summaryrefslogtreecommitdiff
path: root/gcc/floatlib.c
blob: bdcdf57ddd0b9ea42549f21500462766964fc098 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
/*
** libgcc support for software floating point.
** Copyright (C) 1991 by Pipeline Associates, Inc.  All rights reserved.
** Permission is granted to do *anything* you want with this file,
** commercial or otherwise, provided this message remains intact.  So there!
** I would appreciate receiving any updates/patches/changes that anyone
** makes, and am willing to be the repository for said changes (am I
** making a big mistake?).

Warning! Only single-precision is actually implemented.  This file
won't really be much use until double-precision is supported.

However, once that is done, this file might eventually become a
replacement for libgcc1.c.  It might also make possible
cross-compilation for an IEEE target machine from a non-IEEE
host such as a VAX.

If you'd like to work on completing this, please talk to rms@gnu.ai.mit.edu.


**
** Pat Wood
** Pipeline Associates, Inc.
** pipeline!phw@motown.com or
** sun!pipeline!phw or
** uunet!motown!pipeline!phw
**
** 05/01/91 -- V1.0 -- first release to gcc mailing lists
** 05/04/91 -- V1.1 -- added float and double prototypes and return values
**                  -- fixed problems with adding and subtracting zero
**                  -- fixed rounding in truncdfsf2
**                  -- fixed SWAP define and tested on 386
*/

/*
** The following are routines that replace the libgcc soft floating point
** routines that are called automatically when -msoft-float is selected.
** The support single and double precision IEEE format, with provisions
** for byte-swapped machines (tested on 386).  Some of the double-precision
** routines work at full precision, but most of the hard ones simply punt
** and call the single precision routines, producing a loss of accuracy.
** long long support is not assumed or included.
** Overall accuracy is close to IEEE (actually 68882) for single-precision
** arithmetic.  I think there may still be a 1 in 1000 chance of a bit
** being rounded the wrong way during a multiply.  I'm not fussy enough to
** bother with it, but if anyone is, knock yourself out.
**
** Efficiency has only been addressed where it was obvious that something
** would make a big difference.  Anyone who wants to do this right for
** best speed should go in and rewrite in assembler.
**
** I have tested this only on a 68030 workstation and 386/ix integrated
** in with -msoft-float.
*/

/* the following deal with IEEE single-precision numbers */
#define D_PHANTOM_BIT   0x00100000
#define EXCESS		126
#define SIGNBIT		0x80000000
#define HIDDEN		(1 << 23)
#define SIGN(fp)	((fp) & SIGNBIT)
#define EXP(fp)		(((fp) >> 23) & 0xFF)
#define MANT(fp)	(((fp) & 0x7FFFFF) | HIDDEN)
#define PACK(s,e,m)	((s) | ((e) << 23) | (m))

/* the following deal with IEEE double-precision numbers */
#define EXCESSD		1022
#define HIDDEND		(1 << 20)
#define EXPD(fp)	(((fp.l.upper) >> 20) & 0x7FF)
#define SIGND(fp)	((fp.l.upper) & SIGNBIT)
#define MANTD(fp)	(((((fp.l.upper) & 0xFFFFF) | HIDDEND) << 10) | \
				(fp.l.lower >> 22))

/* define SWAP for 386/960 reverse-byte-order brain-damaged CPUs */
union double_long
  {
    double d;
#ifdef SWAP
    struct {
      unsigned long lower;
      long upper;
    } l;
#else
    struct {
      long upper;
      unsigned long lower;
    } l;
#endif
  };

union float_long
  {
    float f;
    long l;
  };

   struct _ieee {
#ifdef SWAP
      unsigned mantissa2 : 32;
      unsigned mantissa1 : 20;
      unsigned exponent  : 11;
      unsigned sign      : 1;
#else
      unsigned exponent  : 11;
      unsigned sign      : 1;
      unsigned mantissa2 : 32;
      unsigned mantissa1 : 20;
#endif
   };

   union _doubleu {
      double d;
      struct _ieee ieee;
#ifdef SWAP
      struct {
         unsigned long lower;
         long upper;
      } l;
#else
      struct {
         long upper;
         unsigned long lower;
      } l;
#endif
   };

/* add two floats */

float
__addsf3 (float a1, float a2)
{
  register long mant1, mant2;
  register union float_long fl1, fl2;
  register int exp1, exp2;
  int sign = 0;

  fl1.f = a1;
  fl2.f = a2;

  /* check for zero args */
  if (!fl1.l)
    return (fl2.f);
  if (!fl2.l)
    return (fl1.f);

  exp1 = EXP (fl1.l);
  exp2 = EXP (fl2.l);

  if (exp1 > exp2 + 25)
    return (fl1.l);
  if (exp2 > exp1 + 25)
    return (fl2.l);

  /* do everything in excess precision so's we can round later */
  mant1 = MANT (fl1.l) << 6;
  mant2 = MANT (fl2.l) << 6;

  if (SIGN (fl1.l))
    mant1 = -mant1;
  if (SIGN (fl2.l))
    mant2 = -mant2;

  if (exp1 > exp2)
    {
      mant2 >>= exp1 - exp2;
    }
  else
    {
      mant1 >>= exp2 - exp1;
      exp1 = exp2;
    }
  mant1 += mant2;

  if (mant1 < 0)
    {
      mant1 = -mant1;
      sign = SIGNBIT;
    }
  else if (!mant1)
    return (0);

  /* normalize up */
  while (!(mant1 & 0xE0000000))
    {
      mant1 <<= 1;
      exp1--;
    }

  /* normalize down? */
  if (mant1 & (1 << 30))
    {
      mant1 >>= 1;
      exp1++;
    }

  /* round to even */
  mant1 += (mant1 & 0x40) ? 0x20 : 0x1F;

  /* normalize down? */
  if (mant1 & (1 << 30))
    {
      mant1 >>= 1;
      exp1++;
    }

  /* lose extra precision */
  mant1 >>= 6;

  /* turn off hidden bit */
  mant1 &= ~HIDDEN;

  /* pack up and go home */
  fl1.l = PACK (sign, exp1, mant1);
  return (fl1.f);
}

/* subtract two floats */

float
__subsf3 (float a1, float a2)
{
  register union float_long fl1, fl2;

  fl1.f = a1;
  fl2.f = a2;

  /* check for zero args */
  if (!fl2.l)
    return (fl1.f);
  if (!fl1.l)
    return (-fl2.f);

  /* twiddle sign bit and add */
  fl2.l ^= SIGNBIT;
  return __addsf3 (a1, fl2.f);
}

/* compare two floats */

long
__cmpsf2 (float a1, float a2)
{
  register union float_long fl1, fl2;

  fl1.f = a1;
  fl2.f = a2;

  if (SIGN (fl1.l) && SIGN (fl2.l))
    {
      fl1.l ^= SIGNBIT;
      fl2.l ^= SIGNBIT;
    }
  if (fl1.l < fl2.l)
    return (-1);
  if (fl1.l > fl2.l)
    return (1);
  return (0);
}

/* multiply two floats */

float
__mulsf3 (float a1, float a2)
{
  register union float_long fl1, fl2;
  register unsigned long result;
  register int exp;
  int sign;

  fl1.f = a1;
  fl2.f = a2;

  if (!fl1.l || !fl2.l)
    return (0);

  /* compute sign and exponent */
  sign = SIGN (fl1.l) ^ SIGN (fl2.l);
  exp = EXP (fl1.l) - EXCESS;
  exp += EXP (fl2.l);

  fl1.l = MANT (fl1.l);
  fl2.l = MANT (fl2.l);

  /* the multiply is done as one 16x16 multiply and two 16x8 multiples */
  result = (fl1.l >> 8) * (fl2.l >> 8);
  result += ((fl1.l & 0xFF) * (fl2.l >> 8)) >> 8;
  result += ((fl2.l & 0xFF) * (fl1.l >> 8)) >> 8;

  if (result & 0x80000000)
    {
      /* round */
      result += 0x80;
      result >>= 8;
    }
  else
    {
      /* round */
      result += 0x40;
      result >>= 7;
      exp--;
    }

  result &= ~HIDDEN;

  /* pack up and go home */
  fl1.l = PACK (sign, exp, result);
  return (fl1.f);
}

/* divide two floats */

float
__divsf3 (float a1, float a2)
{
  register union float_long fl1, fl2;
  register int result;
  register int mask;
  register int exp, sign;

  fl1.f = a1;
  fl2.f = a2;

  /* subtract exponents */
  exp = EXP (fl1.l) - EXP (fl2.l) + EXCESS;

  /* compute sign */
  sign = SIGN (fl1.l) ^ SIGN (fl2.l);

  /* divide by zero??? */
  if (!fl2.l)
    /* return NaN or -NaN */
    return (sign ? 0xFFFFFFFF : 0x7FFFFFFF);

  /* numerator zero??? */
  if (!fl1.l)
    return (0);

  /* now get mantissas */
  fl1.l = MANT (fl1.l);
  fl2.l = MANT (fl2.l);

  /* this assures we have 25 bits of precision in the end */
  if (fl1.l < fl2.l)
    {
      fl1.l <<= 1;
      exp--;
    }

  /* now we perform repeated subtraction of fl2.l from fl1.l */
  mask = 0x1000000;
  result = 0;
  while (mask)
    {
      if (fl1.l >= fl2.l)
	{
	  result |= mask;
	  fl1.l -= fl2.l;
	}
      fl1.l <<= 1;
      mask >>= 1;
    }

  /* round */
  result += 1;

  /* normalize down */
  exp++;
  result >>= 1;

  result &= ~HIDDEN;

  /* pack up and go home */
  fl1.l = PACK (sign, exp, result);
  return (fl1.f);
}

/* convert int to double */

double
__floatsidf (register long a1)
{
  register int sign = 0, exp = 31 + EXCESSD;
  union double_long dl;

  if (!a1)
    {
      dl.l.upper = dl.l.lower = 0;
      return (dl.d);
    }

  if (a1 < 0)
    {
      sign = SIGNBIT;
      a1 = -a1;
    }

  while (a1 < 0x1000000)
    {
      a1 <<= 4;
      exp -= 4;
    }

  while (a1 < 0x40000000)
    {
      a1 <<= 1;
      exp--;
    }

  /* pack up and go home */
  dl.l.upper = sign;
  dl.l.upper |= exp << 20;
  dl.l.upper |= (a1 >> 10) & ~HIDDEND;
  dl.l.lower = a1 << 22;

  return (dl.d);
}

/* negate a float */

float
__negsf2 (float a1)
{
  register union float_long fl1;

  fl1.f = a1;
  if (!fl1.l)
    return (0);

  fl1.l ^= SIGNBIT;
  return (fl1.f);
}

/* negate a double */

double
__negdf2 (double a1)
{
  register union double_long dl1;

  dl1.d = a1;

  if (!dl1.l.upper && !dl1.l.lower)
      return (dl1.d);

  dl1.l.upper ^= SIGNBIT;
  return (dl1.d);
}

/* convert float to double */

double
__extendsfdf2 (float a1)
{
  register union float_long fl1;
  register union double_long dl;
  register int exp;

  fl1.f = a1;

  if (!fl1.l)
    {
      dl.l.upper = dl.l.lower = 0;
      return (dl.d);
    }

  dl.l.upper = SIGN (fl1.l);
  exp = EXP (fl1.l) - EXCESS + EXCESSD;
  dl.l.upper |= exp << 20;
  dl.l.upper |= (MANT (fl1.l) & ~HIDDEN) >> 3;
  dl.l.lower = MANT (fl1.l) << 29;

  return (dl.d);
}

/* convert double to float */

float
__truncdfsf2 (double a1)
{
  register int exp;
  register long mant;
  register union float_long fl;
  register union double_long dl1;

  dl1.d = a1;

  if (!dl1.l.upper && !dl1.l.lower)
    return (0);

  exp = EXPD (dl1) - EXCESSD + EXCESS;

  /* shift double mantissa 6 bits so we can round */
  mant = MANTD (dl1) >> 6;

  /* now round and shift down */
  mant += 1;
  mant >>= 1;

  /* did the round overflow? */
  if (mant & 0xFF000000)
    {
      mant >>= 1;
      exp++;
    }

  mant &= ~HIDDEN;

  /* pack up and go home */
  fl.l = PACK (SIGND (dl1), exp, mant);
  return (fl.f);
}

/* compare two doubles */

long
__cmpdf2 (double a1, double a2)
{
  register union double_long dl1, dl2;

  dl1.d = a1;
  dl2.d = a2;

  if (SIGND (dl1) && SIGND (dl2))
    {
      dl1.l.upper ^= SIGNBIT;
      dl2.l.upper ^= SIGNBIT;
    }
  if (dl1.l.upper < dl2.l.upper)
    return (-1);
  if (dl1.l.upper > dl2.l.upper)
    return (1);
  if (dl1.l.lower < dl2.l.lower)
    return (-1);
  if (dl1.l.lower > dl2.l.lower)
    return (1);
  return (0);
}

/* convert double to int */

long
__fixdfsi (double a1)
{
  register union double_long dl1;
  register int exp;
  register long l;

  dl1.d = a1;

  if (!dl1.l.upper && !dl1.l.lower)
    return (0);

  exp = EXPD (dl1) - EXCESSD - 31;
  l = MANTD (dl1);

  if (exp > 0)
    return (0x7FFFFFFF | SIGND (dl1)); /* largest integer */

  /* shift down until exp = 0 or l = 0 */
  if (exp < 0 && exp > -32 && l)
    l >>= -exp;
  else
    return (0);

  return (SIGND (dl1) ? -l : l);
}

/* convert double to unsigned int */

unsigned
long __fixunsdfsi (double a1)
{
  register union double_long dl1;
  register int exp;
  register unsigned long l;

  dl1.d = a1;

  if (!dl1.l.upper && !dl1.l.lower)
    return (0);

  exp = EXPD (dl1) - EXCESSD - 32;
  l = (((((dl1.l.upper) & 0xFFFFF) | HIDDEND) << 11) | (dl1.l.lower >> 21));

  if (exp > 0)
    return (0xFFFFFFFF);	/* largest integer */

  /* shift down until exp = 0 or l = 0 */
  if (exp < 0 && exp > -32 && l)
    l >>= -exp;
  else
    return (0);

  return (l);
}

/* For now, the hard double-precision routines simply
   punt and do it in single */
/* addtwo doubles */

double
__adddf3 (double a1, double a2)
{
  return ((float) a1 + (float) a2);
}

/* subtract two doubles */

double
__subdf3 (double a1, double a2)
{
  return ((float) a1 - (float) a2);
}

/* multiply two doubles */

double
__muldf3 (double a1, double a2)
{
  return ((float) a1 * (float) a2);
}

/*
 *
 * Name:   Barrett Richardson
 * E-mail: barrett@iglou.com
 * When:   Thu Dec 15 10:31:11 EST 1994
 *
 *    callable function:
 *
 *       double __divdf3(double a1, double a2);
 *
 *       Does software divide of a1 / a2.
 *
 *       Based largely on __divsf3() in floatlib.c in the gcc
 *       distribution.
 *
 *    Purpose: To be used in conjunction with the -msoft-float
 *             option of gcc. You should be able to tack it to the
 *             end of floatlib.c included in the gcc distribution,
 *             and delete the __divdf3() already there which just 
 *             calls the single precision function (or may just
 *             use the floating point processor with some configurations).
 *
 *   You may use this code for whatever your heart desires.
 */




/*
 * Compare the the mantissas of two doubles.
 * Each mantissa is in two longs.
 * 
 *   return      1   if x1's mantissa is greater than x2's
 *              -1   if x1's mantissa is less than x2's
 *               0   if the two mantissa's are equal.
 *
 *   The Mantissas won't fit into a 4 byte word, so they are
 *   broken up into two parts.
 *
 *   This function is used internally by __divdf3()
 */

int
__dcmp (long x1m1, long x1m2, long x2m1, long x2m2)
{
   if (x1m1 > x2m1)
      return 1;

   if (x1m1 < x2m1)
      return -1;

 /*  If the first word in the two mantissas were equal check the second word */

   if (x1m2 > x2m2)
      return 1;

   if (x1m2 < x2m2)
      return -1;

   return 0;
}


/* divide two doubles */

double
__divdf3 (double a1, double a2)
{

   int  sign,
        exponent,
        bit_bucket;

   register unsigned long mantissa1,
                          mantissa2,
                          x1m1,
                          x1m2,
                          x2m1,
                          x2m2,
                          mask;

   union _doubleu x1,
                  x2,
                  result;


   x1.d = a1;
   x2.d = a2;

   exponent = x1.ieee.exponent - x2.ieee.exponent + EXCESSD;

   sign = x1.ieee.sign ^ x2.ieee.sign;

   x2.ieee.sign = 0;  /* don't want the sign bit to affect any zero */
                      /* comparisons when checking for zero divide  */

   if (!x2.l.lower && !x2.l.upper) { /* check for zero divide */
      result.l.lower = 0x0;
      if (sign)
         result.l.upper = 0xFFF00000;   /* negative infinity */
      else
         result.l.upper = 0x7FF00000;   /* positive infinity */
      return result.d;
   }

   if (!x1.l.upper && !x1.l.lower)  /* check for 0.0 numerator */
      return (0.0);

   x1m1 = x1.ieee.mantissa1 | D_PHANTOM_BIT;  /* turn on phantom bit */
   x1m2 = x1.ieee.mantissa2;

   x2m1 = x2.ieee.mantissa1 | D_PHANTOM_BIT;  /* turn on phantom bit */
   x2m2 = x2.ieee.mantissa2;

   if (__dcmp(x1m1,x1m2,x2m1,x2m2) < 0) {

   /* if x1's mantissa is less than x2's shift it left one and decrement */
   /* the exponent to accomodate the change in the mantissa              */

      x1m1 <<= 1;               /*                          */
      bit_bucket = x1m2 >> 31;  /*  Shift mantissa left one */
      x1m1 |= bit_bucket;       /*                          */
      x1m2 <<= 1;               /*                          */

      exponent--;
   }


   mantissa1 = 0;
   mantissa2 = 0;


  /* Get the first part of the results mantissa using successive */
  /* subtraction.                                                */

   mask = 0x00200000;
   while (mask) {

      if (__dcmp(x1m1,x1m2,x2m1,x2m2) >= 0) {

     /* subtract x2's mantissa from x1's */

         mantissa1 |= mask;   /* turn on a bit in the result */

         if (x2m2 > x1m2)
            x1m1--;
         x1m2 -= x2m2; 
         x1m1 -= x2m1;
      }

      x1m1 <<= 1;               /*                          */
      bit_bucket = x1m2 >> 31;  /*  Shift mantissa left one */
      x1m1 |= bit_bucket;       /*                          */
      x1m2 <<= 1;               /*                          */

      mask >>= 1;
   }

  /* Get the second part of the results mantissa using successive */
  /* subtraction.                                                 */

   mask = 0x80000000;
   while (mask) {

      if (__dcmp(x1m1,x1m2,x2m1,x2m2) >= 0) {

     /* subtract x2's mantissa from x1's */

         mantissa2 |= mask;   /* turn on a bit in the result */

         if (x2m2 > x1m2)
            x1m1--;
         x1m2 -= x2m2; 
         x1m1 -= x2m1;
      }
      x1m1 <<= 1;               /*                          */
      bit_bucket = x1m2 >> 31;  /*  Shift mantissa left one */
      x1m1 |= bit_bucket;       /*                          */
      x1m2 <<= 1;               /*                          */

      mask >>= 1;
   }

  /* round up by adding 1 to mantissa */

   if (mantissa2 == 0xFFFFFFFF) {  /* check for over flow */

   /* spill if overflow */

      mantissa2 = 0;
      mantissa1++;
   }
   else
      mantissa2++;

   exponent++;   /* increment exponent (mantissa must be shifted right */
                 /* also)                                              */

 /* shift mantissa right one and assume a phantom bit (which really gives */
 /* 53 bits of precision in the mantissa)                                 */

   mantissa2 >>= 1;
   bit_bucket = mantissa1 & 1;
   mantissa2 |= (bit_bucket << 31);
   mantissa1 >>= 1;

  /* put all the info into the result */

   result.ieee.exponent  = exponent;
   result.ieee.sign      = sign;
   result.ieee.mantissa1 = mantissa1;
   result.ieee.mantissa2 = mantissa2;


   return result.d;
}