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
|
/*
* The implementations contained in this file are heavily based on the
* implementations found in the Berkeley SoftFloat library. As such, they are
* licensed under the same 3-clause BSD license:
*
* License for Berkeley SoftFloat Release 3e
*
* John R. Hauser
* 2018 January 20
*
* The following applies to the whole of SoftFloat Release 3e as well as to
* each source file individually.
*
* Copyright 2011, 2012, 2013, 2014, 2015, 2016, 2017, 2018 The Regents of the
* University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice,
* this list of conditions, and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions, and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS "AS IS", AND ANY
* EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
* WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE, ARE
* DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE FOR ANY
* DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
* LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
* ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
* THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#version 430
#extension GL_ARB_gpu_shader_int64 : enable
#extension GL_ARB_shader_bit_encoding : enable
#extension GL_EXT_shader_integer_mix : enable
#extension GL_MESA_shader_integer_functions : enable
#pragma warning(off)
/* Software IEEE floating-point rounding mode.
* GLSL spec section "4.7.1 Range and Precision":
* The rounding mode cannot be set and is undefined.
* But here, we are able to define the rounding mode at the compilation time.
*/
#define FLOAT_ROUND_NEAREST_EVEN 0
#define FLOAT_ROUND_TO_ZERO 1
#define FLOAT_ROUND_DOWN 2
#define FLOAT_ROUND_UP 3
#define FLOAT_ROUNDING_MODE FLOAT_ROUND_NEAREST_EVEN
/* Absolute value of a Float64 :
* Clear the sign bit
*/
uint64_t
__fabs64(uint64_t __a)
{
uvec2 a = unpackUint2x32(__a);
a.y &= 0x7FFFFFFFu;
return packUint2x32(a);
}
/* Returns 1 if the double-precision floating-point value `a' is a NaN;
* otherwise returns 0.
*/
bool
__is_nan(uint64_t __a)
{
uvec2 a = unpackUint2x32(__a);
return (0xFFE00000u <= (a.y<<1)) &&
((a.x != 0u) || ((a.y & 0x000FFFFFu) != 0u));
}
/* Negate value of a Float64 :
* Toggle the sign bit
*/
uint64_t
__fneg64(uint64_t __a)
{
uvec2 a = unpackUint2x32(__a);
uint t = a.y;
t ^= (1u << 31);
a.y = mix(t, a.y, __is_nan(__a));
return packUint2x32(a);
}
uint64_t
__fsign64(uint64_t __a)
{
uvec2 a = unpackUint2x32(__a);
uvec2 retval;
retval.x = 0u;
retval.y = mix((a.y & 0x80000000u) | 0x3FF00000u, 0u, (a.y << 1 | a.x) == 0u);
return packUint2x32(retval);
}
/* Returns the fraction bits of the double-precision floating-point value `a'.*/
uint
__extractFloat64FracLo(uint64_t a)
{
return unpackUint2x32(a).x;
}
uint
__extractFloat64FracHi(uint64_t a)
{
return unpackUint2x32(a).y & 0x000FFFFFu;
}
/* Returns the exponent bits of the double-precision floating-point value `a'.*/
int
__extractFloat64Exp(uint64_t __a)
{
uvec2 a = unpackUint2x32(__a);
return int((a.y>>20) & 0x7FFu);
}
bool
__feq64_nonnan(uint64_t __a, uint64_t __b)
{
uvec2 a = unpackUint2x32(__a);
uvec2 b = unpackUint2x32(__b);
return (a.x == b.x) &&
((a.y == b.y) || ((a.x == 0u) && (((a.y | b.y)<<1) == 0u)));
}
/* Returns true if the double-precision floating-point value `a' is equal to the
* corresponding value `b', and false otherwise. The comparison is performed
* according to the IEEE Standard for Floating-Point Arithmetic.
*/
bool
__feq64(uint64_t a, uint64_t b)
{
if (__is_nan(a) || __is_nan(b))
return false;
return __feq64_nonnan(a, b);
}
/* Returns true if the double-precision floating-point value `a' is not equal
* to the corresponding value `b', and false otherwise. The comparison is
* performed according to the IEEE Standard for Floating-Point Arithmetic.
*/
bool
__fne64(uint64_t a, uint64_t b)
{
if (__is_nan(a) || __is_nan(b))
return true;
return !__feq64_nonnan(a, b);
}
/* Returns the sign bit of the double-precision floating-point value `a'.*/
uint
__extractFloat64Sign(uint64_t a)
{
return unpackUint2x32(a).y >> 31;
}
/* Returns true if the 64-bit value formed by concatenating `a0' and `a1' is less
* than the 64-bit value formed by concatenating `b0' and `b1'. Otherwise,
* returns false.
*/
bool
lt64(uint a0, uint a1, uint b0, uint b1)
{
return (a0 < b0) || ((a0 == b0) && (a1 < b1));
}
bool
__flt64_nonnan(uint64_t __a, uint64_t __b)
{
uvec2 a = unpackUint2x32(__a);
uvec2 b = unpackUint2x32(__b);
uint aSign = __extractFloat64Sign(__a);
uint bSign = __extractFloat64Sign(__b);
if (aSign != bSign)
return (aSign != 0u) && ((((a.y | b.y)<<1) | a.x | b.x) != 0u);
return mix(lt64(a.y, a.x, b.y, b.x), lt64(b.y, b.x, a.y, a.x), aSign != 0u);
}
/* Returns true if the double-precision floating-point value `a' is less than
* the corresponding value `b', and false otherwise. The comparison is performed
* according to the IEEE Standard for Floating-Point Arithmetic.
*/
bool
__flt64(uint64_t a, uint64_t b)
{
if (__is_nan(a) || __is_nan(b))
return false;
return __flt64_nonnan(a, b);
}
/* Returns true if the double-precision floating-point value `a' is greater
* than or equal to * the corresponding value `b', and false otherwise. The
* comparison is performed * according to the IEEE Standard for Floating-Point
* Arithmetic.
*/
bool
__fge64(uint64_t a, uint64_t b)
{
if (__is_nan(a) || __is_nan(b))
return false;
return !__flt64_nonnan(a, b);
}
/* Adds the 64-bit value formed by concatenating `a0' and `a1' to the 64-bit
* value formed by concatenating `b0' and `b1'. Addition is modulo 2^64, so
* any carry out is lost. The result is broken into two 32-bit pieces which
* are stored at the locations pointed to by `z0Ptr' and `z1Ptr'.
*/
void
__add64(uint a0, uint a1, uint b0, uint b1,
out uint z0Ptr,
out uint z1Ptr)
{
uint z1 = a1 + b1;
z1Ptr = z1;
z0Ptr = a0 + b0 + uint(z1 < a1);
}
/* Subtracts the 64-bit value formed by concatenating `b0' and `b1' from the
* 64-bit value formed by concatenating `a0' and `a1'. Subtraction is modulo
* 2^64, so any borrow out (carry out) is lost. The result is broken into two
* 32-bit pieces which are stored at the locations pointed to by `z0Ptr' and
* `z1Ptr'.
*/
void
__sub64(uint a0, uint a1, uint b0, uint b1,
out uint z0Ptr,
out uint z1Ptr)
{
z1Ptr = a1 - b1;
z0Ptr = a0 - b0 - uint(a1 < b1);
}
/* Shifts the 64-bit value formed by concatenating `a0' and `a1' right by the
* number of bits given in `count'. If any nonzero bits are shifted off, they
* are "jammed" into the least significant bit of the result by setting the
* least significant bit to 1. The value of `count' can be arbitrarily large;
* in particular, if `count' is greater than 64, the result will be either 0
* or 1, depending on whether the concatenation of `a0' and `a1' is zero or
* nonzero. The result is broken into two 32-bit pieces which are stored at
* the locations pointed to by `z0Ptr' and `z1Ptr'.
*/
void
__shift64RightJamming(uint a0,
uint a1,
int count,
out uint z0Ptr,
out uint z1Ptr)
{
uint z0;
uint z1;
int negCount = (-count) & 31;
z0 = mix(0u, a0, count == 0);
z0 = mix(z0, (a0 >> count), count < 32);
z1 = uint((a0 | a1) != 0u); /* count >= 64 */
uint z1_lt64 = (a0>>(count & 31)) | uint(((a0<<negCount) | a1) != 0u);
z1 = mix(z1, z1_lt64, count < 64);
z1 = mix(z1, (a0 | uint(a1 != 0u)), count == 32);
uint z1_lt32 = (a0<<negCount) | (a1>>count) | uint ((a1<<negCount) != 0u);
z1 = mix(z1, z1_lt32, count < 32);
z1 = mix(z1, a1, count == 0);
z1Ptr = z1;
z0Ptr = z0;
}
/* Shifts the 96-bit value formed by concatenating `a0', `a1', and `a2' right
* by 32 _plus_ the number of bits given in `count'. The shifted result is
* at most 64 nonzero bits; these are broken into two 32-bit pieces which are
* stored at the locations pointed to by `z0Ptr' and `z1Ptr'. The bits shifted
* off form a third 32-bit result as follows: The _last_ bit shifted off is
* the most-significant bit of the extra result, and the other 31 bits of the
* extra result are all zero if and only if _all_but_the_last_ bits shifted off
* were all zero. This extra result is stored in the location pointed to by
* `z2Ptr'. The value of `count' can be arbitrarily large.
* (This routine makes more sense if `a0', `a1', and `a2' are considered
* to form a fixed-point value with binary point between `a1' and `a2'. This
* fixed-point value is shifted right by the number of bits given in `count',
* and the integer part of the result is returned at the locations pointed to
* by `z0Ptr' and `z1Ptr'. The fractional part of the result may be slightly
* corrupted as described above, and is returned at the location pointed to by
* `z2Ptr'.)
*/
void
__shift64ExtraRightJamming(uint a0, uint a1, uint a2,
int count,
out uint z0Ptr,
out uint z1Ptr,
out uint z2Ptr)
{
uint z0 = 0u;
uint z1;
uint z2;
int negCount = (-count) & 31;
z2 = mix(uint(a0 != 0u), a0, count == 64);
z2 = mix(z2, a0 << negCount, count < 64);
z2 = mix(z2, a1 << negCount, count < 32);
z1 = mix(0u, (a0 >> (count & 31)), count < 64);
z1 = mix(z1, (a0<<negCount) | (a1>>count), count < 32);
a2 = mix(a2 | a1, a2, count < 32);
z0 = mix(z0, a0 >> count, count < 32);
z2 |= uint(a2 != 0u);
z0 = mix(z0, 0u, (count == 32));
z1 = mix(z1, a0, (count == 32));
z2 = mix(z2, a1, (count == 32));
z0 = mix(z0, a0, (count == 0));
z1 = mix(z1, a1, (count == 0));
z2 = mix(z2, a2, (count == 0));
z2Ptr = z2;
z1Ptr = z1;
z0Ptr = z0;
}
/* Shifts the 64-bit value formed by concatenating `a0' and `a1' left by the
* number of bits given in `count'. Any bits shifted off are lost. The value
* of `count' must be less than 32. The result is broken into two 32-bit
* pieces which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'.
*/
void
__shortShift64Left(uint a0, uint a1,
int count,
out uint z0Ptr,
out uint z1Ptr)
{
z1Ptr = a1<<count;
z0Ptr = mix((a0 << count | (a1 >> ((-count) & 31))), a0, count == 0);
}
/* Packs the sign `zSign', the exponent `zExp', and the significand formed by
* the concatenation of `zFrac0' and `zFrac1' into a double-precision floating-
* point value, returning the result. After being shifted into the proper
* positions, the three fields `zSign', `zExp', and `zFrac0' are simply added
* together to form the most significant 32 bits of the result. This means
* that any integer portion of `zFrac0' will be added into the exponent. Since
* a properly normalized significand will have an integer portion equal to 1,
* the `zExp' input should be 1 less than the desired result exponent whenever
* `zFrac0' and `zFrac1' concatenated form a complete, normalized significand.
*/
uint64_t
__packFloat64(uint zSign, int zExp, uint zFrac0, uint zFrac1)
{
uvec2 z;
z.y = (zSign << 31) + (uint(zExp) << 20) + zFrac0;
z.x = zFrac1;
return packUint2x32(z);
}
/* Takes an abstract floating-point value having sign `zSign', exponent `zExp',
* and extended significand formed by the concatenation of `zFrac0', `zFrac1',
* and `zFrac2', and returns the proper double-precision floating-point value
* corresponding to the abstract input. Ordinarily, the abstract value is
* simply rounded and packed into the double-precision format, with the inexact
* exception raised if the abstract input cannot be represented exactly.
* However, if the abstract value is too large, the overflow and inexact
* exceptions are raised and an infinity or maximal finite value is returned.
* If the abstract value is too small, the input value is rounded to a
* subnormal number, and the underflow and inexact exceptions are raised if the
* abstract input cannot be represented exactly as a subnormal double-precision
* floating-point number.
* The input significand must be normalized or smaller. If the input
* significand is not normalized, `zExp' must be 0; in that case, the result
* returned is a subnormal number, and it must not require rounding. In the
* usual case that the input significand is normalized, `zExp' must be 1 less
* than the "true" floating-point exponent. The handling of underflow and
* overflow follows the IEEE Standard for Floating-Point Arithmetic.
*/
uint64_t
__roundAndPackFloat64(uint zSign,
int zExp,
uint zFrac0,
uint zFrac1,
uint zFrac2)
{
bool roundNearestEven;
bool increment;
roundNearestEven = FLOAT_ROUNDING_MODE == FLOAT_ROUND_NEAREST_EVEN;
increment = int(zFrac2) < 0;
if (!roundNearestEven) {
if (FLOAT_ROUNDING_MODE == FLOAT_ROUND_TO_ZERO) {
increment = false;
} else {
if (zSign != 0u) {
increment = (FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN) &&
(zFrac2 != 0u);
} else {
increment = (FLOAT_ROUNDING_MODE == FLOAT_ROUND_UP) &&
(zFrac2 != 0u);
}
}
}
if (0x7FD <= zExp) {
if ((0x7FD < zExp) ||
((zExp == 0x7FD) &&
(0x001FFFFFu == zFrac0 && 0xFFFFFFFFu == zFrac1) &&
increment)) {
if ((FLOAT_ROUNDING_MODE == FLOAT_ROUND_TO_ZERO) ||
((zSign != 0u) && (FLOAT_ROUNDING_MODE == FLOAT_ROUND_UP)) ||
((zSign == 0u) && (FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN))) {
return __packFloat64(zSign, 0x7FE, 0x000FFFFFu, 0xFFFFFFFFu);
}
return __packFloat64(zSign, 0x7FF, 0u, 0u);
}
if (zExp < 0) {
__shift64ExtraRightJamming(
zFrac0, zFrac1, zFrac2, -zExp, zFrac0, zFrac1, zFrac2);
zExp = 0;
if (roundNearestEven) {
increment = zFrac2 < 0u;
} else {
if (zSign != 0u) {
increment = (FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN) &&
(zFrac2 != 0u);
} else {
increment = (FLOAT_ROUNDING_MODE == FLOAT_ROUND_UP) &&
(zFrac2 != 0u);
}
}
}
}
if (increment) {
__add64(zFrac0, zFrac1, 0u, 1u, zFrac0, zFrac1);
zFrac1 &= ~((zFrac2 + uint(zFrac2 == 0u)) & uint(roundNearestEven));
} else {
zExp = mix(zExp, 0, (zFrac0 | zFrac1) == 0u);
}
return __packFloat64(zSign, zExp, zFrac0, zFrac1);
}
/* Returns the number of leading 0 bits before the most-significant 1 bit of
* `a'. If `a' is zero, 32 is returned.
*/
int
__countLeadingZeros32(uint a)
{
int shiftCount;
shiftCount = mix(31 - findMSB(a), 32, a == 0u);
return shiftCount;
}
/* Takes an abstract floating-point value having sign `zSign', exponent `zExp',
* and significand formed by the concatenation of `zSig0' and `zSig1', and
* returns the proper double-precision floating-point value corresponding
* to the abstract input. This routine is just like `__roundAndPackFloat64'
* except that the input significand has fewer bits and does not have to be
* normalized. In all cases, `zExp' must be 1 less than the "true" floating-
* point exponent.
*/
uint64_t
__normalizeRoundAndPackFloat64(uint zSign,
int zExp,
uint zFrac0,
uint zFrac1)
{
int shiftCount;
uint zFrac2;
if (zFrac0 == 0u) {
zExp -= 32;
zFrac0 = zFrac1;
zFrac1 = 0u;
}
shiftCount = __countLeadingZeros32(zFrac0) - 11;
if (0 <= shiftCount) {
zFrac2 = 0u;
__shortShift64Left(zFrac0, zFrac1, shiftCount, zFrac0, zFrac1);
} else {
__shift64ExtraRightJamming(
zFrac0, zFrac1, 0u, -shiftCount, zFrac0, zFrac1, zFrac2);
}
zExp -= shiftCount;
return __roundAndPackFloat64(zSign, zExp, zFrac0, zFrac1, zFrac2);
}
/* Takes two double-precision floating-point values `a' and `b', one of which
* is a NaN, and returns the appropriate NaN result.
*/
uint64_t
__propagateFloat64NaN(uint64_t __a, uint64_t __b)
{
bool aIsNaN = __is_nan(__a);
bool bIsNaN = __is_nan(__b);
uvec2 a = unpackUint2x32(__a);
uvec2 b = unpackUint2x32(__b);
a.y |= 0x00080000u;
b.y |= 0x00080000u;
return packUint2x32(mix(b, mix(a, b, bvec2(bIsNaN, bIsNaN)), bvec2(aIsNaN, aIsNaN)));
}
/* Returns the result of adding the double-precision floating-point values
* `a' and `b'. The operation is performed according to the IEEE Standard for
* Floating-Point Arithmetic.
*/
uint64_t
__fadd64(uint64_t a, uint64_t b)
{
uint aSign = __extractFloat64Sign(a);
uint bSign = __extractFloat64Sign(b);
uint aFracLo = __extractFloat64FracLo(a);
uint aFracHi = __extractFloat64FracHi(a);
uint bFracLo = __extractFloat64FracLo(b);
uint bFracHi = __extractFloat64FracHi(b);
int aExp = __extractFloat64Exp(a);
int bExp = __extractFloat64Exp(b);
uint zFrac0 = 0u;
uint zFrac1 = 0u;
int expDiff = aExp - bExp;
if (aSign == bSign) {
uint zFrac2 = 0u;
int zExp;
bool orig_exp_diff_is_zero = (expDiff == 0);
if (orig_exp_diff_is_zero) {
if (aExp == 0x7FF) {
bool propagate = (aFracHi | aFracLo | bFracHi | bFracLo) != 0u;
return mix(a, __propagateFloat64NaN(a, b), propagate);
}
__add64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1);
if (aExp == 0)
return __packFloat64(aSign, 0, zFrac0, zFrac1);
zFrac2 = 0u;
zFrac0 |= 0x00200000u;
zExp = aExp;
__shift64ExtraRightJamming(
zFrac0, zFrac1, zFrac2, 1, zFrac0, zFrac1, zFrac2);
} else if (0 < expDiff) {
if (aExp == 0x7FF) {
bool propagate = (aFracHi | aFracLo) != 0u;
return mix(a, __propagateFloat64NaN(a, b), propagate);
}
expDiff = mix(expDiff, expDiff - 1, bExp == 0);
bFracHi = mix(bFracHi | 0x00100000u, bFracHi, bExp == 0);
__shift64ExtraRightJamming(
bFracHi, bFracLo, 0u, expDiff, bFracHi, bFracLo, zFrac2);
zExp = aExp;
} else if (expDiff < 0) {
if (bExp == 0x7FF) {
bool propagate = (bFracHi | bFracLo) != 0u;
return mix(__packFloat64(aSign, 0x7ff, 0u, 0u), __propagateFloat64NaN(a, b), propagate);
}
expDiff = mix(expDiff, expDiff + 1, aExp == 0);
aFracHi = mix(aFracHi | 0x00100000u, aFracHi, aExp == 0);
__shift64ExtraRightJamming(
aFracHi, aFracLo, 0u, - expDiff, aFracHi, aFracLo, zFrac2);
zExp = bExp;
}
if (!orig_exp_diff_is_zero) {
aFracHi |= 0x00100000u;
__add64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1);
--zExp;
if (!(zFrac0 < 0x00200000u)) {
__shift64ExtraRightJamming(zFrac0, zFrac1, zFrac2, 1, zFrac0, zFrac1, zFrac2);
++zExp;
}
}
return __roundAndPackFloat64(aSign, zExp, zFrac0, zFrac1, zFrac2);
} else {
int zExp;
__shortShift64Left(aFracHi, aFracLo, 10, aFracHi, aFracLo);
__shortShift64Left(bFracHi, bFracLo, 10, bFracHi, bFracLo);
if (0 < expDiff) {
if (aExp == 0x7FF) {
bool propagate = (aFracHi | aFracLo) != 0u;
return mix(a, __propagateFloat64NaN(a, b), propagate);
}
expDiff = mix(expDiff, expDiff - 1, bExp == 0);
bFracHi = mix(bFracHi | 0x40000000u, bFracHi, bExp == 0);
__shift64RightJamming(bFracHi, bFracLo, expDiff, bFracHi, bFracLo);
aFracHi |= 0x40000000u;
__sub64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1);
zExp = aExp;
--zExp;
return __normalizeRoundAndPackFloat64(aSign, zExp - 10, zFrac0, zFrac1);
}
if (expDiff < 0) {
if (bExp == 0x7FF) {
bool propagate = (bFracHi | bFracLo) != 0u;
return mix(__packFloat64(aSign ^ 1u, 0x7ff, 0u, 0u), __propagateFloat64NaN(a, b), propagate);
}
expDiff = mix(expDiff, expDiff + 1, aExp == 0);
aFracHi = mix(aFracHi | 0x40000000u, aFracHi, aExp == 0);
__shift64RightJamming(aFracHi, aFracLo, - expDiff, aFracHi, aFracLo);
bFracHi |= 0x40000000u;
__sub64(bFracHi, bFracLo, aFracHi, aFracLo, zFrac0, zFrac1);
zExp = bExp;
aSign ^= 1u;
--zExp;
return __normalizeRoundAndPackFloat64(aSign, zExp - 10, zFrac0, zFrac1);
}
if (aExp == 0x7FF) {
bool propagate = (aFracHi | aFracLo | bFracHi | bFracLo) != 0u;
return mix(0xFFFFFFFFFFFFFFFFUL, __propagateFloat64NaN(a, b), propagate);
}
bExp = mix(bExp, 1, aExp == 0);
aExp = mix(aExp, 1, aExp == 0);
bool zexp_normal = false;
bool blta = true;
if (bFracHi < aFracHi) {
__sub64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1);
zexp_normal = true;
}
else if (aFracHi < bFracHi) {
__sub64(bFracHi, bFracLo, aFracHi, aFracLo, zFrac0, zFrac1);
blta = false;
zexp_normal = true;
}
else if (bFracLo < aFracLo) {
__sub64(aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1);
zexp_normal = true;
}
else if (aFracLo < bFracLo) {
__sub64(bFracHi, bFracLo, aFracHi, aFracLo, zFrac0, zFrac1);
blta = false;
zexp_normal = true;
}
zExp = mix(bExp, aExp, blta);
aSign = mix(aSign ^ 1u, aSign, blta);
uint64_t retval_0 = __packFloat64(uint(FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN), 0, 0u, 0u);
uint64_t retval_1 = __normalizeRoundAndPackFloat64(aSign, zExp - 11, zFrac0, zFrac1);
return mix(retval_0, retval_1, zexp_normal);
}
}
/* Multiplies `a' by `b' to obtain a 64-bit product. The product is broken
* into two 32-bit pieces which are stored at the locations pointed to by
* `z0Ptr' and `z1Ptr'.
*/
void
__mul32To64(uint a, uint b, out uint z0Ptr, out uint z1Ptr)
{
uint aLow = a & 0x0000FFFFu;
uint aHigh = a>>16;
uint bLow = b & 0x0000FFFFu;
uint bHigh = b>>16;
uint z1 = aLow * bLow;
uint zMiddleA = aLow * bHigh;
uint zMiddleB = aHigh * bLow;
uint z0 = aHigh * bHigh;
zMiddleA += zMiddleB;
z0 += ((uint(zMiddleA < zMiddleB)) << 16) + (zMiddleA >> 16);
zMiddleA <<= 16;
z1 += zMiddleA;
z0 += uint(z1 < zMiddleA);
z1Ptr = z1;
z0Ptr = z0;
}
/* Multiplies the 64-bit value formed by concatenating `a0' and `a1' to the
* 64-bit value formed by concatenating `b0' and `b1' to obtain a 128-bit
* product. The product is broken into four 32-bit pieces which are stored at
* the locations pointed to by `z0Ptr', `z1Ptr', `z2Ptr', and `z3Ptr'.
*/
void
__mul64To128(uint a0, uint a1, uint b0, uint b1,
out uint z0Ptr,
out uint z1Ptr,
out uint z2Ptr,
out uint z3Ptr)
{
uint z0 = 0u;
uint z1 = 0u;
uint z2 = 0u;
uint z3 = 0u;
uint more1 = 0u;
uint more2 = 0u;
__mul32To64(a1, b1, z2, z3);
__mul32To64(a1, b0, z1, more2);
__add64(z1, more2, 0u, z2, z1, z2);
__mul32To64(a0, b0, z0, more1);
__add64(z0, more1, 0u, z1, z0, z1);
__mul32To64(a0, b1, more1, more2);
__add64(more1, more2, 0u, z2, more1, z2);
__add64(z0, z1, 0u, more1, z0, z1);
z3Ptr = z3;
z2Ptr = z2;
z1Ptr = z1;
z0Ptr = z0;
}
/* Normalizes the subnormal double-precision floating-point value represented
* by the denormalized significand formed by the concatenation of `aFrac0' and
* `aFrac1'. The normalized exponent is stored at the location pointed to by
* `zExpPtr'. The most significant 21 bits of the normalized significand are
* stored at the location pointed to by `zFrac0Ptr', and the least significant
* 32 bits of the normalized significand are stored at the location pointed to
* by `zFrac1Ptr'.
*/
void
__normalizeFloat64Subnormal(uint aFrac0, uint aFrac1,
out int zExpPtr,
out uint zFrac0Ptr,
out uint zFrac1Ptr)
{
int shiftCount;
uint temp_zfrac0, temp_zfrac1;
shiftCount = __countLeadingZeros32(mix(aFrac0, aFrac1, aFrac0 == 0u)) - 11;
zExpPtr = mix(1 - shiftCount, -shiftCount - 31, aFrac0 == 0u);
temp_zfrac0 = mix(aFrac1<<shiftCount, aFrac1>>(-shiftCount), shiftCount < 0);
temp_zfrac1 = mix(0u, aFrac1<<(shiftCount & 31), shiftCount < 0);
__shortShift64Left(aFrac0, aFrac1, shiftCount, zFrac0Ptr, zFrac1Ptr);
zFrac0Ptr = mix(zFrac0Ptr, temp_zfrac0, aFrac0 == 0);
zFrac1Ptr = mix(zFrac1Ptr, temp_zfrac1, aFrac0 == 0);
}
/* Returns the result of multiplying the double-precision floating-point values
* `a' and `b'. The operation is performed according to the IEEE Standard for
* Floating-Point Arithmetic.
*/
uint64_t
__fmul64(uint64_t a, uint64_t b)
{
uint zFrac0 = 0u;
uint zFrac1 = 0u;
uint zFrac2 = 0u;
uint zFrac3 = 0u;
int zExp;
uint aFracLo = __extractFloat64FracLo(a);
uint aFracHi = __extractFloat64FracHi(a);
uint bFracLo = __extractFloat64FracLo(b);
uint bFracHi = __extractFloat64FracHi(b);
int aExp = __extractFloat64Exp(a);
uint aSign = __extractFloat64Sign(a);
int bExp = __extractFloat64Exp(b);
uint bSign = __extractFloat64Sign(b);
uint zSign = aSign ^ bSign;
if (aExp == 0x7FF) {
if (((aFracHi | aFracLo) != 0u) ||
((bExp == 0x7FF) && ((bFracHi | bFracLo) != 0u))) {
return __propagateFloat64NaN(a, b);
}
if ((uint(bExp) | bFracHi | bFracLo) == 0u)
return 0xFFFFFFFFFFFFFFFFUL;
return __packFloat64(zSign, 0x7FF, 0u, 0u);
}
if (bExp == 0x7FF) {
if ((bFracHi | bFracLo) != 0u)
return __propagateFloat64NaN(a, b);
if ((uint(aExp) | aFracHi | aFracLo) == 0u)
return 0xFFFFFFFFFFFFFFFFUL;
return __packFloat64(zSign, 0x7FF, 0u, 0u);
}
if (aExp == 0) {
if ((aFracHi | aFracLo) == 0u)
return __packFloat64(zSign, 0, 0u, 0u);
__normalizeFloat64Subnormal(aFracHi, aFracLo, aExp, aFracHi, aFracLo);
}
if (bExp == 0) {
if ((bFracHi | bFracLo) == 0u)
return __packFloat64(zSign, 0, 0u, 0u);
__normalizeFloat64Subnormal(bFracHi, bFracLo, bExp, bFracHi, bFracLo);
}
zExp = aExp + bExp - 0x400;
aFracHi |= 0x00100000u;
__shortShift64Left(bFracHi, bFracLo, 12, bFracHi, bFracLo);
__mul64To128(
aFracHi, aFracLo, bFracHi, bFracLo, zFrac0, zFrac1, zFrac2, zFrac3);
__add64(zFrac0, zFrac1, aFracHi, aFracLo, zFrac0, zFrac1);
zFrac2 |= uint(zFrac3 != 0u);
if (0x00200000u <= zFrac0) {
__shift64ExtraRightJamming(
zFrac0, zFrac1, zFrac2, 1, zFrac0, zFrac1, zFrac2);
++zExp;
}
return __roundAndPackFloat64(zSign, zExp, zFrac0, zFrac1, zFrac2);
}
/* Shifts the 64-bit value formed by concatenating `a0' and `a1' right by the
* number of bits given in `count'. Any bits shifted off are lost. The value
* of `count' can be arbitrarily large; in particular, if `count' is greater
* than 64, the result will be 0. The result is broken into two 32-bit pieces
* which are stored at the locations pointed to by `z0Ptr' and `z1Ptr'.
*/
void
__shift64Right(uint a0, uint a1,
int count,
out uint z0Ptr,
out uint z1Ptr)
{
uint z0;
uint z1;
int negCount = (-count) & 31;
z0 = 0u;
z0 = mix(z0, (a0 >> count), count < 32);
z0 = mix(z0, a0, count == 0);
z1 = mix(0u, (a0 >> (count & 31)), count < 64);
z1 = mix(z1, (a0<<negCount) | (a1>>count), count < 32);
z1 = mix(z1, a0, count == 0);
z1Ptr = z1;
z0Ptr = z0;
}
/* Returns the result of converting the double-precision floating-point value
* `a' to the unsigned integer format. The conversion is performed according
* to the IEEE Standard for Floating-Point Arithmetic.
*/
uint
__fp64_to_uint(uint64_t a)
{
uint aFracLo = __extractFloat64FracLo(a);
uint aFracHi = __extractFloat64FracHi(a);
int aExp = __extractFloat64Exp(a);
uint aSign = __extractFloat64Sign(a);
if ((aExp == 0x7FF) && ((aFracHi | aFracLo) != 0u))
return 0xFFFFFFFFu;
aFracHi |= mix(0u, 0x00100000u, aExp != 0);
int shiftDist = 0x427 - aExp;
if (0 < shiftDist)
__shift64RightJamming(aFracHi, aFracLo, shiftDist, aFracHi, aFracLo);
if ((aFracHi & 0xFFFFF000u) != 0u)
return mix(~0u, 0u, (aSign != 0u));
uint z = 0u;
uint zero = 0u;
__shift64Right(aFracHi, aFracLo, 12, zero, z);
uint expt = mix(~0u, 0u, (aSign != 0u));
return mix(z, expt, (aSign != 0u) && (z != 0u));
}
uint64_t
__uint_to_fp64(uint a)
{
if (a == 0u)
return 0ul;
int shiftDist = __countLeadingZeros32(a) + 21;
uint aHigh = 0u;
uint aLow = 0u;
int negCount = (- shiftDist) & 31;
aHigh = mix(0u, a<< shiftDist - 32, shiftDist < 64);
aLow = 0u;
aHigh = mix(aHigh, 0u, shiftDist == 0);
aLow = mix(aLow, a, shiftDist ==0);
aHigh = mix(aHigh, a >> negCount, shiftDist < 32);
aLow = mix(aLow, a << shiftDist, shiftDist < 32);
return __packFloat64(0u, 0x432 - shiftDist, aHigh, aLow);
}
/* Returns the result of converting the double-precision floating-point value
* `a' to the 32-bit two's complement integer format. The conversion is
* performed according to the IEEE Standard for Floating-Point Arithmetic---
* which means in particular that the conversion is rounded according to the
* current rounding mode. If `a' is a NaN, the largest positive integer is
* returned. Otherwise, if the conversion overflows, the largest integer with
* the same sign as `a' is returned.
*/
int
__fp64_to_int(uint64_t a)
{
uint aFracLo = __extractFloat64FracLo(a);
uint aFracHi = __extractFloat64FracHi(a);
int aExp = __extractFloat64Exp(a);
uint aSign = __extractFloat64Sign(a);
uint absZ = 0u;
uint aFracExtra = 0u;
int shiftCount = aExp - 0x413;
if (0 <= shiftCount) {
if (0x41E < aExp) {
if ((aExp == 0x7FF) && bool(aFracHi | aFracLo))
aSign = 0u;
return mix(0x7FFFFFFF, 0x80000000, bool(aSign));
}
__shortShift64Left(aFracHi | 0x00100000u, aFracLo, shiftCount, absZ, aFracExtra);
} else {
if (aExp < 0x3FF)
return 0;
aFracHi |= 0x00100000u;
aFracExtra = ( aFracHi << (shiftCount & 31)) | aFracLo;
absZ = aFracHi >> (- shiftCount);
}
int z = mix(int(absZ), -int(absZ), (aSign != 0u));
int nan = mix(0x7FFFFFFF, 0x80000000, bool(aSign));
return mix(z, nan, bool(aSign ^ uint(z < 0)) && bool(z));
}
/* Returns the result of converting the 32-bit two's complement integer `a'
* to the double-precision floating-point format. The conversion is performed
* according to the IEEE Standard for Floating-Point Arithmetic.
*/
uint64_t
__int_to_fp64(int a)
{
uint zFrac0 = 0u;
uint zFrac1 = 0u;
if (a==0)
return __packFloat64(0u, 0, 0u, 0u);
uint zSign = uint(a < 0);
uint absA = mix(uint(a), uint(-a), a < 0);
int shiftCount = __countLeadingZeros32(absA) - 11;
if (0 <= shiftCount) {
zFrac0 = absA << shiftCount;
zFrac1 = 0u;
} else {
__shift64Right(absA, 0u, -shiftCount, zFrac0, zFrac1);
}
return __packFloat64(zSign, 0x412 - shiftCount, zFrac0, zFrac1);
}
/* Packs the sign `zSign', exponent `zExp', and significand `zFrac' into a
* single-precision floating-point value, returning the result. After being
* shifted into the proper positions, the three fields are simply added
* together to form the result. This means that any integer portion of `zSig'
* will be added into the exponent. Since a properly normalized significand
* will have an integer portion equal to 1, the `zExp' input should be 1 less
* than the desired result exponent whenever `zFrac' is a complete, normalized
* significand.
*/
float
__packFloat32(uint zSign, int zExp, uint zFrac)
{
return uintBitsToFloat((zSign<<31) + (uint(zExp)<<23) + zFrac);
}
/* Takes an abstract floating-point value having sign `zSign', exponent `zExp',
* and significand `zFrac', and returns the proper single-precision floating-
* point value corresponding to the abstract input. Ordinarily, the abstract
* value is simply rounded and packed into the single-precision format, with
* the inexact exception raised if the abstract input cannot be represented
* exactly. However, if the abstract value is too large, the overflow and
* inexact exceptions are raised and an infinity or maximal finite value is
* returned. If the abstract value is too small, the input value is rounded to
* a subnormal number, and the underflow and inexact exceptions are raised if
* the abstract input cannot be represented exactly as a subnormal single-
* precision floating-point number.
* The input significand `zFrac' has its binary point between bits 30
* and 29, which is 7 bits to the left of the usual location. This shifted
* significand must be normalized or smaller. If `zFrac' is not normalized,
* `zExp' must be 0; in that case, the result returned is a subnormal number,
* and it must not require rounding. In the usual case that `zFrac' is
* normalized, `zExp' must be 1 less than the "true" floating-point exponent.
* The handling of underflow and overflow follows the IEEE Standard for
* Floating-Point Arithmetic.
*/
float
__roundAndPackFloat32(uint zSign, int zExp, uint zFrac)
{
bool roundNearestEven;
int roundIncrement;
int roundBits;
roundNearestEven = FLOAT_ROUNDING_MODE == FLOAT_ROUND_NEAREST_EVEN;
roundIncrement = 0x40;
if (!roundNearestEven) {
if (FLOAT_ROUNDING_MODE == FLOAT_ROUND_TO_ZERO) {
roundIncrement = 0;
} else {
roundIncrement = 0x7F;
if (zSign != 0u) {
if (FLOAT_ROUNDING_MODE == FLOAT_ROUND_UP)
roundIncrement = 0;
} else {
if (FLOAT_ROUNDING_MODE == FLOAT_ROUND_DOWN)
roundIncrement = 0;
}
}
}
roundBits = int(zFrac & 0x7Fu);
if (0xFDu <= uint(zExp)) {
if ((0xFD < zExp) || ((zExp == 0xFD) && (int(zFrac) + roundIncrement) < 0))
return __packFloat32(zSign, 0xFF, 0u) - float(roundIncrement == 0);
int count = -zExp;
bool zexp_lt0 = zExp < 0;
uint zFrac_lt0 = mix(uint(zFrac != 0u), (zFrac>>count) | uint((zFrac<<((-count) & 31)) != 0u), (-zExp) < 32);
zFrac = mix(zFrac, zFrac_lt0, zexp_lt0);
roundBits = mix(roundBits, int(zFrac) & 0x7f, zexp_lt0);
zExp = mix(zExp, 0, zexp_lt0);
}
zFrac = (zFrac + uint(roundIncrement))>>7;
zFrac &= ~uint(((roundBits ^ 0x40) == 0) && roundNearestEven);
return __packFloat32(zSign, mix(zExp, 0, zFrac == 0u), zFrac);
}
/* Returns the result of converting the double-precision floating-point value
* `a' to the single-precision floating-point format. The conversion is
* performed according to the IEEE Standard for Floating-Point Arithmetic.
*/
float
__fp64_to_fp32(uint64_t __a)
{
uvec2 a = unpackUint2x32(__a);
uint zFrac = 0u;
uint allZero = 0u;
uint aFracLo = __extractFloat64FracLo(__a);
uint aFracHi = __extractFloat64FracHi(__a);
int aExp = __extractFloat64Exp(__a);
uint aSign = __extractFloat64Sign(__a);
if (aExp == 0x7FF) {
__shortShift64Left(a.y, a.x, 12, a.y, a.x);
float rval = uintBitsToFloat((aSign<<31) | 0x7FC00000u | (a.y>>9));
rval = mix(__packFloat32(aSign, 0xFF, 0u), rval, (aFracHi | aFracLo) != 0u);
return rval;
}
__shift64RightJamming(aFracHi, aFracLo, 22, allZero, zFrac);
zFrac = mix(zFrac, zFrac | 0x40000000u, aExp != 0);
return __roundAndPackFloat32(aSign, aExp - 0x381, zFrac);
}
/* Returns the result of converting the single-precision floating-point value
* `a' to the double-precision floating-point format.
*/
uint64_t
__fp32_to_fp64(float f)
{
uint a = floatBitsToUint(f);
uint aFrac = a & 0x007FFFFFu;
int aExp = int((a>>23) & 0xFFu);
uint aSign = a>>31;
uint zFrac0 = 0u;
uint zFrac1 = 0u;
if (aExp == 0xFF) {
if (aFrac != 0u) {
uint nanLo = 0u;
uint nanHi = a<<9;
__shift64Right(nanHi, nanLo, 12, nanHi, nanLo);
nanHi |= ((aSign<<31) | 0x7FF80000u);
return packUint2x32(uvec2(nanLo, nanHi));
}
return __packFloat64(aSign, 0x7FF, 0u, 0u);
}
if (aExp == 0) {
if (aFrac == 0u)
return __packFloat64(aSign, 0, 0u, 0u);
/* Normalize subnormal */
int shiftCount = __countLeadingZeros32(aFrac) - 8;
aFrac <<= shiftCount;
aExp = 1 - shiftCount;
--aExp;
}
__shift64Right(aFrac, 0u, 3, zFrac0, zFrac1);
return __packFloat64(aSign, aExp + 0x380, zFrac0, zFrac1);
}
|