-
Notifications
You must be signed in to change notification settings - Fork 29
/
fold.cpp
1679 lines (1492 loc) · 54.5 KB
/
fold.cpp
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 <string.h>
#include <math.h>
#include "fold.h"
#include "ast.h"
#include "ir.h"
#include "parser.h"
#define FOLD_STRING_UNTRANSLATE_HTSIZE 1024
#define FOLD_STRING_DOTRANSLATE_HTSIZE 1024
/* The options to use for inexact and arithmetic exceptions */
#define FOLD_ROUNDING SFLOAT_ROUND_NEAREST_EVEN
#define FOLD_TINYNESS SFLOAT_TBEFORE
/*
* Comparing float values is an unsafe operation when the operands to the
* comparison are floating point values that are inexact. For instance 1/3 is an
* inexact value. The FPU is meant to raise exceptions when these sorts of things
* happen, including division by zero, underflows and overflows. The C standard
* library provides us with the <fenv.h> header to gain access to the floating-
* point environment and lets us set the rounding mode and check for these exceptions.
* The problem is the standard C library allows an implementation to leave these
* stubbed out and does not require they be implemented. Furthermore, depending
* on implementations there is no control over the FPU. This is an IEE 754
* conforming implementation in software to compensate.
*/
typedef uint32_t sfloat_t;
union sfloat_cast_t {
qcfloat_t f;
sfloat_t s;
};
/* Exception flags */
enum sfloat_exceptionflags_t {
SFLOAT_NOEXCEPT = 0,
SFLOAT_INVALID = 1,
SFLOAT_DIVBYZERO = 4,
SFLOAT_OVERFLOW = 8,
SFLOAT_UNDERFLOW = 16,
SFLOAT_INEXACT = 32
};
/* Rounding modes */
enum sfloat_roundingmode_t {
SFLOAT_ROUND_NEAREST_EVEN,
SFLOAT_ROUND_DOWN,
SFLOAT_ROUND_UP,
SFLOAT_ROUND_TO_ZERO
};
/* Underflow tininess-detection mode */
enum sfloat_tdetect_t {
SFLOAT_TAFTER,
SFLOAT_TBEFORE
};
struct sfloat_state_t {
sfloat_roundingmode_t roundingmode;
sfloat_exceptionflags_t exceptionflags;
sfloat_tdetect_t tiny;
};
/* Counts the number of leading zero bits before the most-significand one bit. */
#ifdef _MSC_VER
/* MSVC has an intrinsic for this */
static GMQCC_INLINE uint32_t sfloat_clz(uint32_t x) {
int r = 0;
_BitScanForward(&r, x);
return r;
}
# define SFLOAT_CLZ(X, SUB) \
(sfloat_clz((X)) - (SUB))
#elif defined(__GNUC__) || defined(__CLANG__)
/* Clang and GCC have a builtin for this */
# define SFLOAT_CLZ(X, SUB) \
(__builtin_clz((X)) - (SUB))
#else
/* Native fallback */
static GMQCC_INLINE uint32_t sfloat_popcnt(uint32_t x) {
x -= ((x >> 1) & 0x55555555);
x = (((x >> 2) & 0x33333333) + (x & 0x33333333));
x = (((x >> 4) + x) & 0x0F0F0F0F);
x += x >> 8;
x += x >> 16;
return x & 0x0000003F;
}
static GMQCC_INLINE uint32_t sfloat_clz(uint32_t x) {
x |= (x >> 1);
x |= (x >> 2);
x |= (x >> 4);
x |= (x >> 8);
x |= (x >> 16);
return 32 - sfloat_popcnt(x);
}
# define SFLOAT_CLZ(X, SUB) \
(sfloat_clz((X) - (SUB)))
#endif
/* The value of a NaN */
#define SFLOAT_NAN 0xFFFFFFFF
/* Test if NaN */
#define SFLOAT_ISNAN(A) \
(0xFF000000 < (uint32_t)((A) << 1))
/* Test if signaling NaN */
#define SFLOAT_ISSNAN(A) \
(((((A) >> 22) & 0x1FF) == 0x1FE) && ((A) & 0x003FFFFF))
/* Raise exception */
#define SFLOAT_RAISE(STATE, FLAGS) \
((STATE)->exceptionflags = (sfloat_exceptionflags_t)((STATE)->exceptionflags | (FLAGS)))
/*
* Shifts `A' right by the number of bits given in `COUNT'. If any non-zero bits
* are shifted off they are forced into the least significand bit of the result
* by setting it to one. As a result of this, the value of `COUNT' can be
* arbitrarily large; if `COUNT' is greater than 32, the result will be either
* zero or one, depending on whether `A' is a zero or non-zero. The result is
* stored into the value pointed by `Z'.
*/
#define SFLOAT_SHIFT(SIZE, A, COUNT, Z) \
*(Z) = ((COUNT) == 0) \
? 1 \
: (((COUNT) < (SIZE)) \
? ((A) >> (COUNT)) | (((A) << ((-(COUNT)) & ((SIZE) - 1))) != 0) \
: ((A) != 0))
/* Extract fractional component */
#define SFLOAT_EXTRACT_FRAC(X) \
((uint32_t)((X) & 0x007FFFFF))
/* Extract exponent component */
#define SFLOAT_EXTRACT_EXP(X) \
((int16_t)((X) >> 23) & 0xFF)
/* Extract sign bit */
#define SFLOAT_EXTRACT_SIGN(X) \
((X) >> 31)
/*
* Normalizes the subnormal value represented by the denormalized significand
* `SA'. The normalized exponent and significand are stored at the locations
* pointed by `Z' and `SZ' respectively.
*/
#define SFLOAT_SUBNORMALIZE(SA, Z, SZ) \
(void)(*(SZ) = (SA) << SFLOAT_CLZ((SA), 8), *(Z) = 1 - SFLOAT_CLZ((SA), 8))
/*
* Packs the sign `SIGN', exponent `EXP' and significand `SIG' into the value
* giving the result.
*
* After the shifting into their proper positions, the fields are added together
* to form the result. This means any integer portion of `SIG' will be added
* to the exponent. Similarly, because a properly normalized significand will
* always have an integer portion equal to one, the exponent input `EXP' should
* be one less than the desired result exponent whenever the significant input
* `SIG' is a complete, normalized significand.
*/
#define SFLOAT_PACK(SIGN, EXP, SIG) \
(sfloat_t)((((uint32_t)(SIGN)) << 31) + (((uint32_t)(EXP)) << 23) + (SIG))
/*
* Takes two values `a' and `b', one of which is a NaN, and returns the appropriate
* NaN result. If either `a' or `b' is a signaling NaN than an invalid exception is
* raised.
*/
static sfloat_t sfloat_propagate_nan(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
bool isnan_a = SFLOAT_ISNAN(a);
bool issnan_a = SFLOAT_ISSNAN(a);
bool isnan_b = SFLOAT_ISNAN(b);
bool issnan_b = SFLOAT_ISSNAN(b);
a |= 0x00400000;
b |= 0x00400000;
if (issnan_a | issnan_b)
SFLOAT_RAISE(state, SFLOAT_INVALID);
if (isnan_a)
return (issnan_a & isnan_b) ? b : a;
return b;
}
/*
* Takes an abstract value having sign `sign_z', exponent `exp_z', and significand
* `sig_z' and returns the appropriate value corresponding to the abstract input.
*
* The abstract value is simply rounded and packed into the format. If the abstract
* input cannot be represented exactly an inexact exception is raised. If the
* abstract input is too large, the overflow and inexact exceptions are both raised
* and an infinity or maximal finite value is returned. If the abstract value is
* too small, the value is rounded to a subnormal and the underflow and inexact
* exceptions are only raised if the value cannot be represented exactly with
* a subnormal.
*
* The input significand `sig_z' has it's binary point between bits 30 and 29,
* this is seven bits to the left of its usual location. The shifted significand
* must be normalized or smaller than this. If it's not normalized then the exponent
* `exp_z' must be zero; in that case, the result returned is a subnormal number
* which must not require rounding. In the more usual case where the significand
* is normalized, the exponent must be one less than the *true* exponent.
*
* The handling of underflow and overflow is otherwise in alignment with IEC/IEEE.
*/
static sfloat_t SFLOAT_PACK_round(sfloat_state_t *state, bool sign_z, int16_t exp_z, uint32_t sig_z) {
sfloat_roundingmode_t mode = state->roundingmode;
bool even = !!(mode == SFLOAT_ROUND_NEAREST_EVEN);
unsigned char increment = 0x40;
unsigned char bits = sig_z & 0x7F;
if (!even) {
if (mode == SFLOAT_ROUND_TO_ZERO)
increment = 0;
else {
increment = 0x7F;
if (sign_z) {
if (mode == SFLOAT_ROUND_UP)
increment = 0;
} else {
if (mode == SFLOAT_ROUND_DOWN)
increment = 0;
}
}
}
if (0xFD <= (uint16_t)exp_z) {
if ((0xFD < exp_z) || ((exp_z == 0xFD) && ((int32_t)(sig_z + increment) < 0))) {
SFLOAT_RAISE(state, SFLOAT_OVERFLOW | SFLOAT_INEXACT);
return SFLOAT_PACK(sign_z, 0xFF, 0) - (increment == 0);
}
if (exp_z < 0) {
/* Check for underflow */
bool tiny = (state->tiny == SFLOAT_TBEFORE) || (exp_z < -1) || (sig_z + increment < 0x80000000);
SFLOAT_SHIFT(32, sig_z, -exp_z, &sig_z);
exp_z = 0;
bits = sig_z & 0x7F;
if (tiny && bits)
SFLOAT_RAISE(state, SFLOAT_UNDERFLOW);
}
}
if (bits)
SFLOAT_RAISE(state, SFLOAT_INEXACT);
sig_z = (sig_z + increment) >> 7;
sig_z &= ~(((bits ^ 0x40) == 0) & even);
if (sig_z == 0)
exp_z = 0;
return SFLOAT_PACK(sign_z, exp_z, sig_z);
}
/*
* Takes an abstract value having sign `sign_z', exponent `exp_z' and significand
* `sig_z' and returns the appropriate value corresponding to the abstract input.
* This function is exactly like `PACK_round' except the significand does not have
* to be normalized.
*
* Bit 31 of the significand must be zero and the exponent must be one less than
* the *true* exponent.
*/
static sfloat_t SFLOAT_PACK_normal(sfloat_state_t *state, bool sign_z, int16_t exp_z, uint32_t sig_z) {
unsigned char c = SFLOAT_CLZ(sig_z, 1);
return SFLOAT_PACK_round(state, sign_z, exp_z - c, sig_z << c);
}
/*
* Returns the result of adding the absolute values of `a' and `b'. The sign
* `sign_z' is ignored if the result is a NaN.
*/
static sfloat_t sfloat_add_impl(sfloat_state_t *state, sfloat_t a, sfloat_t b, bool sign_z) {
int16_t exp_a = SFLOAT_EXTRACT_EXP(a);
int16_t exp_b = SFLOAT_EXTRACT_EXP(b);
int16_t exp_z = 0;
int16_t exp_d = exp_a - exp_b;
uint32_t sig_a = SFLOAT_EXTRACT_FRAC(a) << 6;
uint32_t sig_b = SFLOAT_EXTRACT_FRAC(b) << 6;
uint32_t sig_z = 0;
if (0 < exp_d) {
if (exp_a == 0xFF)
return sig_a ? sfloat_propagate_nan(state, a, b) : a;
if (exp_b == 0)
--exp_d;
else
sig_b |= 0x20000000;
SFLOAT_SHIFT(32, sig_b, exp_d, &sig_b);
exp_z = exp_a;
} else if (exp_d < 0) {
if (exp_b == 0xFF)
return sig_b ? sfloat_propagate_nan(state, a, b) : SFLOAT_PACK(sign_z, 0xFF, 0);
if (exp_a == 0)
++exp_d;
else
sig_a |= 0x20000000;
SFLOAT_SHIFT(32, sig_a, -exp_d, &sig_a);
exp_z = exp_b;
} else {
if (exp_a == 0xFF)
return (sig_a | sig_b) ? sfloat_propagate_nan(state, a, b) : a;
if (exp_a == 0)
return SFLOAT_PACK(sign_z, 0, (sig_a + sig_b) >> 6);
sig_z = 0x40000000 + sig_a + sig_b;
exp_z = exp_a;
goto end;
}
sig_a |= 0x20000000;
sig_z = (sig_a + sig_b) << 1;
--exp_z;
if ((int32_t)sig_z < 0) {
sig_z = sig_a + sig_b;
++exp_z;
}
end:
return SFLOAT_PACK_round(state, sign_z, exp_z, sig_z);
}
/*
* Returns the result of subtracting the absolute values of `a' and `b'. If the
* sign `sign_z' is one, the difference is negated before being returned. The
* sign is ignored if the result is a NaN.
*/
static sfloat_t sfloat_sub_impl(sfloat_state_t *state, sfloat_t a, sfloat_t b, bool sign_z) {
int16_t exp_a = SFLOAT_EXTRACT_EXP(a);
int16_t exp_b = SFLOAT_EXTRACT_EXP(b);
int16_t exp_z = 0;
int16_t exp_d = exp_a - exp_b;
uint32_t sig_a = SFLOAT_EXTRACT_FRAC(a) << 7;
uint32_t sig_b = SFLOAT_EXTRACT_FRAC(b) << 7;
uint32_t sig_z = 0;
if (0 < exp_d) goto exp_greater_a;
if (exp_d < 0) goto exp_greater_b;
if (exp_a == 0xFF) {
if (sig_a | sig_b)
return sfloat_propagate_nan(state, a, b);
SFLOAT_RAISE(state, SFLOAT_INVALID);
return SFLOAT_NAN;
}
if (exp_a == 0)
exp_a = exp_b = 1;
if (sig_b < sig_a) goto greater_a;
if (sig_a < sig_b) goto greater_b;
return SFLOAT_PACK(state->roundingmode == SFLOAT_ROUND_DOWN, 0, 0);
exp_greater_b:
if (exp_b == 0xFF)
return (sig_b) ? sfloat_propagate_nan(state, a, b) : SFLOAT_PACK(sign_z ^ 1, 0xFF, 0);
if (exp_a == 0)
++exp_d;
else
sig_a |= 0x40000000;
SFLOAT_SHIFT(32, sig_a, -exp_d, &sig_a);
sig_b |= 0x40000000;
greater_b:
sig_z = sig_b - sig_a;
exp_z = exp_b;
sign_z ^= 1;
goto end;
exp_greater_a:
if (exp_a == 0xFF)
return (sig_a) ? sfloat_propagate_nan(state, a, b) : a;
if (exp_b == 0)
--exp_d;
else
sig_b |= 0x40000000;
SFLOAT_SHIFT(32, sig_b, exp_d, &sig_b);
sig_a |= 0x40000000;
greater_a:
sig_z = sig_a - sig_b;
exp_z = exp_a;
end:
--exp_z;
return SFLOAT_PACK_normal(state, sign_z, exp_z, sig_z);
}
static GMQCC_INLINE sfloat_t sfloat_add(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
bool sign_a = SFLOAT_EXTRACT_SIGN(a);
bool sign_b = SFLOAT_EXTRACT_SIGN(b);
return (sign_a == sign_b) ? sfloat_add_impl(state, a, b, sign_a)
: sfloat_sub_impl(state, a, b, sign_a);
}
static GMQCC_INLINE sfloat_t sfloat_sub(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
bool sign_a = SFLOAT_EXTRACT_SIGN(a);
bool sign_b = SFLOAT_EXTRACT_SIGN(b);
return (sign_a == sign_b) ? sfloat_sub_impl(state, a, b, sign_a)
: sfloat_add_impl(state, a, b, sign_a);
}
static sfloat_t sfloat_mul(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
int16_t exp_a = SFLOAT_EXTRACT_EXP(a);
int16_t exp_b = SFLOAT_EXTRACT_EXP(b);
int16_t exp_z = 0;
uint32_t sig_a = SFLOAT_EXTRACT_FRAC(a);
uint32_t sig_b = SFLOAT_EXTRACT_FRAC(b);
uint32_t sig_z = 0;
uint64_t sig_z64 = 0;
bool sign_a = SFLOAT_EXTRACT_SIGN(a);
bool sign_b = SFLOAT_EXTRACT_SIGN(b);
bool sign_z = sign_a ^ sign_b;
if (exp_a == 0xFF) {
if (sig_a || ((exp_b == 0xFF) && sig_b))
return sfloat_propagate_nan(state, a, b);
if ((exp_b | sig_b) == 0) {
SFLOAT_RAISE(state, SFLOAT_INVALID);
return SFLOAT_NAN;
}
return SFLOAT_PACK(sign_z, 0xFF, 0);
}
if (exp_b == 0xFF) {
if (sig_b)
return sfloat_propagate_nan(state, a, b);
if ((exp_a | sig_a) == 0) {
SFLOAT_RAISE(state, SFLOAT_INVALID);
return SFLOAT_NAN;
}
return SFLOAT_PACK(sign_z, 0xFF, 0);
}
if (exp_a == 0) {
if (sig_a == 0)
return SFLOAT_PACK(sign_z, 0, 0);
SFLOAT_SUBNORMALIZE(sig_a, &exp_a, &sig_a);
}
if (exp_b == 0) {
if (sig_b == 0)
return SFLOAT_PACK(sign_z, 0, 0);
SFLOAT_SUBNORMALIZE(sig_b, &exp_b, &sig_b);
}
exp_z = exp_a + exp_b - 0x7F;
sig_a = (sig_a | 0x00800000) << 7;
sig_b = (sig_b | 0x00800000) << 8;
SFLOAT_SHIFT(64, ((uint64_t)sig_a) * sig_b, 32, &sig_z64);
sig_z = sig_z64;
if (0 <= (int32_t)(sig_z << 1)) {
sig_z <<= 1;
--exp_z;
}
return SFLOAT_PACK_round(state, sign_z, exp_z, sig_z);
}
static sfloat_t sfloat_div(sfloat_state_t *state, sfloat_t a, sfloat_t b) {
int16_t exp_a = SFLOAT_EXTRACT_EXP(a);
int16_t exp_b = SFLOAT_EXTRACT_EXP(b);
int16_t exp_z = 0;
uint32_t sig_a = SFLOAT_EXTRACT_FRAC(a);
uint32_t sig_b = SFLOAT_EXTRACT_FRAC(b);
uint32_t sig_z = 0;
bool sign_a = SFLOAT_EXTRACT_SIGN(a);
bool sign_b = SFLOAT_EXTRACT_SIGN(b);
bool sign_z = sign_a ^ sign_b;
if (exp_a == 0xFF) {
if (sig_a)
return sfloat_propagate_nan(state, a, b);
if (exp_b == 0xFF) {
if (sig_b)
return sfloat_propagate_nan(state, a, b);
SFLOAT_RAISE(state, SFLOAT_INVALID);
return SFLOAT_NAN;
}
return SFLOAT_PACK(sign_z, 0xFF, 0);
}
if (exp_b == 0xFF)
return (sig_b) ? sfloat_propagate_nan(state, a, b) : SFLOAT_PACK(sign_z, 0, 0);
if (exp_b == 0) {
if (sig_b == 0) {
if ((exp_a | sig_a) == 0) {
SFLOAT_RAISE(state, SFLOAT_INVALID);
return SFLOAT_NAN;
}
SFLOAT_RAISE(state, SFLOAT_DIVBYZERO);
return SFLOAT_PACK(sign_z, 0xFF, 0);
}
SFLOAT_SUBNORMALIZE(sig_b, &exp_b, &sig_b);
}
if (exp_a == 0) {
if (sig_a == 0)
return SFLOAT_PACK(sign_z, 0, 0);
SFLOAT_SUBNORMALIZE(sig_a, &exp_a, &sig_a);
}
exp_z = exp_a - exp_b + 0x7D;
sig_a = (sig_a | 0x00800000) << 7;
sig_b = (sig_b | 0x00800000) << 8;
if (sig_b <= (sig_a + sig_a)) {
sig_a >>= 1;
++exp_z;
}
sig_z = (((uint64_t)sig_a) << 32) / sig_b;
if ((sig_z & 0x3F) == 0)
sig_z |= ((uint64_t)sig_b * sig_z != ((uint64_t)sig_a) << 32);
return SFLOAT_PACK_round(state, sign_z, exp_z, sig_z);
}
static sfloat_t sfloat_neg(sfloat_state_t *state, sfloat_t a) {
sfloat_cast_t neg;
neg.f = -1;
return sfloat_mul(state, a, neg.s);
}
static GMQCC_INLINE void sfloat_check(lex_ctx_t ctx, sfloat_state_t *state, const char *vec) {
/* Exception comes from vector component */
if (vec) {
if (state->exceptionflags & SFLOAT_DIVBYZERO)
compile_error(ctx, "division by zero in `%s' component", vec);
if (state->exceptionflags & SFLOAT_INVALID)
compile_error(ctx, "undefined (inf) in `%s' component", vec);
if (state->exceptionflags & SFLOAT_OVERFLOW)
compile_error(ctx, "arithmetic overflow in `%s' component", vec);
if (state->exceptionflags & SFLOAT_UNDERFLOW)
compile_error(ctx, "arithmetic underflow in `%s' component", vec);
return;
}
if (state->exceptionflags & SFLOAT_DIVBYZERO)
compile_error(ctx, "division by zero");
if (state->exceptionflags & SFLOAT_INVALID)
compile_error(ctx, "undefined (inf)");
if (state->exceptionflags & SFLOAT_OVERFLOW)
compile_error(ctx, "arithmetic overflow");
if (state->exceptionflags & SFLOAT_UNDERFLOW)
compile_error(ctx, "arithmetic underflow");
}
static GMQCC_INLINE void sfloat_init(sfloat_state_t *state) {
state->exceptionflags = SFLOAT_NOEXCEPT;
state->roundingmode = FOLD_ROUNDING;
state->tiny = FOLD_TINYNESS;
}
/*
* There is two stages to constant folding in GMQCC: there is the parse
* stage constant folding, where, with the help of the AST, operator
* usages can be constant folded. Then there is the constant folding
* in the IR for things like eliding if statements, can occur.
*
* This file is thus, split into two parts.
*/
#define isfloat(X) (((X))->m_vtype == TYPE_FLOAT)
#define isvector(X) (((X))->m_vtype == TYPE_VECTOR)
#define isstring(X) (((X))->m_vtype == TYPE_STRING)
#define isarray(X) (((X))->m_vtype == TYPE_ARRAY)
#define isfloats(X,Y) (isfloat (X) && isfloat (Y))
/*
* Implementation of basic vector math for vec3_t, for trivial constant
* folding.
*
* TODO: gcc/clang hinting for autovectorization
*/
enum vec3_comp_t {
VEC_COMP_X = 1 << 0,
VEC_COMP_Y = 1 << 1,
VEC_COMP_Z = 1 << 2
};
struct vec3_soft_t {
sfloat_cast_t x;
sfloat_cast_t y;
sfloat_cast_t z;
};
struct vec3_soft_state_t {
vec3_comp_t faults;
sfloat_state_t state[3];
};
static GMQCC_INLINE vec3_soft_t vec3_soft_convert(vec3_t vec) {
vec3_soft_t soft;
soft.x.f = vec.x;
soft.y.f = vec.y;
soft.z.f = vec.z;
return soft;
}
static GMQCC_INLINE bool vec3_soft_exception(vec3_soft_state_t *vstate, size_t index) {
sfloat_exceptionflags_t flags = vstate->state[index].exceptionflags;
if (flags & SFLOAT_DIVBYZERO) return true;
if (flags & SFLOAT_INVALID) return true;
if (flags & SFLOAT_OVERFLOW) return true;
if (flags & SFLOAT_UNDERFLOW) return true;
return false;
}
static GMQCC_INLINE void vec3_soft_eval(vec3_soft_state_t *state,
sfloat_t (*callback)(sfloat_state_t *, sfloat_t, sfloat_t),
vec3_t a,
vec3_t b)
{
vec3_soft_t sa = vec3_soft_convert(a);
vec3_soft_t sb = vec3_soft_convert(b);
callback(&state->state[0], sa.x.s, sb.x.s);
if (vec3_soft_exception(state, 0)) state->faults = (vec3_comp_t)(state->faults | VEC_COMP_X);
callback(&state->state[1], sa.y.s, sb.y.s);
if (vec3_soft_exception(state, 1)) state->faults = (vec3_comp_t)(state->faults | VEC_COMP_Y);
callback(&state->state[2], sa.z.s, sb.z.s);
if (vec3_soft_exception(state, 2)) state->faults = (vec3_comp_t)(state->faults | VEC_COMP_Z);
}
static GMQCC_INLINE void vec3_check_except(vec3_t a,
vec3_t b,
lex_ctx_t ctx,
sfloat_t (*callback)(sfloat_state_t *, sfloat_t, sfloat_t))
{
vec3_soft_state_t state = {};
if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
return;
sfloat_init(&state.state[0]);
sfloat_init(&state.state[1]);
sfloat_init(&state.state[2]);
vec3_soft_eval(&state, callback, a, b);
if (state.faults & VEC_COMP_X) sfloat_check(ctx, &state.state[0], "x");
if (state.faults & VEC_COMP_Y) sfloat_check(ctx, &state.state[1], "y");
if (state.faults & VEC_COMP_Z) sfloat_check(ctx, &state.state[2], "z");
}
static GMQCC_INLINE vec3_t vec3_add(lex_ctx_t ctx, vec3_t a, vec3_t b) {
vec3_t out;
vec3_check_except(a, b, ctx, &sfloat_add);
out.x = a.x + b.x;
out.y = a.y + b.y;
out.z = a.z + b.z;
return out;
}
static GMQCC_INLINE vec3_t vec3_sub(lex_ctx_t ctx, vec3_t a, vec3_t b) {
vec3_t out;
vec3_check_except(a, b, ctx, &sfloat_sub);
out.x = a.x - b.x;
out.y = a.y - b.y;
out.z = a.z - b.z;
return out;
}
static GMQCC_INLINE vec3_t vec3_neg(lex_ctx_t ctx, vec3_t a) {
vec3_t out;
sfloat_cast_t v[3];
sfloat_state_t s[3];
if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
goto end;
v[0].f = a.x;
v[1].f = a.y;
v[2].f = a.z;
sfloat_init(&s[0]);
sfloat_init(&s[1]);
sfloat_init(&s[2]);
sfloat_neg(&s[0], v[0].s);
sfloat_neg(&s[1], v[1].s);
sfloat_neg(&s[2], v[2].s);
sfloat_check(ctx, &s[0], nullptr);
sfloat_check(ctx, &s[1], nullptr);
sfloat_check(ctx, &s[2], nullptr);
end:
out.x = -a.x;
out.y = -a.y;
out.z = -a.z;
return out;
}
static GMQCC_INLINE vec3_t vec3_or(vec3_t a, vec3_t b) {
vec3_t out;
out.x = (qcfloat_t)(((qcint_t)a.x) | ((qcint_t)b.x));
out.y = (qcfloat_t)(((qcint_t)a.y) | ((qcint_t)b.y));
out.z = (qcfloat_t)(((qcint_t)a.z) | ((qcint_t)b.z));
return out;
}
static GMQCC_INLINE vec3_t vec3_orvf(vec3_t a, qcfloat_t b) {
vec3_t out;
out.x = (qcfloat_t)(((qcint_t)a.x) | ((qcint_t)b));
out.y = (qcfloat_t)(((qcint_t)a.y) | ((qcint_t)b));
out.z = (qcfloat_t)(((qcint_t)a.z) | ((qcint_t)b));
return out;
}
static GMQCC_INLINE vec3_t vec3_and(vec3_t a, vec3_t b) {
vec3_t out;
out.x = (qcfloat_t)(((qcint_t)a.x) & ((qcint_t)b.x));
out.y = (qcfloat_t)(((qcint_t)a.y) & ((qcint_t)b.y));
out.z = (qcfloat_t)(((qcint_t)a.z) & ((qcint_t)b.z));
return out;
}
static GMQCC_INLINE vec3_t vec3_andvf(vec3_t a, qcfloat_t b) {
vec3_t out;
out.x = (qcfloat_t)(((qcint_t)a.x) & ((qcint_t)b));
out.y = (qcfloat_t)(((qcint_t)a.y) & ((qcint_t)b));
out.z = (qcfloat_t)(((qcint_t)a.z) & ((qcint_t)b));
return out;
}
static GMQCC_INLINE vec3_t vec3_xor(vec3_t a, vec3_t b) {
vec3_t out;
out.x = (qcfloat_t)(((qcint_t)a.x) ^ ((qcint_t)b.x));
out.y = (qcfloat_t)(((qcint_t)a.y) ^ ((qcint_t)b.y));
out.z = (qcfloat_t)(((qcint_t)a.z) ^ ((qcint_t)b.z));
return out;
}
static GMQCC_INLINE vec3_t vec3_xorvf(vec3_t a, qcfloat_t b) {
vec3_t out;
out.x = (qcfloat_t)(((qcint_t)a.x) ^ ((qcint_t)b));
out.y = (qcfloat_t)(((qcint_t)a.y) ^ ((qcint_t)b));
out.z = (qcfloat_t)(((qcint_t)a.z) ^ ((qcint_t)b));
return out;
}
static GMQCC_INLINE vec3_t vec3_not(vec3_t a) {
vec3_t out;
out.x = -1-a.x;
out.y = -1-a.y;
out.z = -1-a.z;
return out;
}
static GMQCC_INLINE qcfloat_t vec3_mulvv(lex_ctx_t ctx, vec3_t a, vec3_t b) {
vec3_soft_t sa;
vec3_soft_t sb;
sfloat_state_t s[5];
sfloat_t r[5];
if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
goto end;
sa = vec3_soft_convert(a);
sb = vec3_soft_convert(b);
sfloat_init(&s[0]);
sfloat_init(&s[1]);
sfloat_init(&s[2]);
sfloat_init(&s[3]);
sfloat_init(&s[4]);
r[0] = sfloat_mul(&s[0], sa.x.s, sb.x.s);
r[1] = sfloat_mul(&s[1], sa.y.s, sb.y.s);
r[2] = sfloat_mul(&s[2], sa.z.s, sb.z.s);
r[3] = sfloat_add(&s[3], r[0], r[1]);
r[4] = sfloat_add(&s[4], r[3], r[2]);
sfloat_check(ctx, &s[0], nullptr);
sfloat_check(ctx, &s[1], nullptr);
sfloat_check(ctx, &s[2], nullptr);
sfloat_check(ctx, &s[3], nullptr);
sfloat_check(ctx, &s[4], nullptr);
end:
return (a.x * b.x + a.y * b.y + a.z * b.z);
}
static GMQCC_INLINE vec3_t vec3_mulvf(lex_ctx_t ctx, vec3_t a, qcfloat_t b) {
vec3_t out;
vec3_soft_t sa;
sfloat_cast_t sb;
sfloat_state_t s[3];
if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
goto end;
sa = vec3_soft_convert(a);
sb.f = b;
sfloat_init(&s[0]);
sfloat_init(&s[1]);
sfloat_init(&s[2]);
sfloat_mul(&s[0], sa.x.s, sb.s);
sfloat_mul(&s[1], sa.y.s, sb.s);
sfloat_mul(&s[2], sa.z.s, sb.s);
sfloat_check(ctx, &s[0], "x");
sfloat_check(ctx, &s[1], "y");
sfloat_check(ctx, &s[2], "z");
end:
out.x = a.x * b;
out.y = a.y * b;
out.z = a.z * b;
return out;
}
static GMQCC_INLINE bool vec3_cmp(vec3_t a, vec3_t b) {
return a.x == b.x &&
a.y == b.y &&
a.z == b.z;
}
static GMQCC_INLINE vec3_t vec3_create(float x, float y, float z) {
vec3_t out;
out.x = x;
out.y = y;
out.z = z;
return out;
}
static GMQCC_INLINE qcfloat_t vec3_notf(vec3_t a) {
return (!a.x && !a.y && !a.z);
}
static GMQCC_INLINE bool vec3_pbool(vec3_t a) {
return (a.x || a.y || a.z);
}
static GMQCC_INLINE vec3_t vec3_cross(lex_ctx_t ctx, vec3_t a, vec3_t b) {
vec3_t out;
vec3_soft_t sa;
vec3_soft_t sb;
sfloat_t r[9];
sfloat_state_t s[9];
if (!OPTS_FLAG(ARITHMETIC_EXCEPTIONS))
goto end;
sa = vec3_soft_convert(a);
sb = vec3_soft_convert(b);
sfloat_init(&s[0]);
sfloat_init(&s[1]);
sfloat_init(&s[2]);
sfloat_init(&s[3]);
sfloat_init(&s[4]);
sfloat_init(&s[5]);
sfloat_init(&s[6]);
sfloat_init(&s[7]);
sfloat_init(&s[8]);
r[0] = sfloat_mul(&s[0], sa.y.s, sb.z.s);
r[1] = sfloat_mul(&s[1], sa.z.s, sb.y.s);
r[2] = sfloat_mul(&s[2], sa.z.s, sb.x.s);
r[3] = sfloat_mul(&s[3], sa.x.s, sb.z.s);
r[4] = sfloat_mul(&s[4], sa.x.s, sb.y.s);
r[5] = sfloat_mul(&s[5], sa.y.s, sb.x.s);
r[6] = sfloat_sub(&s[6], r[0], r[1]);
r[7] = sfloat_sub(&s[7], r[2], r[3]);
r[8] = sfloat_sub(&s[8], r[4], r[5]);
sfloat_check(ctx, &s[0], nullptr);
sfloat_check(ctx, &s[1], nullptr);
sfloat_check(ctx, &s[2], nullptr);
sfloat_check(ctx, &s[3], nullptr);
sfloat_check(ctx, &s[4], nullptr);
sfloat_check(ctx, &s[5], nullptr);
sfloat_check(ctx, &s[6], "x");
sfloat_check(ctx, &s[7], "y");
sfloat_check(ctx, &s[8], "z");
end:
out.x = a.y * b.z - a.z * b.y;
out.y = a.z * b.x - a.x * b.z;
out.z = a.x * b.y - a.y * b.x;
return out;
}
qcfloat_t fold::immvalue_float(ast_value *value) {
return value->m_constval.vfloat;
}
vec3_t fold::immvalue_vector(ast_value *value) {
return value->m_constval.vvec;
}
const char *fold::immvalue_string(ast_value *value) {
return value->m_constval.vstring;
}
lex_ctx_t fold::ctx() {
lex_ctx_t ctx;
if (m_parser->lex)
return parser_ctx(m_parser);
memset(&ctx, 0, sizeof(ctx));
return ctx;
}
bool fold::immediate_true(ast_value *v) {
switch (v->m_vtype) {
case TYPE_FLOAT:
return !!v->m_constval.vfloat;
case TYPE_INTEGER:
return !!v->m_constval.vint;
case TYPE_VECTOR:
if (OPTS_FLAG(CORRECT_LOGIC))
return vec3_pbool(v->m_constval.vvec);
return !!(v->m_constval.vvec.x);
case TYPE_STRING:
if (!v->m_constval.vstring)
return false;
if (OPTS_FLAG(TRUE_EMPTY_STRINGS))
return true;
return !!v->m_constval.vstring[0];
default:
compile_error(ctx(), "internal error: fold_immediate_true on invalid type");
break;
}
return !!v->m_constval.vfunc;
}
/* Handy macros to determine if an ast_value can be constant folded. */
#define fold_can_1(X) \
(ast_istype(((X)), ast_value) && (X)->m_hasvalue && ((X)->m_cvq == CV_CONST) && \
((X))->m_vtype != TYPE_FUNCTION)
#define fold_can_2(X, Y) (fold_can_1(X) && fold_can_1(Y))
fold::fold()
: m_parser(nullptr)
{
}
fold::fold(parser_t *parser)
: m_parser(parser)
{
m_imm_string_untranslate = util_htnew(FOLD_STRING_UNTRANSLATE_HTSIZE);
m_imm_string_dotranslate = util_htnew(FOLD_STRING_DOTRANSLATE_HTSIZE);
constgen_float(0.0f, false);
constgen_float(1.0f, false);
constgen_float(-1.0f, false);
constgen_float(2.0f, false);
constgen_vector(vec3_create(0.0f, 0.0f, 0.0f));
constgen_vector(vec3_create(-1.0f, -1.0f, -1.0f));
}
bool fold::generate(ir_builder *ir) {
// generate globals for immediate folded values
ast_value *cur;
for (auto &it : m_imm_float)
if (!(cur = it)->generateGlobal(ir, false)) goto err;
for (auto &it : m_imm_vector)
if (!(cur = it)->generateGlobal(ir, false)) goto err;
for (auto &it : m_imm_string)
if (!(cur = it)->generateGlobal(ir, false)) goto err;
return true;
err:
con_out("failed to generate global %s\n", cur->m_name.c_str());
delete ir;
return false;
}
fold::~fold() {
// TODO: parser lifetime so this is called when it should be
#if 0
for (auto &it : m_imm_float) ast_delete(it);
for (auto &it : m_imm_vector) ast_delete(it);
for (auto &it : m_imm_string) ast_delete(it);
util_htdel(m_imm_string_untranslate);
util_htdel(m_imm_string_dotranslate);
#endif
}
ast_expression *fold::constgen_float(qcfloat_t value, bool inexact) {
for (auto &it : m_imm_float)
if (!memcmp(&it->m_constval.vfloat, &value, sizeof(qcfloat_t)))
return it;
ast_value *out = new ast_value(ctx(), "#IMMEDIATE", TYPE_FLOAT);
out->m_cvq = CV_CONST;
out->m_hasvalue = true;
out->m_inexact = inexact;
out->m_constval.vfloat = value;
m_imm_float.push_back(out);
return out;
}
ast_expression *fold::constgen_vector(vec3_t value) {
for (auto &it : m_imm_vector)
if (vec3_cmp(it->m_constval.vvec, value))
return it;
ast_value *out = new ast_value(ctx(), "#IMMEDIATE", TYPE_VECTOR);
out->m_cvq = CV_CONST;
out->m_hasvalue = true;
out->m_constval.vvec = value;
m_imm_vector.push_back(out);
return out;
}
ast_expression *fold::constgen_string(const char *str, bool translate) {
hash_table_t *table = translate ? m_imm_string_untranslate : m_imm_string_dotranslate;
ast_value *out = nullptr;
size_t hash = util_hthash(table, str);
if ((out = (ast_value*)util_htgeth(table, str, hash)))
return out;
if (translate) {
char name[32];
util_snprintf(name, sizeof(name), "dotranslate_%zu", m_parser->translated++);
out = new ast_value(ctx(), name, TYPE_STRING);