4' | '5' | '6' | '7' | '8' | '9'
62   <exponent-character> ::= {one character from "string_of_decimal_exponent_marks"}
63   <radix-character> ::= {one character from "string_of_decimal_marks"}
64 
65   */
66 
67 int
atof_generic(char ** address_of_string_pointer,const char * string_of_decimal_marks,const char * string_of_decimal_exponent_marks,FLONUM_TYPE * address_of_generic_floating_point_number)68 atof_generic (/* return pointer to just AFTER number we read.  */
69 	      char **address_of_string_pointer,
70 	      /* At most one per number.  */
71 	      const char *string_of_decimal_marks,
72 	      const char *string_of_decimal_exponent_marks,
73 	      FLONUM_TYPE *address_of_generic_floating_point_number)
74 {
75   int return_value;		/* 0 means OK.  */
76   char *first_digit;
77   unsigned int number_of_digits_before_decimal;
78   unsigned int number_of_digits_after_decimal;
79   long decimal_exponent;
80   unsigned int number_of_digits_available;
81   char digits_sign_char;
82 
83   /*
84    * Scan the input string, abstracting (1)digits (2)decimal mark (3) exponent.
85    * It would be simpler to modify the string, but we don't; just to be nice
86    * to caller.
87    * We need to know how many digits we have, so we can allocate space for
88    * the digits' value.
89    */
90 
91   char *p;
92   char c;
93   int seen_significant_digit;
94 
95 #ifdef ASSUME_DECIMAL_MARK_IS_DOT
96   gas_assert (string_of_decimal_marks[0] == '.'
97 	  && string_of_decimal_marks[1] == 0);
98 #define IS_DECIMAL_MARK(c)	((c) == '.')
99 #else
100 #define IS_DECIMAL_MARK(c)	(0 != strchr (string_of_decimal_marks, (c)))
101 #endif
102 
103   first_digit = *address_of_string_pointer;
104   c = *first_digit;
105 
106   if (c == '-' || c == '+')
107     {
108       digits_sign_char = c;
109       first_digit++;
110     }
111   else
112     digits_sign_char = '+';
113 
114   switch (first_digit[0])
115     {
116     case 'n':
117     case 'N':
118       if (!strncasecmp ("nan", first_digit, 3))
119 	{
120 	  address_of_generic_floating_point_number->sign = 0;
121 	  address_of_generic_floating_point_number->exponent = 0;
122 	  address_of_generic_floating_point_number->leader =
123 	    address_of_generic_floating_point_number->low;
124 	  *address_of_string_pointer = first_digit + 3;
125 	  return 0;
126 	}
127       break;
128 
129     case 'i':
130     case 'I':
131       if (!strncasecmp ("inf", first_digit, 3))
132 	{
133 	  address_of_generic_floating_point_number->sign =
134 	    digits_sign_char == '+' ? 'P' : 'N';
135 	  address_of_generic_floating_point_number->exponent = 0;
136 	  address_of_generic_floating_point_number->leader =
137 	    address_of_generic_floating_point_number->low;
138 
139 	  first_digit += 3;
140 	  if (!strncasecmp ("inity", first_digit, 5))
141 	    first_digit += 5;
142 
143 	  *address_of_string_pointer = first_digit;
144 
145 	  return 0;
146 	}
147       break;
148     }
149 
150   number_of_digits_before_decimal = 0;
151   number_of_digits_after_decimal = 0;
152   decimal_exponent = 0;
153   seen_significant_digit = 0;
154   for (p = first_digit;
155        (((c = *p) != '\0')
156 	&& (!c || !IS_DECIMAL_MARK (c))
157 	&& (!c || !strchr (string_of_decimal_exponent_marks, c)));
158        p++)
159     {
160       if (ISDIGIT (c))
161 	{
162 	  if (seen_significant_digit || c > '0')
163 	    {
164 	      ++number_of_digits_before_decimal;
165 	      seen_significant_digit = 1;
166 	    }
167 	  else
168 	    {
169 	      first_digit++;
170 	    }
171 	}
172       else
173 	{
174 	  break;		/* p -> char after pre-decimal digits.  */
175 	}
176     }				/* For each digit before decimal mark.  */
177 
178 #ifndef OLD_FLOAT_READS
179   /* Ignore trailing 0's after the decimal point.  The original code here
180      (ifdef'd out) does not do this, and numbers like
181     	4.29496729600000000000e+09	(2**31)
182      come out inexact for some reason related to length of the digit
183      string.  */
184 
185   /* The case number_of_digits_before_decimal = 0 is handled for
186      deleting zeros after decimal.  In this case the decimal mark and
187      the first zero digits after decimal mark are skipped.  */
188   seen_significant_digit = 0;
189   signed long subtract_decimal_exponent = 0;
190 
191   if (c && IS_DECIMAL_MARK (c))
192     {
193       unsigned int zeros = 0;	/* Length of current string of zeros.  */
194 
195       if (number_of_digits_before_decimal == 0)
196 	/* Skip decimal mark.  */
197 	first_digit++;
198 
199       for (p++; (c = *p) && ISDIGIT (c); p++)
200 	{
201 	  if (c == '0')
202 	    {
203 	      if (number_of_digits_before_decimal == 0
204 		  && !seen_significant_digit)
205 		{
206 		  /* Skip '0' and the decimal mark.  */
207 		  first_digit++;
208 		  subtract_decimal_exponent--;
209 		}
210 	      else
211 		zeros++;
212 	    }
213 	  else
214 	    {
215 	      seen_significant_digit = 1;
216 	      number_of_digits_after_decimal += 1 + zeros;
217 	      zeros = 0;
218 	    }
219 	}
220     }
221 #else
222   if (c && IS_DECIMAL_MARK (c))
223     {
224       for (p++;
225 	   (((c = *p) != '\0')
226 	    && (!c || !strchr (string_of_decimal_exponent_marks, c)));
227 	   p++)
228 	{
229 	  if (ISDIGIT (c))
230 	    {
231 	      /* This may be retracted below.  */
232 	      number_of_digits_after_decimal++;
233 
234 	      if ( /* seen_significant_digit || */ c > '0')
235 		{
236 		  seen_significant_digit = true;
237 		}
238 	    }
239 	  else
240 	    {
241 	      if (!seen_significant_digit)
242 		{
243 		  number_of_digits_after_decimal = 0;
244 		}
245 	      break;
246 	    }
247 	}			/* For each digit after decimal mark.  */
248     }
249 
250   while (number_of_digits_after_decimal
251 	 && first_digit[number_of_digits_before_decimal
252 			+ number_of_digits_after_decimal] == '0')
253     --number_of_digits_after_decimal;
254 #endif
255 
256   if (flag_m68k_mri)
257     {
258       while (c == '_')
259 	c = *++p;
260     }
261   if (c && strchr (string_of_decimal_exponent_marks, c))
262     {
263       char digits_exponent_sign_char;
264 
265       c = *++p;
266       if (flag_m68k_mri)
267 	{
268 	  while (c == '_')
269 	    c = *++p;
270 	}
271       if (c && strchr ("+-", c))
272 	{
273 	  digits_exponent_sign_char = c;
274 	  c = *++p;
275 	}
276       else
277 	{
278 	  digits_exponent_sign_char = '+';
279 	}
280 
281       for (; (c); c = *++p)
282 	{
283 	  if (ISDIGIT (c))
284 	    {
285 	      decimal_exponent = decimal_exponent * 10 + c - '0';
286 	      /*
287 	       * BUG! If we overflow here, we lose!
288 	       */
289 	    }
290 	  else
291 	    {
292 	      break;
293 	    }
294 	}
295 
296       if (digits_exponent_sign_char == '-')
297 	{
298 	  decimal_exponent = -decimal_exponent;
299 	}
300     }
301 
302 #ifndef OLD_FLOAT_READS
303   /* Subtract_decimal_exponent != 0 when number_of_digits_before_decimal = 0
304      and first digit after decimal is '0'.  */
305   decimal_exponent += subtract_decimal_exponent;
306 #endif
307 
308   *address_of_string_pointer = p;
309 
310   number_of_digits_available =
311     number_of_digits_before_decimal + number_of_digits_after_decimal;
312   return_value = 0;
313   if (number_of_digits_available == 0)
314     {
315       address_of_generic_floating_point_number->exponent = 0;	/* Not strictly necessary */
316       address_of_generic_floating_point_number->leader
317 	= -1 + address_of_generic_floating_point_number->low;
318       address_of_generic_floating_point_number->sign = digits_sign_char;
319       /* We have just concocted (+/-)0.0E0 */
320 
321     }
322   else
323     {
324       int count;		/* Number of useful digits left to scan.  */
325 
326       LITTLENUM_TYPE *temporary_binary_low = NULL;
327       LITTLENUM_TYPE *power_binary_low = NULL;
328       LITTLENUM_TYPE *digits_binary_low;
329       unsigned int precision;
330       unsigned int maximum_useful_digits;
331       unsigned int number_of_digits_to_use;
332       unsigned int more_than_enough_bits_for_digits;
333       unsigned int more_than_enough_littlenums_for_digits;
334       unsigned int size_of_digits_in_littlenums;
335       unsigned int size_of_digits_in_chars;
336       FLONUM_TYPE power_of_10_flonum;
337       FLONUM_TYPE digits_flonum;
338 
339       precision = (address_of_generic_floating_point_number->high
340 		   - address_of_generic_floating_point_number->low
341 		   + 1);	/* Number of destination littlenums.  */
342 
343       /* precision includes two littlenums worth of guard bits,
344 	 so this gives us 10 decimal guard digits here.  */
345       maximum_useful_digits = (precision
346 			       * LITTLENUM_NUMBER_OF_BITS
347 			       * 1000000 / 3321928
348 			       + 1);	/* round up.  */
349 
350       if (number_of_digits_available > maximum_useful_digits)
351 	{
352 	  number_of_digits_to_use = maximum_useful_digits;
353 	}
354       else
355 	{
356 	  number_of_digits_to_use = number_of_digits_available;
357 	}
358 
359       /* Cast these to SIGNED LONG first, otherwise, on systems with
360 	 LONG wider than INT (such as Alpha OSF/1), unsignedness may
361 	 cause unexpected results.  */
362       decimal_exponent += ((long) number_of_digits_before_decimal
363 			   - (long) number_of_digits_to_use);
364 
365       more_than_enough_bits_for_digits
366 	= (number_of_digits_to_use * 3321928 / 1000000 + 1);
367 
368       more_than_enough_littlenums_for_digits
369 	= (more_than_enough_bits_for_digits
370 	   / LITTLENUM_NUMBER_OF_BITS)
371 	+ 2;
372 
373       /* Compute (digits) part. In "12.34E56" this is the "1234" part.
374 	 Arithmetic is exact here. If no digits are supplied then this
375 	 part is a 0 valued binary integer.  Allocate room to build up
376 	 the binary number as littlenums.  We want this memory to
377 	 disappear when we leave this function.  Assume no alignment
378 	 problems => (room for n objects) == n * (room for 1
379 	 object).  */
380 
381       size_of_digits_in_littlenums = more_than_enough_littlenums_for_digits;
382       size_of_digits_in_chars = size_of_digits_in_littlenums
383 	* sizeof (LITTLENUM_TYPE);
384 
385       digits_binary_low = (LITTLENUM_TYPE *)
386 	xmalloc (size_of_digits_in_chars);
387 
388       memset ((char *) digits_binary_low, '\0', size_of_digits_in_chars);
389 
390       /* Digits_binary_low[] is allocated and zeroed.  */
391 
392       /*
393        * Parse the decimal digits as if * digits_low was in the units position.
394        * Emit a binary number into digits_binary_low[].
395        *
396        * Use a large-precision version of:
397        * (((1st-digit) * 10 + 2nd-digit) * 10 + 3rd-digit ...) * 10 + last-digit
398        */
399 
400       for (p = first_digit, count = number_of_digits_to_use; count; p++, --count)
401 	{
402 	  c = *p;
403 	  if (ISDIGIT (c))
404 	    {
405 	      /*
406 	       * Multiply by 10. Assume can never overflow.
407 	       * Add this digit to digits_binary_low[].
408 	       */
409 
410 	      long carry;
411 	      LITTLENUM_TYPE *littlenum_pointer;
412 	      LITTLENUM_TYPE *littlenum_limit;
413 
414 	      littlenum_limit = digits_binary_low
415 		+ more_than_enough_littlenums_for_digits
416 		- 1;
417 
418 	      carry = c - '0';	/* char -> binary */
419 
420 	      for (littlenum_pointer = digits_binary_low;
421 		   littlenum_pointer <= littlenum_limit;
422 		   littlenum_pointer++)
423 		{
424 		  long work;
425 
426 		  work = carry + 10 * (long) (*littlenum_pointer);
427 		  *littlenum_pointer = work & LITTLENUM_MASK;
428 		  carry = work >> LITTLENUM_NUMBER_OF_BITS;
429 		}
430 
431 	      if (carry != 0)
432 		{
433 		  /*
434 		   * We have a GROSS internal error.
435 		   * This should never happen.
436 		   */
437 		  as_fatal (_("failed sanity check"));
438 		}
439 	    }
440 	  else
441 	    {
442 	      ++count;		/* '.' doesn't alter digits used count.  */
443 	    }
444 	}
445 
446       /*
447        * Digits_binary_low[] properly encodes the value of the digits.
448        * Forget about any high-order littlenums that are 0.
449        */
450       while (digits_binary_low[size_of_digits_in_littlenums - 1] == 0
451 	     && size_of_digits_in_littlenums >= 2)
452 	size_of_digits_in_littlenums--;
453 
454       digits_flonum.low = digits_binary_low;
455       digits_flonum.high = digits_binary_low + size_of_digits_in_littlenums - 1;
456       digits_flonum.leader = digits_flonum.high;
457       digits_flonum.exponent = 0;
458       /*
459        * The value of digits_flonum . sign should not be important.
460        * We have already decided the output's sign.
461        * We trust that the sign won't influence the other parts of the number!
462        * So we give it a value for these reasons:
463        * (1) courtesy to humans reading/debugging
464        *     these numbers so they don't get excited about strange values
465        * (2) in future there may be more meaning attached to sign,
466        *     and what was
467        *     harmless noise may become disruptive, ill-conditioned (or worse)
468        *     input.
469        */
470       digits_flonum.sign = '+';
471 
472       {
473 	/*
474 	 * Compute the mantissa (& exponent) of the power of 10.
475 	 * If successful, then multiply the power of 10 by the digits
476 	 * giving return_binary_mantissa and return_binary_exponent.
477 	 */
478 
479 	int decimal_exponent_is_negative;
480 	/* This refers to the "-56" in "12.34E-56".  */
481 	/* FALSE: decimal_exponent is positive (or 0) */
482 	/* TRUE:  decimal_exponent is negative */
483 	FLONUM_TYPE temporary_flonum;
484 	unsigned int size_of_power_in_littlenums;
485 	unsigned int size_of_power_in_chars;
486 
487 	size_of_power_in_littlenums = precision;
488 	/* Precision has a built-in fudge factor so we get a few guard bits.  */
489 
490 	decimal_exponent_is_negative = decimal_exponent < 0;
491 	if (decimal_exponent_is_negative)
492 	  {
493 	    decimal_exponent = -decimal_exponent;
494 	  }
495 
496 	/* From now on: the decimal exponent is > 0. Its sign is separate.  */
497 
498 	size_of_power_in_chars = size_of_power_in_littlenums
499 	  * sizeof (LITTLENUM_TYPE) + 2;
500 
501 	power_binary_low = (LITTLENUM_TYPE *) xmalloc (size_of_power_in_chars);
502 	temporary_binary_low = (LITTLENUM_TYPE *) xmalloc (size_of_power_in_chars);
503 
504 	memset ((char *) power_binary_low, '\0', size_of_power_in_chars);
505 	*power_binary_low = 1;
506 	power_of_10_flonum.exponent = 0;
507 	power_of_10_flonum.low = power_binary_low;
508 	power_of_10_flonum.leader = power_binary_low;
509 	power_of_10_flonum.high = power_binary_low + size_of_power_in_littlenums - 1;
510 	power_of_10_flonum.sign = '+';
511 	temporary_flonum.low = temporary_binary_low;
512 	temporary_flonum.high = temporary_binary_low + size_of_power_in_littlenums - 1;
513 	/*
514 	 * (power) == 1.
515 	 * Space for temporary_flonum allocated.
516 	 */
517 
518 	/*
519 	 * ...
520 	 *
521 	 * WHILE	more bits
522 	 * DO	find next bit (with place value)
523 	 *	multiply into power mantissa
524 	 * OD
525 	 */
526 	{
527 	  int place_number_limit;
528 	  /* Any 10^(2^n) whose "n" exceeds this */
529 	  /* value will fall off the end of */
530 	  /* flonum_XXXX_powers_of_ten[].  */
531 	  int place_number;
532 	  const FLONUM_TYPE *multiplicand;	/* -> 10^(2^n) */
533 
534 	  place_number_limit = table_size_of_flonum_powers_of_ten;
535 
536 	  multiplicand = (decimal_exponent_is_negative
537 			  ? flonum_negative_powers_of_ten
538 			  : flonum_positive_powers_of_ten);
539 
540 	  for (place_number = 1;/* Place value of this bit of exponent.  */
541 	       decimal_exponent;/* Quit when no more 1 bits in exponent.  */
542 	       decimal_exponent >>= 1, place_number++)
543 	    {
544 	      if (decimal_exponent & 1)
545 		{
546 		  if (place_number > place_number_limit)
547 		    {
548 		      /* The decimal exponent has a magnitude so great
549 			 that our tables can't help us fragment it.
550 			 Although this routine is in error because it
551 			 can't imagine a number that big, signal an
552 			 error as if it is the user's fault for
553 			 presenting such a big number.  */
554 		      return_value = ERROR_EXPONENT_OVERFLOW;
555 		      /* quit out of loop gracefully */
556 		      decimal_exponent = 0;
557 		    }
558 		  else
559 		    {
560 #ifdef TRACE
561 		      printf ("before multiply, place_number = %d., power_of_10_flonum:\n",
562 			      place_number);
563 
564 		      flonum_print (&power_of_10_flonum);
565 		      (void) putchar ('\n');
566 #endif
567 #ifdef TRACE
568 		      printf ("multiplier:\n");
569 		      flonum_print (multiplicand + place_number);
570 		      (void) putchar ('\n');
571 #endif
572 		      flonum_multip (multiplicand + place_number,
573 				     &power_of_10_flonum, &temporary_flonum);
574 #ifdef TRACE
575 		      printf ("after multiply:\n");
576 		      flonum_print (&temporary_flonum);
577 		      (void) putchar ('\n');
578 #endif
579 		      flonum_copy (&temporary_flonum, &power_of_10_flonum);
580 #ifdef TRACE
581 		      printf ("after copy:\n");
582 		      flonum_print (&power_of_10_flonum);
583 		      (void) putchar ('\n');
584 #endif
585 		    } /* If this bit of decimal_exponent was computable.*/
586 		} /* If this bit of decimal_exponent was set.  */
587 	    } /* For each bit of binary representation of exponent */
588 #ifdef TRACE
589 	  printf ("after computing power_of_10_flonum:\n");
590 	  flonum_print (&power_of_10_flonum);
591 	  (void) putchar ('\n');
592 #endif
593 	}
594       }
595 
596       /*
597        * power_of_10_flonum is power of ten in binary (mantissa) , (exponent).
598        * It may be the number 1, in which case we don't NEED to multiply.
599        *
600        * Multiply (decimal digits) by power_of_10_flonum.
601        */
602 
603       flonum_multip (&power_of_10_flonum, &digits_flonum, address_of_generic_floating_point_number);
604       /* Assert sign of the number we made is '+'.  */
605       address_of_generic_floating_point_number->sign = digits_sign_char;
606 
607       free (temporary_binary_low);
608       free (power_binary_low);
609       free (digits_binary_low);
610     }
611   return return_value;
612 }
613 
614 #ifdef TRACE
615 static void
flonum_print(f)616 flonum_print (f)
617      const FLONUM_TYPE *f;
618 {
619   LITTLENUM_TYPE *lp;
620   char littlenum_format[10];
621   sprintf (littlenum_format, " %%0%dx", sizeof (LITTLENUM_TYPE) * 2);
622 #define print_littlenum(LP)	(printf (littlenum_format, LP))
623   printf ("flonum @%p %c e%ld", f, f->sign, f->exponent);
624   if (f->low < f->high)
625     for (lp = f->high; lp >= f->low; lp--)
626       print_littlenum (*lp);
627   else
628     for (lp = f->low; lp <= f->high; lp++)
629       print_littlenum (*lp);
630   printf ("\n");
631   fflush (stdout);
632 }
633 #endif
634 
635 /* end of atof_generic.c */
636