-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathcuda.html
1835 lines (1459 loc) · 174 KB
/
cuda.html
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
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
<head>
<meta name="robots" content="noindex">
<meta name="robots" content="noindex">
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>torch.cuda — PyTorch 2.0 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/_modules/torch/cuda.html"/>
<link rel="stylesheet" href="../../_static/css/theme.css" type="text/css" />
<!-- <link rel="stylesheet" href="../../_static/pygments.css" type="text/css" /> -->
<link rel="stylesheet" href="../../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="../../_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="../../_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="../../_static/katex-math.css" type="text/css" />
<link rel="stylesheet" href="../../_static/sphinx-dropdown.css" type="text/css" />
<link rel="stylesheet" href="../../_static/panels-bootstrap.min.css" type="text/css" />
<link rel="stylesheet" href="../../_static/css/jit.css" type="text/css" />
<link rel="index" title="Index" href="../../genindex.html" />
<link rel="search" title="Search" href="../../search.html" />
<!-- Google Analytics -->
<script async src="https://www.googletagmanager.com/gtag/js?id=UA-117752657-2"></script>
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){dataLayer.push(arguments);}
gtag('js', new Date());
gtag('config', 'UA-117752657-2');
</script>
<!-- End Google Analytics -->
<script src="../../_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="../../_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../../_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../../_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../../_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../../_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../../_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/[email protected]/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.15.2/css/all.css" integrity="sha384-vSIIfh2YWi9wW0r9iZe7RJPrKwp6bG+s9QZMoITbCckVJqGCCRhc+ccxNcdpHuYu" crossorigin="anonymous">
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://pytorch.org/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.org/get-started">Get Started</a>
</li>
<li>
<a href="https://pytorch.org/ecosystem">Ecosystem</a>
</li>
<li>
<a href="https://pytorch.org/mobile">Mobile</a>
</li>
<li>
<a href="https://pytorch.org/blog/">Blog</a>
</li>
<li>
<a href="https://pytorch.org/tutorials">Tutorials</a>
</li>
<li class="active docs-active">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="resource-option with-down-orange-arrow">
Docs
</a>
<div class="resources-dropdown-menu">
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/docs/stable/index.html">
<span class="dropdown-title">PyTorch</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/audio/stable/index.html">
<span class="dropdown-title">torchaudio</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/text/stable/index.html">
<span class="dropdown-title">torchtext</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/vision/stable/index.html">
<span class="dropdown-title">torchvision</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/torcharrow">
<span class="dropdown-title">torcharrow</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/data">
<span class="dropdown-title">TorchData</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/torchrec">
<span class="dropdown-title">TorchRec</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/serve/">
<span class="dropdown-title">TorchServe</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/torchx/">
<span class="dropdown-title">TorchX</span>
<p></p>
</a>
<a class="doc-dropdown-option nav-dropdown-item" href="https://pytorch.org/xla">
<span class="dropdown-title">PyTorch on XLA Devices</span>
<p></p>
</a>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="resource-option with-down-arrow">
Resources
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/features">
<span class="dropdown-title">About</span>
<p>Learn about PyTorch’s features and capabilities</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/foundation">
<span class="dropdown-title">PyTorch Foundation</span>
<p>Learn about the PyTorch foundation</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/#community-module">
<span class="dropdown-title">Community</span>
<p>Join the PyTorch developer community to contribute, learn, and get your questions answered.</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/community-stories">
<span class="dropdown-title">Community Stories</span>
<p>Learn how our community solves real, everyday machine learning problems with PyTorch.</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/resources">
<span class="dropdown-title">Developer Resources</span>
<p>Find resources and get questions answered</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/events">
<span class="dropdown-title">Events</span>
<p>Find events, webinars, and podcasts</p>
</a>
<a class="nav-dropdown-item" href="https://discuss.pytorch.org/" target="_blank">
<span class="dropdown-title">Forums</span>
<p>A place to discuss PyTorch code, issues, install, research</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/hub">
<span class="dropdown-title">Models (Beta)</span>
<p>Discover, publish, and reuse pre-trained models</p>
</a>
</div>
</div>
</li>
<li>
<a href="https://github.com/pytorch/pytorch">GitHub</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
<a href='https://pytorch.org/docs/versions.html'>2.0 ▼</a>
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../../search.html" method="get">
<input type="text" name="q" placeholder="Search Docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption" role="heading"><span class="caption-text">Community</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../community/build_ci_governance.html">PyTorch Governance | Build + CI</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../community/contribution_guide.html">PyTorch Contribution Guide</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../community/design.html">PyTorch Design Philosophy</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../community/governance.html">PyTorch Governance | Mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../community/persons_of_interest.html">PyTorch Governance | Maintainers</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Developer Notes</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../notes/amp_examples.html">CUDA Automatic Mixed Precision examples</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/autograd.html">Autograd mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/broadcasting.html">Broadcasting semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/cpu_threading_torchscript_inference.html">CPU threading and TorchScript inference</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/cuda.html">CUDA semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/ddp.html">Distributed Data Parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/extending.html">Extending PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/extending.func.html">Extending torch.func with autograd.Function</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/gradcheck.html">Gradcheck mechanics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/hip.html">HIP (ROCm) semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/large_scale_deployments.html">Features for large-scale deployments</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/modules.html">Modules</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/mps.html">MPS backend</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/multiprocessing.html">Multiprocessing best practices</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/numerical_accuracy.html">Numerical accuracy</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/randomness.html">Reproducibility</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/serialization.html">Serialization semantics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../notes/windows.html">Windows FAQ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">torch.compile</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../dynamo/index.html">TorchDynamo Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../dynamo/installation.html">Installing TorchDynamo</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../dynamo/get-started.html">Getting Started</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../dynamo/guards-overview.html">Guards Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../dynamo/custom-backends.html">Custom Backends</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../dynamo/deep-dive.html">TorchDynamo Deeper Dive</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../dynamo/troubleshooting.html">TorchDynamo Troubleshooting</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../dynamo/faq.html">Frequently Asked Questions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../ir.html">IRs</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Language Bindings</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../cpp_index.html">C++</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/javadoc/">Javadoc</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../deploy.html">torch::deploy</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Python API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../../torch.html">torch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../nn.html">torch.nn</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../nn.functional.html">torch.nn.functional</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../tensors.html">torch.Tensor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../tensor_attributes.html">Tensor Attributes</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../tensor_view.html">Tensor Views</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../amp.html">torch.amp</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../autograd.html">torch.autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../library.html">torch.library</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../cuda.html">torch.cuda</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../mps.html">torch.mps</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../backends.html">torch.backends</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.html">torch.distributed</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.algorithms.join.html">torch.distributed.algorithms.join</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.elastic.html">torch.distributed.elastic</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../fsdp.html">torch.distributed.fsdp</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.optim.html">torch.distributed.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.tensor.parallel.html">torch.distributed.tensor.parallel</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributed.checkpoint.html">torch.distributed.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../distributions.html">torch.distributions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../_dynamo.html">torch._dynamo</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../fft.html">torch.fft</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../func.html">torch.func</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../futures.html">torch.futures</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../fx.html">torch.fx</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../hub.html">torch.hub</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../jit.html">torch.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../linalg.html">torch.linalg</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../monitor.html">torch.monitor</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../signal.html">torch.signal</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../special.html">torch.special</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../torch.overrides.html">torch.overrides</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../package.html">torch.package</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../profiler.html">torch.profiler</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../nn.init.html">torch.nn.init</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../onnx.html">torch.onnx</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../onnx_diagnostics.html">torch.onnx diagnostics</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../optim.html">torch.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../complex_numbers.html">Complex Numbers</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../ddp_comm_hooks.html">DDP Communication Hooks</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../pipeline.html">Pipeline Parallelism</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../quantization.html">Quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../rpc.html">Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../random.html">torch.random</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../masked.html">torch.masked</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../nested.html">torch.nested</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../sparse.html">torch.sparse</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../storage.html">torch.Storage</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../testing.html">torch.testing</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../benchmark_utils.html">torch.utils.benchmark</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../bottleneck.html">torch.utils.bottleneck</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../checkpoint.html">torch.utils.checkpoint</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../cpp_extension.html">torch.utils.cpp_extension</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../data.html">torch.utils.data</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../jit_utils.html">torch.utils.jit</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../dlpack.html">torch.utils.dlpack</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../mobile_optimizer.html">torch.utils.mobile_optimizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../model_zoo.html">torch.utils.model_zoo</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../tensorboard.html">torch.utils.tensorboard</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../type_info.html">Type Info</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../named_tensor.html">Named Tensors</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../name_inference.html">Named Tensors operator coverage</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../config_mod.html">torch.__config__</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Libraries</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/audio/stable">torchaudio</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/data">TorchData</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/torchrec">TorchRec</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/serve">TorchServe</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/text/stable">torchtext</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/vision/stable">torchvision</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/xla/">PyTorch on XLA Devices</a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="../../index.html">
Docs
</a> >
</li>
<li><a href="../index.html">Module code</a> ></li>
<li><a href="../torch.html">torch</a> ></li>
<li>torch.cuda</li>
<li class="pytorch-breadcrumbs-aside">
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<div class="rst-content">
<div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
<article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
<h1>Source code for torch.cuda</h1><div class="highlight"><pre>
<span></span><span class="sa">r</span><span class="sd">"""</span>
<span class="sd">This package adds support for CUDA tensor types, that implement the same</span>
<span class="sd">function as CPU tensors, but they utilize GPUs for computation.</span>
<span class="sd">It is lazily initialized, so you can always import it, and use</span>
<span class="sd">:func:`is_available()` to determine if your system supports CUDA.</span>
<span class="sd">:ref:`cuda-semantics` has more details about working with CUDA.</span>
<span class="sd">"""</span>
<span class="kn">import</span> <span class="nn">contextlib</span>
<span class="kn">import</span> <span class="nn">os</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">from</span> <span class="nn">torch.types</span> <span class="kn">import</span> <span class="n">Device</span>
<span class="kn">import</span> <span class="nn">traceback</span>
<span class="kn">import</span> <span class="nn">warnings</span>
<span class="kn">import</span> <span class="nn">threading</span>
<span class="kn">from</span> <span class="nn">functools</span> <span class="kn">import</span> <span class="n">lru_cache</span>
<span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Any</span><span class="p">,</span> <span class="n">List</span><span class="p">,</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Tuple</span><span class="p">,</span> <span class="n">Union</span><span class="p">,</span> <span class="n">cast</span>
<span class="kn">from</span> <span class="nn">._utils</span> <span class="kn">import</span> <span class="n">_get_device_index</span><span class="p">,</span> <span class="n">_dummy_type</span>
<span class="kn">from</span> <span class="nn">.._utils</span> <span class="kn">import</span> <span class="n">classproperty</span>
<span class="kn">from</span> <span class="nn">.graphs</span> <span class="kn">import</span> <span class="n">CUDAGraph</span><span class="p">,</span> <span class="n">graph_pool_handle</span><span class="p">,</span> <span class="n">graph</span><span class="p">,</span> \
<span class="n">make_graphed_callables</span><span class="p">,</span> <span class="n">is_current_stream_capturing</span>
<span class="kn">from</span> <span class="nn">.streams</span> <span class="kn">import</span> <span class="n">ExternalStream</span><span class="p">,</span> <span class="n">Stream</span><span class="p">,</span> <span class="n">Event</span>
<span class="kn">from</span> <span class="nn">..</span> <span class="kn">import</span> <span class="n">device</span> <span class="k">as</span> <span class="n">_device</span>
<span class="kn">import</span> <span class="nn">torch._C</span>
<span class="k">try</span><span class="p">:</span>
<span class="kn">from</span> <span class="nn">torch._C</span> <span class="kn">import</span> <span class="n">_cudart</span> <span class="c1"># type: ignore[attr-defined]</span>
<span class="k">except</span> <span class="ne">ImportError</span><span class="p">:</span>
<span class="n">_cudart</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">_initialized</span> <span class="o">=</span> <span class="kc">False</span>
<span class="n">_tls</span> <span class="o">=</span> <span class="n">threading</span><span class="o">.</span><span class="n">local</span><span class="p">()</span>
<span class="n">_initialization_lock</span> <span class="o">=</span> <span class="n">threading</span><span class="o">.</span><span class="n">Lock</span><span class="p">()</span>
<span class="n">_queued_calls</span> <span class="o">=</span> <span class="p">[]</span> <span class="c1"># don't invoke these until initialization occurs</span>
<span class="n">_is_in_bad_fork</span> <span class="o">=</span> <span class="nb">getattr</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="p">,</span> <span class="s2">"_cuda_isInBadFork"</span><span class="p">,</span> <span class="k">lambda</span><span class="p">:</span> <span class="kc">False</span><span class="p">)</span>
<span class="n">_device_t</span> <span class="o">=</span> <span class="n">Union</span><span class="p">[</span><span class="n">_device</span><span class="p">,</span> <span class="nb">str</span><span class="p">,</span> <span class="nb">int</span><span class="p">,</span> <span class="kc">None</span><span class="p">]</span>
<span class="k">class</span> <span class="nc">_LazySeedTracker</span><span class="p">:</span>
<span class="c1"># Since seeding is memory-less, only track the latest seed.</span>
<span class="c1"># Note: `manual_seed_all` followed by `manual_seed` overwrites</span>
<span class="c1"># the seed on current device. We track the order of **latest**</span>
<span class="c1"># calls between these two API.</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">manual_seed_all_cb</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">manual_seed_cb</span> <span class="o">=</span> <span class="kc">None</span>
<span class="bp">self</span><span class="o">.</span><span class="n">call_order</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">def</span> <span class="nf">queue_seed_all</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">cb</span><span class="p">,</span> <span class="n">traceback</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">manual_seed_all_cb</span> <span class="o">=</span> <span class="p">(</span><span class="n">cb</span><span class="p">,</span> <span class="n">traceback</span><span class="p">)</span>
<span class="c1"># update seed_all to be latest</span>
<span class="bp">self</span><span class="o">.</span><span class="n">call_order</span> <span class="o">=</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">manual_seed_cb</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">manual_seed_all_cb</span><span class="p">]</span>
<span class="k">def</span> <span class="nf">queue_seed</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">cb</span><span class="p">,</span> <span class="n">traceback</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">manual_seed_cb</span> <span class="o">=</span> <span class="p">(</span><span class="n">cb</span><span class="p">,</span> <span class="n">traceback</span><span class="p">)</span>
<span class="c1"># update seed to be latest</span>
<span class="bp">self</span><span class="o">.</span><span class="n">call_order</span> <span class="o">=</span> <span class="p">[</span><span class="bp">self</span><span class="o">.</span><span class="n">manual_seed_all_cb</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">manual_seed_cb</span><span class="p">]</span>
<span class="k">def</span> <span class="nf">get_calls</span><span class="p">(</span><span class="bp">self</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">:</span>
<span class="k">return</span> <span class="bp">self</span><span class="o">.</span><span class="n">call_order</span>
<span class="n">_lazy_seed_tracker</span> <span class="o">=</span> <span class="n">_LazySeedTracker</span><span class="p">()</span>
<span class="c1"># Define dummy _CudaDeviceProperties type if PyTorch was compiled without CUDA</span>
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="p">,</span> <span class="s1">'_CudaDeviceProperties'</span><span class="p">):</span>
<span class="n">_CudaDeviceProperties</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_CudaDeviceProperties</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">_CudaDeviceProperties</span> <span class="o">=</span> <span class="n">_dummy_type</span><span class="p">(</span><span class="s1">'_CudaDeviceProperties'</span><span class="p">)</span> <span class="c1"># type: ignore[assignment, misc]</span>
<span class="k">if</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="p">,</span> <span class="s1">'_cuda_exchangeDevice'</span><span class="p">):</span>
<span class="n">_exchange_device</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_exchangeDevice</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">def</span> <span class="nf">_exchange_device</span><span class="p">(</span><span class="n">device</span><span class="p">:</span> <span class="nb">int</span><span class="p">)</span> <span class="o">-></span> <span class="nb">int</span><span class="p">:</span>
<span class="k">if</span> <span class="n">device</span> <span class="o"><</span> <span class="mi">0</span><span class="p">:</span>
<span class="k">return</span> <span class="o">-</span><span class="mi">1</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s2">"PyTorch was compiled without CUDA support"</span><span class="p">)</span>
<span class="c1"># Global variables dynamically populated by native code</span>
<span class="n">has_magma</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span>
<span class="n">has_half</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span>
<span class="n">default_generators</span><span class="p">:</span> <span class="n">Tuple</span><span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">Generator</span><span class="p">]</span> <span class="o">=</span> <span class="p">()</span> <span class="c1"># type: ignore[assignment]</span>
<span class="k">def</span> <span class="nf">_is_compiled</span><span class="p">()</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Returns true if compile with CUDA support."""</span>
<span class="k">return</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="p">,</span> <span class="s1">'_cuda_getDeviceCount'</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">_nvml_based_avail</span><span class="p">()</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
<span class="k">return</span> <span class="n">os</span><span class="o">.</span><span class="n">getenv</span><span class="p">(</span><span class="s1">'PYTORCH_NVML_BASED_CUDA_CHECK'</span><span class="p">)</span> <span class="o">==</span> <span class="s1">'1'</span>
<div class="viewcode-block" id="is_available"><a class="viewcode-back" href="../../generated/torch.cuda.is_available.html#torch.cuda.is_available">[docs]</a><span class="k">def</span> <span class="nf">is_available</span><span class="p">()</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Returns a bool indicating if CUDA is currently available."""</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">_is_compiled</span><span class="p">():</span>
<span class="k">return</span> <span class="kc">False</span>
<span class="k">if</span> <span class="n">_nvml_based_avail</span><span class="p">():</span>
<span class="c1"># The user has set an env variable to request this availability check that attempts to avoid fork poisoning by</span>
<span class="c1"># using NVML at the cost of a weaker CUDA availability assessment. Note that if NVML discovery/initialization</span>
<span class="c1"># fails, this assessment falls back to the default CUDA Runtime API assessment (`cudaGetDeviceCount`)</span>
<span class="k">return</span> <span class="n">device_count</span><span class="p">()</span> <span class="o">></span> <span class="mi">0</span>
<span class="k">else</span><span class="p">:</span>
<span class="c1"># The default availability inspection never throws and returns 0 if the driver is missing or can't</span>
<span class="c1"># be initialized. This uses the CUDA Runtime API `cudaGetDeviceCount` which in turn initializes the CUDA Driver</span>
<span class="c1"># API via `cuInit`</span>
<span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_getDeviceCount</span><span class="p">()</span> <span class="o">></span> <span class="mi">0</span></div>
<span class="k">def</span> <span class="nf">is_bf16_supported</span><span class="p">():</span>
<span class="sa">r</span><span class="sd">"""Returns a bool indicating if the current CUDA/ROCm device supports dtype bfloat16"""</span>
<span class="c1"># Check for ROCm, if true return true, no ROCM_VERSION check required,</span>
<span class="c1"># since it is supported on AMD GPU archs.</span>
<span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">version</span><span class="o">.</span><span class="n">hip</span><span class="p">:</span>
<span class="k">return</span> <span class="kc">True</span>
<span class="n">cu_vers</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">version</span><span class="o">.</span><span class="n">cuda</span>
<span class="k">if</span> <span class="n">cu_vers</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">cuda_maj_decide</span> <span class="o">=</span> <span class="nb">int</span><span class="p">(</span><span class="n">cu_vers</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">'.'</span><span class="p">)[</span><span class="mi">0</span><span class="p">])</span> <span class="o">>=</span> <span class="mi">11</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">cuda_maj_decide</span> <span class="o">=</span> <span class="kc">False</span>
<span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">get_device_properties</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">current_device</span><span class="p">())</span><span class="o">.</span><span class="n">major</span> <span class="o">>=</span> <span class="mi">8</span> <span class="ow">and</span> <span class="n">cuda_maj_decide</span>
<span class="k">def</span> <span class="nf">_sleep</span><span class="p">(</span><span class="n">cycles</span><span class="p">):</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_sleep</span><span class="p">(</span><span class="n">cycles</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">_check_capability</span><span class="p">():</span>
<span class="n">incorrect_binary_warn</span> <span class="o">=</span> <span class="s2">"""</span>
<span class="s2"> Found GPU</span><span class="si">%d</span><span class="s2"> </span><span class="si">%s</span><span class="s2"> which requires CUDA_VERSION >= </span><span class="si">%d</span><span class="s2"> to</span>
<span class="s2"> work properly, but your PyTorch was compiled</span>
<span class="s2"> with CUDA_VERSION </span><span class="si">%d</span><span class="s2">. Please install the correct PyTorch binary</span>
<span class="s2"> using instructions from https://pytorch.org</span>
<span class="s2"> """</span>
<span class="n">old_gpu_warn</span> <span class="o">=</span> <span class="s2">"""</span>
<span class="s2"> Found GPU</span><span class="si">%d</span><span class="s2"> </span><span class="si">%s</span><span class="s2"> which is of cuda capability </span><span class="si">%d</span><span class="s2">.</span><span class="si">%d</span><span class="s2">.</span>
<span class="s2"> PyTorch no longer supports this GPU because it is too old.</span>
<span class="s2"> The minimum cuda capability supported by this library is </span><span class="si">%d</span><span class="s2">.</span><span class="si">%d</span><span class="s2">.</span>
<span class="s2"> """</span>
<span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">version</span><span class="o">.</span><span class="n">cuda</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span> <span class="c1"># on ROCm we don't want this check</span>
<span class="n">CUDA_VERSION</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_getCompiledVersion</span><span class="p">()</span>
<span class="k">for</span> <span class="n">d</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">device_count</span><span class="p">()):</span>
<span class="n">capability</span> <span class="o">=</span> <span class="n">get_device_capability</span><span class="p">(</span><span class="n">d</span><span class="p">)</span>
<span class="n">major</span> <span class="o">=</span> <span class="n">capability</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="n">minor</span> <span class="o">=</span> <span class="n">capability</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span>
<span class="n">name</span> <span class="o">=</span> <span class="n">get_device_name</span><span class="p">(</span><span class="n">d</span><span class="p">)</span>
<span class="n">current_arch</span> <span class="o">=</span> <span class="n">major</span> <span class="o">*</span> <span class="mi">10</span> <span class="o">+</span> <span class="n">minor</span>
<span class="n">min_arch</span> <span class="o">=</span> <span class="nb">min</span><span class="p">((</span><span class="nb">int</span><span class="p">(</span><span class="n">arch</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s2">"_"</span><span class="p">)[</span><span class="mi">1</span><span class="p">])</span> <span class="k">for</span> <span class="n">arch</span> <span class="ow">in</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">get_arch_list</span><span class="p">()),</span> <span class="n">default</span><span class="o">=</span><span class="mi">35</span><span class="p">)</span>
<span class="k">if</span> <span class="n">current_arch</span> <span class="o"><</span> <span class="n">min_arch</span><span class="p">:</span>
<span class="n">warnings</span><span class="o">.</span><span class="n">warn</span><span class="p">(</span><span class="n">old_gpu_warn</span> <span class="o">%</span> <span class="p">(</span><span class="n">d</span><span class="p">,</span> <span class="n">name</span><span class="p">,</span> <span class="n">major</span><span class="p">,</span> <span class="n">minor</span><span class="p">,</span> <span class="n">min_arch</span> <span class="o">//</span> <span class="mi">10</span><span class="p">,</span> <span class="n">min_arch</span> <span class="o">%</span> <span class="mi">10</span><span class="p">))</span>
<span class="k">def</span> <span class="nf">_check_cubins</span><span class="p">():</span>
<span class="n">incompatible_device_warn</span> <span class="o">=</span> <span class="s2">"""</span>
<span class="si">{}</span><span class="s2"> with CUDA capability sm_</span><span class="si">{}</span><span class="s2"> is not compatible with the current PyTorch installation.</span>
<span class="s2">The current PyTorch install supports CUDA capabilities </span><span class="si">{}</span><span class="s2">.</span>
<span class="s2">If you want to use the </span><span class="si">{}</span><span class="s2"> GPU with PyTorch, please check the instructions at https://pytorch.org/get-started/locally/</span>
<span class="s2">"""</span>
<span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">version</span><span class="o">.</span><span class="n">cuda</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span> <span class="c1"># on ROCm we don't want this check</span>
<span class="k">return</span>
<span class="n">arch_list</span> <span class="o">=</span> <span class="n">get_arch_list</span><span class="p">()</span>
<span class="k">if</span> <span class="nb">len</span><span class="p">(</span><span class="n">arch_list</span><span class="p">)</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="k">return</span>
<span class="n">supported_sm</span> <span class="o">=</span> <span class="p">[</span><span class="nb">int</span><span class="p">(</span><span class="n">arch</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s1">'_'</span><span class="p">)[</span><span class="mi">1</span><span class="p">])</span> <span class="k">for</span> <span class="n">arch</span> <span class="ow">in</span> <span class="n">arch_list</span> <span class="k">if</span> <span class="s1">'sm_'</span> <span class="ow">in</span> <span class="n">arch</span><span class="p">]</span>
<span class="k">for</span> <span class="n">idx</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">device_count</span><span class="p">()):</span>
<span class="n">cap_major</span><span class="p">,</span> <span class="n">cap_minor</span> <span class="o">=</span> <span class="n">get_device_capability</span><span class="p">(</span><span class="n">idx</span><span class="p">)</span>
<span class="c1"># NVIDIA GPU compute architectures are backward compatible within major version</span>
<span class="n">supported</span> <span class="o">=</span> <span class="nb">any</span><span class="p">([</span><span class="n">sm</span> <span class="o">//</span> <span class="mi">10</span> <span class="o">==</span> <span class="n">cap_major</span> <span class="k">for</span> <span class="n">sm</span> <span class="ow">in</span> <span class="n">supported_sm</span><span class="p">])</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">supported</span><span class="p">:</span>
<span class="n">device_name</span> <span class="o">=</span> <span class="n">get_device_name</span><span class="p">(</span><span class="n">idx</span><span class="p">)</span>
<span class="n">capability</span> <span class="o">=</span> <span class="n">cap_major</span> <span class="o">*</span> <span class="mi">10</span> <span class="o">+</span> <span class="n">cap_minor</span>
<span class="n">warnings</span><span class="o">.</span><span class="n">warn</span><span class="p">(</span><span class="n">incompatible_device_warn</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">device_name</span><span class="p">,</span> <span class="n">capability</span><span class="p">,</span> <span class="s2">" "</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">arch_list</span><span class="p">),</span> <span class="n">device_name</span><span class="p">))</span>
<div class="viewcode-block" id="is_initialized"><a class="viewcode-back" href="../../generated/torch.cuda.is_initialized.html#torch.cuda.is_initialized">[docs]</a><span class="k">def</span> <span class="nf">is_initialized</span><span class="p">():</span>
<span class="sa">r</span><span class="sd">"""Returns whether PyTorch's CUDA state has been initialized."""</span>
<span class="k">return</span> <span class="n">_initialized</span> <span class="ow">and</span> <span class="ow">not</span> <span class="n">_is_in_bad_fork</span><span class="p">()</span></div>
<span class="k">def</span> <span class="nf">_lazy_call</span><span class="p">(</span><span class="n">callable</span><span class="p">,</span> <span class="o">**</span><span class="n">kwargs</span><span class="p">):</span>
<span class="k">if</span> <span class="n">is_initialized</span><span class="p">():</span>
<span class="n">callable</span><span class="p">()</span>
<span class="k">else</span><span class="p">:</span>
<span class="c1"># TODO(torch_deploy): this accesses linecache, which attempts to read the</span>
<span class="c1"># file system to get traceback info. Patch linecache or do something</span>
<span class="c1"># else here if this ends up being important.</span>
<span class="k">global</span> <span class="n">_lazy_seed_tracker</span>
<span class="k">if</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"seed_all"</span><span class="p">,</span> <span class="kc">False</span><span class="p">):</span>
<span class="n">_lazy_seed_tracker</span><span class="o">.</span><span class="n">queue_seed_all</span><span class="p">(</span><span class="n">callable</span><span class="p">,</span> <span class="n">traceback</span><span class="o">.</span><span class="n">format_stack</span><span class="p">())</span>
<span class="k">elif</span> <span class="n">kwargs</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="s2">"seed"</span><span class="p">,</span> <span class="kc">False</span><span class="p">):</span>
<span class="n">_lazy_seed_tracker</span><span class="o">.</span><span class="n">queue_seed</span><span class="p">(</span><span class="n">callable</span><span class="p">,</span> <span class="n">traceback</span><span class="o">.</span><span class="n">format_stack</span><span class="p">())</span>
<span class="k">else</span><span class="p">:</span>
<span class="c1"># Don't store the actual traceback to avoid memory cycle</span>
<span class="n">_queued_calls</span><span class="o">.</span><span class="n">append</span><span class="p">((</span><span class="n">callable</span><span class="p">,</span> <span class="n">traceback</span><span class="o">.</span><span class="n">format_stack</span><span class="p">()))</span>
<span class="n">_lazy_call</span><span class="p">(</span><span class="n">_check_capability</span><span class="p">)</span>
<span class="n">_lazy_call</span><span class="p">(</span><span class="n">_check_cubins</span><span class="p">)</span>
<span class="k">class</span> <span class="nc">DeferredCudaCallError</span><span class="p">(</span><span class="ne">Exception</span><span class="p">):</span>
<span class="k">pass</span>
<span class="n">OutOfMemoryError</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_OutOfMemoryError</span>
<div class="viewcode-block" id="init"><a class="viewcode-back" href="../../generated/torch.cuda.init.html#torch.cuda.init">[docs]</a><span class="k">def</span> <span class="nf">init</span><span class="p">():</span>
<span class="sa">r</span><span class="sd">"""Initialize PyTorch's CUDA state. You may need to call</span>
<span class="sd"> this explicitly if you are interacting with PyTorch via</span>
<span class="sd"> its C API, as Python bindings for CUDA functionality will not</span>
<span class="sd"> be available until this initialization takes place. Ordinary users</span>
<span class="sd"> should not need this, as all of PyTorch's CUDA methods</span>
<span class="sd"> automatically initialize CUDA state on-demand.</span>
<span class="sd"> Does nothing if the CUDA state is already initialized.</span>
<span class="sd"> """</span>
<span class="n">_lazy_init</span><span class="p">()</span></div>
<span class="k">def</span> <span class="nf">_lazy_init</span><span class="p">():</span>
<span class="k">global</span> <span class="n">_initialized</span><span class="p">,</span> <span class="n">_queued_calls</span>
<span class="k">if</span> <span class="n">is_initialized</span><span class="p">()</span> <span class="ow">or</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">_tls</span><span class="p">,</span> <span class="s1">'is_initializing'</span><span class="p">):</span>
<span class="k">return</span>
<span class="k">with</span> <span class="n">_initialization_lock</span><span class="p">:</span>
<span class="c1"># We be double-checked locking, boys! This is OK because</span>
<span class="c1"># the above test was GIL protected anyway. The inner test</span>
<span class="c1"># is for when a thread blocked on some other thread which was</span>
<span class="c1"># doing the initialization; when they get the lock, they will</span>
<span class="c1"># find there is nothing left to do.</span>
<span class="k">if</span> <span class="n">is_initialized</span><span class="p">():</span>
<span class="k">return</span>
<span class="c1"># It is important to prevent other threads from entering _lazy_init</span>
<span class="c1"># immediately, while we are still guaranteed to have the GIL, because some</span>
<span class="c1"># of the C calls we make below will release the GIL</span>
<span class="k">if</span> <span class="n">_is_in_bad_fork</span><span class="p">():</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span>
<span class="s2">"Cannot re-initialize CUDA in forked subprocess. To use CUDA with "</span>
<span class="s2">"multiprocessing, you must use the 'spawn' start method"</span><span class="p">)</span>
<span class="k">if</span> <span class="ow">not</span> <span class="nb">hasattr</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="p">,</span> <span class="s1">'_cuda_getDeviceCount'</span><span class="p">):</span>
<span class="k">raise</span> <span class="ne">AssertionError</span><span class="p">(</span><span class="s2">"Torch not compiled with CUDA enabled"</span><span class="p">)</span>
<span class="k">if</span> <span class="n">_cudart</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">AssertionError</span><span class="p">(</span>
<span class="s2">"libcudart functions unavailable. It looks like you have a broken build?"</span><span class="p">)</span>
<span class="c1"># This function throws if there's a driver initialization error, no GPUs</span>
<span class="c1"># are found or any other error occurs</span>
<span class="k">if</span> <span class="s1">'CUDA_MODULE_LOADING'</span> <span class="ow">not</span> <span class="ow">in</span> <span class="n">os</span><span class="o">.</span><span class="n">environ</span><span class="p">:</span>
<span class="n">os</span><span class="o">.</span><span class="n">environ</span><span class="p">[</span><span class="s1">'CUDA_MODULE_LOADING'</span><span class="p">]</span> <span class="o">=</span> <span class="s1">'LAZY'</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_init</span><span class="p">()</span>
<span class="c1"># Some of the queued calls may reentrantly call _lazy_init();</span>
<span class="c1"># we need to just return without initializing in that case.</span>
<span class="c1"># However, we must not let any *other* threads in!</span>
<span class="n">_tls</span><span class="o">.</span><span class="n">is_initializing</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">for</span> <span class="n">calls</span> <span class="ow">in</span> <span class="n">_lazy_seed_tracker</span><span class="o">.</span><span class="n">get_calls</span><span class="p">():</span>
<span class="k">if</span> <span class="n">calls</span><span class="p">:</span>
<span class="n">_queued_calls</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">calls</span><span class="p">)</span>
<span class="k">try</span><span class="p">:</span>
<span class="k">for</span> <span class="n">queued_call</span><span class="p">,</span> <span class="n">orig_traceback</span> <span class="ow">in</span> <span class="n">_queued_calls</span><span class="p">:</span>
<span class="k">try</span><span class="p">:</span>
<span class="n">queued_call</span><span class="p">()</span>
<span class="k">except</span> <span class="ne">Exception</span> <span class="k">as</span> <span class="n">e</span><span class="p">:</span>
<span class="n">msg</span> <span class="o">=</span> <span class="p">(</span><span class="sa">f</span><span class="s2">"CUDA call failed lazily at initialization with error: </span><span class="si">{</span><span class="nb">str</span><span class="p">(</span><span class="n">e</span><span class="p">)</span><span class="si">}</span><span class="se">\n\n</span><span class="s2">"</span>
<span class="sa">f</span><span class="s2">"CUDA call was originally invoked at:</span><span class="se">\n\n</span><span class="si">{</span><span class="n">orig_traceback</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span>
<span class="k">raise</span> <span class="n">DeferredCudaCallError</span><span class="p">(</span><span class="n">msg</span><span class="p">)</span> <span class="kn">from</span> <span class="nn">e</span>
<span class="k">finally</span><span class="p">:</span>
<span class="nb">delattr</span><span class="p">(</span><span class="n">_tls</span><span class="p">,</span> <span class="s1">'is_initializing'</span><span class="p">)</span>
<span class="n">_initialized</span> <span class="o">=</span> <span class="kc">True</span>
<span class="k">def</span> <span class="nf">cudart</span><span class="p">():</span>
<span class="n">_lazy_init</span><span class="p">()</span>
<span class="k">return</span> <span class="n">_cudart</span>
<span class="k">class</span> <span class="nc">cudaStatus</span><span class="p">:</span>
<span class="n">SUCCESS</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">ERROR_NOT_READY</span><span class="p">:</span> <span class="nb">int</span> <span class="o">=</span> <span class="mi">34</span>
<span class="k">class</span> <span class="nc">CudaError</span><span class="p">(</span><span class="ne">RuntimeError</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">code</span><span class="p">:</span> <span class="nb">int</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="n">msg</span> <span class="o">=</span> <span class="n">_cudart</span><span class="o">.</span><span class="n">cudaGetErrorString</span><span class="p">(</span><span class="n">_cudart</span><span class="o">.</span><span class="n">cudaError</span><span class="p">(</span><span class="n">code</span><span class="p">))</span>
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="s1">'</span><span class="si">{0}</span><span class="s1"> (</span><span class="si">{1}</span><span class="s1">)'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">msg</span><span class="p">,</span> <span class="n">code</span><span class="p">))</span>
<span class="k">def</span> <span class="nf">check_error</span><span class="p">(</span><span class="n">res</span><span class="p">:</span> <span class="nb">int</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="k">if</span> <span class="n">res</span> <span class="o">!=</span> <span class="n">_cudart</span><span class="o">.</span><span class="n">cudaError</span><span class="o">.</span><span class="n">success</span><span class="p">:</span>
<span class="k">raise</span> <span class="n">CudaError</span><span class="p">(</span><span class="n">res</span><span class="p">)</span>
<span class="k">class</span> <span class="nc">_DeviceGuard</span><span class="p">:</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">index</span><span class="p">:</span> <span class="nb">int</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">idx</span> <span class="o">=</span> <span class="n">index</span>
<span class="bp">self</span><span class="o">.</span><span class="n">prev_idx</span> <span class="o">=</span> <span class="o">-</span><span class="mi">1</span>
<span class="k">def</span> <span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">prev_idx</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">_exchange_device</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">idx</span><span class="p">)</span>
<span class="k">def</span> <span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">type</span><span class="p">:</span> <span class="n">Any</span><span class="p">,</span> <span class="n">value</span><span class="p">:</span> <span class="n">Any</span><span class="p">,</span> <span class="n">traceback</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
<span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">_exchange_device</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">prev_idx</span><span class="p">)</span>
<span class="k">return</span> <span class="kc">False</span>
<div class="viewcode-block" id="device"><a class="viewcode-back" href="../../generated/torch.cuda.device.html#torch.cuda.device">[docs]</a><span class="k">class</span> <span class="nc">device</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Context-manager that changes the selected device.</span>
<span class="sd"> Args:</span>
<span class="sd"> device (torch.device or int): device index to select. It's a no-op if</span>
<span class="sd"> this argument is a negative integer or ``None``.</span>
<span class="sd"> """</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">device</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">idx</span> <span class="o">=</span> <span class="n">_get_device_index</span><span class="p">(</span><span class="n">device</span><span class="p">,</span> <span class="n">optional</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">prev_idx</span> <span class="o">=</span> <span class="o">-</span><span class="mi">1</span>
<span class="k">def</span> <span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">prev_idx</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">_exchange_device</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">idx</span><span class="p">)</span>
<span class="k">def</span> <span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">type</span><span class="p">:</span> <span class="n">Any</span><span class="p">,</span> <span class="n">value</span><span class="p">:</span> <span class="n">Any</span><span class="p">,</span> <span class="n">traceback</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
<span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">_exchange_device</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">prev_idx</span><span class="p">)</span>
<span class="k">return</span> <span class="kc">False</span></div>
<div class="viewcode-block" id="device_of"><a class="viewcode-back" href="../../generated/torch.cuda.device_of.html#torch.cuda.device_of">[docs]</a><span class="k">class</span> <span class="nc">device_of</span><span class="p">(</span><span class="n">device</span><span class="p">):</span>
<span class="sa">r</span><span class="sd">"""Context-manager that changes the current device to that of given object.</span>
<span class="sd"> You can use both tensors and storages as arguments. If a given object is</span>
<span class="sd"> not allocated on a GPU, this is a no-op.</span>
<span class="sd"> Args:</span>
<span class="sd"> obj (Tensor or Storage): object allocated on the selected device.</span>
<span class="sd"> """</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">obj</span><span class="p">):</span>
<span class="n">idx</span> <span class="o">=</span> <span class="n">obj</span><span class="o">.</span><span class="n">get_device</span><span class="p">()</span> <span class="k">if</span> <span class="n">obj</span><span class="o">.</span><span class="n">is_cuda</span> <span class="k">else</span> <span class="o">-</span><span class="mi">1</span>
<span class="nb">super</span><span class="p">()</span><span class="o">.</span><span class="fm">__init__</span><span class="p">(</span><span class="n">idx</span><span class="p">)</span></div>
<div class="viewcode-block" id="set_device"><a class="viewcode-back" href="../../generated/torch.cuda.set_device.html#torch.cuda.set_device">[docs]</a><span class="k">def</span> <span class="nf">set_device</span><span class="p">(</span><span class="n">device</span><span class="p">:</span> <span class="n">_device_t</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Sets the current device.</span>
<span class="sd"> Usage of this function is discouraged in favor of :any:`device`. In most</span>
<span class="sd"> cases it's better to use ``CUDA_VISIBLE_DEVICES`` environmental variable.</span>
<span class="sd"> Args:</span>
<span class="sd"> device (torch.device or int): selected device. This function is a no-op</span>
<span class="sd"> if this argument is negative.</span>
<span class="sd"> """</span>
<span class="n">device</span> <span class="o">=</span> <span class="n">_get_device_index</span><span class="p">(</span><span class="n">device</span><span class="p">)</span>
<span class="k">if</span> <span class="n">device</span> <span class="o">>=</span> <span class="mi">0</span><span class="p">:</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_setDevice</span><span class="p">(</span><span class="n">device</span><span class="p">)</span></div>
<div class="viewcode-block" id="get_device_name"><a class="viewcode-back" href="../../generated/torch.cuda.get_device_name.html#torch.cuda.get_device_name">[docs]</a><span class="k">def</span> <span class="nf">get_device_name</span><span class="p">(</span><span class="n">device</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">_device_t</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="nb">str</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Gets the name of a device.</span>
<span class="sd"> Args:</span>
<span class="sd"> device (torch.device or int, optional): device for which to return the</span>
<span class="sd"> name. This function is a no-op if this argument is a negative</span>
<span class="sd"> integer. It uses the current device, given by :func:`~torch.cuda.current_device`,</span>
<span class="sd"> if :attr:`device` is ``None`` (default).</span>
<span class="sd"> Returns:</span>
<span class="sd"> str: the name of the device</span>
<span class="sd"> """</span>
<span class="k">return</span> <span class="n">get_device_properties</span><span class="p">(</span><span class="n">device</span><span class="p">)</span><span class="o">.</span><span class="n">name</span></div>
<div class="viewcode-block" id="get_device_capability"><a class="viewcode-back" href="../../generated/torch.cuda.get_device_capability.html#torch.cuda.get_device_capability">[docs]</a><span class="k">def</span> <span class="nf">get_device_capability</span><span class="p">(</span><span class="n">device</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="n">_device_t</span><span class="p">]</span> <span class="o">=</span> <span class="kc">None</span><span class="p">)</span> <span class="o">-></span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">int</span><span class="p">,</span> <span class="nb">int</span><span class="p">]:</span>
<span class="sa">r</span><span class="sd">"""Gets the cuda capability of a device.</span>
<span class="sd"> Args:</span>
<span class="sd"> device (torch.device or int, optional): device for which to return the</span>
<span class="sd"> device capability. This function is a no-op if this argument is</span>
<span class="sd"> a negative integer. It uses the current device, given by</span>
<span class="sd"> :func:`~torch.cuda.current_device`, if :attr:`device` is ``None``</span>
<span class="sd"> (default).</span>
<span class="sd"> Returns:</span>
<span class="sd"> tuple(int, int): the major and minor cuda capability of the device</span>
<span class="sd"> """</span>
<span class="n">prop</span> <span class="o">=</span> <span class="n">get_device_properties</span><span class="p">(</span><span class="n">device</span><span class="p">)</span>
<span class="k">return</span> <span class="n">prop</span><span class="o">.</span><span class="n">major</span><span class="p">,</span> <span class="n">prop</span><span class="o">.</span><span class="n">minor</span></div>
<div class="viewcode-block" id="get_device_properties"><a class="viewcode-back" href="../../generated/torch.cuda.get_device_properties.html#torch.cuda.get_device_properties">[docs]</a><span class="k">def</span> <span class="nf">get_device_properties</span><span class="p">(</span><span class="n">device</span><span class="p">:</span> <span class="n">_device_t</span><span class="p">)</span> <span class="o">-></span> <span class="n">_CudaDeviceProperties</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Gets the properties of a device.</span>
<span class="sd"> Args:</span>
<span class="sd"> device (torch.device or int or str): device for which to return the</span>
<span class="sd"> properties of the device.</span>
<span class="sd"> Returns:</span>
<span class="sd"> _CudaDeviceProperties: the properties of the device</span>
<span class="sd"> """</span>
<span class="n">_lazy_init</span><span class="p">()</span> <span class="c1"># will define _get_device_properties</span>
<span class="n">device</span> <span class="o">=</span> <span class="n">_get_device_index</span><span class="p">(</span><span class="n">device</span><span class="p">,</span> <span class="n">optional</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="k">if</span> <span class="n">device</span> <span class="o"><</span> <span class="mi">0</span> <span class="ow">or</span> <span class="n">device</span> <span class="o">>=</span> <span class="n">device_count</span><span class="p">():</span>
<span class="k">raise</span> <span class="ne">AssertionError</span><span class="p">(</span><span class="s2">"Invalid device id"</span><span class="p">)</span>
<span class="k">return</span> <span class="n">_get_device_properties</span><span class="p">(</span><span class="n">device</span><span class="p">)</span> <span class="c1"># type: ignore[name-defined]</span></div>
<div class="viewcode-block" id="can_device_access_peer"><a class="viewcode-back" href="../../generated/torch.cuda.can_device_access_peer.html#torch.cuda.can_device_access_peer">[docs]</a><span class="k">def</span> <span class="nf">can_device_access_peer</span><span class="p">(</span><span class="n">device</span><span class="p">:</span> <span class="n">_device_t</span><span class="p">,</span> <span class="n">peer_device</span><span class="p">:</span> <span class="n">_device_t</span><span class="p">)</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Checks if peer access between two devices is possible.</span>
<span class="sd"> """</span>
<span class="n">_lazy_init</span><span class="p">()</span>
<span class="n">device</span> <span class="o">=</span> <span class="n">_get_device_index</span><span class="p">(</span><span class="n">device</span><span class="p">,</span> <span class="n">optional</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span>
<span class="n">peer_device</span> <span class="o">=</span> <span class="n">_get_device_index</span><span class="p">(</span><span class="n">peer_device</span><span class="p">)</span>
<span class="k">if</span> <span class="n">device</span> <span class="o"><</span> <span class="mi">0</span> <span class="ow">or</span> <span class="n">device</span> <span class="o">>=</span> <span class="n">device_count</span><span class="p">():</span>
<span class="k">raise</span> <span class="ne">AssertionError</span><span class="p">(</span><span class="s2">"Invalid device id"</span><span class="p">)</span>
<span class="k">if</span> <span class="n">peer_device</span> <span class="o"><</span> <span class="mi">0</span> <span class="ow">or</span> <span class="n">peer_device</span> <span class="o">>=</span> <span class="n">device_count</span><span class="p">():</span>
<span class="k">raise</span> <span class="ne">AssertionError</span><span class="p">(</span><span class="s2">"Invalid peer device id"</span><span class="p">)</span>
<span class="k">return</span> <span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_canDeviceAccessPeer</span><span class="p">(</span><span class="n">device</span><span class="p">,</span> <span class="n">peer_device</span><span class="p">)</span></div>
<div class="viewcode-block" id="StreamContext"><a class="viewcode-back" href="../../generated/torch.cuda.StreamContext.html#torch.cuda.StreamContext">[docs]</a><span class="k">class</span> <span class="nc">StreamContext</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Context-manager that selects a given stream.</span>
<span class="sd"> All CUDA kernels queued within its context will be enqueued on a selected</span>
<span class="sd"> stream.</span>
<span class="sd"> Args:</span>
<span class="sd"> Stream (Stream): selected stream. This manager is a no-op if it's</span>
<span class="sd"> ``None``.</span>
<span class="sd"> .. note:: Streams are per-device.</span>
<span class="sd"> """</span>
<span class="n">cur_stream</span> <span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="s1">'torch.cuda.Stream'</span><span class="p">]</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">stream</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="s1">'torch.cuda.Stream'</span><span class="p">]):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">stream</span> <span class="o">=</span> <span class="n">stream</span>
<span class="bp">self</span><span class="o">.</span><span class="n">idx</span> <span class="o">=</span> <span class="n">_get_device_index</span><span class="p">(</span><span class="kc">None</span><span class="p">,</span> <span class="kc">True</span><span class="p">)</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">is_scripting</span><span class="p">():</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">idx</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="bp">self</span><span class="o">.</span><span class="n">idx</span> <span class="o">=</span> <span class="o">-</span><span class="mi">1</span>
<span class="bp">self</span><span class="o">.</span><span class="n">src_prev_stream</span> <span class="o">=</span> <span class="kc">None</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">is_scripting</span><span class="p">()</span> <span class="k">else</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">default_stream</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dst_prev_stream</span> <span class="o">=</span> <span class="kc">None</span> <span class="k">if</span> <span class="ow">not</span> <span class="n">torch</span><span class="o">.</span><span class="n">jit</span><span class="o">.</span><span class="n">is_scripting</span><span class="p">()</span> <span class="k">else</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">default_stream</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
<span class="k">def</span> <span class="fm">__enter__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="c1"># Local cur_stream variable for type refinement</span>
<span class="n">cur_stream</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">stream</span>
<span class="c1"># Return if stream is None or CUDA device not available</span>
<span class="k">if</span> <span class="n">cur_stream</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">idx</span> <span class="o">==</span> <span class="o">-</span><span class="mi">1</span><span class="p">:</span>
<span class="k">return</span>
<span class="bp">self</span><span class="o">.</span><span class="n">src_prev_stream</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">current_stream</span><span class="p">(</span><span class="kc">None</span><span class="p">)</span>
<span class="c1"># If the stream is not on the current device, then</span>
<span class="c1"># set the current stream on the device</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">src_prev_stream</span><span class="o">.</span><span class="n">device</span> <span class="o">!=</span> <span class="n">cur_stream</span><span class="o">.</span><span class="n">device</span><span class="p">:</span>
<span class="k">with</span> <span class="n">device</span><span class="p">(</span><span class="n">cur_stream</span><span class="o">.</span><span class="n">device</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">dst_prev_stream</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">current_stream</span><span class="p">(</span><span class="n">cur_stream</span><span class="o">.</span><span class="n">device</span><span class="p">)</span>
<span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">set_stream</span><span class="p">(</span><span class="n">cur_stream</span><span class="p">)</span>
<span class="k">def</span> <span class="fm">__exit__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">type</span><span class="p">:</span> <span class="n">Any</span><span class="p">,</span> <span class="n">value</span><span class="p">:</span> <span class="n">Any</span><span class="p">,</span> <span class="n">traceback</span><span class="p">:</span> <span class="n">Any</span><span class="p">):</span>
<span class="c1"># Local cur_stream variable for type refinement</span>
<span class="n">cur_stream</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">stream</span>
<span class="c1"># If stream is None or no CUDA device available, return</span>
<span class="k">if</span> <span class="n">cur_stream</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="bp">self</span><span class="o">.</span><span class="n">idx</span> <span class="o">==</span> <span class="o">-</span><span class="mi">1</span><span class="p">:</span>
<span class="k">return</span>
<span class="c1"># Reset the stream on the original device</span>
<span class="c1"># and destination device</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">src_prev_stream</span><span class="o">.</span><span class="n">device</span> <span class="o">!=</span> <span class="n">cur_stream</span><span class="o">.</span><span class="n">device</span><span class="p">:</span> <span class="c1"># type: ignore[union-attr]</span>
<span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">set_stream</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">dst_prev_stream</span><span class="p">)</span> <span class="c1"># type: ignore[arg-type]</span>
<span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">set_stream</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">src_prev_stream</span><span class="p">)</span> <span class="c1"># type: ignore[arg-type]</span></div>
<div class="viewcode-block" id="stream"><a class="viewcode-back" href="../../generated/torch.cuda.stream.html#torch.cuda.stream">[docs]</a><span class="k">def</span> <span class="nf">stream</span><span class="p">(</span><span class="n">stream</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="s1">'torch.cuda.Stream'</span><span class="p">])</span> <span class="o">-></span> <span class="n">StreamContext</span><span class="p">:</span>
<span class="sa">r</span><span class="sd">"""Wrapper around the Context-manager StreamContext that</span>
<span class="sd"> selects a given stream.</span>
<span class="sd"> Arguments:</span>
<span class="sd"> stream (Stream): selected stream. This manager is a no-op if it's</span>
<span class="sd"> ``None``.</span>
<span class="sd"> ..Note:: In eager mode stream is of type Stream class while in JIT it is</span>
<span class="sd"> an object of the custom class ``torch.classes.cuda.Stream``.</span>
<span class="sd"> """</span>
<span class="k">return</span> <span class="n">StreamContext</span><span class="p">(</span><span class="n">stream</span><span class="p">)</span></div>
<div class="viewcode-block" id="set_stream"><a class="viewcode-back" href="../../generated/torch.cuda.set_stream.html#torch.cuda.set_stream">[docs]</a><span class="k">def</span> <span class="nf">set_stream</span><span class="p">(</span><span class="n">stream</span><span class="p">:</span> <span class="n">Stream</span><span class="p">):</span>
<span class="sa">r</span><span class="sd">"""Sets the current stream.This is a wrapper API to set the stream.</span>
<span class="sd"> Usage of this function is discouraged in favor of the ``stream``</span>
<span class="sd"> context manager.</span>
<span class="sd"> Args:</span>
<span class="sd"> stream (Stream): selected stream. This function is a no-op</span>
<span class="sd"> if this argument is ``None``.</span>
<span class="sd"> """</span>
<span class="k">if</span> <span class="n">stream</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_setStream</span><span class="p">(</span><span class="n">stream_id</span><span class="o">=</span><span class="n">stream</span><span class="o">.</span><span class="n">stream_id</span><span class="p">,</span> <span class="n">device_index</span><span class="o">=</span><span class="n">stream</span><span class="o">.</span><span class="n">device_index</span><span class="p">,</span> <span class="n">device_type</span><span class="o">=</span><span class="n">stream</span><span class="o">.</span><span class="n">device_type</span><span class="p">)</span></div>
<span class="k">def</span> <span class="nf">_parse_visible_devices</span><span class="p">()</span> <span class="o">-></span> <span class="n">Union</span><span class="p">[</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span> <span class="n">List</span><span class="p">[</span><span class="nb">str</span><span class="p">]]:</span>
<span class="sd">"""Parse CUDA_VISIBLE_DEVICES environment variable."""</span>
<span class="n">var</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">getenv</span><span class="p">(</span><span class="s2">"CUDA_VISIBLE_DEVICES"</span><span class="p">)</span>
<span class="k">if</span> <span class="n">var</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="k">return</span> <span class="nb">list</span><span class="p">(</span><span class="nb">range</span><span class="p">(</span><span class="mi">64</span><span class="p">))</span>
<span class="k">def</span> <span class="nf">_strtoul</span><span class="p">(</span><span class="n">s</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="nb">int</span><span class="p">:</span>
<span class="sd">"""Return -1 or positive integer sequence string starts with,"""</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">s</span><span class="p">:</span>
<span class="k">return</span> <span class="o">-</span><span class="mi">1</span>
<span class="k">for</span> <span class="n">idx</span><span class="p">,</span> <span class="n">c</span> <span class="ow">in</span> <span class="nb">enumerate</span><span class="p">(</span><span class="n">s</span><span class="p">):</span>
<span class="k">if</span> <span class="ow">not</span> <span class="p">(</span><span class="n">c</span><span class="o">.</span><span class="n">isdigit</span><span class="p">()</span> <span class="ow">or</span> <span class="p">(</span><span class="n">idx</span> <span class="o">==</span> <span class="mi">0</span> <span class="ow">and</span> <span class="n">c</span> <span class="ow">in</span> <span class="s1">'+-'</span><span class="p">)):</span>
<span class="k">break</span>
<span class="k">if</span> <span class="n">idx</span> <span class="o">+</span> <span class="mi">1</span> <span class="o">==</span> <span class="nb">len</span><span class="p">(</span><span class="n">s</span><span class="p">):</span>
<span class="n">idx</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="k">return</span> <span class="nb">int</span><span class="p">(</span><span class="n">s</span><span class="p">[:</span><span class="n">idx</span><span class="p">])</span> <span class="k">if</span> <span class="n">idx</span> <span class="o">></span> <span class="mi">0</span> <span class="k">else</span> <span class="o">-</span><span class="mi">1</span>
<span class="k">def</span> <span class="nf">parse_list_with_prefix</span><span class="p">(</span><span class="n">lst</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">prefix</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="n">List</span><span class="p">[</span><span class="nb">str</span><span class="p">]:</span>
<span class="n">rcs</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">str</span><span class="p">]</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">elem</span> <span class="ow">in</span> <span class="n">lst</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s2">","</span><span class="p">):</span>
<span class="c1"># Repeated id results in empty set</span>
<span class="k">if</span> <span class="n">elem</span> <span class="ow">in</span> <span class="n">rcs</span><span class="p">:</span>
<span class="k">return</span> <span class="n">cast</span><span class="p">(</span><span class="n">List</span><span class="p">[</span><span class="nb">str</span><span class="p">],</span> <span class="p">[])</span>
<span class="c1"># Anything other but prefix is ignored</span>
<span class="k">if</span> <span class="ow">not</span> <span class="n">elem</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="n">prefix</span><span class="p">):</span>
<span class="k">break</span>
<span class="n">rcs</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">elem</span><span class="p">)</span>
<span class="k">return</span> <span class="n">rcs</span>
<span class="k">if</span> <span class="n">var</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="s2">"GPU-"</span><span class="p">):</span>
<span class="k">return</span> <span class="n">parse_list_with_prefix</span><span class="p">(</span><span class="n">var</span><span class="p">,</span> <span class="s2">"GPU-"</span><span class="p">)</span>
<span class="k">if</span> <span class="n">var</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="s2">"MIG-"</span><span class="p">):</span>
<span class="k">return</span> <span class="n">parse_list_with_prefix</span><span class="p">(</span><span class="n">var</span><span class="p">,</span> <span class="s2">"MIG-"</span><span class="p">)</span>
<span class="c1"># CUDA_VISIBLE_DEVICES uses something like strtoul</span>
<span class="c1"># which makes `1gpu2,2ampere` is equivalent to `1,2`</span>
<span class="n">rc</span><span class="p">:</span> <span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">]</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">for</span> <span class="n">elem</span> <span class="ow">in</span> <span class="n">var</span><span class="o">.</span><span class="n">split</span><span class="p">(</span><span class="s2">","</span><span class="p">):</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">_strtoul</span><span class="p">(</span><span class="n">elem</span><span class="o">.</span><span class="n">strip</span><span class="p">())</span>
<span class="c1"># Repeated ordinal results in empty set</span>
<span class="k">if</span> <span class="n">x</span> <span class="ow">in</span> <span class="n">rc</span><span class="p">:</span>
<span class="k">return</span> <span class="n">cast</span><span class="p">(</span><span class="n">List</span><span class="p">[</span><span class="nb">int</span><span class="p">],</span> <span class="p">[])</span>
<span class="c1"># Negative value aborts the sequence</span>
<span class="k">if</span> <span class="n">x</span> <span class="o"><</span> <span class="mi">0</span><span class="p">:</span>
<span class="k">break</span>
<span class="n">rc</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">x</span><span class="p">)</span>
<span class="k">return</span> <span class="n">rc</span>
<span class="k">def</span> <span class="nf">_raw_device_count_nvml</span><span class="p">()</span> <span class="o">-></span> <span class="nb">int</span><span class="p">:</span>
<span class="sd">"""Return number of devices as reported by NVML</span>
<span class="sd"> or negative value if NVML discovery/initialization failed."""</span>
<span class="kn">from</span> <span class="nn">ctypes</span> <span class="kn">import</span> <span class="n">CDLL</span><span class="p">,</span> <span class="n">c_int</span><span class="p">,</span> <span class="n">byref</span>
<span class="n">nvml_h</span> <span class="o">=</span> <span class="n">CDLL</span><span class="p">(</span><span class="s2">"libnvidia-ml.so.1"</span><span class="p">)</span>
<span class="n">rc</span> <span class="o">=</span> <span class="n">nvml_h</span><span class="o">.</span><span class="n">nvmlInit</span><span class="p">()</span>