aboutsummaryrefslogtreecommitdiff
path: root/sim/ppc/dp-bit.c
blob: d1c51583beecf43ccdf1e29a9744d8b51e7a06eb (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
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
/* 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, 2007, 2008 Free Software Foundation, Inc.

This program 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 3 of the License, or
(at your option) any later version.

This program 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.  If not, see <http://www.gnu.org/licenses/>.  */

/* 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.  */

/* 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 */

#ifndef SFtype
typedef SFtype __attribute__ ((mode (SF)));
#endif
#ifndef DFtype
typedef DFtype __attribute__ ((mode (DF)));
#endif

#ifndef HItype
typedef int HItype __attribute__ ((mode (HI)));
#endif
#ifndef SItype
typedef int SItype __attribute__ ((mode (SI)));
#endif
#ifndef DItype
typedef int DItype __attribute__ ((mode (DI)));
#endif

/* The type of the result of a fp compare */
#ifndef CMPtype
#define CMPtype SItype
#endif

#ifndef UHItype
typedef unsigned int UHItype __attribute__ ((mode (HI)));
#endif
#ifndef USItype
typedef unsigned int USItype __attribute__ ((mode (SI)));
#endif
#ifndef UDItype
typedef unsigned int UDItype __attribute__ ((mode (DI)));
#endif

#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
	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
	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::NGARGS-(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;
#ifdef _DEBUG_BITFLOAT
  int l[2];
#endif
  struct
    {
#ifndef FLOAT_BIT_ORDER_MISMATCH
      unsigned int sign:1 __attribute__ ((packed));
      unsigned int exp:EXPBITS __attribute__ ((packed));
      fractype fraction:FRACBITS __attribute__ ((packed));
#else
      fractype fraction:FRACBITS __attribute__ ((packed));
      unsigned int exp:EXPBITS __attribute__ ((packed));
      unsigned int sign:1 __attribute__ ((packed));
#endif
    }
  bits;
}
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;
}

static FLO_type
pack_d ( fp_number_type *  src)
{
  FLO_union_type dst;
  fractype fraction = src->fraction.ll;	/* wasn't unsigned before? */

  dst.bits.sign = src->sign;

  if (isnan (src))
    {
      dst.bits.exp = EXPMAX;
      dst.bits.fraction = src->fraction.ll;
      if (src->class == CLASS_QNAN || 1)
	{
	  dst.bits.fraction |= QUIET_NAN;
	}
    }
  else if (isinf (src))
    {
      dst.bits.exp = EXPMAX;
      dst.bits.fraction = 0;
    }
  else if (iszero (src))
    {
      dst.bits.exp = 0;
      dst.bits.fraction = 0;
    }
  else if (fraction == 0)
    {
      dst.value = 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;

	  dst.bits.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;
	  dst.bits.fraction = fraction;
	}
      else if (src->normal_exp > EXPBIAS)
	{
	  dst.bits.exp = EXPMAX;
	  dst.bits.fraction = 0;
	}
      else
	{
	  dst.bits.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;
	      dst.bits.exp += 1;
	    }
	  fraction >>= NGARDS;
	  dst.bits.fraction = fraction;
	}
    }
  return dst.value;
}

static void
unpack_d (FLO_union_type * src, fp_number_type * dst)
{
  fractype fraction = src->bits.fraction;

  dst->sign = src->bits.sign;
  if (src->bits.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 = src->bits.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 (src->bits.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 (dst->sign)
	    {
	      dst->class = CLASS_SNAN;
	    }
	  else
	    {
	      dst->class = CLASS_QNAN;
	    }
	  /* Keep the fraction part as the nan number */
	  dst->fraction.ll = fraction;
	}
    }
  else
    {
      /* Nothing strange about this number */
      dst->normal_exp = src->bits.exp - EXPBIAS;
      dst->class = CLASS_NUMBER;
      dst->fraction.ll = (fraction << NGARDS) | IMPLICIT_1;
    }
}

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))
    {
      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);
}

static 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 */
  {
    fractype x = a->fraction.ll;
    fractype ylow = b->fraction.ll;
    fractype yhigh = 0;
    int bit;

#if defined(NO_DI_MODE)
    {
      /* ??? 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);
}

static fp_number_type *
_fpdiv_parts (fp_number_type * a,
	      fp_number_type * b,
	      fp_number_type * tmp)
{
  fractype low = 0;
  fractype high = 0;
  fractype r0, r1, y0, y1, bit;
  fractype q;
  fractype numerator;
  fractype denominator;
  fractype quotient;
  fractype remainder;

  if (isnan (a))
    {
      return a;
    }
  if (isnan (b))
    {
      return b;
    }
  if (isinf (a) || iszero (a))
    {
      if (a->class == b->class)
	return nan ();
      return a;
    }
  a->sign = a->sign ^ b->sign;

  if (isinf (b))
    {
      a->fraction.ll = 0;
      a->normal_exp = 0;
      return a;
    }
  if (iszero (b))
    {
      a->class = CLASS_INFINITY;
      return b;
    }

  /* Calculate the mantissa by multiplying both 64bit numbers to get a
     128 bit number */
  {
    int carry;
    intfrac d0, d1;		/* weren't unsigned before ??? */

    /* 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 tmp;
  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, &tmp);

  return pack_d (res);
}

/* according to the demo, fpcmp returns a comparison with 0... thus
   a<b -> -1
   a==b -> 0
   a>b -> +1
 */

static 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;
}

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);
}

#ifndef US_SOFTWARE_GOFAST

/* These should be optimized for their specific tasks someday.  */

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) ;
}

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) ;
}

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);
}

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) ;
}

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);
}

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 /* ! US_SOFTWARE_GOFAST */

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 == 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);
}

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 : (-MAX_SI_INT)-1;
  /* 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);
}

#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;
  /* get reasonable MAX_USI_INT... */
  if (isinf (&a))
    return a.sign ? MAX_USI_INT : 0;
  /* it is a negative number */
  if (a.sign)
    return 0;
  /* 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 << ((FRACBITS + NGARDS) - a.normal_exp);
  else
    return a.fraction.ll >> ((FRACBITS + NGARDS) - a.normal_exp);
}
#endif

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);
}

#ifdef FLOAT

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);
}

#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);

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

#ifndef FLOAT

extern SFtype __make_fp (fp_class_type, unsigned int, int, USItype);

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);
}

SFtype
df_to_sf (DFtype arg_a)
{
  fp_number_type in;

  unpack_d ((FLO_union_type *) & arg_a, &in);
  return __make_fp (in.class, in.sign, in.normal_exp,
		    in.fraction.ll >> F_D_BITOFF);
}

#endif