forked from shattered/macro11
-
Notifications
You must be signed in to change notification settings - Fork 5
/
parse.c
1503 lines (1265 loc) · 42.3 KB
/
parse.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
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <math.h>
#include <float.h>
#include <ctype.h>
#include "parse.h" /* my own definitions */
#include "util.h"
#include "rad50.h"
#include "listing.h"
#include "assemble_globals.h"
/* skipwhite - used everywhere to advance a char pointer past spaces */
char *skipwhite(
char *cp)
{
while (*cp == ' ' || *cp == '\t')
cp++;
return cp;
}
/* skipdelim - used everywhere to advance between tokens. Whitespace
and one comma are allowed delims. */
char *skipdelim(
char *cp)
{
cp = skipwhite(cp);
if (*cp == ',')
cp = skipwhite(cp + 1);
return cp;
}
/* skipdelim_comma - used to advance between tokens. Whitespace
and one comma are allowed delims.
Set *comma depending on whether a comma was skipped. */
char *skipdelim_comma(
char *cp,
int *comma)
{
cp = skipwhite(cp);
if ((*comma = (*cp == ','))) {
cp = skipwhite(cp + 1);
}
return cp;
}
/*
* check_eol - check that we're at the end of a line.
* Complain if not.
*/
int check_eol(
STACK *stack,
char *cp)
{
cp = skipwhite(cp);
if (EOL(*cp)) {
return 1;
}
report(stack->top, "Junk at end of line ('%.20s')\n", cp);
return 0;
}
/* Parses a string from the input stream. */
/* If not bracketed by <...> or ^/.../, then */
/* the string is delimited by trailing comma or whitespace. */
/* Allows nested <>'s */
char *getstring(
char *cp,
char **endp)
{
int len;
int start;
char *str;
if (!brackrange(cp, &start, &len, endp)) {
start = 0;
len = strcspn(cp, " \t\n,;");
if (endp)
*endp = cp + len;
}
str = memcheck(malloc(len + 1));
memcpy(str, cp + start, len);
str[len] = 0;
return str;
}
/* Parses a string from the input stream for .include and .library.
* These have a special kind of delimiters. It likes
* .include /name/ ?name? \name\ "name"
* but not
* .include ^/name/ <name> name =name= :name:
* .include :name: seems to be silently ignored.
*
* This should probably follow the exact same rules as .ASCII
* although that is not mentioned in the manual,
*/
char *getstring_fn(
char *cp,
char **endp)
{
char endstr[4];
int len;
char *str;
switch (*cp) {
case '<':
case ':':
return NULL;
}
if (!ispunct((unsigned char)*cp)) {
return NULL;
}
endstr[0] = *cp;
endstr[1] = '\n';
endstr[2] = '\0';
cp++;
len = strcspn(cp, endstr);
if (endp)
*endp = cp + len + 1;
str = memcheck(malloc(len + 1));
memcpy(str, cp, len);
str[len] = 0;
return str;
}
/* Get what would be the operation code from the line. */
/* Used to find the ends of streams without evaluating them, like
finding the closing .ENDM on a macro definition */
SYMBOL *get_op(
char *cp,
char **endp)
{
int local;
char *label;
SYMBOL *op;
cp = skipwhite(cp);
if (EOL(*cp))
return NULL;
label = get_symbol(cp, &cp, &local);
if (label == NULL)
return NULL; /* No operation code. */
cp = skipwhite(cp);
if (*cp == ':') { /* A label definition? */
cp++;
if (*cp == ':')
cp++; /* Skip it */
free(label);
label = get_symbol(cp, &cp, NULL);
if (label == NULL)
return NULL;
}
op = lookup_sym(label, &system_st);
free(label);
if (endp)
*endp = cp;
return op;
}
/* get_mode - parse a general addressing mode. */
int get_mode(
char *cp,
char **endp,
ADDR_MODE *mode,
char **error)
{
EX_TREE *value;
mode->offset = NULL;
mode->pcrel = 0;
mode->type = MODE_REG;
cp = skipwhite(cp);
/* @ means "indirect," sets bit 3 */
if (*cp == '@') {
cp++;
mode->type |= MODE_INDIRECT;
}
/* Immediate modes #imm and @#imm */
if (*cp == '#') {
cp++;
mode->type |= MODE_AUTO_INCR | MODE_PC;
mode->offset = parse_expr(cp, 0);
if (endp)
*endp = mode->offset->cp;
int ok = expr_ok(mode->offset);
if (!ok) {
*error = "Invalid expression after '#'";
free_tree(mode->offset);
mode->offset = NULL;
}
return ok;
}
/* Check for -(Rn) */
if (*cp == '-') {
char *tcp = skipwhite(cp + 1);
if (*tcp++ == '(') {
unsigned reg;
/* It's -(Rn) */
value = parse_expr(tcp, 0);
reg = get_register(value);
if (reg == NO_REG) {
*error = "Register expected after '-('";
free_tree(value);
return FALSE;
}
if (tcp = skipwhite(value->cp), *tcp++ != ')') {
*error = "')' expected after register";
free_tree(value);
return FALSE;
}
mode->type |= MODE_AUTO_DECR | reg;
if (endp)
*endp = tcp;
free_tree(value);
return TRUE;
}
}
/* Check for (Rn) */
if (*cp == '(') {
char *tcp;
unsigned reg;
value = parse_expr(cp + 1, 0);
reg = get_register(value);
if (reg == NO_REG) {
*error = "Register expected after '('";
free_tree(value);
return FALSE;
}
if (tcp = skipwhite(value->cp), *tcp++ != ')') {
*error = "')' expected after register";
free_tree(value);
return FALSE;
}
tcp = skipwhite(tcp);
if (*tcp == '+') {
tcp++; /* It's (Rn)+ */
if (endp)
*endp = tcp;
mode->type |= MODE_AUTO_INCR | reg;
free_tree(value);
return TRUE;
}
if (mode->type == MODE_INDIRECT) { /* For @(Rn) there's an
implied 0 offset */
mode->offset = new_ex_lit(0);
mode->type |= MODE_OFFSET | reg;
free_tree(value);
if (endp)
*endp = tcp;
return TRUE;
}
mode->type |= MODE_INDIRECT | reg; /* Mode 10 is register indirect
as in (Rn) */
free_tree(value);
if (endp)
*endp = tcp;
return TRUE;
}
/* Modes with an offset */
mode->offset = parse_expr(cp, 0);
if (!expr_ok(mode->offset)) {
*error = "Invalid expression";
free_tree(mode->offset);
mode->offset = NULL;
return FALSE;
}
cp = skipwhite(mode->offset->cp);
if (*cp == '(') {
unsigned reg;
/* indirect register plus offset */
value = parse_expr(cp + 1, 0);
reg = get_register(value);
if (reg == NO_REG) {
*error = "Register expected after 'offset('";
free_tree(value);
free_tree(mode->offset);
mode->offset = NULL;
return FALSE; /* Syntax error in addressing mode */
}
if (cp = skipwhite(value->cp), *cp++ != ')') {
*error = "')' expected after 'offset(register'";
free_tree(value);
free_tree(mode->offset);
mode->offset = NULL;
return FALSE; /* Syntax error in addressing mode */
}
mode->type |= MODE_OFFSET | reg;
free_tree(value);
if (endp)
*endp = cp;
return TRUE;
}
/* Plain old expression. */
if (endp)
*endp = cp;
/* It might be a register, though. */
if (mode->offset->type == EX_SYM) {
SYMBOL *sym = mode->offset->data.symbol;
if (sym->section->type == SECTION_REGISTER) {
free_tree(mode->offset);
mode->offset = NULL;
mode->type |= sym->value;
return TRUE;
}
}
/* It's either 067 (PC-relative) or 037 (absolute) mode, depending */
/* on user option. */
if (mode->type & MODE_INDIRECT) { /* Have already noted indirection? */
mode->type |= MODE_OFFSET|MODE_PC;/* If so, then PC-relative is the only
option */
mode->pcrel = 1; /* Note PC-relative */
} else if (enabl_ama) { /* User asked for absolute adressing? */
mode->type |= MODE_INDIRECT|MODE_AUTO_INCR|MODE_PC;
/* Give it to him. */
} else {
mode->type |= MODE_OFFSET|MODE_PC; /* PC-relative */
mode->pcrel = 1; /* Note PC-relative */
}
return TRUE;
}
/* get_fp_src_mode - parse an immediate fp literal or a general mode */
int get_fp_src_mode(
char *cp,
char **endp,
ADDR_MODE *mode,
char **error)
{
cp = skipwhite(cp);
char *savecp = cp;
if (cp[0] == '#') {
unsigned flt[1];
char *fltendp = NULL;
cp = skipwhite(cp + 1);
int ret = parse_float(cp, &fltendp, 1, flt);
if (ret) {
mode->type = MODE_AUTO_INCR | MODE_PC;
mode->pcrel = 0;
mode->offset = new_ex_lit(flt[0]);
mode->offset->cp = fltendp;
if (endp)
*endp = mode->offset->cp;
return TRUE;
} else if (fltendp) {
/* it looked like a fp number but something was wrong with it */
}
}
int ret = get_mode(savecp, endp, mode, error);
return ret;
}
#define DEBUG_FLOAT 0
void
printflt(unsigned *flt, int size)
{
printf("%06o: ", flt[0]);
printf("sign: %d ", (flt[0] & 0x8000) >> 15);
printf("uexp: %x ", (flt[0] & 0x7F80) >> 7);
printf("ufrac: %02x", flt[0] & 0x007F);
for (int i = 1; i < size; i++) {
printf(" %04x", flt[i]);
}
printf("\n");
}
#if DEBUG_FLOAT
#define DF(...) printf(__VA_ARGS__)
#else
#define DF(...)
#endif
/*
* We need 56 bits of mantissa.
*
* Try to detect if it is needed, possible and useful to use
* long double instead of double, when parsing floating point numbers.
*/
#if DBL_MANT_DIG >= 56
/* plain double seems big enough */
# define USE_LONG_DOUBLE 0
/* long double exists and seems big enough */
#elif LDBL_MANT_DIG >= 56
# define USE_LONG_DOUBLE 1
#elif defined(LDBL_MANT_DIG)
/* long double exists but is probably still too small */
# define USE_LONG_DOUBLE 1
#else
/* long double does not exist and plain double is too small */
# define USE_LONG_DOUBLE 0
#endif
#if USE_LONG_DOUBLE
# define DOUBLE long double
# define SCANF_FMT "%Lf"
# define FREXP frexpl
#else
# define DOUBLE double
# define SCANF_FMT "%lf"
# define FREXP frexp
#endif
#define PARSE_FLOAT_WITH_FLOATS 0
#define PARSE_FLOAT_WITH_INTS 1
#define PARSE_FLOAT_DIVIDE_BY_MULT_LOOP 0
/* Parse PDP-11 64-bit floating point format. */
/* Give a pointer to "size" words to receive the result. */
/* Note: there are probably degenerate cases that store incorrect
results. For example, I think rounding up a FLT2 might cause
exponent overflow. Sorry. */
#if PARSE_FLOAT_WITH_FLOATS
/* Note also that the full 56 bits of precision probably aren't always
available on the source platform, given the widespread application
of IEEE floating point formats, so expect some differences. Sorry
again. */
int parse_float(
char *cp,
char **endp,
int size,
unsigned *flt)
{
DOUBLE d; /* value */
DOUBLE frac; /* fractional value */
uint64_t ufrac; /* fraction converted to 56 bit
unsigned integer */
uint64_t onehalf; /* one half of the smallest bit
(used for rounding) */
int i; /* Number of fields converted by sscanf */
int n; /* Number of characters converted by sscanf */
int sexp; /* Signed exponent */
unsigned uexp; /* Unsigned excess-128 exponent */
unsigned sign = 0; /* Sign mask */
i = sscanf(cp, SCANF_FMT "%n", &d, &n);
if (i == 0)
return 0; /* Wasn't able to convert */
DF("LDBL_MANT_DIG: %d\n", LDBL_MANT_DIG);
DF("%Lf input: %s", d, cp);
cp += n;
if (endp)
*endp = cp;
if (d == 0.0) {
for (i = 0; i < size; i++) {
flt[i] = 0; /* All-bits-zero equals zero */
}
return 1; /* Good job. */
}
frac = FREXP(d, &sexp); /* Separate into exponent and mantissa */
DF("frac: %Lf %La sexp: %d\n", frac, frac, sexp);
if (sexp < -127 || sexp > 127)
return 0; /* Exponent out of range. */
uexp = sexp + 128; /* Make excess-128 mode */
uexp &= 0xff; /* express in 8 bits */
DF("uexp: %02x\n", uexp);
/*
* frexp guarantees its fractional return value is
* abs(frac) >= 0.5 and abs(frac) < 1.0
* Another way to think of this is that:
* abs(frac) >= 2**-1 and abs(frac) < 2**0
*/
if (frac < 0) {
sign = (1 << 15); /* Negative sign */
frac = -frac; /* fix the mantissa */
}
/*
* For the PDP-11 floating point representation the
* fractional part is 7 bits (for 16-bit floating point
* literals), 23 bits (for 32-bit floating point values),
* or 55 bits (for 64-bit floating point values).
* However the bit immediately above the MSB is always 1
* because the value is normalized. So it's actually
* 8 bits, 24 bits, or 56 bits.
* We effectively multiply the fractional part of our value by
* 2**56 to fully expose all of those bits (including
* the MSB which is 1).
* However as an intermediate step, we really multiply by
* 2**57, so we get one lsb for possible later rounding
* purposes. After that, we divide by 2 again.
*/
/* The following big literal is 2 to the 57th power: */
ufrac = (uint64_t) (frac * 144115188075855872.0); /* Align fraction bits */
DF("ufrac: %016lx\n", ufrac);
DF("56 : %016lx\n", (1UL<<57) - 2);
/*
* ufrac is now >= 2**56 and < 2**57.
* This means it's normalized: bit [56] is 1
* and all higher bits are 0.
*/
/* Round from 57-bits to 56, 24, or 8.
* We do this by:
* + first adding a value equal to one half of the
* least significant bit (the value 'onehalf')
* + (possibly) dealing with any carrying that
* causes the value to no longer be normalized
* (with bit [56] = 1 and all higher bits = 0)
* + shifting right by 1 bit (which throws away
* the 0 bit). Note this step could be rolled
* into the next step.
* + taking the remaining highest order 8,
* 24, or 56 bits.
*
* +--+--------+-------+ +--------+--------+
* |15|14 7|6 0| |15 | 0|
* +--+--------+-------+ +--------+--------+
* | S|EEEEEEEE|MMMMMMM| |MMMMMMMM|MMMMMMMM| ...maybe 2 more words...
* +--+--------+-------+ +--------+--------+
* Sign (1 bit)
* Exponent (8 bits)
* Mantissa (7 bits)
*/
onehalf = 1ULL << (16 * (4-size));
ufrac += onehalf;
DF("onehalf=%016lx, ufrac+onehalf: %016lx\n", onehalf, ufrac);
/* Did it roll up to a value 2**56? */
if ((ufrac >> 57) > 0) { /* Overflow? */
if (uexp < 0xFF) {
ufrac >>= 1; /* Normalize */
uexp++;
DF("ufrac: %016lx uexp: %02x (normalized)\n", ufrac, uexp);
} else {
/*
* If rounding and then normalisation would cause the exponent to
* overflow, just don't round: the cure is worse than the disease.
* We could detect ahead of time but the conditions for all size
* values may be a bit complicated, and so rare, that it is more
* readable to just undo it here.
*/
ufrac -= onehalf;
DF("don't round: exponent overflow\n");
}
}
ufrac >>= 1; /* Go from 57 bits to 56 */
flt[0] = (unsigned) (sign | (uexp << 7) | ((ufrac >> 48) & 0x7F));
if (size > 1) {
flt[1] = (unsigned) ((ufrac >> 32) & 0xffff);
if (size > 2) {
flt[2] = (unsigned) ((ufrac >> 16) & 0xffff);
flt[3] = (unsigned) ((ufrac >> 0) & 0xffff);
}
}
return 1;
}
#elif PARSE_FLOAT_WITH_INTS
#define DUMP3 DF("exp: %d. %d %016llx\n", \
float_dec_exponent, float_bin_exponent, float_buf)
/*
* Parse floating point denotations using integer operations only.
* Follows the reference version's implementation fairly closely.
*/
int parse_float(
char *cp,
char **endp,
int size,
unsigned *flt)
{
int float_sign = 0;
int float_dot = 0;
int float_dec_exponent = 0;
int float_bin_exponent = 0;
int ok_chars = 0;
uint64_t float_buf = 0;
float_bin_exponent = 65;
cp = skipwhite(cp);
if (*cp == '+') {
cp++;
} else if (*cp == '-') {
float_sign = 1;
cp++;
}
DF("float_sign: %d\n", float_sign);
while (!EOL(*cp)) {
if (isdigit(*cp)) {
/* Can we multiply by 10? */
DF("digit: %c\n", *cp);
ok_chars++;
if (float_buf & 0xF800000000000000ULL) { /* [0] & 0174000, 0xF800 */
/* No, that would overflow */
float_dec_exponent++; /* no, compensate for the snub */
/*
* Explanation of the above comment:
* - after the decimal separator, we should ignore extra digits
* completely. Since float_dot == -1, the exponent will be
* decremented below, and compensate for that here.
* - before the decimal separator, we don't add the digit
* (which would overflow) so we lose some lesser significant
* bits. float_dot == 0 in this case, and we do want the
* exponent increased to compensate for the ignored digit.
* So in both cases the right thing happens.
*/
} else {
/* Multiply by 10 */
float_buf *= 10;
/* Add digit */
float_buf += *cp - '0';
}
float_dec_exponent -= float_dot;
DUMP3;
cp++;
} else if (*cp == '.') {
DF("dot: %c\n", *cp);
ok_chars++;
if (float_dot) {
return 0; /* Error: repeated decimal separator */
}
float_dot = 1;
cp++;
} else {
DF("Other char: %c\n", *cp);
if (ok_chars == 0) {
return 0; /* No valid number found */
}
break;
}
}
if (toupper(*cp) == 'E') {
cp++;
int exp = strtol(cp, &cp, 10);
float_dec_exponent += exp;
DF("E%d -> dec_exp %d\n", exp, float_dec_exponent);
}
if (endp)
*endp = cp;
/* FLTG3 */
if (float_buf) {
/* Non-zero */
DF("Non-zero: decimal exponent: %d\n", float_dec_exponent);
while (float_dec_exponent > 0) { /* 31$ */
DUMP3;
if (float_buf <= 0x3316000000000000ULL) { /* 031426, 13078, 0x3316, 65390 / 5 */
/* Multiply by 5 and 2 */
DF("Multiply by 5 and 2\n");
float_buf *= 5;
float_bin_exponent++;
DUMP3;
} else {
/* Multiply by 5/4 and 8 32$ */
DF("Multiply by 5/4 and 8\n");
if (float_buf >= 0xCCCC000000000000ULL) {
float_buf >>= 1;
float_bin_exponent++;
}
float_buf += (float_buf >> 2);
float_bin_exponent += 3;
DUMP3;
}
float_dec_exponent--;
}
while (float_dec_exponent < 0) { /* 41$ ish */
DUMP3;
DF("Prepare for division by left-shifting the bits\n");
/* Prepare for division by left-shifting the bits */
while (((float_buf >> 63) & 1) == 0) { /* 41$ */
float_bin_exponent--; /* 40$ */
float_buf <<= 1;
DUMP3;
}
/* Divide by 2 */
float_buf >>= 1;
#if PARSE_FLOAT_DIVIDE_BY_MULT_LOOP
uint64_t float_save = float_buf;
DUMP3;
DF("float_save: %016llx\n", float_save);
/*
* Divide by 5: this is done by the trick of "dividing by
* multiplying". In order to keep as many significant bits as
* possible, we multiply by 8/5, and adjust the binary exponent to
* compensate for the factor of 8.
* The result is found when we drop the 64 low bits.
*
* So we multiply with the 65-bit number
* 0x19999999999999999
* 1 1001 1001 1001 ...
* which is 8 * 0011 0011 0011 ... aka 0x333...
* which is (2**64 - 1) / 5 aka 0xFFF... / 5.
*
* The rightmost (1 * float_save << 0) is contributed to the total
* because float_buf already contains that value.
* In loop i=32, (float_save << 3) is added:
* due to the two extra conditional shifts.
* In loop i=31, (float_save << 4) is added.
* In loop i=30, (float_save << 7) is added.
* etc, etc,
* so forming the repeating bit-pattern 1100 of the multiplier.
*
* Instead of shifting float_save left, we shift float_buf right,
* which over time drops precisely the desired 64 low bits.
*
* This is nearly exact but exact enough.
*
* The final result = start * 8 / 5.
*/
for (int i = 16 * 2; i > 0; i--) {
if ((i & 1) == 0) { /* 42$ */
float_buf >>= 2;
}
float_buf >>= 1; /* 43$ */
float_buf += float_save;
DF("Loop i=%2d: ", i); DUMP3;
}
#else
int round = float_buf % 5;
float_buf = float_buf / 5 * 8;
/*
* Try to fill in some of the lesser significant bits.
* This is not always bitwise identical to the original method
* but probably more accurate.
*/
if (round) {
float_buf += round * 8 / 5;
}
#endif
/* It's not simply dividing by 5, it also multiplies by 8,
* so we need to adjust the exponent here. */
float_bin_exponent -= 3;
float_dec_exponent++;
DUMP3;
}
/* Normalize the mantissa: shift a single 1 out to the left */
DF("Normalize the mantissa: shift a single 1 out to the left\n");
int carry;
do {
/* FLTG5 */
float_bin_exponent--;
carry = (float_buf >> 63) & 1;
float_buf <<= 1;
DUMP3;
} while (carry == 0);
/* Set excess 128. */
DF("Set excess 128.\n");
float_bin_exponent += 0200;
DUMP3;
if (float_bin_exponent & 0xFF00) {
/* Error N. Underflow. 2$ */
report(NULL, "Error N (underflow)\n");
return 0;
}
/* Shift right 9 positions to make room for sign and exponent 3$ */
DF("Shift right 9 positions to make room for sign and exponent\n");
int round = (float_buf >> 8) & 0x0001;
float_buf >>= 9;
float_buf |= (uint64_t)float_bin_exponent << (64-9);
DUMP3;
/*
* This rounding step seems always needed to make the result the same
* as the implementation with long doubles.
*
* This may be because of the slight imprecision of the divisions by 10?
*
* It is needed to get some exact results for values that are indeed
* exactly representable. Examples:
*
* (2**9-3)/2**9 = 0.994140625 = 0,11111101
* 407e 7fff ffff ffff -> 407e 8000 0000 0000 (correct)
* 1.00 (or 100E-2) divides 100 by 100 and gets
* 407f ffff ffff ffff -> 4080 0000 0000 0000 (correct)
*
* The reference implementation omits this rounding for size != 4:
* it has only one rounding step, which always depends on the size.
*/
float_buf += round;
DF("round: size = 4, round = %d\n", round);
/* Round (there is a truncation option to omit this step) */
if (1) {
uint64_t onehalf;
if (size < 4) {
/* 1 << 31 or 1 << 47 */
onehalf = 1ULL << ((16 * (4-size)) - 1);
DF("round: size = %d, onehalf = %016llx\n", size, onehalf);
float_buf += onehalf;
DUMP3;
/* The rounding bit is the lesser significant bit that's just
* outside the returned result. If we round, we add it to the
* returned value.
*
* If there is a carry-out of the mantissa, it gets added to
* the exponent (increasing it by 1).
*
* If that also overflows, we truely have overflow.
*/
}
DF("After rounding\n");
DUMP3;
}
if (float_buf & 0x8000000000000000ULL) {
// 6$ error T: exponent overflow
report(NULL, "error T: exponent overflow\n");
return 0;
}
/* 7$ */
float_buf |= (uint64_t)float_sign << 63;
DF("Put in float_sign: "); DUMP3;
}
/* Now put together the result from the parts */
flt[0] = (float_buf >> 48) & 0xFFFF;
if (size > 1) {
flt[1] = (float_buf >> 32) & 0xFFFF;
if (size > 2) {
flt[2] = (float_buf >> 16) & 0xFFFF;
flt[3] = (float_buf >> 0) & 0xFFFF;
}
}
return 1;
}
#else
# error How are you going to parse floats?
#endif
/* The recursive-descent expression parser parse_expr. */
/* This parser was designed for expressions with operator precedence.
However, MACRO-11 doesn't observe any sort of operator precedence.
If you feel your source deserves better, give the operators
appropriate precedence values right here. */
#define ADD_PREC 1
#define MUL_PREC 1
#define AND_PREC 1
#define OR_PREC 1
#define LSH_PREC 1
EX_TREE *parse_unary(
char *cp); /* Prototype for forward calls */
EX_TREE *parse_binary(
char *cp,
char term,
int depth)
{
EX_TREE *leftp,
*rightp,
*tp;
leftp = parse_unary(cp);
while (leftp->type != EX_ERR) {
cp = skipwhite(leftp->cp);
if (*cp == term)
return leftp;
switch (*cp) {
case '+':
if (depth >= ADD_PREC)
return leftp;
rightp = parse_binary(cp + 1, term, ADD_PREC);
tp = new_ex_bin(EX_ADD, leftp, rightp);
leftp = tp;
break;
case '-':
if (depth >= ADD_PREC)
return leftp;
rightp = parse_binary(cp + 1, term, ADD_PREC);
tp = new_ex_bin(EX_SUB, leftp, rightp);
leftp = tp;
break;
case '*':
if (depth >= MUL_PREC)
return leftp;
rightp = parse_binary(cp + 1, term, MUL_PREC);
tp = new_ex_bin(EX_MUL, leftp, rightp);
leftp = tp;
break;
case '/':
if (depth >= MUL_PREC)
return leftp;
rightp = parse_binary(cp + 1, term, MUL_PREC);
tp = new_ex_bin(EX_DIV, leftp, rightp);
leftp = tp;
break;
case '!':
if (depth >= OR_PREC)
return leftp;
rightp = parse_binary(cp + 1, term, OR_PREC);