forked from Xilinx/linux-xlnx
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathfpu_trig.c
1644 lines (1402 loc) · 38.3 KB
/
fpu_trig.c
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
// SPDX-License-Identifier: GPL-2.0
/*---------------------------------------------------------------------------+
| fpu_trig.c |
| |
| Implementation of the FPU "transcendental" functions. |
| |
| Copyright (C) 1992,1993,1994,1997,1999 |
| W. Metzenthen, 22 Parker St, Ormond, Vic 3163, |
| Australia. E-mail [email protected] |
| |
| |
+---------------------------------------------------------------------------*/
#include "fpu_system.h"
#include "exception.h"
#include "fpu_emu.h"
#include "status_w.h"
#include "control_w.h"
#include "reg_constant.h"
static void rem_kernel(unsigned long long st0, unsigned long long *y,
unsigned long long st1, unsigned long long q, int n);
#define BETTER_THAN_486
#define FCOS 4
/* Used only by fptan, fsin, fcos, and fsincos. */
/* This routine produces very accurate results, similar to
using a value of pi with more than 128 bits precision. */
/* Limited measurements show no results worse than 64 bit precision
except for the results for arguments close to 2^63, where the
precision of the result sometimes degrades to about 63.9 bits */
static int trig_arg(FPU_REG *st0_ptr, int even)
{
FPU_REG tmp;
u_char tmptag;
unsigned long long q;
int old_cw = control_word, saved_status = partial_status;
int tag, st0_tag = TAG_Valid;
if (exponent(st0_ptr) >= 63) {
partial_status |= SW_C2; /* Reduction incomplete. */
return -1;
}
control_word &= ~CW_RC;
control_word |= RC_CHOP;
setpositive(st0_ptr);
tag = FPU_u_div(st0_ptr, &CONST_PI2, &tmp, PR_64_BITS | RC_CHOP | 0x3f,
SIGN_POS);
FPU_round_to_int(&tmp, tag); /* Fortunately, this can't overflow
to 2^64 */
q = significand(&tmp);
if (q) {
rem_kernel(significand(st0_ptr),
&significand(&tmp),
significand(&CONST_PI2),
q, exponent(st0_ptr) - exponent(&CONST_PI2));
setexponent16(&tmp, exponent(&CONST_PI2));
st0_tag = FPU_normalize(&tmp);
FPU_copy_to_reg0(&tmp, st0_tag);
}
if ((even && !(q & 1)) || (!even && (q & 1))) {
st0_tag =
FPU_sub(REV | LOADED | TAG_Valid, (int)&CONST_PI2,
FULL_PRECISION);
#ifdef BETTER_THAN_486
/* So far, the results are exact but based upon a 64 bit
precision approximation to pi/2. The technique used
now is equivalent to using an approximation to pi/2 which
is accurate to about 128 bits. */
if ((exponent(st0_ptr) <= exponent(&CONST_PI2extra) + 64)
|| (q > 1)) {
/* This code gives the effect of having pi/2 to better than
128 bits precision. */
significand(&tmp) = q + 1;
setexponent16(&tmp, 63);
FPU_normalize(&tmp);
tmptag =
FPU_u_mul(&CONST_PI2extra, &tmp, &tmp,
FULL_PRECISION, SIGN_POS,
exponent(&CONST_PI2extra) +
exponent(&tmp));
setsign(&tmp, getsign(&CONST_PI2extra));
st0_tag = FPU_add(&tmp, tmptag, 0, FULL_PRECISION);
if (signnegative(st0_ptr)) {
/* CONST_PI2extra is negative, so the result of the addition
can be negative. This means that the argument is actually
in a different quadrant. The correction is always < pi/2,
so it can't overflow into yet another quadrant. */
setpositive(st0_ptr);
q++;
}
}
#endif /* BETTER_THAN_486 */
}
#ifdef BETTER_THAN_486
else {
/* So far, the results are exact but based upon a 64 bit
precision approximation to pi/2. The technique used
now is equivalent to using an approximation to pi/2 which
is accurate to about 128 bits. */
if (((q > 0)
&& (exponent(st0_ptr) <= exponent(&CONST_PI2extra) + 64))
|| (q > 1)) {
/* This code gives the effect of having p/2 to better than
128 bits precision. */
significand(&tmp) = q;
setexponent16(&tmp, 63);
FPU_normalize(&tmp); /* This must return TAG_Valid */
tmptag =
FPU_u_mul(&CONST_PI2extra, &tmp, &tmp,
FULL_PRECISION, SIGN_POS,
exponent(&CONST_PI2extra) +
exponent(&tmp));
setsign(&tmp, getsign(&CONST_PI2extra));
st0_tag = FPU_sub(LOADED | (tmptag & 0x0f), (int)&tmp,
FULL_PRECISION);
if ((exponent(st0_ptr) == exponent(&CONST_PI2)) &&
((st0_ptr->sigh > CONST_PI2.sigh)
|| ((st0_ptr->sigh == CONST_PI2.sigh)
&& (st0_ptr->sigl > CONST_PI2.sigl)))) {
/* CONST_PI2extra is negative, so the result of the
subtraction can be larger than pi/2. This means
that the argument is actually in a different quadrant.
The correction is always < pi/2, so it can't overflow
into yet another quadrant. */
st0_tag =
FPU_sub(REV | LOADED | TAG_Valid,
(int)&CONST_PI2, FULL_PRECISION);
q++;
}
}
}
#endif /* BETTER_THAN_486 */
FPU_settag0(st0_tag);
control_word = old_cw;
partial_status = saved_status & ~SW_C2; /* Reduction complete. */
return (q & 3) | even;
}
/* Convert a long to register */
static void convert_l2reg(long const *arg, int deststnr)
{
int tag;
long num = *arg;
u_char sign;
FPU_REG *dest = &st(deststnr);
if (num == 0) {
FPU_copy_to_regi(&CONST_Z, TAG_Zero, deststnr);
return;
}
if (num > 0) {
sign = SIGN_POS;
} else {
num = -num;
sign = SIGN_NEG;
}
dest->sigh = num;
dest->sigl = 0;
setexponent16(dest, 31);
tag = FPU_normalize(dest);
FPU_settagi(deststnr, tag);
setsign(dest, sign);
return;
}
static void single_arg_error(FPU_REG *st0_ptr, u_char st0_tag)
{
if (st0_tag == TAG_Empty)
FPU_stack_underflow(); /* Puts a QNaN in st(0) */
else if (st0_tag == TW_NaN)
real_1op_NaN(st0_ptr); /* return with a NaN in st(0) */
#ifdef PARANOID
else
EXCEPTION(EX_INTERNAL | 0x0112);
#endif /* PARANOID */
}
static void single_arg_2_error(FPU_REG *st0_ptr, u_char st0_tag)
{
int isNaN;
switch (st0_tag) {
case TW_NaN:
isNaN = (exponent(st0_ptr) == EXP_OVER)
&& (st0_ptr->sigh & 0x80000000);
if (isNaN && !(st0_ptr->sigh & 0x40000000)) { /* Signaling ? */
EXCEPTION(EX_Invalid);
if (control_word & CW_Invalid) {
/* The masked response */
/* Convert to a QNaN */
st0_ptr->sigh |= 0x40000000;
push();
FPU_copy_to_reg0(st0_ptr, TAG_Special);
}
} else if (isNaN) {
/* A QNaN */
push();
FPU_copy_to_reg0(st0_ptr, TAG_Special);
} else {
/* pseudoNaN or other unsupported */
EXCEPTION(EX_Invalid);
if (control_word & CW_Invalid) {
/* The masked response */
FPU_copy_to_reg0(&CONST_QNaN, TAG_Special);
push();
FPU_copy_to_reg0(&CONST_QNaN, TAG_Special);
}
}
break; /* return with a NaN in st(0) */
#ifdef PARANOID
default:
EXCEPTION(EX_INTERNAL | 0x0112);
#endif /* PARANOID */
}
}
/*---------------------------------------------------------------------------*/
static void f2xm1(FPU_REG *st0_ptr, u_char tag)
{
FPU_REG a;
clear_C1();
if (tag == TAG_Valid) {
/* For an 80486 FPU, the result is undefined if the arg is >= 1.0 */
if (exponent(st0_ptr) < 0) {
denormal_arg:
FPU_to_exp16(st0_ptr, &a);
/* poly_2xm1(x) requires 0 < st(0) < 1. */
poly_2xm1(getsign(st0_ptr), &a, st0_ptr);
}
set_precision_flag_up(); /* 80486 appears to always do this */
return;
}
if (tag == TAG_Zero)
return;
if (tag == TAG_Special)
tag = FPU_Special(st0_ptr);
switch (tag) {
case TW_Denormal:
if (denormal_operand() < 0)
return;
goto denormal_arg;
case TW_Infinity:
if (signnegative(st0_ptr)) {
/* -infinity gives -1 (p16-10) */
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
setnegative(st0_ptr);
}
return;
default:
single_arg_error(st0_ptr, tag);
}
}
static void fptan(FPU_REG *st0_ptr, u_char st0_tag)
{
FPU_REG *st_new_ptr;
int q;
u_char arg_sign = getsign(st0_ptr);
/* Stack underflow has higher priority */
if (st0_tag == TAG_Empty) {
FPU_stack_underflow(); /* Puts a QNaN in st(0) */
if (control_word & CW_Invalid) {
st_new_ptr = &st(-1);
push();
FPU_stack_underflow(); /* Puts a QNaN in the new st(0) */
}
return;
}
if (STACK_OVERFLOW) {
FPU_stack_overflow();
return;
}
if (st0_tag == TAG_Valid) {
if (exponent(st0_ptr) > -40) {
if ((q = trig_arg(st0_ptr, 0)) == -1) {
/* Operand is out of range */
return;
}
poly_tan(st0_ptr);
setsign(st0_ptr, (q & 1) ^ (arg_sign != 0));
set_precision_flag_up(); /* We do not really know if up or down */
} else {
/* For a small arg, the result == the argument */
/* Underflow may happen */
denormal_arg:
FPU_to_exp16(st0_ptr, st0_ptr);
st0_tag =
FPU_round(st0_ptr, 1, 0, FULL_PRECISION, arg_sign);
FPU_settag0(st0_tag);
}
push();
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
return;
}
if (st0_tag == TAG_Zero) {
push();
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
setcc(0);
return;
}
if (st0_tag == TAG_Special)
st0_tag = FPU_Special(st0_ptr);
if (st0_tag == TW_Denormal) {
if (denormal_operand() < 0)
return;
goto denormal_arg;
}
if (st0_tag == TW_Infinity) {
/* The 80486 treats infinity as an invalid operand */
if (arith_invalid(0) >= 0) {
st_new_ptr = &st(-1);
push();
arith_invalid(0);
}
return;
}
single_arg_2_error(st0_ptr, st0_tag);
}
static void fxtract(FPU_REG *st0_ptr, u_char st0_tag)
{
FPU_REG *st_new_ptr;
u_char sign;
register FPU_REG *st1_ptr = st0_ptr; /* anticipate */
if (STACK_OVERFLOW) {
FPU_stack_overflow();
return;
}
clear_C1();
if (st0_tag == TAG_Valid) {
long e;
push();
sign = getsign(st1_ptr);
reg_copy(st1_ptr, st_new_ptr);
setexponent16(st_new_ptr, exponent(st_new_ptr));
denormal_arg:
e = exponent16(st_new_ptr);
convert_l2reg(&e, 1);
setexponentpos(st_new_ptr, 0);
setsign(st_new_ptr, sign);
FPU_settag0(TAG_Valid); /* Needed if arg was a denormal */
return;
} else if (st0_tag == TAG_Zero) {
sign = getsign(st0_ptr);
if (FPU_divide_by_zero(0, SIGN_NEG) < 0)
return;
push();
FPU_copy_to_reg0(&CONST_Z, TAG_Zero);
setsign(st_new_ptr, sign);
return;
}
if (st0_tag == TAG_Special)
st0_tag = FPU_Special(st0_ptr);
if (st0_tag == TW_Denormal) {
if (denormal_operand() < 0)
return;
push();
sign = getsign(st1_ptr);
FPU_to_exp16(st1_ptr, st_new_ptr);
goto denormal_arg;
} else if (st0_tag == TW_Infinity) {
sign = getsign(st0_ptr);
setpositive(st0_ptr);
push();
FPU_copy_to_reg0(&CONST_INF, TAG_Special);
setsign(st_new_ptr, sign);
return;
} else if (st0_tag == TW_NaN) {
if (real_1op_NaN(st0_ptr) < 0)
return;
push();
FPU_copy_to_reg0(st0_ptr, TAG_Special);
return;
} else if (st0_tag == TAG_Empty) {
/* Is this the correct behaviour? */
if (control_word & EX_Invalid) {
FPU_stack_underflow();
push();
FPU_stack_underflow();
} else
EXCEPTION(EX_StackUnder);
}
#ifdef PARANOID
else
EXCEPTION(EX_INTERNAL | 0x119);
#endif /* PARANOID */
}
static void fdecstp(void)
{
clear_C1();
top--;
}
static void fincstp(void)
{
clear_C1();
top++;
}
static void fsqrt_(FPU_REG *st0_ptr, u_char st0_tag)
{
int expon;
clear_C1();
if (st0_tag == TAG_Valid) {
u_char tag;
if (signnegative(st0_ptr)) {
arith_invalid(0); /* sqrt(negative) is invalid */
return;
}
/* make st(0) in [1.0 .. 4.0) */
expon = exponent(st0_ptr);
denormal_arg:
setexponent16(st0_ptr, (expon & 1));
/* Do the computation, the sign of the result will be positive. */
tag = wm_sqrt(st0_ptr, 0, 0, control_word, SIGN_POS);
addexponent(st0_ptr, expon >> 1);
FPU_settag0(tag);
return;
}
if (st0_tag == TAG_Zero)
return;
if (st0_tag == TAG_Special)
st0_tag = FPU_Special(st0_ptr);
if (st0_tag == TW_Infinity) {
if (signnegative(st0_ptr))
arith_invalid(0); /* sqrt(-Infinity) is invalid */
return;
} else if (st0_tag == TW_Denormal) {
if (signnegative(st0_ptr)) {
arith_invalid(0); /* sqrt(negative) is invalid */
return;
}
if (denormal_operand() < 0)
return;
FPU_to_exp16(st0_ptr, st0_ptr);
expon = exponent16(st0_ptr);
goto denormal_arg;
}
single_arg_error(st0_ptr, st0_tag);
}
static void frndint_(FPU_REG *st0_ptr, u_char st0_tag)
{
int flags, tag;
if (st0_tag == TAG_Valid) {
u_char sign;
denormal_arg:
sign = getsign(st0_ptr);
if (exponent(st0_ptr) > 63)
return;
if (st0_tag == TW_Denormal) {
if (denormal_operand() < 0)
return;
}
/* Fortunately, this can't overflow to 2^64 */
if ((flags = FPU_round_to_int(st0_ptr, st0_tag)))
set_precision_flag(flags);
setexponent16(st0_ptr, 63);
tag = FPU_normalize(st0_ptr);
setsign(st0_ptr, sign);
FPU_settag0(tag);
return;
}
if (st0_tag == TAG_Zero)
return;
if (st0_tag == TAG_Special)
st0_tag = FPU_Special(st0_ptr);
if (st0_tag == TW_Denormal)
goto denormal_arg;
else if (st0_tag == TW_Infinity)
return;
else
single_arg_error(st0_ptr, st0_tag);
}
static int fsin(FPU_REG *st0_ptr, u_char tag)
{
u_char arg_sign = getsign(st0_ptr);
if (tag == TAG_Valid) {
int q;
if (exponent(st0_ptr) > -40) {
if ((q = trig_arg(st0_ptr, 0)) == -1) {
/* Operand is out of range */
return 1;
}
poly_sine(st0_ptr);
if (q & 2)
changesign(st0_ptr);
setsign(st0_ptr, getsign(st0_ptr) ^ arg_sign);
/* We do not really know if up or down */
set_precision_flag_up();
return 0;
} else {
/* For a small arg, the result == the argument */
set_precision_flag_up(); /* Must be up. */
return 0;
}
}
if (tag == TAG_Zero) {
setcc(0);
return 0;
}
if (tag == TAG_Special)
tag = FPU_Special(st0_ptr);
if (tag == TW_Denormal) {
if (denormal_operand() < 0)
return 1;
/* For a small arg, the result == the argument */
/* Underflow may happen */
FPU_to_exp16(st0_ptr, st0_ptr);
tag = FPU_round(st0_ptr, 1, 0, FULL_PRECISION, arg_sign);
FPU_settag0(tag);
return 0;
} else if (tag == TW_Infinity) {
/* The 80486 treats infinity as an invalid operand */
arith_invalid(0);
return 1;
} else {
single_arg_error(st0_ptr, tag);
return 1;
}
}
static int f_cos(FPU_REG *st0_ptr, u_char tag)
{
u_char st0_sign;
st0_sign = getsign(st0_ptr);
if (tag == TAG_Valid) {
int q;
if (exponent(st0_ptr) > -40) {
if ((exponent(st0_ptr) < 0)
|| ((exponent(st0_ptr) == 0)
&& (significand(st0_ptr) <=
0xc90fdaa22168c234LL))) {
poly_cos(st0_ptr);
/* We do not really know if up or down */
set_precision_flag_down();
return 0;
} else if ((q = trig_arg(st0_ptr, FCOS)) != -1) {
poly_sine(st0_ptr);
if ((q + 1) & 2)
changesign(st0_ptr);
/* We do not really know if up or down */
set_precision_flag_down();
return 0;
} else {
/* Operand is out of range */
return 1;
}
} else {
denormal_arg:
setcc(0);
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
#ifdef PECULIAR_486
set_precision_flag_down(); /* 80486 appears to do this. */
#else
set_precision_flag_up(); /* Must be up. */
#endif /* PECULIAR_486 */
return 0;
}
} else if (tag == TAG_Zero) {
FPU_copy_to_reg0(&CONST_1, TAG_Valid);
setcc(0);
return 0;
}
if (tag == TAG_Special)
tag = FPU_Special(st0_ptr);
if (tag == TW_Denormal) {
if (denormal_operand() < 0)
return 1;
goto denormal_arg;
} else if (tag == TW_Infinity) {
/* The 80486 treats infinity as an invalid operand */
arith_invalid(0);
return 1;
} else {
single_arg_error(st0_ptr, tag); /* requires st0_ptr == &st(0) */
return 1;
}
}
static void fcos(FPU_REG *st0_ptr, u_char st0_tag)
{
f_cos(st0_ptr, st0_tag);
}
static void fsincos(FPU_REG *st0_ptr, u_char st0_tag)
{
FPU_REG *st_new_ptr;
FPU_REG arg;
u_char tag;
/* Stack underflow has higher priority */
if (st0_tag == TAG_Empty) {
FPU_stack_underflow(); /* Puts a QNaN in st(0) */
if (control_word & CW_Invalid) {
st_new_ptr = &st(-1);
push();
FPU_stack_underflow(); /* Puts a QNaN in the new st(0) */
}
return;
}
if (STACK_OVERFLOW) {
FPU_stack_overflow();
return;
}
if (st0_tag == TAG_Special)
tag = FPU_Special(st0_ptr);
else
tag = st0_tag;
if (tag == TW_NaN) {
single_arg_2_error(st0_ptr, TW_NaN);
return;
} else if (tag == TW_Infinity) {
/* The 80486 treats infinity as an invalid operand */
if (arith_invalid(0) >= 0) {
/* Masked response */
push();
arith_invalid(0);
}
return;
}
reg_copy(st0_ptr, &arg);
if (!fsin(st0_ptr, st0_tag)) {
push();
FPU_copy_to_reg0(&arg, st0_tag);
f_cos(&st(0), st0_tag);
} else {
/* An error, so restore st(0) */
FPU_copy_to_reg0(&arg, st0_tag);
}
}
/*---------------------------------------------------------------------------*/
/* The following all require two arguments: st(0) and st(1) */
/* A lean, mean kernel for the fprem instructions. This relies upon
the division and rounding to an integer in do_fprem giving an
exact result. Because of this, rem_kernel() needs to deal only with
the least significant 64 bits, the more significant bits of the
result must be zero.
*/
static void rem_kernel(unsigned long long st0, unsigned long long *y,
unsigned long long st1, unsigned long long q, int n)
{
int dummy;
unsigned long long x;
x = st0 << n;
/* Do the required multiplication and subtraction in the one operation */
/* lsw x -= lsw st1 * lsw q */
asm volatile ("mull %4; subl %%eax,%0; sbbl %%edx,%1":"=m"
(((unsigned *)&x)[0]), "=m"(((unsigned *)&x)[1]),
"=a"(dummy)
:"2"(((unsigned *)&st1)[0]), "m"(((unsigned *)&q)[0])
:"%dx");
/* msw x -= msw st1 * lsw q */
asm volatile ("mull %3; subl %%eax,%0":"=m" (((unsigned *)&x)[1]),
"=a"(dummy)
:"1"(((unsigned *)&st1)[1]), "m"(((unsigned *)&q)[0])
:"%dx");
/* msw x -= lsw st1 * msw q */
asm volatile ("mull %3; subl %%eax,%0":"=m" (((unsigned *)&x)[1]),
"=a"(dummy)
:"1"(((unsigned *)&st1)[0]), "m"(((unsigned *)&q)[1])
:"%dx");
*y = x;
}
/* Remainder of st(0) / st(1) */
/* This routine produces exact results, i.e. there is never any
rounding or truncation, etc of the result. */
static void do_fprem(FPU_REG *st0_ptr, u_char st0_tag, int round)
{
FPU_REG *st1_ptr = &st(1);
u_char st1_tag = FPU_gettagi(1);
if (!((st0_tag ^ TAG_Valid) | (st1_tag ^ TAG_Valid))) {
FPU_REG tmp, st0, st1;
u_char st0_sign, st1_sign;
u_char tmptag;
int tag;
int old_cw;
int expdif;
long long q;
unsigned short saved_status;
int cc;
fprem_valid:
/* Convert registers for internal use. */
st0_sign = FPU_to_exp16(st0_ptr, &st0);
st1_sign = FPU_to_exp16(st1_ptr, &st1);
expdif = exponent16(&st0) - exponent16(&st1);
old_cw = control_word;
cc = 0;
/* We want the status following the denorm tests, but don't want
the status changed by the arithmetic operations. */
saved_status = partial_status;
control_word &= ~CW_RC;
control_word |= RC_CHOP;
if (expdif < 64) {
/* This should be the most common case */
if (expdif > -2) {
u_char sign = st0_sign ^ st1_sign;
tag = FPU_u_div(&st0, &st1, &tmp,
PR_64_BITS | RC_CHOP | 0x3f,
sign);
setsign(&tmp, sign);
if (exponent(&tmp) >= 0) {
FPU_round_to_int(&tmp, tag); /* Fortunately, this can't
overflow to 2^64 */
q = significand(&tmp);
rem_kernel(significand(&st0),
&significand(&tmp),
significand(&st1),
q, expdif);
setexponent16(&tmp, exponent16(&st1));
} else {
reg_copy(&st0, &tmp);
q = 0;
}
if ((round == RC_RND)
&& (tmp.sigh & 0xc0000000)) {
/* We may need to subtract st(1) once more,
to get a result <= 1/2 of st(1). */
unsigned long long x;
expdif =
exponent16(&st1) - exponent16(&tmp);
if (expdif <= 1) {
if (expdif == 0)
x = significand(&st1) -
significand(&tmp);
else /* expdif is 1 */
x = (significand(&st1)
<< 1) -
significand(&tmp);
if ((x < significand(&tmp)) ||
/* or equi-distant (from 0 & st(1)) and q is odd */
((x == significand(&tmp))
&& (q & 1))) {
st0_sign = !st0_sign;
significand(&tmp) = x;
q++;
}
}
}
if (q & 4)
cc |= SW_C0;
if (q & 2)
cc |= SW_C3;
if (q & 1)
cc |= SW_C1;
} else {
control_word = old_cw;
setcc(0);
return;
}
} else {
/* There is a large exponent difference ( >= 64 ) */
/* To make much sense, the code in this section should
be done at high precision. */
int exp_1, N;
u_char sign;
/* prevent overflow here */
/* N is 'a number between 32 and 63' (p26-113) */
reg_copy(&st0, &tmp);
tmptag = st0_tag;
N = (expdif & 0x0000001f) + 32; /* This choice gives results
identical to an AMD 486 */
setexponent16(&tmp, N);
exp_1 = exponent16(&st1);
setexponent16(&st1, 0);
expdif -= N;
sign = getsign(&tmp) ^ st1_sign;
tag =
FPU_u_div(&tmp, &st1, &tmp,
PR_64_BITS | RC_CHOP | 0x3f, sign);
setsign(&tmp, sign);
FPU_round_to_int(&tmp, tag); /* Fortunately, this can't
overflow to 2^64 */
rem_kernel(significand(&st0),
&significand(&tmp),
significand(&st1),
significand(&tmp), exponent(&tmp)
);
setexponent16(&tmp, exp_1 + expdif);
/* It is possible for the operation to be complete here.
What does the IEEE standard say? The Intel 80486 manual
implies that the operation will never be completed at this
point, and the behaviour of a real 80486 confirms this.
*/
if (!(tmp.sigh | tmp.sigl)) {
/* The result is zero */
control_word = old_cw;
partial_status = saved_status;
FPU_copy_to_reg0(&CONST_Z, TAG_Zero);
setsign(&st0, st0_sign);
#ifdef PECULIAR_486
setcc(SW_C2);
#else
setcc(0);
#endif /* PECULIAR_486 */
return;
}
cc = SW_C2;
}
control_word = old_cw;
partial_status = saved_status;
tag = FPU_normalize_nuo(&tmp);
reg_copy(&tmp, st0_ptr);
/* The only condition to be looked for is underflow,
and it can occur here only if underflow is unmasked. */
if ((exponent16(&tmp) <= EXP_UNDER) && (tag != TAG_Zero)
&& !(control_word & CW_Underflow)) {
setcc(cc);
tag = arith_underflow(st0_ptr);
setsign(st0_ptr, st0_sign);
FPU_settag0(tag);
return;
} else if ((exponent16(&tmp) > EXP_UNDER) || (tag == TAG_Zero)) {
stdexp(st0_ptr);
setsign(st0_ptr, st0_sign);
} else {
tag =
FPU_round(st0_ptr, 0, 0, FULL_PRECISION, st0_sign);
}
FPU_settag0(tag);
setcc(cc);
return;
}
if (st0_tag == TAG_Special)
st0_tag = FPU_Special(st0_ptr);
if (st1_tag == TAG_Special)
st1_tag = FPU_Special(st1_ptr);
if (((st0_tag == TAG_Valid) && (st1_tag == TW_Denormal))
|| ((st0_tag == TW_Denormal) && (st1_tag == TAG_Valid))
|| ((st0_tag == TW_Denormal) && (st1_tag == TW_Denormal))) {
if (denormal_operand() < 0)
return;
goto fprem_valid;
} else if ((st0_tag == TAG_Empty) || (st1_tag == TAG_Empty)) {
FPU_stack_underflow();
return;
} else if (st0_tag == TAG_Zero) {
if (st1_tag == TAG_Valid) {
setcc(0);
return;
} else if (st1_tag == TW_Denormal) {
if (denormal_operand() < 0)
return;
setcc(0);
return;
} else if (st1_tag == TAG_Zero) {
arith_invalid(0);
return;
} /* fprem(?,0) always invalid */
else if (st1_tag == TW_Infinity) {
setcc(0);
return;
}
} else if ((st0_tag == TAG_Valid) || (st0_tag == TW_Denormal)) {
if (st1_tag == TAG_Zero) {
arith_invalid(0); /* fprem(Valid,Zero) is invalid */
return;
} else if (st1_tag != TW_NaN) {
if (((st0_tag == TW_Denormal)
|| (st1_tag == TW_Denormal))
&& (denormal_operand() < 0))
return;
if (st1_tag == TW_Infinity) {
/* fprem(Valid,Infinity) is o.k. */
setcc(0);
return;
}