forked from pytorch/pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
BinaryOps.cpp
1199 lines (983 loc) · 44.6 KB
/
BinaryOps.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 <ATen/native/BinaryOps.h>
#include <type_traits>
#include <ATen/ATen.h>
#include <ATen/Dispatch.h>
#include <ATen/MemoryOverlap.h>
#include <ATen/NativeFunctions.h>
#include <ATen/native/TensorIterator.h>
#include <ATen/ExpandUtils.h>
#include <ATen/RedispatchFunctions.h>
#include <torch/library.h>
namespace at {
namespace native {
// These are still needed because we don't have C++ conversions from number
// types (int, float, etc.) to Tensor (only to Scalar). They're not exposed
// to Python.
static void check_convert(const Scalar& scalar, ScalarType scalarType) {
// Validate that is possible to convert scalar to tensor dtype without
// overflow
AT_DISPATCH_ALL_TYPES_AND_COMPLEX_AND3(
at::ScalarType::Bool,
at::ScalarType::BFloat16,
at::ScalarType::Half,
scalarType,
"check_convert",
[&] { scalar.to<scalar_t>(); });
}
static Tensor wrapped_scalar_tensor_and_check_convert(
const Scalar& scalar,
Tensor tensor) {
check_convert(scalar, tensor.scalar_type());
return at::native::wrapped_scalar_tensor(scalar);
}
} // namespace native
namespace meta {
TORCH_META_FUNC2(add, Tensor) (
const Tensor& self, const Tensor& other, const Scalar& alpha
) {
build_borrowing_binary_op(maybe_get_output(), self, other);
native::alpha_check(dtype(), alpha);
}
TORCH_META_FUNC2(sub, Tensor) (
const Tensor& self, const Tensor& other, const Scalar& alpha
) {
native::sub_check(self, other);
build_borrowing_binary_op(maybe_get_output(), self, other);
native::alpha_check(dtype(), alpha);
}
TORCH_META_FUNC2(mul, Tensor) (
const Tensor& self, const Tensor& other
) {
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(div, Tensor) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_float_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(div, Tensor_mode) (const Tensor& self, const Tensor& other, c10::optional<c10::string_view> rounding_mode) {
if (!rounding_mode.has_value()) {
build_borrowing_binary_float_op(maybe_get_output(), self, other);
// NOLINTNEXTLINE(bugprone-branch-clone)
} else if (*rounding_mode == "trunc") {
build_borrowing_binary_op(maybe_get_output(), self, other);
} else if (*rounding_mode == "floor") {
build_borrowing_binary_op(maybe_get_output(), self, other);
} else {
TORCH_CHECK(false,
"div expected rounding_mode to be one of None, 'trunc', or 'floor' "
"but found '", *rounding_mode, "'");
}
}
TORCH_META_FUNC(special_xlog1py) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_float_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC(special_zeta) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_float_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(copysign, Tensor) (
const Tensor& self, const Tensor& other
) {
build_borrowing_binary_float_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC(heaviside) (
const Tensor& self, const Tensor& other
) {
TORCH_CHECK(!self.is_complex() && !other.is_complex() &&
(maybe_get_output().defined() ? !maybe_get_output().is_complex() : true),
"heaviside is not yet implemented for complex tensors.");
TORCH_CHECK(self.dtype() == other.dtype() &&
(maybe_get_output().defined() ? maybe_get_output().dtype() == self.dtype() : true),
"heaviside is not yet implemented for tensors with different dtypes.");
build_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC(atan2) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_float_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(remainder, Tensor)(const Tensor& self, const Tensor& other) {
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(bitwise_left_shift, Tensor) (
const Tensor& self, const Tensor& other
) {
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(bitwise_right_shift, Tensor) (
const Tensor& self, const Tensor& other
) {
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(bitwise_and, Tensor) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(bitwise_or, Tensor) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(bitwise_xor, Tensor) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(fmod, Tensor) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC2(xlogy, Tensor) (const Tensor& self, const Tensor& other) {
build_borrowing_binary_float_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC(logit_backward) (const Tensor& grad_output, const Tensor& input, c10::optional<double> eps) {
build_borrowing_binary_op(maybe_get_output(), grad_output, input);
}
TORCH_META_FUNC(sigmoid_backward) (const Tensor& grad_output, const Tensor& output) {
build_borrowing_binary_op(maybe_get_output(), grad_output, output);
}
TORCH_META_FUNC(tanh_backward) (const Tensor& grad_output, const Tensor& output) {
build_borrowing_binary_op(maybe_get_output(), grad_output, output);
}
// These are normal binary ops that preserve dtype
#define CREATE_BINARY_META_FUNC(func) \
TORCH_META_FUNC(func) (const Tensor& self, const Tensor& other) { \
build_borrowing_binary_op(maybe_get_output(), self, other); \
}
CREATE_BINARY_META_FUNC(logaddexp);
CREATE_BINARY_META_FUNC(logaddexp2);
CREATE_BINARY_META_FUNC(gcd);
CREATE_BINARY_META_FUNC(lcm);
CREATE_BINARY_META_FUNC(hypot);
CREATE_BINARY_META_FUNC(igamma);
CREATE_BINARY_META_FUNC(igammac);
CREATE_BINARY_META_FUNC(nextafter);
TORCH_META_FUNC(maximum) (const Tensor& self, const Tensor& other) {
TORCH_CHECK(!self.is_complex() && !other.is_complex(), "maximum not implemented for complex tensors.");
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC(minimum) (const Tensor& self, const Tensor& other) {
TORCH_CHECK(!self.is_complex() && !other.is_complex(), "minimum not implemented for complex tensors.");
build_borrowing_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC(fmax) (const Tensor& self, const Tensor& other) {
TORCH_CHECK(!self.is_complex() && !other.is_complex(), "fmax not implemented for complex tensors.");
build_binary_op(maybe_get_output(), self, other);
}
TORCH_META_FUNC(fmin) (const Tensor& self, const Tensor& other) {
TORCH_CHECK(!self.is_complex() && !other.is_complex(), "fmin not implemented for complex tensors.");
build_binary_op(maybe_get_output(), self, other);
}
void comparison_op_check(const Tensor& self, const Tensor& other, const Tensor& result) {
// Validate that is possible to convert zero-dim tensor's dtype to other dtype
// without overflow
if (self.scalar_type() != other.scalar_type()) {
if (self.dim() != 0 && other.dim() == 0) {
native::check_convert(other.item(), self.scalar_type());
} else if (self.dim() == 0 && other.dim() != 0) {
native::check_convert(self.item(), other.scalar_type());
}
}
}
#define CREATE_COMPARISON_SCALAR_TENSOR_META_FUNC(func) \
TORCH_META_FUNC2(func, Tensor)(const Tensor& self, const Tensor& other) { \
const Tensor& result = maybe_get_output(); \
comparison_op_check(self, other, result); \
build_borrowing_comparison_op(result, self, other); \
} \
\
TORCH_META_FUNC2(func, Scalar)(const Tensor& self, const Scalar& other) { \
auto other_tensor = \
native::wrapped_scalar_tensor_and_check_convert(other, self); \
build_borrowing_except_last_argument_comparison_op(maybe_get_output(), self, other_tensor); \
}
CREATE_COMPARISON_SCALAR_TENSOR_META_FUNC(eq);
CREATE_COMPARISON_SCALAR_TENSOR_META_FUNC(ne);
CREATE_COMPARISON_SCALAR_TENSOR_META_FUNC(lt);
CREATE_COMPARISON_SCALAR_TENSOR_META_FUNC(le);
CREATE_COMPARISON_SCALAR_TENSOR_META_FUNC(gt);
CREATE_COMPARISON_SCALAR_TENSOR_META_FUNC(ge);
} // namespace meta
namespace native {
DEFINE_DISPATCH(add_clamp_stub);
DEFINE_DISPATCH(mul_stub);
DEFINE_DISPATCH(sub_stub);
DEFINE_DISPATCH(div_true_stub);
DEFINE_DISPATCH(div_floor_stub);
DEFINE_DISPATCH(div_trunc_stub);
DEFINE_DISPATCH(remainder_stub);
DEFINE_DISPATCH(atan2_stub);
DEFINE_DISPATCH(bitwise_and_stub);
DEFINE_DISPATCH(bitwise_or_stub);
DEFINE_DISPATCH(bitwise_xor_stub);
DEFINE_DISPATCH(lshift_stub);
DEFINE_DISPATCH(rshift_stub);
DEFINE_DISPATCH(logical_and_stub);
DEFINE_DISPATCH(logical_or_stub);
DEFINE_DISPATCH(logical_xor_stub);
DEFINE_DISPATCH(lt_stub);
DEFINE_DISPATCH(le_stub);
DEFINE_DISPATCH(gt_stub);
DEFINE_DISPATCH(ge_stub);
DEFINE_DISPATCH(eq_stub);
DEFINE_DISPATCH(ne_stub);
DEFINE_DISPATCH(sigmoid_backward_stub);
DEFINE_DISPATCH(logit_backward_stub);
DEFINE_DISPATCH(tanh_backward_stub);
DEFINE_DISPATCH(maximum_stub);
DEFINE_DISPATCH(minimum_stub);
DEFINE_DISPATCH(fmax_stub);
DEFINE_DISPATCH(fmin_stub);
DEFINE_DISPATCH(fmod_stub);
DEFINE_DISPATCH(logaddexp_stub);
DEFINE_DISPATCH(logaddexp2_stub);
DEFINE_DISPATCH(gcd_stub);
DEFINE_DISPATCH(lcm_stub);
DEFINE_DISPATCH(hypot_stub);
DEFINE_DISPATCH(igamma_stub);
DEFINE_DISPATCH(igammac_stub);
DEFINE_DISPATCH(nextafter_stub);
DEFINE_DISPATCH(heaviside_stub);
DEFINE_DISPATCH(copysign_stub);
DEFINE_DISPATCH(xlogy_stub);
DEFINE_DISPATCH(xlog1py_stub);
DEFINE_DISPATCH(zeta_stub);
TORCH_IMPL_FUNC(sub_out) (
const Tensor& self, const Tensor& other, const Scalar& alpha, const Tensor& result
) {
add_stub(device_type(), *this, -alpha);
TORCH_INTERNAL_ASSERT(result.scalar_type() == output().dtype());
}
TORCH_IMPL_FUNC(mul_out) (
const Tensor& self, const Tensor& other, const Tensor& result
) {
mul_stub(device_type(), *this);
}
TORCH_IMPL_FUNC(div_out) (const Tensor& self, const Tensor& other, const Tensor& result) {
div_true_stub(device_type(), *this);
}
TORCH_IMPL_FUNC(div_out_mode) (
const Tensor& self, const Tensor& other, c10::optional<c10::string_view> rounding_mode, const Tensor& result
) {
if (!rounding_mode.has_value()) {
div_true_stub(device_type(), *this);
} else if (*rounding_mode == "trunc") {
div_trunc_stub(device_type(), *this);
} else if (*rounding_mode == "floor") {
div_floor_stub(device_type(), *this);
}
}
TORCH_IMPL_FUNC(logit_backward_out) (const Tensor& grad_output, const Tensor& input, c10::optional<double> eps, const Tensor& result) {
logit_backward_stub(device_type(), *this, Scalar(eps ? eps.value() : -1.0));
}
TORCH_IMPL_FUNC(sigmoid_backward_out) (const Tensor& grad_output, const Tensor& output, const Tensor& result) {
sigmoid_backward_stub(device_type(), *this);
}
TORCH_IMPL_FUNC(special_xlog1py_out) (const Tensor& self, const Tensor& other, const Tensor& result) {
xlog1py_stub(device_type(), *this);
}
TORCH_IMPL_FUNC(special_zeta_out) (const Tensor& self, const Tensor& other, const Tensor& result) {
zeta_stub(device_type(), *this);
}
TORCH_IMPL_FUNC(tanh_backward_out) (const Tensor& grad_output, const Tensor& output, const Tensor& result) {
tanh_backward_stub(device_type(), *this);
}
#define CREATE_BINARY_TORCH_IMPL_FUNC(func_out, func_stub) \
TORCH_IMPL_FUNC(func_out) (const Tensor& self, const Tensor& other, const Tensor& result) { \
func_stub(device_type(), *this); \
}
CREATE_BINARY_TORCH_IMPL_FUNC(bitwise_and_out, bitwise_and_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(bitwise_or_out, bitwise_or_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(bitwise_xor_out, bitwise_xor_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(maximum_out, maximum_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(minimum_out, minimum_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(fmax_out, fmax_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(fmin_out, fmin_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(fmod_out, fmod_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(logaddexp_out, logaddexp_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(logaddexp2_out, logaddexp2_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(gcd_out, gcd_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(lcm_out, lcm_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(hypot_out, hypot_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(igamma_out, igamma_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(igammac_out, igammac_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(nextafter_out, nextafter_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(remainder_out, remainder_stub);
CREATE_BINARY_TORCH_IMPL_FUNC(xlogy_out, xlogy_stub);
Tensor special_xlog1py(const Scalar& x, const Tensor& y) {
return at::special_xlog1py(wrapped_scalar_tensor(x), y);
}
Tensor special_xlog1py(const Tensor& x, const Scalar& y) {
return at::special_xlog1py(x, wrapped_scalar_tensor(y));
}
Tensor& special_xlog1py_out(const Scalar& self, const Tensor& other, Tensor& result) {
return at::special_xlog1py_out(result, wrapped_scalar_tensor(self), other);
}
Tensor& special_xlog1py_out(const Tensor& self, const Scalar& other, Tensor& result) {
return at::special_xlog1py_out(result, self, wrapped_scalar_tensor(other));
}
Tensor special_zeta(const Scalar& x, const Tensor& y) {
return at::special_zeta(wrapped_scalar_tensor(x), y);
}
Tensor special_zeta(const Tensor& x, const Scalar& y) {
return at::special_zeta(x, wrapped_scalar_tensor(y));
}
Tensor& special_zeta_out(const Scalar& self, const Tensor& other, Tensor& result) {
return at::special_zeta_out(result, wrapped_scalar_tensor(self), other);
}
Tensor& special_zeta_out(const Tensor& self, const Scalar& other, Tensor& result) {
return at::special_zeta_out(result, self, wrapped_scalar_tensor(other));
}
Tensor& special_gammainc_out(const Tensor& self, const Tensor& other, Tensor& result) {
return at::igamma_out(result, self, other);
}
Tensor special_gammainc(const Tensor& self, const Tensor& other) {
return at::igamma(self, other);
}
Tensor& special_gammaincc_out(const Tensor& self, const Tensor& other, Tensor& result) {
return at::igammac_out(result, self, other);
}
Tensor special_gammaincc(const Tensor& self, const Tensor& other) {
return at::igammac(self, other);
}
TORCH_IMPL_FUNC(atan2_out) (const Tensor& self, const Tensor& other, const Tensor& result) {
atan2_stub(device_type(), *this);
}
Tensor arctan2(const Tensor& self, const Tensor& other) {
return at::atan2(self, other);
}
Tensor& arctan2_(Tensor& self, const Tensor& other) {
return self.atan2_(other);
}
Tensor& arctan2_out(const Tensor& self, const Tensor& other, Tensor& result) {
return at::atan2_out(result, self, other);
}
Tensor& add_relu_impl(
Tensor& result, const Tensor& self, const Tensor& other, const Scalar& alpha) {
auto iter = TensorIterator::binary_op(result, self, other);
Scalar min_val;
Scalar max_val;
if (self.dtype() == at::kInt) {
min_val = 0;
max_val = std::numeric_limits<int32_t>::max();
} else if (self.dtype() == at::kLong) {
min_val = 0;
max_val = std::numeric_limits<int64_t>::max();
} else if (self.dtype() == at::kShort) {
min_val = 0;
max_val = std::numeric_limits<int16_t>::max();
} else if (self.dtype() == at::kChar) {
min_val = 0;
max_val = std::numeric_limits<int8_t>::max();
} else if (self.dtype() == at::kFloat) {
min_val = 0.0;
max_val = std::numeric_limits<float>::max();
} else if (self.dtype() == at::kDouble) {
min_val = 0.0;
max_val = std::numeric_limits<double>::max();
} else {
TORCH_INTERNAL_ASSERT(
false, "Unsupported datatype for add_relu:", self.dtype().name());
}
result = iter.output();
add_clamp_stub(iter.device_type(), iter, alpha, min_val, max_val);
return result;
}
Tensor& add_relu_out(const Tensor& self, const Tensor& other, const Scalar& alpha, Tensor& result) {
return add_relu_impl(result, self, other, alpha);
}
Tensor add_relu(const Tensor& self, const Tensor& other, const Scalar& alpha) {
Tensor result;
return add_relu_impl(result, self, other, alpha);
}
Tensor add_relu(const Tensor& self, const Scalar& other, const Scalar& alpha) {
return add_relu(self, wrapped_scalar_tensor(other), alpha);
}
Tensor& add_relu_(Tensor& self, const Tensor& other, const Scalar& alpha) {
return add_relu_impl(self, self, other, alpha);
}
Tensor& add_relu_(Tensor& self, const Scalar& other, const Scalar& alpha) {
return add_relu_(self, wrapped_scalar_tensor(other), alpha);
}
TORCH_IMPL_FUNC(copysign_out) (
const Tensor& self, const Tensor& other, const Tensor& result
) {
copysign_stub(device_type(), *this);
}
Tensor copysign(const Tensor& self, const Scalar& other) {
// redispatch!
return at::copysign(self, wrapped_scalar_tensor(other));
}
Tensor& copysign_(Tensor& self, const Scalar& other) {
// redispatch!
return self.copysign_(wrapped_scalar_tensor(other));
}
Tensor& copysign_out(const Tensor& self, const Scalar& other, Tensor& result) {
// redispatch!
return at::copysign_out(result, self, wrapped_scalar_tensor(other));
}
// WARNING: There doesn't appear to be any testing for this function
// with sparse self input.
Tensor div(const Tensor& self, const Scalar& other) {
return self.div(wrapped_scalar_tensor(other)); // redispatch!
}
// WARNING: This function, with a sparse self, is currently only
// exercised by DistributedDataParallelTest.test_sparse_gradients
// (you need to exercise it from C++, because this overload is never
// used for Python)
Tensor& div_(Tensor& self, const Scalar& other) {
return self.div_(wrapped_scalar_tensor(other)); // redispatch!
}
Tensor div(const Tensor& self, const Scalar& other, c10::optional<c10::string_view> rounding_mode) {
return self.div(wrapped_scalar_tensor(other), std::move(rounding_mode)); // redispatch!
}
Tensor& div_(Tensor& self, const Scalar& other, c10::optional<c10::string_view> rounding_mode) {
return self.div_(wrapped_scalar_tensor(other), std::move(rounding_mode)); // redispatch!
}
// divide, alias for div
Tensor& divide_out(const Tensor& self, const Tensor& other, Tensor& result) {
return at::div_out(result, self, other);
}
Tensor divide(const Tensor& self, const Tensor& other) {
return self.div(other);
}
Tensor& divide_(Tensor& self, const Tensor& other) {
return self.div_(other);
}
Tensor divide(const Tensor& self, const Scalar& other) {
return self.div(other);
}
Tensor& divide_(Tensor& self, const Scalar& other) {
return self.div_(other);
}
Tensor& divide_out(const Tensor& self, const Tensor& other, c10::optional<c10::string_view> rounding_mode, Tensor& result) {
return at::div_out(result, self, other, std::move(rounding_mode));
}
Tensor divide(const Tensor& self, const Tensor& other, c10::optional<c10::string_view> rounding_mode) {
return self.div(other, std::move(rounding_mode));
}
Tensor& divide_(Tensor& self, const Tensor& other, c10::optional<c10::string_view> rounding_mode) {
return self.div_(other, std::move(rounding_mode));
}
Tensor divide(const Tensor& self, const Scalar& other, c10::optional<c10::string_view> rounding_mode) {
return self.div(other, std::move(rounding_mode));
}
Tensor& divide_(Tensor& self, const Scalar& other, c10::optional<c10::string_view> rounding_mode) {
return self.div_(other, std::move(rounding_mode));
}
// true_divide, an alias for div
Tensor& true_divide_out(const Tensor& self, const Tensor& divisor, Tensor& result) {
return at::div_out(result, self, divisor);
}
Tensor true_divide(const Tensor& self, const Tensor& divisor) {
return self.div(divisor);
}
Tensor& true_divide_(Tensor& self, const Tensor& divisor) {
return self.div_(divisor);
}
Tensor true_divide(const Tensor& self, const Scalar& divisor) {
return self.div(divisor);
}
Tensor& true_divide_(Tensor& self, const Scalar& divisor) {
return self.div_(divisor);
}
Tensor& floor_divide_out(const Tensor& self, const Tensor& other, Tensor& result) {
TORCH_WARN_ONCE(
"floor_divide is deprecated, and will be removed in a future version of pytorch. "
"It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). "
"This results in incorrect rounding for negative values.\n"
"To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), "
"or for actual floor division, use torch.div(a, b, rounding_mode='floor')."
);
// FIXME: Not actually doing floor division (#43874)
auto iter = TensorIterator::binary_op(result, self, other);
div_trunc_stub(iter.device_type(), iter);
if (!result.defined()) {
result = iter.output();
}
return result;
}
Tensor floor_divide(const Tensor& self, const Tensor& other) {
TORCH_WARN_ONCE(
"floor_divide is deprecated, and will be removed in a future version of pytorch. "
"It currently rounds toward 0 (like the 'trunc' function NOT 'floor'). "
"This results in incorrect rounding for negative values.\n"
"To keep the current behavior, use torch.div(a, b, rounding_mode='trunc'), "
"or for actual floor division, use torch.div(a, b, rounding_mode='floor')."
);
// FIXME: Not actually doing floor division (#43874)
Tensor result;
auto iter = TensorIterator::binary_op(result, self, other);
div_trunc_stub(iter.device_type(), iter);
return iter.output();
}
Tensor& floor_divide_(Tensor& self, const Tensor& other) {
return native::floor_divide_out(self, other, self);
}
// TODO: Make this structured to undo the perf regression from native:: removal
// in call here
Tensor mul(const Tensor& self, const Scalar& other) {
return at::mul(self, wrapped_scalar_tensor(other)); // redispatch!
}
Tensor& mul_(Tensor& self, const Scalar& other) {
return at::mul_out(self, wrapped_scalar_tensor(other), self); // redispatch!
}
Device correct_out_device(const Tensor& self, const Tensor& other) {
if (self.device() == at::kCPU){
return other.device();
} else {
return self.device();
}
}
Tensor mul_zerotensor(const Tensor& self, const Tensor& other) {
auto out_device = correct_out_device(self, other);
// hack to use the TensorIterator to get the correct broadcasting and type promotion logic
auto device_ = Device(DeviceType::Meta);
auto meta_out = at::redispatch::mul(c10::DispatchKeySet(at::DispatchKey::Meta), self.to(device_), other.to(device_));
return at::_efficientzerotensor(meta_out.sizes(), meta_out.options().device(out_device));
}
Tensor div_zerotensor(const Tensor& self, const Tensor& other) {
TORCH_INTERNAL_ASSERT(self._is_zerotensor() || other._is_zerotensor());
auto out_device = correct_out_device(self, other);
// hack to use the TensorIterator to get the correct broadcasting and type promotion logic
auto device_ = Device(DeviceType::Meta);
auto meta_out = at::redispatch::div(c10::DispatchKeySet(at::DispatchKey::Meta), self.to(device_), other.to(device_));
if (self._is_zerotensor()) {
if (other._is_zerotensor()) {
// 0/0, return full NAN
return at::full(meta_out.sizes(), std::numeric_limits<float>::quiet_NaN(), meta_out.options().device(out_device));
}
else {
// 0/x, return zero tensor
return at::_efficientzerotensor(meta_out.sizes(), meta_out.options().device(out_device));
}
}
else {
if (other._is_zerotensor()) {
// x/0, return full INF
return at::full(meta_out.sizes(), std::numeric_limits<float>::infinity(), meta_out.options().device(out_device));
}
else {
// x/y -- unreachable, see TORCH_INTERNAL_ASSERT above
return at::_efficientzerotensor(meta_out.sizes(), meta_out.options().device(out_device));
}
}
}
Tensor add_zerotensor(const Tensor& self, const Tensor& other, const Scalar& alpha) {
auto out_device = correct_out_device(self, other);
// hack to use the TensorIterator to get the correct broadcasting and type promotion logic
auto device_ = Device(DeviceType::Meta);
auto meta_out = at::redispatch::add(c10::DispatchKeySet(at::DispatchKey::Meta), self.to(device_), other.to(device_));
auto get_out_like = [&] (const Tensor& tensor)
{
auto sizes = meta_out.sizes();
return at::_to_copy(tensor.expand(sizes), meta_out.options().device(out_device));
};
if (self._is_zerotensor()) {
if (other._is_zerotensor()) {
return at::_efficientzerotensor(meta_out.sizes(), meta_out.options().device(out_device));
}
auto res = get_out_like(other);
return alpha.equal(1) ? res : res.mul(alpha);
} else {
return get_out_like(self);
}
}
// multiply, alias for mul
Tensor& multiply_out(const Tensor& self, const Tensor& other, Tensor& result) {
return at::mul_out(result, self, other);
}
Tensor multiply(const Tensor& self, const Tensor& other) {
return self.mul(other);
}
Tensor& multiply_(Tensor& self, const Tensor& other) {
return self.mul_(other);
}
Tensor multiply(const Tensor& self, const Scalar& other) {
return self.mul(other);
}
Tensor& multiply_(Tensor& self, const Scalar& other) {
return self.mul_(other);
}
Tensor sub(const Tensor& self, const Scalar& other, const Scalar& alpha) {
return at::sub(self, wrapped_scalar_tensor(other), alpha); // redispatch!
}
Tensor& sub_(Tensor& self, const Scalar& other, const Scalar& alpha) {
return self.sub_(wrapped_scalar_tensor(other), alpha); // redispatch!
}
// subtract, alias for sub
Tensor& subtract_out(const Tensor& self, const Tensor& other, const Scalar& alpha, Tensor& result) {
return at::sub_out(result, self, other, alpha);
}
Tensor subtract(const Tensor& self, const Tensor& other, const Scalar& alpha) {
return self.sub(other, alpha);
}
Tensor& subtract_(Tensor& self, const Tensor& other, const Scalar& alpha) {
return self.sub_(other, alpha);
}
Tensor subtract(const Tensor& self, const Scalar& other, const Scalar& alpha) {
return self.sub(other, alpha);
}
Tensor& subtract_(Tensor& self, const Scalar& other, const Scalar& alpha) {
return self.sub_(other, alpha);
}
Tensor rsub(const Tensor& self, const Tensor& other, const Scalar& alpha) {
return at::sub(other, self, alpha); // redispatch!
}
// TODO: Make this structured to undo the perf regression from native:: removal
// in call here
Tensor add(const Tensor& self, const Scalar& other, const Scalar& alpha) {
return at::add(self, wrapped_scalar_tensor(other), alpha);
}
Tensor& add_(Tensor& self, const Scalar& other, const Scalar& alpha) {
return self.add_(wrapped_scalar_tensor(other), alpha);
}
Tensor remainder(const Tensor& self, const Scalar& other) {
// redispatch
return at::remainder(self, wrapped_scalar_tensor(other));
}
Tensor& remainder_(Tensor& self, const Scalar& other) {
// redispatch
return self.remainder_(wrapped_scalar_tensor(other));
}
Tensor& remainder_out(const Tensor& self, const Scalar& other, Tensor& result) {
// redispatch
return at::remainder_out(result, self, wrapped_scalar_tensor(other));
}
Tensor remainder(const Scalar& self, const Tensor& other) {
return at::remainder(wrapped_scalar_tensor(self), other);
}
Tensor rsub(const Tensor& self, const Scalar& other, const Scalar& alpha) {
return native::rsub(self, wrapped_scalar_tensor(other), alpha);
}
Tensor& bitwise_and_out(const Tensor& self, const Scalar& other, Tensor& result) {
return at::bitwise_and_out(result, self, wrapped_scalar_tensor(other));
}
Tensor bitwise_and(const Tensor& self, const Scalar& other) {
return at::bitwise_and(self, wrapped_scalar_tensor(other));
}
Tensor& bitwise_and_(Tensor& self, const Scalar& other) {
return self.bitwise_and_(wrapped_scalar_tensor(other));
}
// Legacy and interfaces. They are aliased to bitwise_and* functions
Tensor __and__(const Tensor& self, const Tensor& other) {
return at::bitwise_and(self, other);
}
Tensor __and__(const Tensor& self, const Scalar& other) {
return at::bitwise_and(self, other);
}
Tensor& __iand__(Tensor& self, const Tensor& other) {
return self.bitwise_and_(other);
}
Tensor& __iand__(Tensor& self, const Scalar& other) {
return self.bitwise_and_(other);
}
Tensor& bitwise_or_out(const Tensor& self, const Scalar& other, Tensor& result) {
return at::bitwise_or_out(result, self, wrapped_scalar_tensor(other));
}
Tensor bitwise_or(const Tensor& self, const Scalar& other) {
return at::bitwise_or(self, wrapped_scalar_tensor(other));
}
Tensor& bitwise_or_(Tensor& self, const Scalar& other) {
return self.bitwise_or_(wrapped_scalar_tensor(other));
}
// Legacy or interfaces. They are aliased to bitwise_or* functions
Tensor __or__(const Tensor& self, const Tensor& other) {
return at::bitwise_or(self, other);
}
Tensor __or__(const Tensor& self, const Scalar& other) {
return at::bitwise_or(self, other);
}
Tensor& __ior__(Tensor& self, const Tensor& other) {
return self.bitwise_or_(other);
}
Tensor& __ior__(Tensor& self, const Scalar& other) {
return self.bitwise_or_(other);
}
Tensor& bitwise_xor_out(const Tensor& self, const Scalar& other, Tensor& result) {
return at::bitwise_xor_out(result, self, wrapped_scalar_tensor(other));
}
Tensor bitwise_xor(const Tensor& self, const Scalar& other) {
return at::bitwise_xor(self, wrapped_scalar_tensor(other));
}
Tensor& bitwise_xor_(Tensor& self, const Scalar& other) {
return self.bitwise_xor_(wrapped_scalar_tensor(other));
}
// Legacy xor interfaces. They are aliased to bitwise_xor* functions
Tensor __xor__(const Tensor& self, const Tensor& other) {
return at::bitwise_xor(self, other);
}
Tensor __xor__(const Tensor& self, const Scalar& other) {
return at::bitwise_xor(self, other);
}
Tensor& __ixor__(Tensor& self, const Tensor& other) {
return self.bitwise_xor_(other);
}
Tensor& __ixor__(Tensor& self, const Scalar& other) {
return self.bitwise_xor_(other);
}
Tensor __lshift__(const Tensor& self, const Tensor& other) {
Tensor result;
auto iter = TensorIterator::binary_op(result, self, other);
lshift_stub(iter.device_type(), iter);
return iter.output();
}
Tensor __lshift__(const Tensor& self, const Scalar& other) {
Tensor result;
auto wrapper = wrapped_scalar_tensor(other).toType(self.scalar_type());
auto iter = TensorIterator::binary_op(result, self, wrapper);
lshift_stub(iter.device_type(), iter);
return iter.output();
}
Tensor& __ilshift__(Tensor& self, const Tensor& other) {
auto iter = TensorIterator::binary_op(self, self, other);
lshift_stub(iter.device_type(), iter);
return self;
}
Tensor& __ilshift__(Tensor& self, const Scalar& other) {
auto wrapper = wrapped_scalar_tensor(other).toType(self.scalar_type());
auto iter = TensorIterator::binary_op(self, self, wrapper);
lshift_stub(iter.device_type(), iter);
return self;
}
TORCH_IMPL_FUNC(bitwise_left_shift_out) (const Tensor& self, const Tensor& other, const Tensor& result) {
lshift_stub(device_type(), *this);
}
Tensor& bitwise_left_shift_out(const Tensor& self, const Scalar& other, Tensor& result) {
return at::bitwise_left_shift_out(result, self, wrapped_scalar_tensor(other).toType(self.scalar_type()));
}
Tensor bitwise_left_shift(const Tensor& self, const Scalar& other) {
return at::bitwise_left_shift(self, wrapped_scalar_tensor(other).toType(self.scalar_type()));
}
Tensor& bitwise_left_shift_(Tensor& self, const Scalar& other) {
return at::bitwise_left_shift_out(self, self, wrapped_scalar_tensor(other).toType(self.scalar_type()));
}
Tensor bitwise_left_shift(const Scalar& self, const Tensor& other) {
return at::bitwise_left_shift(wrapped_scalar_tensor(self).toType(other.scalar_type()), other);
}
Tensor __rshift__(const Tensor& self, const Tensor& other) {
Tensor result;
auto iter = TensorIterator::binary_op(result, self, other);
rshift_stub(iter.device_type(), iter);
return iter.output();
}
Tensor __rshift__(const Tensor& self, const Scalar& other) {
Tensor result;
auto wrapper = wrapped_scalar_tensor(other).toType(self.scalar_type());
auto iter = TensorIterator::binary_op(result, self, wrapper);
rshift_stub(iter.device_type(), iter);
return iter.output();
}
Tensor& __irshift__(Tensor& self, const Tensor& other) {
auto iter = TensorIterator::binary_op(self, self, other);
rshift_stub(iter.device_type(), iter);
return self;
}
Tensor& __irshift__(Tensor& self, const Scalar& other) {
auto wrapper = wrapped_scalar_tensor(other).toType(self.scalar_type());
auto iter = TensorIterator::binary_op(self, self, wrapper);
rshift_stub(iter.device_type(), iter);
return self;
}
TORCH_IMPL_FUNC(bitwise_right_shift_out) (const Tensor& self, const Tensor& other, const Tensor& result) {
rshift_stub(device_type(), *this);
}
Tensor& bitwise_right_shift_out(const Tensor& self, const Scalar& other, Tensor& result) {
return at::bitwise_right_shift_out(result, self, wrapped_scalar_tensor(other).toType(self.scalar_type()));
}
Tensor bitwise_right_shift(const Tensor& self, const Scalar& other) {
return at::bitwise_right_shift(self, wrapped_scalar_tensor(other).toType(self.scalar_type()));
}
Tensor& bitwise_right_shift_(Tensor& self, const Scalar& other) {
return at::bitwise_right_shift_out(self, self, wrapped_scalar_tensor(other).toType(self.scalar_type()));
}
Tensor bitwise_right_shift(const Scalar& self, const Tensor& other) {
return at::bitwise_right_shift(wrapped_scalar_tensor(self).toType(other.scalar_type()), other);
}
template <typename Stub>
Tensor& comparison_op_out(Tensor& result, const Tensor& self, const Tensor& other, Stub& stub) {
// Validate that is possible to convert zero-dim tensor's dtype to other dtype without overflow
if (self.scalar_type() != other.scalar_type()) {
if (self.dim() != 0 && other.dim() == 0) {
check_convert(other.item(), self.scalar_type());
} else if (self.dim() == 0 && other.dim() != 0) {
check_convert(self.item(), other.scalar_type());
}
}
auto iter = TensorIterator::comparison_op(result, self, other);
stub(iter.device_type(), iter);
return result;
}
template <typename OutImpl>
Tensor comparison_op(const Tensor& self, const Tensor& other, OutImpl& out_impl) {
Tensor result = at::empty({0}, self.options().dtype(kBool));
return out_impl(result, self, other);
}
// To avoid overflow during type promotion we will check that both dtypes of self and other are same
template <typename OutImpl>
Tensor& comparison_op_(Tensor& self, const Tensor& other, OutImpl& out_impl) {
return out_impl(self, self, other);
}
// validates that is possible to convert Scalar other to self's dtype without overflow.
// This behavior is unique to comparison ops; arithmetic operations don't do this.
// In the future, we should reconsider this inconsistency and decide if we want to add the same check to arithmetic ops.
template <typename OutImpl>
Tensor& comparison_op_out(Tensor& result, const Tensor& self, const Scalar& other, OutImpl& out_impl) {
return out_impl(result, self, wrapped_scalar_tensor_and_check_convert(other, self));
}
template <typename OutImpl>
Tensor comparison_op(const Tensor& self, const Scalar& other, OutImpl& out_impl) {
return comparison_op(self, wrapped_scalar_tensor_and_check_convert(other, self), out_impl);
}