-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathtunable.html
1446 lines (1167 loc) · 116 KB
/
tunable.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 charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>torch.cuda.tunable — PyTorch 2.6 documentation</title>
<link rel="canonical" href="https://pytorch.org/docs/stable/_modules/torch/cuda/tunable.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="stylesheet" href="../../../_static/css/custom.css" type="text/css" />
<link rel="index" title="Index" href="../../../genindex.html" />
<link rel="search" title="Search" href="../../../search.html" />
<!-- Google Tag Manager -->
<script>(function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':
new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],
j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src=
'https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);
})(window,document,'script','dataLayer','GTM-T8XT4PS');</script>
<!-- End Google Tag Manager -->
<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 class="main-menu-item">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Learn
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/get-started">
<span class=dropdown-title>Get Started</span>
<p>Run PyTorch locally or get started quickly with one of the supported cloud platforms</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials">
<span class="dropdown-title">Tutorials</span>
<p>Whats new in PyTorch tutorials</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/beginner/basics/intro.html">
<span class="dropdown-title">Learn the Basics</span>
<p>Familiarize yourself with PyTorch concepts and modules</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/recipes/recipes_index.html">
<span class="dropdown-title">PyTorch Recipes</span>
<p>Bite-size, ready-to-deploy PyTorch code examples</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/tutorials/beginner/introyt.html">
<span class="dropdown-title">Intro to PyTorch - YouTube Series</span>
<p>Master PyTorch basics with our engaging YouTube tutorial series</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Ecosystem
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/ecosystem">
<span class="dropdown-title">Tools</span>
<p>Learn about the tools and frameworks in the PyTorch Ecosystem</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://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/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/ecosystem/contributor-awards-2023">
<span class="dropdown-title">Contributor Awards - 2023</span>
<p>Award winners announced at this year's PyTorch Conference</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Edge
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/edge">
<span class="dropdown-title">About PyTorch Edge</span>
<p>Build innovative and privacy-aware AI experiences for edge devices</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/executorch-overview">
<span class="dropdown-title">ExecuTorch</span>
<p>End-to-end solution for enabling on-device inference capabilities across mobile and edge devices</p>
</a>
</div>
</div>
</li>
<li class="main-menu-item">
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Docs
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/docs/stable/index.html">
<span class="dropdown-title">PyTorch</span>
<p>Explore the documentation for comprehensive guidance on how to use PyTorch</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/pytorch-domains">
<span class="dropdown-title">PyTorch Domains</span>
<p>Read the PyTorch Domains documentation to learn more about domain-specific libraries</p>
</a>
</div>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
Blogs & News
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/blog/">
<span class="dropdown-title">PyTorch Blog</span>
<p>Catch up on the latest technical news and happenings</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/community-blog">
<span class="dropdown-title">Community Blog</span>
<p>Stories from the PyTorch ecosystem</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/videos">
<span class="dropdown-title">Videos</span>
<p>Learn about the latest PyTorch tutorials, new, and more </p>
<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/events">
<span class="dropdown-title">Events</span>
<p>Find events, webinars, and podcasts</p>
</a>
</div>
</li>
<li>
<div id="resourcesDropdownButton" data-toggle="resources-dropdown" class="resources-dropdown">
<a class="with-down-arrow">
About
</a>
<div class="resources-dropdown-menu">
<a class="nav-dropdown-item" href="https://pytorch.org/foundation">
<span class="dropdown-title">PyTorch Foundation</span>
<p>Learn more about the PyTorch Foundation</p>
</a>
<a class="nav-dropdown-item" href="https://pytorch.org/governing-board">
<span class="dropdown-title">Governing Board</span>
<p></p>
</a>
</div>
</div>
</li>
<li class="main-menu-item">
<div class="no-dropdown">
<a href="https://pytorch.org/join" data-cta="join">
Become a Member
</a>
</div>
</li>
<li>
<div class="main-menu-item">
<a href="https://github.com/pytorch/pytorch" class="github-icon">
</a>
</div>
</li>
<!--- TODO: This block adds the search icon to the nav bar. We will enable it later.
<li>
<div class="main-menu-item">
<a href="https://github.com/pytorch/pytorch" class="search-icon">
</a>
</div>
</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'>2.6 ▼</a>
</div>
<div id="searchBox">
<div class="searchbox" id="googleSearchBox">
<script async src="https://cse.google.com/cse.js?cx=e65585f8c3ea1440e"></script>
<div class="gcse-search"></div>
</div>
<div id="sphinxSearchBox" style="display: none;">
<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>
</div>
<form id="searchForm">
<label style="margin-bottom: 1rem">
<input type="radio" name="searchType" value="google" checked>
Google Search
</label>
<label style="margin-bottom: 1rem">
<input type="radio" name="searchType" value="sphinx">
Classic Search
</label>
</form>
<script>
document.addEventListener('DOMContentLoaded', function() {
const searchForm = document.getElementById('searchForm');
const googleSearchBox = document.getElementById('googleSearchBox');
const sphinxSearchBox = document.getElementById('sphinxSearchBox');
// Function to toggle search box visibility
function toggleSearchBox(searchType) {
googleSearchBox.style.display = searchType === 'google' ? 'block' : 'none';
sphinxSearchBox.style.display = searchType === 'sphinx' ? 'block' : 'none';
}
// Determine the default search type
let defaultSearchType;
const currentUrl = window.location.href;
if (currentUrl.startsWith('https://pytorch.org/docs/stable')) {
// For the stable documentation, default to Google
defaultSearchType = localStorage.getItem('searchType') || 'google';
} else {
// For any other version, including docs-preview, default to Sphinx
defaultSearchType = 'sphinx';
}
// Set the default search type
document.querySelector(`input[name="searchType"][value="${defaultSearchType}"]`).checked = true;
toggleSearchBox(defaultSearchType);
// Event listener for changes in search type
searchForm.addEventListener('change', function(event) {
const selectedSearchType = event.target.value;
localStorage.setItem('searchType', selectedSearchType);
toggleSearchBox(selectedSearchType);
});
// Set placeholder text for Google search box
window.onload = function() {
var placeholderText = "Search Docs";
var googleSearchboxText = document.querySelector("#gsc-i-id1");
if (googleSearchboxText) {
googleSearchboxText.placeholder = placeholderText;
googleSearchboxText.style.fontFamily = 'FreightSans';
googleSearchboxText.style.fontSize = "1.2rem";
googleSearchboxText.style.color = '#262626';
}
};
});
</script>
<!--temporarily add a link to survey -->
<div class="survey-link">
<p><i class="fas fa-poll" aria-hidden="true">  </i><a href="https://forms.gle/tdrnwJhaQ9tUePxz9">Share Your Feedback</a> about our new search</p>
</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">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/custom_operators.html">PyTorch Custom Operators Landing Page</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/fsdp.html">FSDP Notes</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../notes/get_start_xpu.html">Getting Started on Intel GPU</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">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="../../../accelerator.html">torch.accelerator</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../cpu.html">torch.cpu</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="../../../torch_cuda_memory.html">Understanding CUDA Memory Usage</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../torch_cuda_memory.html#generating-a-snapshot">Generating a Snapshot</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../torch_cuda_memory.html#using-the-visualizer">Using the visualizer</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../torch_cuda_memory.html#snapshot-api-reference">Snapshot API Reference</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="../../../xpu.html">torch.xpu</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../mtia.html">torch.mtia</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../mtia.memory.html">torch.mtia.memory</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../meta.html">Meta device</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="../../../export.html">torch.export</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.tensor.html">torch.distributed.tensor</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.fsdp.fully_shard.html">torch.distributed.fsdp.fully_shard</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.optim.html">torch.distributed.optim</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../distributed.pipelining.html">torch.distributed.pipelining</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="../../../torch.compiler.html">torch.compiler</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="../../../fx.experimental.html">torch.fx.experimental</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="../../../nn.attention.html">torch.nn.attention</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="../../../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="../../../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="../../../size.html">torch.Size</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="../../../utils.html">torch.utils</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="../../../deterministic.html">torch.utils.deterministic</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="../../../module_tracker.html">torch.utils.module_tracker</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>
<li class="toctree-l1"><a class="reference internal" href="../../../future_mod.html">torch.__future__</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../logging.html">torch._logging</a></li>
<li class="toctree-l1"><a class="reference internal" href="../../../torch_environment_variables.html">Torch Environment Variables</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><a href="../cuda.html">torch.cuda</a> ></li>
<li>torch.cuda.tunable</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">
<!-- Google Tag Manager (noscript) -->
<noscript><iframe src="https://www.googletagmanager.com/ns.html?id=GTM-T8XT4PS"
height="0" width="0" style="display:none;visibility:hidden"></iframe></noscript>
<!-- End Google Tag Manager (noscript) -->
<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.tunable</h1><div class="highlight"><pre>
<span></span><span class="sa">r</span><span class="sd">"""</span>
<span class="sd">This module exposes a TunableOp interface.</span>
<span class="sd">Some operations, such as GEMMs, could be implemented using more than one library</span>
<span class="sd">or more than one technique. For example, a GEMM could be implemented for CUDA or</span>
<span class="sd">ROCm using either the blas or blasLt libraries. Further, ROCm's rocblas and</span>
<span class="sd">hipblaslt libraries allow the user to query for all possible algorithms and then</span>
<span class="sd">choose one. How does one know which implementation is the fastest and should be</span>
<span class="sd">chosen? That's what TunableOp provides.</span>
<span class="sd">Enabling TunableOp and Tuning Separately</span>
<span class="sd">========================================</span>
<span class="sd">The TunableOp feature is enabled separately from enabling the tuning phase</span>
<span class="sd">itself. Enabling TunableOp means that PyTorch will replace any standard</span>
<span class="sd">operators with their Tunable implementations. Any call to a TunableOp first</span>
<span class="sd">checks whether it has already been tuned for the given operator inputs. If so,</span>
<span class="sd">it will immediately call the tuned operation; no further tuning will take place</span>
<span class="sd">even when the tuning setting is enabled. Instead if no tuning result is found,</span>
<span class="sd">and tuning is enabled, the TunableOp will benchmark every registered</span>
<span class="sd">implementation of that operator for the given set of inputs and select the</span>
<span class="sd">fastest.</span>
<span class="sd">File Input and Output</span>
<span class="sd">=====================</span>
<span class="sd">The first time any TunableOp is invoked, the internal database of tuned</span>
<span class="sd">operations will be prepared by attempting to read the results from the given</span>
<span class="sd">file. The default filename is 'tunableop_results.csv'. To support tuning when</span>
<span class="sd">multiple GPUs are used across multiple processes, the GPU device ordinal is</span>
<span class="sd">automatically inserted into the filename to avoid multiple processes overwriting</span>
<span class="sd">the same file.</span>
<span class="sd">If tuning is enabled and new tunings are discovered during the course of your</span>
<span class="sd">workload, it will also write out to this same filename with all tunings, both</span>
<span class="sd">the ones it read in at startup as well as the new ones found at runtime. This</span>
<span class="sd">can be used, for example, to build up a tunings file across many workloads by</span>
<span class="sd">reusing the same file. The output file is automatically created when the</span>
<span class="sd">application terminates. This behavior can be controlled by the C++ and Python</span>
<span class="sd">APIs but not the environment variables.</span>
<span class="sd">Assuming you specified a filename, you'll end up with a CSV file with contents</span>
<span class="sd">like so::</span>
<span class="sd"> Validator,PT_VERSION,2.2.0</span>
<span class="sd"> Validator,ROCM_VERSION,6.0.0.0-12969-1544e39</span>
<span class="sd"> Validator,HIPBLASLT_VERSION,0.6.0-a9c5cc7</span>
<span class="sd"> Validator,ROCBLAS_VERSION,4.0.0-72e57364-dirty</span>
<span class="sd"> GemmTunableOp_float_NT,nt_25088_4096_64,1219,1.262</span>
<span class="sd"> GemmTunableOp_float_NT,nt_4096_4096_64,1216,0.033</span>
<span class="sd">Note the "Validator" lines. If you change a library version, or ROCm version, or</span>
<span class="sd">PyTorch version, TunableOp will detect this and reject the tunings file because</span>
<span class="sd">the prior tunings are likely affected by other software changes.</span>
<span class="sd">The remaining lines are the tuned solutions for each TunableOp encountered</span>
<span class="sd">during your execution. Each line consists of 4 comma-separated fields: operator</span>
<span class="sd">name, operator parameters, solution name, and average execution time. The</span>
<span class="sd">execution time is an optional field. The CSV file can be edited, but with</span>
<span class="sd">caution. For example, the solution name (field 3) can be changed to "Default"</span>
<span class="sd">and it will fall back to the original PyTorch untuned implementation. Or, in the</span>
<span class="sd">case of ROCm's hipBLAS or hipBLASLt libraries, if you know the specific solution</span>
<span class="sd">index you can override the solution that TunableOp selected by replacing the</span>
<span class="sd">value. The operator name and parameters (fields 1 and 2) are internally named</span>
<span class="sd">and should not be modified. In the case of GemmTunableOp, field 1 indicates the</span>
<span class="sd">datatype and whether the inputs are transposed (T) or not (N) and field 2</span>
<span class="sd">indicates the M, N, K input shapes.</span>
<span class="sd">There is an option to enable verbose output but it is only recommended for</span>
<span class="sd">debugging purposes. This will produce a lot of diagnostic messages but may be</span>
<span class="sd">useful to see if TunableOp is being used at all. Otherwise, TunableOp is</span>
<span class="sd">completely silent, besides file output, unless there is a warning or error</span>
<span class="sd">during its use. The verbose option is only available by setting the environment</span>
<span class="sd">variable PYTORCH_TUNABLEOP_VEROBSE=1.</span>
<span class="sd">A Note on Tuning Behavior</span>
<span class="sd">=========================</span>
<span class="sd">Tuning an operator consists of iterating through the list or registered</span>
<span class="sd">implementations and profiling each one. The profile is established by running a</span>
<span class="sd">single implementation in a loop multiple times and taking the average execution</span>
<span class="sd">time.</span>
<span class="sd">By default, each possible solution for a given operator will be run for either</span>
<span class="sd">100 iterations or as many iterations that can be run within 30ms, whichever is</span>
<span class="sd">smaller, and its average execution will be calculated. The fastest solution</span>
<span class="sd">among all that were successfully profiled will be chosen. A profile might fail</span>
<span class="sd">if the given solution doesn't achieve the same accuracy as the default</span>
<span class="sd">implementation or if the solution returns an error code.</span>
<span class="sd">Current Tunable Operators</span>
<span class="sd">=========================</span>
<span class="sd">TunableGemm for ROCm</span>
<span class="sd">--------------------</span>
<span class="sd">Currently only a TunableGemm for ROCm is implemented. Note that CUDA builds of</span>
<span class="sd">PyTorch will function correctly when using TunableOp but the only solution</span>
<span class="sd">available to CUDA builds is the 'Default' implementation i.e. the original</span>
<span class="sd">cuBLAS default, now called through TunableOp. Any call to at::cuda::blas::gemm()</span>
<span class="sd">or ::bgemm() will be routed through TunableOp when enabled. Calling gemm() for a</span>
<span class="sd">given set of input arguments (transa, transb, m, n, k) will attempt to use the</span>
<span class="sd">fastest available implementation across both rocblas and hipblaslt.</span>
<span class="sd">Tuning Context</span>
<span class="sd">==============</span>
<span class="sd">The behavior of TunableOp is currently manipulated through environment</span>
<span class="sd">variables, the C++ interface of at::cuda::tunable::getTuningContext(), or the</span>
<span class="sd">torch.cuda.tunable python interfaces that wrap the C++ TuningContext. The</span>
<span class="sd">environment variables take precedence over any setting you manipulate using the</span>
<span class="sd">C++ or Python APIs.</span>
<span class="sd">"""</span>
<span class="kn">import</span> <span class="nn">concurrent.futures</span>
<span class="kn">import</span> <span class="nn">glob</span>
<span class="kn">import</span> <span class="nn">multiprocessing</span> <span class="k">as</span> <span class="nn">mp</span>
<span class="kn">import</span> <span class="nn">os</span>
<span class="kn">import</span> <span class="nn">shutil</span>
<span class="kn">import</span> <span class="nn">warnings</span>
<span class="kn">from</span> <span class="nn">typing</span> <span class="kn">import</span> <span class="n">Optional</span><span class="p">,</span> <span class="n">Tuple</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="n">__all__</span> <span class="o">=</span> <span class="p">[</span>
<span class="s2">"enable"</span><span class="p">,</span>
<span class="s2">"is_enabled"</span><span class="p">,</span>
<span class="s2">"tuning_enable"</span><span class="p">,</span>
<span class="s2">"tuning_is_enabled"</span><span class="p">,</span>
<span class="s2">"record_untuned_enable"</span><span class="p">,</span>
<span class="s2">"record_untuned_is_enabled"</span><span class="p">,</span>
<span class="s2">"set_max_tuning_duration"</span><span class="p">,</span>
<span class="s2">"get_max_tuning_duration"</span><span class="p">,</span>
<span class="s2">"set_max_tuning_iterations"</span><span class="p">,</span>
<span class="s2">"get_max_tuning_iterations"</span><span class="p">,</span>
<span class="s2">"set_filename"</span><span class="p">,</span>
<span class="s2">"get_filename"</span><span class="p">,</span>
<span class="s2">"get_results"</span><span class="p">,</span>
<span class="s2">"get_validators"</span><span class="p">,</span>
<span class="s2">"write_file_on_exit"</span><span class="p">,</span>
<span class="s2">"write_file"</span><span class="p">,</span>
<span class="s2">"read_file"</span><span class="p">,</span>
<span class="s2">"tune_gemm_in_file"</span><span class="p">,</span>
<span class="s2">"mgpu_tune_gemm_in_file"</span><span class="p">,</span>
<span class="p">]</span>
<div class="viewcode-block" id="enable"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.enable">[docs]</a><span class="k">def</span> <span class="nf">enable</span><span class="p">(</span><span class="n">val</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""This is the big on/off switch for all TunableOp implementations."""</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_tunableop_enable</span><span class="p">(</span><span class="n">val</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="is_enabled"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.is_enabled">[docs]</a><span class="k">def</span> <span class="nf">is_enabled</span><span class="p">()</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Returns whether the TunableOp feature is enabled."""</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_tunableop_is_enabled</span><span class="p">()</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="tuning_enable"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.tuning_enable">[docs]</a><span class="k">def</span> <span class="nf">tuning_enable</span><span class="p">(</span><span class="n">val</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Enable tuning of TunableOp implementations.</span>
<span class="sd"> When enabled, if a tuned entry isn't found, run the tuning step and record</span>
<span class="sd"> the entry.</span>
<span class="sd"> """</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_tunableop_tuning_enable</span><span class="p">(</span><span class="n">val</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="tuning_is_enabled"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.tuning_is_enabled">[docs]</a><span class="k">def</span> <span class="nf">tuning_is_enabled</span><span class="p">()</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Returns whether TunableOp implementations can be tuned."""</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_tunableop_tuning_is_enabled</span><span class="p">()</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="record_untuned_enable"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.record_untuned_enable">[docs]</a><span class="k">def</span> <span class="nf">record_untuned_enable</span><span class="p">(</span><span class="n">val</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">True</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Enable recording untuned of TunableOp perations for offline tuning.</span>
<span class="sd"> When enabled, if a tuned entry isn't found, write it to the untuned file.</span>
<span class="sd"> """</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_record_untuned_enable</span><span class="p">(</span><span class="n">val</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="record_untuned_is_enabled"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.record_untuned_is_enabled">[docs]</a><span class="k">def</span> <span class="nf">record_untuned_is_enabled</span><span class="p">()</span> <span class="o">-></span> <span class="nb">bool</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Returns whether TunableOp operations are recorded for offline tuning."""</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_record_untuned_is_enabled</span><span class="p">()</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="set_max_tuning_duration"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.set_max_tuning_duration">[docs]</a><span class="k">def</span> <span class="nf">set_max_tuning_duration</span><span class="p">(</span><span class="n">duration</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="w"> </span><span class="sa">r</span><span class="sd">"""Set max time in milliseconds to spend tuning a given solution.</span>
<span class="sd"> If both max tuning duration and iterations are set, the smaller of the two</span>
<span class="sd"> will be honored. At minimum 1 tuning iteration will always be run.</span>
<span class="sd"> """</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_tunableop_set_max_tuning_duration</span><span class="p">(</span><span class="n">duration</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="get_max_tuning_duration"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.get_max_tuning_duration">[docs]</a><span class="k">def</span> <span class="nf">get_max_tuning_duration</span><span class="p">()</span> <span class="o">-></span> <span class="nb">int</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Get max time to spend tuning a given solution."""</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_tunableop_get_max_tuning_duration</span><span class="p">()</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="set_max_tuning_iterations"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.set_max_tuning_iterations">[docs]</a><span class="k">def</span> <span class="nf">set_max_tuning_iterations</span><span class="p">(</span><span class="n">iterations</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="w"> </span><span class="sa">r</span><span class="sd">"""Set max number of iterations to spend tuning a given solution.</span>
<span class="sd"> If both max tuning duration and iterations are set, the smaller of the two</span>
<span class="sd"> will be honored. At minimum 1 tuning iteration will always be run.</span>
<span class="sd"> """</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_tunableop_set_max_tuning_iterations</span><span class="p">(</span><span class="n">iterations</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="get_max_tuning_iterations"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.get_max_tuning_iterations">[docs]</a><span class="k">def</span> <span class="nf">get_max_tuning_iterations</span><span class="p">()</span> <span class="o">-></span> <span class="nb">int</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Get max iterations to spend tuning a given solution."""</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_tunableop_get_max_tuning_iterations</span><span class="p">()</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="set_filename"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.set_filename">[docs]</a><span class="k">def</span> <span class="nf">set_filename</span><span class="p">(</span><span class="n">filename</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">insert_device_ordinal</span><span class="p">:</span> <span class="nb">bool</span> <span class="o">=</span> <span class="kc">False</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Set the filename to use for input/output of tuning results.</span>
<span class="sd"> If :attr:`insert_device_ordinal` is ``True`` then the current device ordinal</span>
<span class="sd"> will be added to the given filename automatically. This can be used in a</span>
<span class="sd"> 1-process-per-gpu cenario to ensure all processes write to a separate file.</span>
<span class="sd"> """</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_tunableop_set_filename</span><span class="p">(</span><span class="n">filename</span><span class="p">,</span> <span class="n">insert_device_ordinal</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="get_filename"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.get_filename">[docs]</a><span class="k">def</span> <span class="nf">get_filename</span><span class="p">()</span> <span class="o">-></span> <span class="nb">str</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Get the results filename."""</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_tunableop_get_filename</span><span class="p">()</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="get_results"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.get_results">[docs]</a><span class="k">def</span> <span class="nf">get_results</span><span class="p">()</span> <span class="o">-></span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">str</span><span class="p">,</span> <span class="nb">str</span><span class="p">,</span> <span class="nb">float</span><span class="p">]:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Return all TunableOp results."""</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_tunableop_get_results</span><span class="p">()</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="get_validators"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.get_validators">[docs]</a><span class="k">def</span> <span class="nf">get_validators</span><span class="p">()</span> <span class="o">-></span> <span class="n">Tuple</span><span class="p">[</span><span class="nb">str</span><span class="p">,</span> <span class="nb">str</span><span class="p">]:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Return the TunableOp validators."""</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_tunableop_get_validators</span><span class="p">()</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="write_file_on_exit"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.write_file_on_exit">[docs]</a><span class="k">def</span> <span class="nf">write_file_on_exit</span><span class="p">(</span><span class="n">val</span><span class="p">:</span> <span class="nb">bool</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""During Tuning Context destruction, write file to disk.</span>
<span class="sd"> This is useful as a final flush of your results to disk if your application</span>
<span class="sd"> terminates as result of normal operation or an error. Manual flushing of</span>
<span class="sd"> your results can be achieved by manually calling ``write_file()``."""</span>
<span class="n">torch</span><span class="o">.</span><span class="n">_C</span><span class="o">.</span><span class="n">_cuda_tunableop_write_file_on_exit</span><span class="p">(</span><span class="n">val</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="write_file"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.write_file">[docs]</a><span class="k">def</span> <span class="nf">write_file</span><span class="p">(</span><span class="n">filename</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</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">bool</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Write results to a CSV file.</span>
<span class="sd"> If :attr:`filename` is not given, ``get_filename()`` is called.</span>
<span class="sd"> """</span>
<span class="k">if</span> <span class="n">filename</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">filename</span> <span class="o">=</span> <span class="n">get_filename</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_tunableop_write_file</span><span class="p">(</span><span class="n">filename</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="read_file"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.read_file">[docs]</a><span class="k">def</span> <span class="nf">read_file</span><span class="p">(</span><span class="n">filename</span><span class="p">:</span> <span class="n">Optional</span><span class="p">[</span><span class="nb">str</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">bool</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Read results from a TunableOp CSV file.</span>
<span class="sd"> If :attr:`filename` is not given, ``get_filename()`` is called.</span>
<span class="sd"> """</span>
<span class="k">if</span> <span class="n">filename</span> <span class="ow">is</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">filename</span> <span class="o">=</span> <span class="n">get_filename</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_tunableop_read_file</span><span class="p">(</span><span class="n">filename</span><span class="p">)</span> <span class="c1"># type: ignore[attr-defined]</span></div>
<div class="viewcode-block" id="tune_gemm_in_file"><a class="viewcode-back" href="../../../cuda.tunable.html#torch.cuda.tunable.tune_gemm_in_file">[docs]</a><span class="k">def</span> <span class="nf">tune_gemm_in_file</span><span class="p">(</span><span class="n">filename</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""tune GEMM in file."""</span>
<span class="k">assert</span> <span class="n">is_enabled</span><span class="p">()</span>
<span class="k">assert</span> <span class="n">tuning_is_enabled</span><span class="p">()</span>
<span class="n">deviceid</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_device</span><span class="p">()</span>
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="n">filename</span><span class="p">)</span> <span class="k">as</span> <span class="n">file</span><span class="p">:</span>
<span class="k">for</span> <span class="n">line</span> <span class="ow">in</span> <span class="n">file</span><span class="p">:</span>
<span class="k">if</span> <span class="n">line</span><span class="o">.</span><span class="n">startswith</span><span class="p">((</span><span class="s2">"Gemm"</span><span class="p">,</span> <span class="s2">"ScaledGemm"</span><span class="p">)):</span>
<span class="n">_process_single_offline_gemm</span><span class="p">(</span><span class="n">line</span><span class="p">,</span> <span class="n">deviceid</span><span class="p">)</span></div>
<span class="k">def</span> <span class="nf">_gather_unique_untuned_gemm_from_files</span><span class="p">(</span><span class="n">filename_pattern</span><span class="p">:</span> <span class="nb">str</span><span class="p">)</span> <span class="o">-></span> <span class="nb">set</span><span class="p">[</span><span class="nb">str</span><span class="p">]:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Process multiple untuned results file and return a set with duplicates removed."""</span>
<span class="n">unique_gemm_entries</span> <span class="o">=</span> <span class="nb">set</span><span class="p">()</span> <span class="c1"># set will avoid duplicates</span>
<span class="k">for</span> <span class="n">file_path</span> <span class="ow">in</span> <span class="n">glob</span><span class="o">.</span><span class="n">glob</span><span class="p">(</span><span class="n">filename_pattern</span><span class="p">):</span>
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="n">file_path</span><span class="p">)</span> <span class="k">as</span> <span class="n">file</span><span class="p">:</span>
<span class="k">for</span> <span class="n">line</span> <span class="ow">in</span> <span class="n">file</span><span class="p">:</span>
<span class="k">if</span> <span class="n">line</span><span class="o">.</span><span class="n">startswith</span><span class="p">((</span><span class="s2">"Gemm"</span><span class="p">,</span> <span class="s2">"ScaledGemm"</span><span class="p">)):</span>
<span class="n">unique_gemm_entries</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="n">line</span><span class="p">)</span>
<span class="k">return</span> <span class="n">unique_gemm_entries</span>
<span class="k">def</span> <span class="nf">_gather_tunableop_results</span><span class="p">()</span> <span class="o">-></span> <span class="kc">None</span><span class="p">:</span>
<span class="w"> </span><span class="sa">r</span><span class="sd">"""Gather results from multiple tunableop results file and create a single file."""</span>
<span class="n">gemm_lines</span> <span class="o">=</span> <span class="nb">set</span><span class="p">()</span>
<span class="n">validator_lines</span> <span class="o">=</span> <span class="p">[]</span>
<span class="c1"># Need to allow for the possibility that results filename was</span>
<span class="c1"># set with the Python API instead of with environment variable.</span>
<span class="c1"># Also possible that results filename was not set at all.</span>
<span class="c1"># There are several test cases to check, but ultimately we</span>
<span class="c1"># need a glob-able expression</span>
<span class="n">results_filename</span> <span class="o">=</span> <span class="n">get_filename</span><span class="p">()</span> <span class="c1"># Note empty string could be returned here</span>
<span class="k">if</span> <span class="p">(</span>
<span class="n">results_filename</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span> <span class="ow">and</span> <span class="n">results_filename</span> <span class="o">!=</span> <span class="s2">""</span>
<span class="p">):</span> <span class="c1"># Case were the Python API was used to set the filename</span>
<span class="n">dot_pos</span> <span class="o">=</span> <span class="n">results_filename</span><span class="o">.</span><span class="n">find</span><span class="p">(</span><span class="s2">"."</span><span class="p">)</span>
<span class="k">if</span> <span class="n">dot_pos</span> <span class="o">!=</span> <span class="o">-</span><span class="mi">1</span> <span class="ow">and</span> <span class="n">dot_pos</span> <span class="o">></span> <span class="mi">0</span><span class="p">:</span>
<span class="c1"># Replace the character just to the left of the dot</span>
<span class="n">filename_pattern</span> <span class="o">=</span> <span class="p">(</span>
<span class="n">results_filename</span><span class="p">[:</span> <span class="n">dot_pos</span> <span class="o">-</span> <span class="mi">1</span><span class="p">]</span> <span class="o">+</span> <span class="s2">"?"</span> <span class="o">+</span> <span class="n">results_filename</span><span class="p">[</span><span class="n">dot_pos</span><span class="p">:]</span>
<span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">filename_pattern</span> <span class="o">=</span> <span class="s2">""</span> <span class="c1"># Needed to make linter happy</span>
<span class="k">else</span><span class="p">:</span> <span class="c1"># Case where the environment variable was used to set the filename.</span>
<span class="n">results_filename_env</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">"PYTORCH_TUNABLEOP_FILENAME"</span><span class="p">)</span>
<span class="k">if</span> <span class="n">results_filename_env</span> <span class="ow">is</span> <span class="kc">None</span> <span class="ow">or</span> <span class="n">results_filename_env</span> <span class="o">==</span> <span class="s2">""</span><span class="p">:</span>
<span class="n">filename_pattern</span> <span class="o">=</span> <span class="s2">"tunableop_results?.csv"</span>
<span class="k">elif</span> <span class="s2">"</span><span class="si">%d</span><span class="s2">"</span> <span class="ow">in</span> <span class="n">results_filename_env</span><span class="p">:</span>
<span class="n">filename_pattern</span> <span class="o">=</span> <span class="n">results_filename_env</span><span class="o">.</span><span class="n">replace</span><span class="p">(</span><span class="s2">"</span><span class="si">%d</span><span class="s2">"</span><span class="p">,</span> <span class="s2">"?"</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">filename_pattern</span> <span class="o">=</span> <span class="n">results_filename_env</span><span class="o">.</span><span class="n">replace</span><span class="p">(</span><span class="s2">"."</span><span class="p">,</span> <span class="s2">"?."</span><span class="p">)</span>
<span class="k">assert</span> <span class="s2">"?"</span> <span class="ow">in</span> <span class="n">filename_pattern</span>
<span class="n">FirstFile</span> <span class="o">=</span> <span class="kc">False</span>
<span class="n">matching_files</span> <span class="o">=</span> <span class="n">glob</span><span class="o">.</span><span class="n">glob</span><span class="p">(</span><span class="n">filename_pattern</span><span class="p">)</span>
<span class="n">num_matching_files</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">matching_files</span><span class="p">)</span>
<span class="k">for</span> <span class="n">file_path</span> <span class="ow">in</span> <span class="n">matching_files</span><span class="p">:</span>
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="n">file_path</span><span class="p">)</span> <span class="k">as</span> <span class="n">file</span><span class="p">:</span>
<span class="k">for</span> <span class="n">line</span> <span class="ow">in</span> <span class="n">file</span><span class="p">:</span>
<span class="k">if</span> <span class="n">line</span><span class="o">.</span><span class="n">startswith</span><span class="p">(</span><span class="s2">"Validator"</span><span class="p">):</span>
<span class="k">if</span> <span class="ow">not</span> <span class="p">(</span><span class="n">FirstFile</span><span class="p">):</span>
<span class="c1"># Only read Validator from first file</span>
<span class="n">validator_lines</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">line</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="n">gemm_lines</span><span class="o">.</span><span class="n">add</span><span class="p">(</span><span class="n">line</span><span class="p">)</span>
<span class="n">FirstFile</span> <span class="o">=</span> <span class="kc">True</span>
<span class="n">output_file</span> <span class="o">=</span> <span class="n">filename_pattern</span><span class="o">.</span><span class="n">replace</span><span class="p">(</span><span class="s2">"?"</span><span class="p">,</span> <span class="s2">"_full0"</span><span class="p">)</span>
<span class="k">with</span> <span class="nb">open</span><span class="p">(</span><span class="n">output_file</span><span class="p">,</span> <span class="s2">"w"</span><span class="p">)</span> <span class="k">as</span> <span class="n">out_file</span><span class="p">:</span>
<span class="k">for</span> <span class="n">line</span> <span class="ow">in</span> <span class="n">validator_lines</span><span class="p">:</span>
<span class="n">out_file</span><span class="o">.</span><span class="n">write</span><span class="p">(</span><span class="n">line</span><span class="p">)</span>
<span class="k">for</span> <span class="n">line</span> <span class="ow">in</span> <span class="n">gemm_lines</span><span class="p">:</span>
<span class="n">out_file</span><span class="o">.</span><span class="n">write</span><span class="p">(</span><span class="n">line</span><span class="p">)</span>
<span class="c1"># Create num_matching_copies of the results file</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="n">num_matching_files</span><span class="p">):</span>
<span class="n">duplicate_file</span> <span class="o">=</span> <span class="n">output_file</span><span class="o">.</span><span class="n">replace</span><span class="p">(</span><span class="s2">"0"</span><span class="p">,</span> <span class="nb">str</span><span class="p">(</span><span class="n">i</span><span class="p">))</span>
<span class="n">shutil</span><span class="o">.</span><span class="n">copy</span><span class="p">(</span><span class="n">output_file</span><span class="p">,</span> <span class="n">duplicate_file</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">_process_single_offline_gemm</span><span class="p">(</span><span class="n">untuned_gemm_line</span><span class="p">:</span> <span class="nb">str</span><span class="p">,</span> <span class="n">gpu_id</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="w"> </span><span class="sa">r</span><span class="sd">"""Process a single untuned GEMM."""</span>
<span class="n">deviceid</span> <span class="o">=</span> <span class="s2">"cuda:"</span> <span class="o">+</span> <span class="nb">str</span><span class="p">(</span><span class="n">gpu_id</span><span class="p">)</span>
<span class="n">dtype_dict</span> <span class="o">=</span> <span class="p">{</span>
<span class="s2">"float"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">float32</span><span class="p">,</span>
<span class="s2">"double"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">float64</span><span class="p">,</span>
<span class="s2">"BFloat16"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">bfloat16</span><span class="p">,</span>
<span class="s2">"Half"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">half</span><span class="p">,</span>
<span class="s2">"c10::complex<double>"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">complex128</span><span class="p">,</span>
<span class="s2">"c10::complex<float>"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">complex64</span><span class="p">,</span>
<span class="s2">"Float8_e4m3fn"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">float8_e4m3fn</span><span class="p">,</span>
<span class="s2">"Float8_e5m2"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">float8_e5m2</span><span class="p">,</span>
<span class="s2">"Float8_e4m3fnuz"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">float8_e4m3fnuz</span><span class="p">,</span>
<span class="s2">"Float8_e5m2fnuz"</span><span class="p">:</span> <span class="n">torch</span><span class="o">.</span><span class="n">float8_e5m2fnuz</span><span class="p">,</span>
<span class="p">}</span>
<span class="n">untuned_gemm</span> <span class="o">=</span> <span class="n">untuned_gemm_line</span><span class="o">.</span><span class="n">strip</span><span class="p">()</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">underscore_count</span> <span class="o">=</span> <span class="n">untuned_gemm</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span><span class="o">.</span><span class="n">count</span><span class="p">(</span><span class="s2">"_"</span><span class="p">)</span>
<span class="c1"># Initialize dtype to make linter happy</span>
<span class="n">dtype</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">dtypeA</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">dtypeB</span> <span class="o">=</span> <span class="kc">None</span>
<span class="n">dtypeC</span> <span class="o">=</span> <span class="kc">None</span>
<span class="k">if</span> <span class="n">underscore_count</span> <span class="o">==</span> <span class="mi">2</span><span class="p">:</span>
<span class="p">[</span><span class="n">op_sig</span><span class="p">,</span> <span class="n">data_type</span><span class="p">,</span> <span class="n">layout</span><span class="p">]</span> <span class="o">=</span> <span class="n">untuned_gemm</span><span class="p">[</span><span class="mi">0</span><span class="p">]</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">transA</span> <span class="o">=</span> <span class="n">layout</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">==</span> <span class="s2">"T"</span>
<span class="n">transB</span> <span class="o">=</span> <span class="n">layout</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">==</span> <span class="s2">"T"</span>
<span class="n">dtype</span> <span class="o">=</span> <span class="n">dtype_dict</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">data_type</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span> <span class="c1"># ScaledGEMM</span>
<span class="n">untuned_gemm_temp</span> <span class="o">=</span> <span class="n">untuned_gemm</span><span class="p">[</span><span class="mi">0</span><span class="p">]</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">op_sig</span> <span class="o">=</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="n">data_typeA</span> <span class="o">=</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">1</span><span class="p">]</span> <span class="o">+</span> <span class="s2">"_"</span> <span class="o">+</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">2</span><span class="p">]</span>
<span class="n">data_typeB</span> <span class="o">=</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">3</span><span class="p">]</span> <span class="o">+</span> <span class="s2">"_"</span> <span class="o">+</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">4</span><span class="p">]</span>
<span class="n">data_typeC</span> <span class="o">=</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">5</span><span class="p">]</span> <span class="o">+</span> <span class="s2">"_"</span> <span class="o">+</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">6</span><span class="p">]</span>
<span class="n">transA</span> <span class="o">=</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">7</span><span class="p">][</span><span class="mi">0</span><span class="p">]</span> <span class="o">==</span> <span class="s2">"T"</span>
<span class="n">transB</span> <span class="o">=</span> <span class="n">untuned_gemm_temp</span><span class="p">[</span><span class="mi">7</span><span class="p">][</span><span class="mi">1</span><span class="p">]</span> <span class="o">==</span> <span class="s2">"T"</span>
<span class="n">dtypeA</span> <span class="o">=</span> <span class="n">dtype_dict</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">data_typeA</span><span class="p">)</span>
<span class="n">dtypeB</span> <span class="o">=</span> <span class="n">dtype_dict</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">data_typeB</span><span class="p">)</span>
<span class="n">dtypeC</span> <span class="o">=</span> <span class="n">dtype_dict</span><span class="o">.</span><span class="n">get</span><span class="p">(</span><span class="n">data_typeC</span><span class="p">)</span>
<span class="p">[</span><span class="n">n</span><span class="p">,</span> <span class="n">m</span><span class="p">,</span> <span class="n">k</span><span class="p">]</span> <span class="o">=</span> <span class="p">[</span><span class="nb">int</span><span class="p">(</span><span class="n">g</span><span class="p">)</span> <span class="k">for</span> <span class="n">g</span> <span class="ow">in</span> <span class="n">untuned_gemm</span><span class="p">[</span><span class="mi">1</span><span class="p">]</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="mi">4</span><span class="p">]]</span>
<span class="k">if</span> <span class="n">op_sig</span> <span class="o">==</span> <span class="s2">"GemmTunableOp"</span><span class="p">:</span>
<span class="n">matA</span> <span class="o">=</span> <span class="p">(</span>
<span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="n">k</span><span class="p">,</span> <span class="n">m</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">deviceid</span><span class="p">)</span><span class="o">.</span><span class="n">t</span><span class="p">()</span>
<span class="k">if</span> <span class="n">transB</span>
<span class="k">else</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="n">m</span><span class="p">,</span> <span class="n">k</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">deviceid</span><span class="p">)</span>
<span class="p">)</span>
<span class="n">matB</span> <span class="o">=</span> <span class="p">(</span>
<span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="n">n</span><span class="p">,</span> <span class="n">k</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">deviceid</span><span class="p">)</span><span class="o">.</span><span class="n">t</span><span class="p">()</span>
<span class="k">if</span> <span class="n">transA</span>
<span class="k">else</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="n">k</span><span class="p">,</span> <span class="n">n</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">deviceid</span><span class="p">)</span>
<span class="p">)</span>
<span class="n">torch</span><span class="o">.</span><span class="n">mm</span><span class="p">(</span><span class="n">matA</span><span class="p">,</span> <span class="n">matB</span><span class="p">)</span>
<span class="k">elif</span> <span class="n">op_sig</span> <span class="o">==</span> <span class="s2">"GemmStridedBatchedTunableOp"</span><span class="p">:</span>
<span class="p">[</span><span class="n">b</span><span class="p">]</span> <span class="o">=</span> <span class="p">[</span><span class="nb">int</span><span class="p">(</span><span class="n">g</span><span class="p">)</span> <span class="k">for</span> <span class="n">g</span> <span class="ow">in</span> <span class="n">untuned_gemm</span><span class="p">[</span><span class="mi">1</span><span class="p">]</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">5</span><span class="p">:</span><span class="mi">6</span><span class="p">]]</span>
<span class="n">matA</span> <span class="o">=</span> <span class="p">(</span>
<span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="n">b</span><span class="p">,</span> <span class="n">k</span><span class="p">,</span> <span class="n">m</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">deviceid</span><span class="p">)</span>
<span class="k">if</span> <span class="n">transB</span>
<span class="k">else</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="n">b</span><span class="p">,</span> <span class="n">m</span><span class="p">,</span> <span class="n">k</span><span class="p">,</span> <span class="n">dtype</span><span class="o">=</span><span class="n">dtype</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">deviceid</span><span class="p">)</span>
<span class="p">)</span>