-
Notifications
You must be signed in to change notification settings - Fork 362
/
Copy pathlowering.html
1031 lines (1030 loc) · 35.9 KB
/
lowering.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>
<html>
<head>
<meta charset="utf-8"/>
<meta content="width=device-width, initial-scale=1.0" name="viewport"/>
<meta content="width=device-width,initial-scale=1" name="viewport"/>
<meta content="ie=edge" http-equiv="x-ua-compatible"/>
<meta content="Copy to clipboard" name="lang:clipboard.copy"/>
<meta content="Copied to clipboard" name="lang:clipboard.copied"/>
<meta content="en" name="lang:search.language"/>
<meta content="True" name="lang:search.pipeline.stopwords"/>
<meta content="True" name="lang:search.pipeline.trimmer"/>
<meta content="No matching documents" name="lang:search.result.none"/>
<meta content="1 matching document" name="lang:search.result.one"/>
<meta content="# matching documents" name="lang:search.result.other"/>
<meta content="[\s\-]+" name="lang:search.tokenizer"/>
<link crossorigin="" href="https://fonts.gstatic.com/" rel="preconnect"/>
<link href="https://fonts.googleapis.com/css?family=Roboto+Mono:400,500,700|Roboto:300,400,400i,700&display=fallback" rel="stylesheet"/>
<style>
body,
input {
font-family: "Roboto", "Helvetica Neue", Helvetica, Arial, sans-serif
}
code,
kbd,
pre {
font-family: "Roboto Mono", "Courier New", Courier, monospace
}
</style>
<link href="../_static/stylesheets/application.css" rel="stylesheet"/>
<link href="../_static/stylesheets/application-palette.css" rel="stylesheet"/>
<link href="../_static/stylesheets/application-fixes.css" rel="stylesheet"/>
<link href="../_static/fonts/material-icons.css" rel="stylesheet"/>
<meta content="84bd00" name="theme-color"/>
<script src="../_static/javascripts/modernizr.js">
</script>
<title>
Lowering Phase — TRTorch 0.0.3 documentation
</title>
<link href="../_static/material.css" rel="stylesheet" type="text/css"/>
<link href="../_static/pygments.css" rel="stylesheet" type="text/css"/>
<link href="../_static/collapsible-lists/css/tree_view.css" rel="stylesheet" type="text/css"/>
<script data-url_root="../" id="documentation_options" src="../_static/documentation_options.js">
</script>
<script src="../_static/jquery.js">
</script>
<script src="../_static/underscore.js">
</script>
<script src="../_static/doctools.js">
</script>
<script src="../_static/language_data.js">
</script>
<script src="../_static/collapsible-lists/js/CollapsibleLists.compressed.js">
</script>
<script src="../_static/collapsible-lists/js/apply-collapsible-lists.js">
</script>
<script crossorigin="anonymous" integrity="sha256-Ae2Vz/4ePdIu6ZyI/5ZGsYnb+m0JlOmKPjt6XZ9JJkA=" src="https://cdnjs.cloudflare.com/ajax/libs/require.js/2.3.4/require.min.js">
</script>
<link href="../genindex.html" rel="index" title="Index"/>
<link href="../search.html" rel="search" title="Search"/>
<link href="conversion.html" rel="next" title="Conversion Phase"/>
<link href="system_overview.html" rel="prev" title="System Overview"/>
</head>
<body data-md-color-accent="light-green" data-md-color-primary="light-green" dir="ltr">
<svg class="md-svg">
<defs data-children-count="0">
<svg height="448" id="__github" viewbox="0 0 416 448" width="416" xmlns="http://www.w3.org/2000/svg">
<path d="M160 304q0 10-3.125 20.5t-10.75 19T128 352t-18.125-8.5-10.75-19T96 304t3.125-20.5 10.75-19T128 256t18.125 8.5 10.75 19T160 304zm160 0q0 10-3.125 20.5t-10.75 19T288 352t-18.125-8.5-10.75-19T256 304t3.125-20.5 10.75-19T288 256t18.125 8.5 10.75 19T320 304zm40 0q0-30-17.25-51T296 232q-10.25 0-48.75 5.25Q229.5 240 208 240t-39.25-2.75Q130.75 232 120 232q-29.5 0-46.75 21T56 304q0 22 8 38.375t20.25 25.75 30.5 15 35 7.375 37.25 1.75h42q20.5 0 37.25-1.75t35-7.375 30.5-15 20.25-25.75T360 304zm56-44q0 51.75-15.25 82.75-9.5 19.25-26.375 33.25t-35.25 21.5-42.5 11.875-42.875 5.5T212 416q-19.5 0-35.5-.75t-36.875-3.125-38.125-7.5-34.25-12.875T37 371.5t-21.5-28.75Q0 312 0 260q0-59.25 34-99-6.75-20.5-6.75-42.5 0-29 12.75-54.5 27 0 47.5 9.875t47.25 30.875Q171.5 96 212 96q37 0 70 8 26.25-20.5 46.75-30.25T376 64q12.75 25.5 12.75 54.5 0 21.75-6.75 42 34 40 34 99.5z" fill="currentColor">
</path>
</svg>
</defs>
</svg>
<input class="md-toggle" data-md-toggle="drawer" id="__drawer" type="checkbox"/>
<input class="md-toggle" data-md-toggle="search" id="__search" type="checkbox"/>
<label class="md-overlay" data-md-component="overlay" for="__drawer">
</label>
<a class="md-skip" href="#contributors/lowering" tabindex="1">
Skip to content
</a>
<header class="md-header" data-md-component="header">
<nav class="md-header-nav md-grid">
<div class="md-flex navheader">
<div class="md-flex__cell md-flex__cell--shrink">
<a class="md-header-nav__button md-logo" href="../index.html" title="TRTorch 0.0.3 documentation">
<i class="md-icon">
</i>
</a>
</div>
<div class="md-flex__cell md-flex__cell--shrink">
<label class="md-icon md-icon--menu md-header-nav__button" for="__drawer">
</label>
</div>
<div class="md-flex__cell md-flex__cell--stretch">
<div class="md-flex__ellipsis md-header-nav__title" data-md-component="title">
<span class="md-header-nav__topic">
TRTorch
</span>
<span class="md-header-nav__topic">
Lowering Phase
</span>
</div>
</div>
<div class="md-flex__cell md-flex__cell--shrink">
<label class="md-icon md-icon--search md-header-nav__button" for="__search">
</label>
<div class="md-search" data-md-component="search" role="dialog">
<label class="md-search__overlay" for="__search">
</label>
<div class="md-search__inner" role="search">
<form action="../search.html" class="md-search__form" method="GET" name="search">
<input autocapitalize="off" autocomplete="off" class="md-search__input" data-md-component="query" data-md-state="active" name="q" placeholder="Search" spellcheck="false" type="text"/>
<label class="md-icon md-search__icon" for="__search">
</label>
<button class="md-icon md-search__icon" data-md-component="reset" tabindex="-1" type="reset">
</button>
</form>
<div class="md-search__output">
<div class="md-search__scrollwrap" data-md-scrollfix="">
<div class="md-search-result" data-md-component="result">
<div class="md-search-result__meta">
Type to start searching
</div>
<ol class="md-search-result__list">
</ol>
</div>
</div>
</div>
</div>
</div>
</div>
<div class="md-flex__cell md-flex__cell--shrink">
<div class="md-header-nav__source">
<a class="md-source" data-md-source="github" href="https://github.com/nvidia/TRTorch/" title="Go to repository">
<div class="md-source__icon">
<svg height="28" viewbox="0 0 24 24" width="28" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
<use height="24" width="24" xlink:href="#__github">
</use>
</svg>
</div>
<div class="md-source__repository">
TRTorch
</div>
</a>
</div>
</div>
<div class="md-flex__cell md-flex__cell--shrink dropdown">
<button class="dropdownbutton">
Versions
</button>
<div class="dropdown-content md-hero">
<a href="https://nvidia.github.io/TRTorch/" title="master">
master
</a>
<a href="https://nvidia.github.io/TRTorch/v0.0.3/" title="v0.0.3">
v0.0.3
</a>
<a href="https://nvidia.github.io/TRTorch/v0.0.2/" title="v0.0.2">
v0.0.2
</a>
<a href="https://nvidia.github.io/TRTorch/v0.0.1/" title="v0.0.1">
v0.0.1
</a>
</div>
</div>
</div>
</nav>
</header>
<div class="md-container">
<nav class="md-tabs" data-md-component="tabs">
<div class="md-tabs__inner md-grid">
<ul class="md-tabs__list">
<li class="md-tabs__item">
<a class="md-tabs__link" href="../index.html">
TRTorch 0.0.3 documentation
</a>
</li>
<li class="md-tabs__item">
<a class="md-tabs__link" href="system_overview.html">
System Overview
</a>
</li>
</ul>
</div>
</nav>
<main class="md-main">
<div class="md-main__inner md-grid" data-md-component="container">
<div class="md-sidebar md-sidebar--primary" data-md-component="navigation">
<div class="md-sidebar__scrollwrap">
<div class="md-sidebar__inner">
<nav class="md-nav md-nav--primary" data-md-level="0">
<label class="md-nav__title md-nav__title--site" for="__drawer">
<a class="md-nav__button md-logo" href="../index.html" title="TRTorch 0.0.3 documentation">
<i class="md-icon">
</i>
</a>
<a href="../index.html" title="TRTorch 0.0.3 documentation">
TRTorch
</a>
</label>
<div class="md-nav__source">
<a class="md-source" data-md-source="github" href="https://github.com/nvidia/TRTorch/" title="Go to repository">
<div class="md-source__icon">
<svg height="28" viewbox="0 0 24 24" width="28" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
<use height="24" width="24" xlink:href="#__github">
</use>
</svg>
</div>
<div class="md-source__repository">
TRTorch
</div>
</a>
</div>
<ul class="md-nav__list">
<li class="md-nav__item">
<span class="md-nav__link caption">
<span class="caption-text">
Getting Started
</span>
</span>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="../tutorials/installation.html">
Installation
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="../tutorials/getting_started.html">
Getting Started
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="../tutorials/ptq.html">
Post Training Quantization (PTQ)
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="../tutorials/trtorchc.html">
trtorchc
</a>
</li>
<li class="md-nav__item">
<span class="md-nav__link caption">
<span class="caption-text">
Notebooks
</span>
</span>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="../_notebooks/lenet-getting-started.html">
TRTorch Getting Started - LeNet
</a>
</li>
<li class="md-nav__item">
<span class="md-nav__link caption">
<span class="caption-text">
Python API Documenation
</span>
</span>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="../py_api/trtorch.html">
trtorch
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="../py_api/logging.html">
trtorch.logging
</a>
</li>
<li class="md-nav__item">
<span class="md-nav__link caption">
<span class="caption-text">
C++ API Documenation
</span>
</span>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="../_cpp_api/trtorch_cpp.html">
TRTorch C++ API
</a>
</li>
<li class="md-nav__item">
<span class="md-nav__link caption">
<span class="caption-text">
Contributor Documentation
</span>
</span>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="system_overview.html">
System Overview
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="writing_converters.html">
Writing Converters
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="useful_links.html">
Useful Links for TRTorch Development
</a>
</li>
</ul>
</nav>
</div>
</div>
</div>
<div class="md-sidebar md-sidebar--secondary" data-md-component="toc">
<div class="md-sidebar__scrollwrap">
<div class="md-sidebar__inner">
<nav class="md-nav md-nav--secondary">
<label class="md-nav__title" for="__toc">
Contents
</label>
<ul class="md-nav__list" data-md-scrollfix="">
<li class="md-nav__item">
<a class="md-nav__link" href="#contributors-lowering--page-root">
Lowering Phase
</a>
<nav class="md-nav">
<ul class="md-nav__list">
<li class="md-nav__item">
<a class="md-nav__link" href="#passes-used">
Passes Used
</a>
<nav class="md-nav">
<ul class="md-nav__list">
<li class="md-nav__item">
<a class="md-nav__link" href="#eliminatecommonsubexpression">
EliminateCommonSubexpression
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#eliminate-dead-code">
Eliminate Dead Code
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#eliminate-exeception-or-pass-pattern">
Eliminate Exeception Or Pass Pattern
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#eliminate-redundant-gaurds">
Eliminate Redundant Gaurds
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#freeze-module">
Freeze Module
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#fuse-addmm-branches">
Fuse AddMM Branches
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#fuse-linear">
Fuse Linear
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#fuse-flatten-linear">
Fuse Flatten Linear
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#lower-graph">
Lower Graph
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#lower-tuples">
Lower Tuples
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#peephole-optimze">
Peephole Optimze
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#remove-contiguous">
Remove Contiguous
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#remove-dropout">
Remove Dropout
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#remove-to">
Remove To
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#unpack-addmm">
Unpack AddMM
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#unpack-logsoftmax">
Unpack LogSoftmax
</a>
</li>
<li class="md-nav__item">
<a class="md-nav__link" href="#unroll-loops">
Unroll Loops
</a>
</li>
</ul>
</nav>
</li>
</ul>
</nav>
</li>
<li class="md-nav__item">
<a class="md-nav__extra_link" href="../_sources/contributors/lowering.rst.txt">
Show Source
</a>
</li>
<li class="md-nav__item" id="searchbox">
</li>
</ul>
</nav>
</div>
</div>
</div>
<div class="md-content">
<article class="md-content__inner md-typeset" role="main">
<span id="lowering">
</span>
<h1 id="contributors-lowering--page-root">
Lowering Phase
<a class="headerlink" href="#contributors-lowering--page-root" title="Permalink to this headline">
¶
</a>
</h1>
<p>
The lowering phase is made up out of passes which are operations which map a graph from a high level representation
to a lower level one. Each pass does something specific for instance inlining method calls. The idea is to
significantly reduce what the conversion phase needs to be able to handle when actually mapping to TensorRT.
We aim for closer to 1->1 op conversion vs looking for applicable subgraphs, limiting the number of converters and
reduce the scope of each converter.
</p>
<p>
You can see the effects of each pass by setting the log level to
<code class="docutils literal notranslate">
<span class="pre">
Level::kGraph
</span>
</code>
</p>
<h2 id="passes-used">
Passes Used
<a class="headerlink" href="#passes-used" title="Permalink to this headline">
¶
</a>
</h2>
<h3 id="eliminatecommonsubexpression">
EliminateCommonSubexpression
<a class="headerlink" href="#eliminatecommonsubexpression" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/common_subexpression_elimination.h">
torch/csrc/jit/passes/common_subexpression_elimination.h
</a>
</p>
</div>
</blockquote>
<p>
Removes common subexpressions in the graph
</p>
<h3 id="eliminate-dead-code">
Eliminate Dead Code
<a class="headerlink" href="#eliminate-dead-code" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/dead_code_elimination.h">
torch/csrc/jit/passes/dead_code_elimination.h
</a>
</p>
</div>
</blockquote>
<p>
Dead code elimination will check if a node has side effects and not delete it if it does.
</p>
<h3 id="eliminate-exeception-or-pass-pattern">
Eliminate Exeception Or Pass Pattern
<a class="headerlink" href="#eliminate-exeception-or-pass-pattern" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/nvidia/trtorch/blob/master/core/lowering/passes/exception_elimination.cpp">
trtorch/core/lowering/passes/exception_elimination.cpp
</a>
</p>
</div>
</blockquote>
<p>
A common pattern in scripted modules are dimension gaurds which will throw execptions if
the input dimension is not what was expected.
</p>
<div class="highlight-none notranslate">
<div class="highlight">
<pre><span></span>%1013 : bool = aten::ne(%1012, %24) # ~/.local/lib/python3.6/site-packages/torch/nn/modules/batchnorm.py:248:11
= prim::If(%1013) # ~/.local/lib/python3.6/site-packages/torch/nn/modules/batchnorm.py:248:8
block0():
= prim::RaiseException(%23) # ~/.local/lib/python3.6/site-packages/torch/nn/modules/batchnorm.py:249:12
-> ()
block1():
-> ()
</pre>
</div>
</div>
<p>
Since we are resolving all of this at compile time and there are no execptions in the TensorRT graph, we just remove it.
</p>
<h3 id="eliminate-redundant-gaurds">
Eliminate Redundant Gaurds
<a class="headerlink" href="#eliminate-redundant-gaurds" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/guard_elimination.h">
torch/csrc/jit/passes/guard_elimination.h
</a>
</p>
</div>
</blockquote>
<p>
Eliminate redundant guards for ops whose outputs are fully determined by their inputs i.e. if inputs to such ops are
guarded we are allowed to remove a guard on ops’ outputs
</p>
<h3 id="freeze-module">
Freeze Module
<a class="headerlink" href="#freeze-module" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/freeze_module.h">
torch/csrc/jit/passes/freeze_module.h
</a>
</p>
</div>
</blockquote>
<p>
Freeze attributes and inline constants and modules. Propogates constants in the graph.
</p>
<h3 id="fuse-addmm-branches">
Fuse AddMM Branches
<a class="headerlink" href="#fuse-addmm-branches" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/nvidia/trtorch/blob/master/core/lowering/passes/fuse_addmm_branches.cpp">
trtorch/core/lowering/passes/fuse_addmm_branches.cpp
</a>
</p>
</div>
</blockquote>
<p>
A common pattern in scripted modules is tensors of different dimensions use different constructions for implementing linear layers. We fuse these
different varients into a single one that will get caught by the Unpack AddMM pass.
</p>
<div class="highlight-none notranslate">
<div class="highlight">
<pre><span></span>%ret : Tensor = prim::If(%622)
block0():
%ret.1 : Tensor = aten::addmm(%self.fc.bias, %x9.1, %3677, %3, %3)
-> (%ret.1)
block1():
%output.1 : Tensor = aten::matmul(%x9.1, %3677)
%output0.1 : Tensor = aten::add_(%output.1, %self.fc.bias, %3)
-> (%output0.1)
</pre>
</div>
</div>
<p>
We fuse this set of blocks into a graph like this:
</p>
<div class="highlight-none notranslate">
<div class="highlight">
<pre><span></span>%ret : Tensor = aten::addmm(%self.fc.bias, %x9.1, %3677, %3, %3)
</pre>
</div>
</div>
<h3 id="fuse-linear">
Fuse Linear
<a class="headerlink" href="#fuse-linear" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/fuse_linear.h">
torch/csrc/jit/passes/fuse_linear.h
</a>
</p>
</div>
</blockquote>
<p>
Match the
<code class="docutils literal notranslate">
<span class="pre">
aten::linear
</span>
</code>
pattern and fuse it into a single
<code class="docutils literal notranslate">
<span class="pre">
aten::linear
</span>
</code>
This pass fuse the addmm or matmul + add generated by JIT back to linear
</p>
<h3 id="fuse-flatten-linear">
Fuse Flatten Linear
<a class="headerlink" href="#fuse-flatten-linear" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/nvidia/trtorch/blob/master/core/lowering/passes/fuse_flatten_linear.cpp">
trtorch/core/lowering/passes/fuse_flatten_linear.cpp
</a>
</p>
</div>
</blockquote>
<p>
TensorRT implicity flattens input layers into fully connected layers when they are higher than 1D. So when there is a
<code class="docutils literal notranslate">
<span class="pre">
aten::flatten
</span>
</code>
->
<code class="docutils literal notranslate">
<span class="pre">
aten::linear
</span>
</code>
pattern we remove the
<code class="docutils literal notranslate">
<span class="pre">
aten::flatten
</span>
</code>
.
</p>
<h3 id="lower-graph">
Lower Graph
<a class="headerlink" href="#lower-graph" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/lower_graph.h">
torch/csrc/jit/passes/lower_graph.h
</a>
</p>
</div>
</blockquote>
<p>
Given a graph with of a method which first argument is %self, lower it to a graph where
all attributes accesses are replaced with explicit inputs of the graph
(rather than results of prim::GetAttr executed on %self). Returns a tuple
(graph, parameters) where the last module.parameters.size() inputs to the
graph are the trainable parameters used in this method. The remaining inputs
are the true inputs to the function.
</p>
<h3 id="lower-tuples">
Lower Tuples
<a class="headerlink" href="#lower-tuples" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/lower_tuples.h">
torch/csrc/jit/passes/lower_tuples.h
</a>
</p>
</div>
</blockquote>
<ul class="simple">
<li>
<p>
<code class="docutils literal notranslate">
<span class="pre">
LowerSimpleTuples
</span>
</code>
:
</p>
</li>
</ul>
<p>
Removes tuples where TupleConstruct and TupleUnpack are matched but leaves tuples in place across if statements, loops, and as inputs/outputs
</p>
<ul class="simple">
<li>
<p>
<code class="docutils literal notranslate">
<span class="pre">
LowerAllTuples
</span>
</code>
:
</p>
</li>
</ul>
<p>
Removes _all_ tuples and raises an error if some cannot be removed, this is used by ONNX to ensure there are not tuples before conversion, but will not work on graphs whose inputs contain tuples.
</p>
<h3 id="peephole-optimze">
Peephole Optimze
<a class="headerlink" href="#peephole-optimze" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/ppeephole_optimze.h">
torch/csrc/jit/passes/peephole_optimze.h
</a>
</p>
</div>
</blockquote>
<p>
The intent for this optimization pass is to catch all of the small, easy to catch peephole optimizations you might be interested in doing.
</p>
<dl class="simple">
<dt>
Right now, it does:
</dt>
<dd>
<ul class="simple">
<li>
<p>
Eliminate no-op ‘expand’ nodes
</p>
</li>
<li>
<p>
Simply x.t().t() to x
</p>
</li>
</ul>
</dd>
</dl>
<h3 id="remove-contiguous">
Remove Contiguous
<a class="headerlink" href="#remove-contiguous" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/nvidia/trtorch/blob/master/core/lowering/passes/remove_contiguous.cpp">
trtorch/core/lowering/passes/remove_contiguous.cpp
</a>
</p>
</div>
</blockquote>
<p>
Removes contiguous operators since we are doing TensorRT memory is already contiguous.
</p>
<h3 id="remove-dropout">
Remove Dropout
<a class="headerlink" href="#remove-dropout" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/nvidia/trtorch/blob/master/core/lowering/passes/remove_dropout.cpp">
trtorch/core/lowering/passes/remove_dropout.cpp
</a>
</p>
</div>
</blockquote>
<p>
Removes dropout operators since we are doing inference.
</p>
<h3 id="remove-to">
Remove To
<a class="headerlink" href="#remove-to" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/nvidia/trtorch/blob/master/core/lowering/passes/remove_to.cpp">
trtorch/core/lowering/passes/remove_to.cpp
</a>
</p>
</div>
</blockquote>
<p>
Removes
<code class="docutils literal notranslate">
<span class="pre">
aten::to
</span>
</code>
operators that do casting, since TensorRT mangages it itself. It is important that this is one of the last passes run so that
other passes have a change to move required cast operators out of the main namespace.
</p>
<h3 id="unpack-addmm">
Unpack AddMM
<a class="headerlink" href="#unpack-addmm" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/nvidia/trtorch/blob/master/core/lowering/passes/unpack_addmm.cpp">
trtorch/core/lowering/passes/unpack_addmm.cpp
</a>
</p>
</div>
</blockquote>
<p>
Unpacks
<code class="docutils literal notranslate">
<span class="pre">
aten::addmm
</span>
</code>
into
<code class="docutils literal notranslate">
<span class="pre">
aten::matmul
</span>
</code>
and
<code class="docutils literal notranslate">
<span class="pre">
aten::add_
</span>
</code>
(with an additional
<code class="docutils literal notranslate">
<span class="pre">
trt::const
</span>
</code>
op to freeze the bias in the TensorRT graph). This lets us reuse the
<code class="docutils literal notranslate">
<span class="pre">
aten::matmul
</span>
</code>
and
<code class="docutils literal notranslate">
<span class="pre">
aten::add_
</span>
</code>
converters instead of needing a dedicated converter.
</p>
<h3 id="unpack-logsoftmax">
Unpack LogSoftmax
<a class="headerlink" href="#unpack-logsoftmax" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/nvidia/trtorch/blob/master/core/lowering/passes/unpack_log_softmax.cpp">
trtorch/core/lowering/passes/unpack_log_softmax.cpp
</a>
</p>
</div>
</blockquote>
<p>
Unpacks
<code class="docutils literal notranslate">
<span class="pre">
aten::logsoftmax
</span>
</code>
into
<code class="docutils literal notranslate">
<span class="pre">
aten::softmax
</span>
</code>
and
<code class="docutils literal notranslate">
<span class="pre">
aten::log
</span>
</code>
. This lets us reuse the
<code class="docutils literal notranslate">
<span class="pre">
aten::softmax
</span>
</code>
and
<code class="docutils literal notranslate">
<span class="pre">
aten::log
</span>
</code>
converters instead of needing a dedicated converter.
</p>
<h3 id="unroll-loops">
Unroll Loops
<a class="headerlink" href="#unroll-loops" title="Permalink to this headline">
¶
</a>
</h3>
<blockquote>
<div>
<p>
<a class="reference external" href="https://github.com/pytorch/pytorch/blob/master/torch/csrc/jit/passes/loop_unrolling.h">
torch/csrc/jit/passes/loop_unrolling.h
</a>
</p>
</div>
</blockquote>
<p>
Unrolls the operations of compatable loops (e.g. sufficently short) so that you only have to go through the loop once.
</p>
</article>
</div>
</div>
</main>
</div>
<footer class="md-footer">
<div class="md-footer-nav">
<nav class="md-footer-nav__inner md-grid">
<a class="md-flex md-footer-nav__link md-footer-nav__link--prev" href="system_overview.html" rel="prev" title="System Overview">
<div class="md-flex__cell md-flex__cell--shrink">
<i class="md-icon md-icon--arrow-back md-footer-nav__button">
</i>
</div>
<div class="md-flex__cell md-flex__cell--stretch md-footer-nav__title">
<span class="md-flex__ellipsis">
<span class="md-footer-nav__direction">
Previous
</span>
System Overview
</span>
</div>
</a>
<a class="md-flex md-footer-nav__link md-footer-nav__link--next" href="conversion.html" rel="next" title="Conversion Phase">
<div class="md-flex__cell md-flex__cell--stretch md-footer-nav__title">
<span class="md-flex__ellipsis">
<span class="md-footer-nav__direction">
Next
</span>
Conversion Phase
</span>
</div>
<div class="md-flex__cell md-flex__cell--shrink">
<i class="md-icon md-icon--arrow-forward md-footer-nav__button">
</i>