-
Notifications
You must be signed in to change notification settings - Fork 1
/
simd_abi.h
1704 lines (1438 loc) · 71.7 KB
/
simd_abi.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/* SPDX-License-Identifier: GPL-3.0-or-later WITH GCC-exception-3.1 */
/* Copyright © 2023-2024 GSI Helmholtzzentrum fuer Schwerionenforschung GmbH
* Matthias Kretz <[email protected]>
*/
#ifndef PROTOTYPE_SIMD_ABI_H_
#define PROTOTYPE_SIMD_ABI_H_
#include "simd_converter.h"
#include "simd_scalar.h"
#include "simd_builtin.h"
#include "simd_x86.h"
#include <cstdint>
namespace std
{
namespace __detail
{
template <int _Bytes>
using _ScalarAbiWrapper = _ScalarAbi;
template <template <int> class...>
struct _AbiList
{
template <typename, int> static constexpr bool _S_has_valid_abi = false;
template <typename, int> using _FirstValidAbi = void;
template <typename, int> using _BestPartialAbi = void;
};
template <template <int> class _A0, template <int> class... _Rest>
struct _AbiList<_A0, _Rest...>
{
template <typename _Tp, int _Np>
static constexpr bool _S_A0_is_valid
= _A0<_Np>::template _IsValid<_Tp>::value and _A0<_Np>::_S_size == _Np;
template <typename _Tp, int _Np>
static constexpr bool _S_has_valid_abi
= _S_A0_is_valid<_Tp, _Np> or _AbiList<_Rest...>::template _S_has_valid_abi<_Tp, _Np>;
template <typename _Tp, int _Np>
using _FirstValidAbi = conditional_t<
_S_A0_is_valid<_Tp, _Np>, _A0<_Np>,
typename _AbiList<_Rest...>::template _FirstValidAbi<_Tp, _Np>>;
template <typename _Tp, int _Np>
static consteval auto
_S_find_next_valid_abi()
{
constexpr int _Next = std::__bit_ceil(_Np) / 2;
using _NextAbi = _A0<_Next>;
if constexpr (_Next <= 1) // break recursion
return _A0<_Np>();
else if constexpr (_NextAbi::_S_is_partial == false
and _NextAbi::template _IsValid<_Tp>::value)
return _NextAbi();
else
return _S_find_next_valid_abi<_Tp, _Next>();
}
template <typename _Tp, int _Np>
static constexpr auto
_S_determine_best_abi()
{
static_assert(_Np >= 1);
if constexpr (_Np == 1)
return _ScalarAbi();
else
{
constexpr int __fullsize = _A0<_Np>::_S_full_size;
constexpr bool __excess_padding = __fullsize / 2 >= _Np;
// _A0<_Np> is good if:
// 1. The ABI tag is valid for _Tp
// 2. The storage overhead is no more than padding to fill the next
// power-of-2 number of bytes
if constexpr (_S_A0_is_valid<_Tp, _Np> and not __excess_padding)
return _A0<_Np>();
else if constexpr (__fullsize == 1)
return _A0<_Np>();
else
{
using _Bp = decltype(_S_find_next_valid_abi<_Tp, _Np>());
if constexpr (_Bp::template _IsValid<_Tp>::value
and _Bp::_S_size <= _Np)
return _Bp{};
else
return _AbiList<_Rest...>::template _S_determine_best_abi<_Tp, _Np>();
}
}
}
/**
* Alias for the best ABI tag that can either hold all _Np values of type _Tp or as much as
* possible.
*/
template <typename _Tp, int _Np>
using _BestPartialAbi = decltype(_S_determine_best_abi<_Tp, _Np>());
};
using _AllNativeAbis = _AbiList<_Avx512Abi, _VecAbi, _ScalarAbiWrapper>;
using namespace vir::literals;
template <typename _Abi0, size_t... _Is>
struct _SimdImplArray;
template <typename _Abi0, size_t... _Is>
struct _MaskImplArray;
template <_SimdSizeType _Np, typename _Tag, auto = __build_flags()>
struct _SimdImplAbiCombine;
template <_SimdSizeType _Np, typename _Tag, auto = __build_flags()>
struct _MaskImplAbiCombine;
template <typename _Tp, __valid_abi_tag<_Tp>... _As>
struct _SimdTuple;
template <typename _Tp, _SimdSizeType _Np>
struct _NextAbiTuple
{
using _Native = _AllNativeAbis::_BestPartialAbi<_Tp, _Np>;
static constexpr int _S_native_size = simd_size_v<_Tp, _Native>;
static constexpr int _S_array_size = _Np / _S_native_size;
using type
= std::conditional_t<_S_array_size >= 2, _AbiArray<_Native, _S_array_size>, _Native>;
};
template <typename _Tp, int _Np, typename _Tuple,
typename _Next = typename _NextAbiTuple<_Tp, _Np>::type,
int _Remain = _Np - int(simd_size_v<_Tp, _Next>)>
struct __fixed_size_storage_builder;
template <typename _Tp, int _Np, typename... _As, typename _Next>
struct __fixed_size_storage_builder<_Tp, _Np, _SimdTuple<_Tp, _As...>, _Next, 0>
{ using type = _SimdTuple<_Tp, _As..., _Next>; };
template <typename _Tp, int _Np, typename... _As, typename _Next, int _Remain>
struct __fixed_size_storage_builder<_Tp, _Np, _SimdTuple<_Tp, _As...>, _Next, _Remain>
{
using type = typename __fixed_size_storage_builder<
_Tp, _Remain, _SimdTuple<_Tp, _As..., _Next>>::type;
};
template <typename _Tp, int _Np>
using __fixed_size_storage_t
= typename __fixed_size_storage_builder<_Tp, _Np, _SimdTuple<_Tp>>::type;
}
template <typename _Abi0, __detail::_SimdSizeType _Np>
struct _AbiArray
{
using _SimdSizeType = __detail::_SimdSizeType;
static constexpr _SimdSizeType _S_abiarray_size = _Np;
static constexpr _SimdSizeType _S_size = _Np * _Abi0::_S_size;
static constexpr _SimdSizeType _S_full_size = _Np * _Abi0::_S_size;
static constexpr bool _S_is_partial = false;
static constexpr _SimdSizeType _S_max_size
= std::max(_SimdSizeType(64), _SimdSizeType(8 * _Abi0::_S_size));
struct _IsValidAbiTag
: bool_constant<_Abi0::_IsValidAbiTag::value and not _Abi0::_S_is_partial
and _Np >= 2 and _S_size <= _S_max_size>
{};
template <typename _Tp>
using _IsValidSizeFor = _Abi0::template _IsValidSizeFor<_Tp>;
template <typename _Tp>
struct _IsValid
: conjunction<_IsValidAbiTag, __detail::__is_vectorizable<_Tp>, _IsValidSizeFor<_Tp>>
{};
using _SimdImpl = decltype([]<size_t... _Is>(index_sequence<_Is...>)
-> __detail::_SimdImplArray<_Abi0, _Is...> {
return {};
}(make_index_sequence<_Np>()));
using _MaskImpl = decltype([]<size_t... _Is>(index_sequence<_Is...>)
-> __detail::_MaskImplArray<_Abi0, _Is...> {
return {};
}(make_index_sequence<_Np>()));
template <typename _Tp>
using _SimdMember = std::array<typename _Abi0::template _SimdMember<_Tp>, _Np>;
template <typename _Tp>
using _MaskMember = std::array<typename _Abi0::template _MaskMember<_Tp>, _Np>;
template <typename _Tp>
struct __traits
: __detail::_InvalidTraits
{};
template <typename _Tp>
requires _IsValid<_Tp>::value
struct __traits<_Tp>
{
// conversions to _AbiArray should be explicit unless _FromAbi is _AbiCombine
template <typename _FromAbi>
static constexpr bool _S_explicit_mask_conversion
= not is_same_v<_FromAbi, _AbiCombine<_S_size, __detail::_NativeAbi<_Tp>>>;
using _SimdImpl = _AbiArray::_SimdImpl;
using _MaskImpl = _AbiArray::_MaskImpl;
using _SimdMember = std::array<typename _Abi0::template _SimdMember<_Tp>, _Np>;
using _MaskMember = std::array<typename _Abi0::template _MaskMember<_Tp>, _Np>;
static constexpr size_t _S_simd_align = alignof(_SimdMember);
static constexpr size_t _S_mask_align = alignof(_MaskMember);
static constexpr _SimdSizeType _S_size = _Np * _Abi0::_S_size;
static constexpr _SimdSizeType _S_full_size = _Np * _Abi0::_S_size;
static constexpr bool _S_is_partial = false;
class _MaskCastType
{};
class _SimdCastType
{};
};
};
template <__detail::_SimdSizeType _Np, typename _Tag>
struct _AbiCombine
{
static constexpr __detail::_SimdSizeType _S_size = _Np;
static constexpr __detail::_SimdSizeType _S_full_size = _Np;
struct _IsValidAbiTag
: public bool_constant<(_Np > 1)>
{};
template <typename _Tp>
static constexpr __detail::_SimdSizeType _S_max_size = 64;
template <typename _Tp>
struct _IsValidSizeFor
: bool_constant<(_Np <= _S_max_size<_Tp>)>
{};
template <typename _Tp>
struct _IsValid
: conjunction<_IsValidAbiTag, __detail::__is_vectorizable<_Tp>, _IsValidSizeFor<_Tp>>
{};
_GLIBCXX_SIMD_INTRINSIC static constexpr __detail::_SanitizedBitMask<_Np>
_S_masked(__detail::_BitMask<_Np> __x)
{ return __x._M_sanitized(); }
_GLIBCXX_SIMD_INTRINSIC static constexpr __detail::_SanitizedBitMask<_Np>
_S_masked(__detail::_SanitizedBitMask<_Np> __x)
{ return __x; }
struct _CommonImpl {};
using _SimdImpl = __detail::_SimdImplAbiCombine<_Np, _Tag>;
using _MaskImpl = __detail::_MaskImplAbiCombine<_Np, _Tag>;
template <typename _Tp>
using _SimdMember = __detail::__fixed_size_storage_t<_Tp, _Np>;
template <typename>
using _MaskMember = __detail::_SanitizedBitMask<_Np>;
template <typename _Tp>
struct __traits
: __detail::_InvalidTraits
{};
template <typename _Tp>
requires _IsValid<_Tp>::value
struct __traits<_Tp>
{
// conversions to _AbiCombine should always be explicit
template <typename _FromAbi>
static constexpr bool _S_explicit_mask_conversion = true;
using _SimdImpl = __detail::_SimdImplAbiCombine<_Np, _Tag>;
using _MaskImpl = __detail::_MaskImplAbiCombine<_Np, _Tag>;
using _SimdMember = __detail::__fixed_size_storage_t<_Tp, _Np>;
using _MaskMember = __detail::_SanitizedBitMask<_Np>;
static constexpr __detail::_SimdSizeType _S_size = _Np;
static constexpr __detail::_SimdSizeType _S_full_size = _Np;
static constexpr size_t _S_simd_align = alignof(_SimdMember);
static constexpr size_t _S_mask_align = alignof(_MaskMember);
struct _SimdCastType
{
_GLIBCXX_SIMD_ALWAYS_INLINE constexpr
_SimdCastType(const array<_Tp, _Np>& __a)
{ __builtin_memcpy(&_M_data, __a.data(), _Np * sizeof(_Tp)); }
_GLIBCXX_SIMD_ALWAYS_INLINE constexpr
_SimdCastType(const _SimdMember& __dd)
: _M_data(__dd)
{}
_GLIBCXX_SIMD_ALWAYS_INLINE constexpr explicit
operator const _SimdMember &() const { return _M_data; }
private:
_SimdMember _M_data;
};
class _MaskCastType
{
_MaskCastType() = delete;
};
};
};
namespace __detail
{
template <typename _Abi0, size_t... _Is>
struct _SimdImplArray
{
static constexpr _SimdSizeType _Np = sizeof...(_Is);
using abi_type = _AbiArray<_Abi0, _Np>;
template <typename _Tp>
using _TypeTag = _Tp*;
template <__vectorizable _Tp>
using _SimdMember = typename abi_type::template _SimdMember<_Tp>;
template <typename _Tp> // std::array of __vec_builtin
using _ValueTypeOf = __value_type_of<typename _Tp::value_type>;
template <typename _Tp>
struct _MaskMemberImpl
{ using type = typename abi_type::template _MaskMember<_ValueTypeOf<_Tp>>; };
template <__vectorizable _Tp>
struct _MaskMemberImpl<_Tp>
{ using type = typename abi_type::template _MaskMember<_Tp>; };
template <typename _Tp>
using _MaskMember = typename _MaskMemberImpl<_Tp>::type;
static constexpr _SimdSizeType _S_size = abi_type::_S_size;
static constexpr _SimdSizeType _S_chunk_size = _Abi0::_S_size;
using _Impl0 = typename _Abi0::_SimdImpl;
using _MaskImpl = typename abi_type::_MaskImpl;
template <__vectorizable _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _SimdMember<_Tp>
_S_broadcast(_Tp __x) noexcept
{ return {((void)_Is, _Impl0::_S_broadcast(__x))...}; }
template <__vectorizable _Tp, typename _Fp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _SimdMember<_Tp>
_S_generator(_Fp&& __gen)
{
return {_Impl0::template _S_generator<_Tp>([&] [[__gnu__::__always_inline__]] (auto __i) {
return __gen(__ic<__i + _Is * _S_chunk_size>);
})...};
}
template <__vectorizable _Tp, typename _Up>
_GLIBCXX_SIMD_INTRINSIC static constexpr _SimdMember<_Tp>
_S_load(const _Up* __mem, _TypeTag<_Tp> __tag) noexcept
{ return {_Impl0::_S_load(__mem + _Is * _S_chunk_size, __tag)...}; }
template <typename _Tp, typename _Up>
static constexpr inline _Tp
_S_masked_load(_Tp const& __merge, _MaskMember<_Tp> const& __k,
const _Up* __mem) noexcept
{
_SimdMember<_Tp> __ret = __merge;
(_Impl0::_S_masked_load(__ret[_Is], __k[_Is], __mem + _Is * _S_chunk_size), ...);
return __ret;
}
template <__vec_builtin _TV, typename _Up>
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_store(const array<_TV, _Np>& __v, _Up* __mem, _TypeTag<__value_type_of<_TV>> __tag) noexcept
{ (_Impl0::_S_store(__v[_Is], __mem + _Is * _S_chunk_size, __tag), ...); }
template <typename _Tp, typename _Up>
static constexpr inline void
_S_masked_store(_Tp const& __v, _Up* __mem, const _MaskMember<_Tp> __k) noexcept
{ (_Impl0::_S_masked_store(__v[_Is], __mem + _Is * _S_chunk_size, __k[_Is]), ...); }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_complement(_Tp const& __x) noexcept
{ return {_Impl0::_S_complement(__x[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_unary_minus(_Tp const& __x) noexcept
{ return {_Impl0::_S_unary_minus(__x[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_plus(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_plus(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_minus(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_minus(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_multiplies(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_multiplies(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_divides(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_divides(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_modulus(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_modulus(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_and(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_bit_and(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_or(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_bit_or(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_xor(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_bit_xor(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_shift_left(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_bit_shift_left(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_shift_right(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_bit_shift_right(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_shift_left(_Tp const& __x, int __y) noexcept
{ return {_Impl0::_S_bit_shift_left(__x[_Is], __y)...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_shift_right(_Tp const& __x, int __y) noexcept
{ return {_Impl0::_S_bit_shift_right(__x[_Is], __y)...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember<_Tp>
_S_equal_to(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_equal_to(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember<_Tp>
_S_not_equal_to(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_not_equal_to(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember<_Tp>
_S_less(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_less(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember<_Tp>
_S_less_equal(_Tp const& __x, _Tp const& __y) noexcept
{ return {_Impl0::_S_less_equal(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember<_Tp>
_S_negate(_Tp const& __x) noexcept
{ return {_Impl0::_S_negate(__x[_Is])...}; }
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_increment(auto& __x)
{ (_Impl0::_S_increment(__x[_Is]), ...); }
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_decrement(auto& __x)
{ (_Impl0::_S_decrement(__x[_Is]), ...); }
template <typename _TV>
_GLIBCXX_SIMD_INTRINSIC static constexpr __value_type_of<_TV>
_S_get(const array<_TV, _Np>& __v, _SimdSizeType __i)
{ return _Impl0::_S_get(__v[__i / _S_chunk_size], __i % _S_chunk_size); }
template <typename _TV>
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_set(array<_TV, _Np>& __v, _SimdSizeType __i, auto&& __x)
{ _Impl0::_S_set(__v[__i / _S_chunk_size], __i % _S_chunk_size, __x); }
// There's a curious case of _SimdMember and _MaskMember mismatch for the AVX w/o AVX2 case:
// The mask type can either be array<int8, 2> or array<int4, 4> depending on whether it
// originates for an integer or floating-point simd comparison. E.g.:
// !simd<float, 16> -> array<int8, 2>
// !simd<unsigned, 16> -> array<int4, 4>
template <typename _Tp, typename _UV, size_t _KSize>
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_masked_assign(const std::array<_UV, _KSize>& __k, _Tp& __lhs,
const __type_identity_t<_Tp>& __rhs)
{
if constexpr (_Np == _KSize)
(_Impl0::_S_masked_assign(__k[_Is], __lhs[_Is], __rhs[_Is]), ...);
else if constexpr (_Np == _KSize * 2)
(_Impl0::_S_masked_assign(__vec_extract_part<_Is % 1, 2>(__k[_Is / 2]),
__lhs[_Is], __rhs[_Is]), ...);
else
__assert_unreachable<_Tp>();
}
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_masked_assign(const __vec_builtin auto __k, _Tp& __lhs,
const __type_identity_t<_Tp>& __rhs)
{
static_assert(_Np == 2);
_Impl0::_S_masked_assign(__vec_extract_part<0, 2>(__k), __lhs[0], __rhs[0]);
_Impl0::_S_masked_assign(__vec_extract_part<1, 2>(__k), __lhs[1], __rhs[1]);
}
template <typename _KV, typename _TV>
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_masked_assign(const array<_KV, _Np>& __k, array<_TV, _Np>& __lhs, __value_type_of<_TV> __rhs)
{ (_Impl0::_S_masked_assign(__k[_Is], __lhs[_Is], __rhs), ...); }
template <__vec_builtin _KV, typename _TV>
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_masked_assign(const array<_KV, _Np / 2>& __k, array<_TV, _Np>& __lhs, __value_type_of<_TV> __rhs)
{
(_Impl0::_S_masked_assign(__vec_extract_part<_Is % 2, 2>(__k[_Is / 2]),
__lhs[_Is], __rhs), ...);
}
template <__vec_builtin _KV, typename _TV>
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_masked_assign(const _KV __k, array<_TV, _Np>& __lhs, __value_type_of<_TV> __rhs)
{ (_Impl0::_S_masked_assign(__vec_extract_part<_Is, _Np>(__k), __lhs[_Is], __rhs), ...); }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr bool
_S_is_constprop(const _Tp& __x)
{
if constexpr (requires {_Impl0::_S_is_constprop(__x[0]);})
return (_Impl0::_S_is_constprop(__x[_Is]) and ...);
else if constexpr (requires {__x[0]._M_is_constprop();})
return (__x[_Is]._M_is_constprop() and ...);
else
return (__builtin_constant_p(__x[_Is]) and ...);
}
template <typename _Tp, typename _BinaryOperation>
static constexpr _Tp
_S_reduce(basic_simd<_Tp, abi_type> __xx, const _BinaryOperation& __binary_op)
{
auto& __x = __data(__xx); // the array was copied by the caller - we're not changing it
(((_Is % 2) == 1 ? (__x[_Is - 1] = __binary_op(__x[_Is - 1], __x[_Is])) : __x[0]), ...);
if constexpr (_Np > 2)
(((_Is % 4) == 2 ? (__x[_Is - 2] = __binary_op(__x[_Is - 2], __x[_Is])) : __x[0]),
...);
if constexpr (_Np > 4)
(((_Is % 8) == 4 ? (__x[_Is - 4] = __binary_op(__x[_Is - 4], __x[_Is])) : __x[0]),
...);
if constexpr (_Np > 8)
(((_Is % 16) == 8 ? (__x[_Is - 8] = __binary_op(__x[_Is - 8], __x[_Is])) : __x[0]),
...);
if constexpr (_Np > 16)
(((_Is % 32) == 16 ? (__x[_Is - 16] = __binary_op(__x[_Is - 16], __x[_Is])) : __x[0]),
...);
if constexpr (_Np > 32)
(((_Is % 64) == 32 ? (__x[_Is - 32] = __binary_op(__x[_Is - 32], __x[_Is])) : __x[0]),
...);
if constexpr (_Np > 64)
(((_Is % 128) == 64 ? (__x[_Is - 64] = __binary_op(__x[_Is - 64], __x[_Is]))
: __x[0]), ...);
static_assert(_Np <= 128);
return std::reduce(basic_simd<_Tp, _Abi0>(__private_init, __x[0]), __binary_op);
}
};
template <typename _Abi0, size_t... _Is>
struct _MaskImplArray
{
static constexpr _SimdSizeType _Np = sizeof...(_Is);
using abi_type = _AbiArray<_Abi0, _Np>;
template <typename _Ts>
using mask_type = std::basic_simd_mask<sizeof(_Ts), _Abi0>;
template <__vectorizable _Tp>
using _MaskMember0 = typename _Abi0::template _MaskMember<_Tp>;
template <__vectorizable _Tp>
using _MaskMember = std::array<_MaskMember0<_Tp>, _Np>;
template <__vectorizable _Tp>
using _SimdMember = typename abi_type::template _SimdMember<_Tp>;
static constexpr _SimdSizeType _S_size = abi_type::_S_size;
// equal to full_size because _AbiArray ensures not-partial _Abi0
static constexpr _SimdSizeType _S_chunk_size = _Abi0::_S_size;
using _Impl0 = typename _Abi0::_MaskImpl;
template <typename _Tp, bool _Sanitized>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_convert_mask(_BitMask<_S_size, _Sanitized> __x)
{
return {_Impl0::template _S_convert_mask<typename _Tp::value_type>(
__x.template _M_extract<_S_chunk_size * _Is, _S_chunk_size>())...};
}
template <typename _Tp, size_t _Bs, typename _UAbi>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember<_Tp>
_S_convert(std::basic_simd_mask<_Bs, _UAbi> __x)
{ return _S_convert_mask<_MaskMember<_Tp>>(__data(__x)); }
_GLIBCXX_SIMD_INTRINSIC static constexpr auto
_S_to_bits_recursive(auto&& __part0, auto&& __part1, auto&&... __rest)
{
if constexpr (sizeof...(__rest))
return _S_to_bits_recursive(__part1._M_prepend(__part0), __rest...);
else
return __part1._M_prepend(__part0);
}
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr auto
_S_to_bits(const std::array<_Tp, _Np>& __k)
{ return _S_to_bits_recursive(_Impl0::_S_to_bits(__k[_Is])...); }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC
static constexpr _Tp
_S_logical_and(const _Tp& __k0, const _Tp& __k1)
{ return {_Impl0::_S_logical_and(__k0[_Is], __k1[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC
static constexpr _Tp
_S_logical_or(const _Tp& __k0, const _Tp& __k1)
{ return {_Impl0::_S_logical_or(__k0[_Is], __k1[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_not(const _Tp& __x)
{ return {_Impl0::_S_bit_not(__x[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_and(const _Tp& __x, const _Tp& __y)
{ return {_Impl0::_S_bit_and(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_or(const _Tp& __x, const _Tp& __y)
{ return {_Impl0::_S_bit_or(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _Tp
_S_bit_xor(const _Tp& __x, const _Tp& __y)
{ return {_Impl0::_S_bit_xor(__x[_Is], __y[_Is])...}; }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr bool
_S_get(const _Tp& __k, _SimdSizeType __i)
{ return _Impl0::_S_get(__k[__i / _S_chunk_size], __i % _S_chunk_size); }
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr void
_S_set(_Tp& __k, _SimdSizeType __i, bool __x)
{ return _Impl0::_S_set(__k[__i / _S_chunk_size], __i % _S_chunk_size, __x); }
template <size_t _Bs>
static constexpr basic_simd_mask<_Bs, _Abi0>
_S_submask(basic_simd_mask<_Bs, abi_type> const& __masks, int __i)
{ return {__private_init, __data(__masks)[__i]}; }
template <size_t _Bs>
_GLIBCXX_SIMD_INTRINSIC
static constexpr bool
_S_any_of(basic_simd_mask<_Bs, abi_type> const& __masks)
{ return (std::any_of(_S_submask(__masks, _Is)) or ...); }
template <size_t _Bs>
_GLIBCXX_SIMD_INTRINSIC
static constexpr bool
_S_all_of(basic_simd_mask<_Bs, abi_type> const& __masks)
{ return (std::all_of(_S_submask(__masks, _Is)) and ...); }
template <size_t _Bs>
_GLIBCXX_SIMD_INTRINSIC
static constexpr bool
_S_none_of(basic_simd_mask<_Bs, abi_type> const& __masks)
{ return (std::none_of(_S_submask(__masks, _Is)) and ...); }
// TODO: benchmark whether it's more efficient to do element-wise reduction of array members
// on -__k or +__k or a sum of popcount on each sub-mask.
// Beware of overflow with _Bs == 1.
template <size_t _Bs>
requires (_Bs == 1)
_GLIBCXX_SIMD_INTRINSIC static constexpr _SimdSizeType
_S_popcount(const basic_simd_mask<_Bs, abi_type> & __k)
{
using _Ip = __mask_integer_from<_Bs>;
using _Up = __make_unsigned_int_t<_Ip>;
if constexpr (-_S_size >= __finite_min_v<_Ip>)
return -_SimdImplArray<_Abi0, _Is...>::_S_reduce(-__k, std::plus<>());
else if constexpr (_S_size <= __finite_max_v<_Up>)
return _SimdImplArray<_Abi0, _Is...>::_S_reduce(
static_cast<basic_simd<_Up, abi_type>>(__k), std::plus<>());
else
return (std::reduce_count(_S_submask(__k, _Is)) + ...);
}
template <size_t _Bs>
static constexpr _SimdSizeType
_S_find_first_set(basic_simd_mask<_Bs, abi_type> const& __masks)
{
if (std::any_of(_S_submask(__masks, 0)))
return std::reduce_min_index(_S_submask(__masks, 0));
for (int __i = 1; __i < _Np - 1; ++__i)
{
if (std::any_of(_S_submask(__masks, __i)))
return __i * _S_chunk_size + std::reduce_min_index(_S_submask(__masks, __i));
}
return (_Np - 1) * _S_chunk_size
+ std::reduce_min_index(_S_submask(__masks, _Np - 1));
}
template <size_t _Bs>
static constexpr _SimdSizeType
_S_find_last_set(basic_simd_mask<_Bs, abi_type> const& __masks)
{
if (std::any_of(_S_submask(__masks, _Np - 1)))
return (_Np - 1) * _S_chunk_size
+ std::reduce_max_index(_S_submask(__masks, _Np - 1));
for (int __i = _Np - 2; __i > 0; --__i)
{
if (std::any_of(_S_submask(__masks, __i)))
return __i * _S_chunk_size + std::reduce_max_index(_S_submask(__masks, __i));
}
return std::reduce_max_index(_S_submask(__masks, 0));
}
template <__vectorizable _Tp, typename _Fp>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember<_Tp>
_S_mask_generator(_Fp&& __gen)
{
if constexpr (requires {_Impl0::template _S_mask_generator<_Tp>(__gen);})
return {_Impl0::template _S_mask_generator<_Tp>([&](auto __i) {
return __gen(__ic<_Is * _S_chunk_size + __i>);
})...};
else
return {[&]<size_t... _Js>(vir::constexpr_value<int> auto __i,
std::index_sequence<_Js...>) {
return _MaskMember0<_Tp>{ -__gen(__ic<__i * _S_chunk_size + _Js>)... };
}(vir::cw<_Is>, std::make_index_sequence<_S_chunk_size>())...};
}
template <typename _Tp>
_GLIBCXX_SIMD_INTRINSIC static constexpr bool
_S_is_constprop(const _Tp& __x)
{
if constexpr (requires {_Impl0::_S_is_constprop(__x[0]);})
return (_Impl0::_S_is_constprop(__x[_Is]) and ...);
else if constexpr (requires {__x[0]._M_is_constprop();})
return (__x[_Is]._M_is_constprop() and ...);
else
return (__builtin_constant_p(__x[_Is]) and ...);
}
};
template <typename _Tp, __valid_abi_tag<_Tp> _A0, _SimdSizeType __offset>
struct _SimdTupleMeta : public _A0::_SimdImpl
{
using _Abi = _A0;
using _Traits = _SimdTraits<_Tp, _Abi>;
using _MaskImpl = typename _Abi::_MaskImpl;
using _MaskMember = typename _Traits::_MaskMember;
static constexpr auto _S_offset = vir::cw<__offset>;
static constexpr auto _S_size = vir::cw<_SimdSizeType(simd_size_v<_Tp, _Abi>)>;
static constexpr _MaskImpl _S_mask_impl = {};
template <size_t _Np, bool _Sanitized>
_GLIBCXX_SIMD_INTRINSIC static constexpr auto
_S_submask(_BitMask<_Np, _Sanitized> __bits)
{ return __bits.template _M_extract<_S_offset, _S_size>(); }
template <size_t _Np, bool _Sanitized>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember
_S_make_mask(_BitMask<_Np, _Sanitized> __bits)
{
return _MaskImpl::template _S_convert_mask<_MaskMember>(
_S_submask(__bits)._M_sanitized());
}
template <__vec_builtin _KV>
_GLIBCXX_SIMD_INTRINSIC static constexpr _MaskMember
_S_make_mask(_KV __k)
{
if constexpr (_S_size == 1)
return __vec_get(__k, __offset) != 0;
else
return __vec_extract_part<__offset, __width_of<_KV>, _S_size>(__k);
}
_GLIBCXX_SIMD_INTRINSIC static constexpr unsigned long long
_S_mask_to_shifted_ullong(_MaskMember __k)
{
unsigned long long __bits = _MaskImpl::_S_to_bits(__k)._M_sanitized()._M_to_bits();
return __bits << _S_offset;
}
_GLIBCXX_SIMD_INTRINSIC static constexpr basic_simd<_Tp, _A0>
_S_to_simd(const _Traits::_SimdMember& __x0)
{ return basic_simd<_Tp, _A0>(__private_init, __x0); }
};
template <typename _Tp, __valid_abi_tag<_Tp>... _As>
struct _SimdTuple
{
static_assert(sizeof...(_As) == 0);
static constexpr auto _S_size = 0_cw;
static constexpr auto _S_tuple_size = 0_cw;
_GLIBCXX_SIMD_INTRINSIC constexpr bool
_M_is_constprop() const
{ return true; }
};
template <typename _Tp, __valid_abi_tag<_Tp> _A0, __valid_abi_tag<_Tp>... _As>
struct _SimdTupleData
{
static_assert(sizeof...(_As) != 0);
typename _A0::template _SimdMember<_Tp> _M_x;
_SimdTuple<_Tp, _As...> _M_tail;
};
template <typename _Tp, __valid_abi_tag<_Tp> _A0>
struct _SimdTupleData<_Tp, _A0>
{
typename _A0::template _SimdMember<_Tp> _M_x;
static constexpr _SimdTuple<_Tp> _M_tail = {};
};
template <typename _Tp, __valid_abi_tag<_Tp> _A0, __valid_abi_tag<_Tp>... _As>
struct _SimdTuple<_Tp, _A0, _As...> : _SimdTupleData<_Tp, _A0, _As...>
{
using _Base = _SimdTupleData<_Tp, _A0, _As...>;
using _Base::_M_x;
using _Base::_M_tail;
using _SimdType = basic_simd<_Tp, _A0>;
using _Simpl0 = _A0::_SimdImpl;
using _Mimpl0 = _A0::_MaskImpl;
static constexpr bool _S_recurse = sizeof...(_As) != 0;
static constexpr auto _S_size = vir::cw<(simd_size_v<_Tp, _A0>)>;
static constexpr auto _S_total_size
= vir::cw<(simd_size_v<_Tp, _A0> + ... + simd_size_v<_Tp, _As>)>;
static constexpr auto _S_tail_size = _SimdTuple<_Tp, _As...>::_S_size;
static constexpr auto _S_tuple_size = vir::cw<_SimdSizeType(sizeof...(_As) + 1)>;
_GLIBCXX_SIMD_INTRINSIC constexpr bool
_M_is_constprop() const
{
bool __r;
if constexpr (requires {_Simpl0::_S_is_constprop(_M_x);})
__r = _Simpl0::_S_is_constprop(_M_x);
else if constexpr (requires {_M_x._M_is_constprop();})
__r = _M_x._M_is_constprop();
else
__r = __builtin_constant_p(_M_x);
return __r and _M_tail._M_is_constprop();
}
_GLIBCXX_SIMD_INTRINSIC constexpr _Tp
operator[](_SimdSizeType __i) const
{
if (not __builtin_is_constant_evaluated())
{
const auto* __ptr = reinterpret_cast<const std::byte*>(this);
_Tp __r;
__builtin_memcpy(&__r, __ptr + __i * sizeof(_Tp), sizeof(_Tp));
return __r;
}
else if (__i < _S_size)
return _Simpl0::_S_get(_M_x, __i);
else if constexpr (_S_recurse)
return _M_tail[__i - _S_size];
else
__invoke_ub("out-of-bounds subscript");
}
_GLIBCXX_SIMD_INTRINSIC constexpr auto
_M_simd_at(vir::constexpr_value<int> auto __i) const
{
if constexpr (__i == 0)
return _SimdType(__private_init, _M_x);
else if constexpr (_S_recurse)
return _M_tail._M_simd_at(__i - 1_cw);
}
template <vir::constexpr_value<int> _Cv = decltype(0_cw)>
_GLIBCXX_SIMD_INTRINSIC constexpr _SimdTuple&
_M_forall(auto&& __fun, _Cv __total_offset = {})
{
__fun(_SimdTupleMeta<_Tp, _A0, __total_offset>(), _M_x);
if constexpr (_S_recurse)
_M_tail._M_forall(__fun, __total_offset + _S_size);
return *this;
}
template <vir::constexpr_value<int> _Cv = decltype(0_cw)>
_GLIBCXX_SIMD_INTRINSIC constexpr _SimdTuple&
_M_forall(const _SimdTuple& __a, auto&& __fun, _Cv __total_offset = {})
{
__fun(_SimdTupleMeta<_Tp, _A0, __total_offset>(), _M_x, __a._M_x);
if constexpr (_S_recurse)
_M_tail._M_forall(__a._M_tail, __fun, __total_offset + _S_size);
return *this;
}
template <vir::constexpr_value<int> _Cv = decltype(0_cw)>
_GLIBCXX_SIMD_INTRINSIC constexpr const _SimdTuple&
_M_forall(auto&& __fun, _Cv __total_offset = {}) const
{
__fun(_SimdTupleMeta<_Tp, _A0, __total_offset>(), _M_x);
if constexpr (_S_recurse)
_M_tail._M_forall(__fun, __total_offset + _S_size);
return *this;
}
template <int __total_offset = 0>
_GLIBCXX_SIMD_INTRINSIC constexpr _SanitizedBitMask<_S_total_size>
_M_test(auto&& __fun, auto&&... __more) const
{
const _SanitizedBitMask<_S_size> __first
= _A0::_MaskImpl::_S_to_bits(__fun(_SimdTupleMeta<_Tp, _A0, __total_offset>(),
_M_x, __more._M_x...));
if constexpr (_S_recurse)
return _M_tail.template _M_test<__total_offset + _S_size>(__fun, __more._M_tail...)
._M_prepend(__first);
else
return __first;
}
constexpr
_SimdTuple() = default;
_GLIBCXX_SIMD_INTRINSIC constexpr explicit
_SimdTuple(auto &&__fun)
{ _M_forall(__fun); }
};