-
Notifications
You must be signed in to change notification settings - Fork 6
/
nep-0011-deferred-ufunc-evaluation.html
875 lines (677 loc) · 45.4 KB
/
nep-0011-deferred-ufunc-evaluation.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
<!DOCTYPE html>
<html lang="en" data-content_root="./" >
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" /><meta name="viewport" content="width=device-width, initial-scale=1" />
<title>NEP 11 — Deferred UFunc evaluation — NumPy Enhancement Proposals</title>
<script data-cfasync="false">
document.documentElement.dataset.mode = localStorage.getItem("mode") || "";
document.documentElement.dataset.theme = localStorage.getItem("theme") || "";
</script>
<!--
this give us a css class that will be invisible only if js is disabled
-->
<noscript>
<style>
.pst-js-only { display: none !important; }
</style>
</noscript>
<!-- Loaded before other Sphinx assets -->
<link href="_static/styles/theme.css?digest=8878045cc6db502f8baf" rel="stylesheet" />
<link href="_static/styles/pydata-sphinx-theme.css?digest=8878045cc6db502f8baf" rel="stylesheet" />
<link rel="stylesheet" type="text/css" href="_static/pygments.css?v=fa44fd50" />
<!-- So that users can add custom icons -->
<script src="_static/scripts/fontawesome.js?digest=8878045cc6db502f8baf"></script>
<!-- Pre-loaded scripts that we'll load fully later -->
<link rel="preload" as="script" href="_static/scripts/bootstrap.js?digest=8878045cc6db502f8baf" />
<link rel="preload" as="script" href="_static/scripts/pydata-sphinx-theme.js?digest=8878045cc6db502f8baf" />
<script src="_static/documentation_options.js?v=7f41d439"></script>
<script src="_static/doctools.js?v=888ff710"></script>
<script src="_static/sphinx_highlight.js?v=dc90522c"></script>
<script>DOCUMENTATION_OPTIONS.pagename = 'nep-0011-deferred-ufunc-evaluation';</script>
<link rel="icon" href="_static/favicon.ico"/>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="NEP 12 — Missing data functionality in NumPy" href="nep-0012-missing-data.html" />
<link rel="prev" title="NEP 9 — Structured array extensions" href="nep-0009-structured_array_extensions.html" />
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<meta name="docsearch:language" content="en"/>
<meta name="docsearch:version" content="" />
<meta name="docbuild:last-update" content="Dec 25, 2024"/>
</head>
<body data-bs-spy="scroll" data-bs-target=".bd-toc-nav" data-offset="180" data-bs-root-margin="0px 0px -60%" data-default-mode="">
<div id="pst-skip-link" class="skip-link d-print-none"><a href="#main-content">Skip to main content</a></div>
<div id="pst-scroll-pixel-helper"></div>
<button type="button" class="btn rounded-pill" id="pst-back-to-top">
<i class="fa-solid fa-arrow-up"></i>Back to top</button>
<dialog id="pst-search-dialog">
<form class="bd-search d-flex align-items-center"
action="search.html"
method="get">
<i class="fa-solid fa-magnifying-glass"></i>
<input type="search"
class="form-control"
name="q"
placeholder="Search the docs ..."
aria-label="Search the docs ..."
autocomplete="off"
autocorrect="off"
autocapitalize="off"
spellcheck="false"/>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd>K</kbd></span>
</form>
</dialog>
<div class="pst-async-banner-revealer d-none">
<aside id="bd-header-version-warning" class="d-none d-print-none" aria-label="Version warning"></aside>
</div>
<header class="bd-header navbar navbar-expand-lg bd-navbar d-print-none">
<div class="bd-header__inner bd-page-width">
<button class="pst-navbar-icon sidebar-toggle primary-toggle" aria-label="Site navigation">
<span class="fa-solid fa-bars"></span>
</button>
<div class="col-lg-3 navbar-header-items__start">
<div class="navbar-item">
<a class="navbar-brand logo" href="content.html">
<img src="_static/numpylogo.svg" class="logo__image only-light" alt="NumPy Enhancement Proposals - Home"/>
<img src="_static/numpylogo.svg" class="logo__image only-dark pst-js-only" alt="NumPy Enhancement Proposals - Home"/>
</a></div>
</div>
<div class="col-lg-9 navbar-header-items">
<div class="me-auto navbar-header-items__center">
<div class="navbar-item">
<nav>
<ul class="bd-navbar-elements navbar-nav">
<li class="nav-item current active">
<a class="nav-link nav-internal" href="index.html">
Index
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="scope.html">
The Scope of NumPy
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="roadmap.html">
Current roadmap
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://github.com/numpy/numpy/issues?q=is%3Aopen+is%3Aissue+label%3A%2223+-+Wish+List%22">
Wish list
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://github.com/numpy/numpy/issues?q=is%3Aopen+is%3Aissue+label%3A%2223+-+Wish+List%22">
Wishlist
</a>
</li>
</ul>
</nav></div>
</div>
<div class="navbar-header-items__end">
<div class="navbar-item navbar-persistent--container">
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass"></i>
<span class="search-button__default-text">Search</span>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
</button>
</div>
<div class="navbar-item">
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
</button></div>
<div class="navbar-item"><ul class="navbar-icon-links"
aria-label="Icon Links">
<li class="nav-item">
<a href="https://github.com/numpy/numpy" title="GitHub" class="nav-link pst-navbar-icon" rel="noopener" target="_blank" data-bs-toggle="tooltip" data-bs-placement="bottom"><i class="fa-brands fa-square-github fa-lg" aria-hidden="true"></i>
<span class="sr-only">GitHub</span></a>
</li>
</ul></div>
</div>
</div>
<div class="navbar-persistent--mobile">
<button class="btn search-button-field search-button__button pst-js-only" title="Search" aria-label="Search" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="fa-solid fa-magnifying-glass"></i>
<span class="search-button__default-text">Search</span>
<span class="search-button__kbd-shortcut"><kbd class="kbd-shortcut__modifier">Ctrl</kbd>+<kbd class="kbd-shortcut__modifier">K</kbd></span>
</button>
</div>
<button class="pst-navbar-icon sidebar-toggle secondary-toggle" aria-label="On this page">
<span class="fa-solid fa-outdent"></span>
</button>
</div>
</header>
<div class="bd-container">
<div class="bd-container__inner bd-page-width">
<dialog id="pst-primary-sidebar-modal"></dialog>
<div id="pst-primary-sidebar" class="bd-sidebar-primary bd-sidebar">
<div class="sidebar-header-items sidebar-primary__section">
<div class="sidebar-header-items__center">
<div class="navbar-item">
<nav>
<ul class="bd-navbar-elements navbar-nav">
<li class="nav-item current active">
<a class="nav-link nav-internal" href="index.html">
Index
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="scope.html">
The Scope of NumPy
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-internal" href="roadmap.html">
Current roadmap
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://github.com/numpy/numpy/issues?q=is%3Aopen+is%3Aissue+label%3A%2223+-+Wish+List%22">
Wish list
</a>
</li>
<li class="nav-item ">
<a class="nav-link nav-external" href="https://github.com/numpy/numpy/issues?q=is%3Aopen+is%3Aissue+label%3A%2223+-+Wish+List%22">
Wishlist
</a>
</li>
</ul>
</nav></div>
</div>
<div class="sidebar-header-items__end">
<div class="navbar-item">
<button class="btn btn-sm nav-link pst-navbar-icon theme-switch-button pst-js-only" aria-label="Color mode" data-bs-title="Color mode" data-bs-placement="bottom" data-bs-toggle="tooltip">
<i class="theme-switch fa-solid fa-sun fa-lg" data-mode="light" title="Light"></i>
<i class="theme-switch fa-solid fa-moon fa-lg" data-mode="dark" title="Dark"></i>
<i class="theme-switch fa-solid fa-circle-half-stroke fa-lg" data-mode="auto" title="System Settings"></i>
</button></div>
<div class="navbar-item"><ul class="navbar-icon-links"
aria-label="Icon Links">
<li class="nav-item">
<a href="https://github.com/numpy/numpy" title="GitHub" class="nav-link pst-navbar-icon" rel="noopener" target="_blank" data-bs-toggle="tooltip" data-bs-placement="bottom"><i class="fa-brands fa-square-github fa-lg" aria-hidden="true"></i>
<span class="sr-only">GitHub</span></a>
</li>
</ul></div>
</div>
</div>
<div class="sidebar-primary-items__start sidebar-primary__section">
<div class="sidebar-primary-item">
<nav class="bd-docs-nav bd-links"
aria-label="Section Navigation">
<p class="bd-links__title" role="heading" aria-level="1">Section Navigation</p>
<div class="bd-toc-item navbar-nav"><ul class="nav bd-sidenav">
<li class="toctree-l1"><a class="reference internal" href="scope.html">The Scope of NumPy</a></li>
<li class="toctree-l1"><a class="reference internal" href="roadmap.html">Current roadmap</a></li>
<li class="toctree-l1"><a class="reference external" href="https://github.com/numpy/numpy/issues?q=is%3Aopen+is%3Aissue+label%3A%2223+-+Wish+List%22">Wish list</a></li>
</ul>
<ul class="current nav bd-sidenav">
<li class="toctree-l1 has-children"><a class="reference internal" href="meta.html">Meta-NEPs (NEPs about NEPs or active Processes)</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0000.html">NEP 0 — Purpose and process</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0023-backwards-compatibility.html">NEP 23 — Backwards compatibility and deprecation policy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0036-fair-play.html">NEP 36 — Fair play</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0045-c_style_guide.html">NEP 45 — C style guide</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0046-sponsorship-guidelines.html">NEP 46 — NumPy sponsorship guidelines</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0048-spending-project-funds.html">NEP 48 — Spending NumPy project funds</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-template.html">NEP X — Template and instructions</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="provisional.html">Provisional NEPs (provisionally accepted; interface may change)</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul class="simple">
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="accepted.html">Accepted NEPs (implementation in progress)</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0041-improved-dtype-support.html">NEP 41 — First step towards a new datatype system</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0042-new-dtypes.html">NEP 42 — New and extensible DTypes</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0044-restructuring-numpy-docs.html">NEP 44 — Restructuring the NumPy documentation</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0051-scalar-representation.html">NEP 51 — Changing the representation of NumPy scalars</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="open.html">Open NEPs (under consideration)</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0043-extensible-ufuncs.html">NEP 43 — Enhancing the extensibility of UFuncs</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0053-c-abi-evolution.html">NEP 53 — Evolving the NumPy C-API for NumPy 2.0</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0054-simd-cpp-highway.html">NEP 54 — SIMD infrastructure evolution: adopting Google Highway when moving to C++?</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="finished.html">Finished NEPs</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0001-npy-format.html">NEP 1 — A simple file format for NumPy arrays</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0005-generalized-ufuncs.html">NEP 5 — Generalized universal functions</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0007-datetime-proposal.html">NEP 7 — A proposal for implementing some date/time types in NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0010-new-iterator-ufunc.html">NEP 10 — Optimizing iterator/UFunc performance</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0013-ufunc-overrides.html">NEP 13 — A mechanism for overriding Ufuncs</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0014-dropping-python2.7-proposal.html">NEP 14 — Plan for dropping Python 2.7 support</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0015-merge-multiarray-umath.html">NEP 15 — Merging multiarray and umath</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0018-array-function-protocol.html">NEP 18 — A dispatch mechanism for NumPy's high level array functions</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0019-rng-policy.html">NEP 19 — Random number generator policy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0020-gufunc-signature-enhancement.html">NEP 20 — Expansion of generalized universal function signatures</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0022-ndarray-duck-typing-overview.html">NEP 22 — Duck typing for NumPy arrays – high level overview</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0027-zero-rank-arrarys.html">NEP 27 — Zero rank arrays</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0028-website-redesign.html">NEP 28 — numpy.org website redesign</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0029-deprecation_policy.html">NEP 29 — Recommend Python and NumPy version support as a community policy standard</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0032-remove-financial-functions.html">NEP 32 — Remove the financial functions from NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0034-infer-dtype-is-object.html">NEP 34 — Disallow inferring ``dtype=object`` from sequences</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0035-array-creation-dispatch-with-array-function.html">NEP 35 — Array creation dispatching with __array_function__</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0038-SIMD-optimizations.html">NEP 38 — Using SIMD optimization instructions for performance</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0040-legacy-datatype-impl.html">NEP 40 — Legacy datatype implementation in NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0049.html">NEP 49 — Data allocation strategies</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0050-scalar-promotion.html">NEP 50 — Promotion rules for Python scalars</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0052-python-api-cleanup.html">NEP 52 — Python API cleanup for NumPy 2.0</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0055-string_dtype.html">NEP 55 — Add a UTF-8 variable-width string DType to NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0056-array-api-main-namespace.html">NEP 56 — Array API standard support in NumPy's main namespace</a></li>
</ul>
</details></li>
<li class="toctree-l1 current active has-children"><a class="reference internal" href="deferred.html">Deferred and Superseded NEPs</a><details open="open"><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul class="current">
<li class="toctree-l2"><a class="reference internal" href="nep-0002-warnfix.html">NEP 2 — A proposal to build numpy without warning with a big set of warning flags</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0003-math_config_clean.html">NEP 3 — Cleaning the math configuration of numpy.core</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0004-datetime-proposal3.html">NEP 4 — A (third) proposal for implementing some date/time types in NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0006-newbugtracker.html">NEP 6 — Replacing Trac with a different bug tracker</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0008-groupby_additions.html">NEP 8 — A proposal for adding groupby functionality to NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0009-structured_array_extensions.html">NEP 9 — Structured array extensions</a></li>
<li class="toctree-l2 current active"><a class="current reference internal" href="#">NEP 11 — Deferred UFunc evaluation</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0012-missing-data.html">NEP 12 — Missing data functionality in NumPy</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0021-advanced-indexing.html">NEP 21 — Simplified and explicit advanced indexing</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0024-missing-data-2.html">NEP 24 — Missing data functionality - alternative 1 to NEP 12</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0025-missing-data-3.html">NEP 25 — NA support via special dtypes</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0026-missing-data-summary.html">NEP 26 — Summary of missing data NEPs and discussion</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0030-duck-array-protocol.html">NEP 30 — Duck typing for NumPy arrays - implementation</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0031-uarray.html">NEP 31 — Context-local and global overrides of the NumPy API</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0037-array-module.html">NEP 37 — A dispatch protocol for NumPy-like modules</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0047-array-api-standard.html">NEP 47 — Adopting the array API standard</a></li>
</ul>
</details></li>
<li class="toctree-l1 has-children"><a class="reference internal" href="rejected.html">Rejected and Withdrawn NEPs</a><details><summary><span class="toctree-toggle" role="presentation"><i class="fa-solid fa-chevron-down"></i></span></summary><ul>
<li class="toctree-l2"><a class="reference internal" href="nep-0016-abstract-array.html">NEP 16 — An abstract base class for identifying "duck arrays"</a></li>
<li class="toctree-l2"><a class="reference internal" href="nep-0017-split-out-maskedarray.html">NEP 17 — Split out masked arrays</a></li>
</ul>
</details></li>
</ul>
</div>
</nav></div>
</div>
<div class="sidebar-primary-items__end sidebar-primary__section">
<div class="sidebar-primary-item">
<div id="ethical-ad-placement"
class="flat"
data-ea-publisher="readthedocs"
data-ea-type="readthedocs-sidebar"
data-ea-manual="true">
</div></div>
</div>
</div>
<main id="main-content" class="bd-main" role="main">
<div class="bd-content">
<div class="bd-article-container">
<div class="bd-header-article d-print-none">
<div class="header-article-items header-article__inner">
<div class="header-article-items__start">
<div class="header-article-item">
<nav aria-label="Breadcrumb" class="d-print-none">
<ul class="bd-breadcrumbs">
<li class="breadcrumb-item breadcrumb-home">
<a href="content.html" class="nav-link" aria-label="Home">
<i class="fa-solid fa-home"></i>
</a>
</li>
<li class="breadcrumb-item"><a href="index.html" class="nav-link">Roadmap & NumPy enhancement proposals</a></li>
<li class="breadcrumb-item"><a href="deferred.html" class="nav-link">Deferred and Superseded NEPs</a></li>
<li class="breadcrumb-item active" aria-current="page"><span class="ellipsis">NEP 11 — Deferred UFunc evaluation</span></li>
</ul>
</nav>
</div>
</div>
</div>
</div>
<div id="searchbox"></div>
<article class="bd-article">
<section id="nep-11-deferred-ufunc-evaluation">
<span id="nep11"></span><h1>NEP 11 — Deferred UFunc evaluation<a class="headerlink" href="#nep-11-deferred-ufunc-evaluation" title="Link to this heading">#</a></h1>
<dl class="field-list simple">
<dt class="field-odd">Author<span class="colon">:</span></dt>
<dd class="field-odd"><p>Mark Wiebe <<a class="reference external" href="mailto:mwwiebe%40gmail.com">mwwiebe<span>@</span>gmail<span>.</span>com</a>></p>
</dd>
<dt class="field-even">Content-Type<span class="colon">:</span></dt>
<dd class="field-even"><p>text/x-rst</p>
</dd>
<dt class="field-odd">Created<span class="colon">:</span></dt>
<dd class="field-odd"><p>30-Nov-2010</p>
</dd>
<dt class="field-even">Status<span class="colon">:</span></dt>
<dd class="field-even"><p>Deferred</p>
</dd>
</dl>
<section id="abstract">
<h2>Abstract<a class="headerlink" href="#abstract" title="Link to this heading">#</a></h2>
<p>This NEP describes a proposal to add deferred evaluation to NumPy’s
UFuncs. This will allow Python expressions like
“a[:] = b + c + d + e” to be evaluated in a single pass through all
the variables at once, with no temporary arrays. The resulting
performance will likely be comparable to the <em>numexpr</em> library,
but with a more natural syntax.</p>
<p>This idea has some interaction with UFunc error handling and
the UPDATEIFCOPY flag, affecting the design and implementation,
but the result allows for the usage of deferred evaluation
with minimal effort from the Python user’s perspective.</p>
</section>
<section id="motivation">
<h2>Motivation<a class="headerlink" href="#motivation" title="Link to this heading">#</a></h2>
<p>NumPy’s style of UFunc execution causes suboptimal performance for
large expressions, because multiple temporaries are allocated and
the inputs are swept through in multiple passes. The <em>numexpr</em> library
can outperform NumPy for such large expressions, by doing the execution
in small cache-friendly blocks, and evaluating the whole expression
per element. This results in one sweep through each input, which
is significantly better for the cache.</p>
<p>For an idea of how to get this kind of behavior in NumPy without
changing the Python code, consider the C++ technique of
expression templates. These can be used to quite arbitrarily
rearrange expressions using
vectors or other data structures, example:</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span><span class="n">A</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">B</span><span class="w"> </span><span class="o">+</span><span class="w"> </span><span class="n">C</span><span class="w"> </span><span class="o">+</span><span class="w"> </span><span class="n">D</span><span class="p">;</span>
</pre></div>
</div>
<p>can be transformed into something equivalent to:</p>
<div class="highlight-cpp notranslate"><div class="highlight"><pre><span></span><span class="k">for</span><span class="p">(</span><span class="n">i</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="mi">0</span><span class="p">;</span><span class="w"> </span><span class="n">i</span><span class="w"> </span><span class="o"><</span><span class="w"> </span><span class="n">A</span><span class="p">.</span><span class="n">size</span><span class="p">;</span><span class="w"> </span><span class="o">++</span><span class="n">i</span><span class="p">)</span><span class="w"> </span><span class="p">{</span>
<span class="w"> </span><span class="n">A</span><span class="p">[</span><span class="n">i</span><span class="p">]</span><span class="w"> </span><span class="o">=</span><span class="w"> </span><span class="n">B</span><span class="p">[</span><span class="n">i</span><span class="p">]</span><span class="w"> </span><span class="o">+</span><span class="w"> </span><span class="n">C</span><span class="p">[</span><span class="n">i</span><span class="p">]</span><span class="w"> </span><span class="o">+</span><span class="w"> </span><span class="n">D</span><span class="p">[</span><span class="n">i</span><span class="p">];</span>
<span class="p">}</span>
</pre></div>
</div>
<p>This is done by returning a proxy object that knows how to calculate
the result instead of returning the actual object. With modern C++
optimizing compilers, the resulting machine code is often the same
as hand-written loops. For an example of this, see the
<a class="reference external" href="http://www.oonumerics.org/blitz/docs/blitz_3.html">Blitz++ Library</a>.
A more recently created library for helping write expression templates
is <a class="reference external" href="http://beta.boost.org/doc/libs/1_44_0/doc/html/proto.html">Boost Proto</a>.</p>
<p>By using the same idea of returning a proxy object in Python, we
can accomplish the same thing dynamically. The return object is
an ndarray without its buffer allocated, and with enough knowledge
to calculate itself when needed. When a “deferred array” is
finally evaluated, we can use the expression tree made up of
all the operand deferred arrays, effectively creating a single new
UFunc to evaluate on the fly.</p>
</section>
<section id="example-python-code">
<h2>Example Python code<a class="headerlink" href="#example-python-code" title="Link to this heading">#</a></h2>
<p>Here’s how it might be used in NumPy.:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># a, b, c are large ndarrays</span>
<span class="k">with</span> <span class="n">np</span><span class="o">.</span><span class="n">deferredstate</span><span class="p">(</span><span class="kc">True</span><span class="p">):</span>
<span class="n">d</span> <span class="o">=</span> <span class="n">a</span> <span class="o">+</span> <span class="n">b</span> <span class="o">+</span> <span class="n">c</span>
<span class="c1"># Now d is a 'deferred array,' a, b, and c are marked READONLY</span>
<span class="c1"># similar to the existing UPDATEIFCOPY mechanism.</span>
<span class="nb">print</span> <span class="n">d</span>
<span class="c1"># Since the value of d was required, it is evaluated so d becomes</span>
<span class="c1"># a regular ndarray and gets printed.</span>
<span class="n">d</span><span class="p">[:]</span> <span class="o">=</span> <span class="n">a</span><span class="o">*</span><span class="n">b</span><span class="o">*</span><span class="n">c</span>
<span class="c1"># Here, the automatically combined "ufunc" that computes</span>
<span class="c1"># a*b*c effectively gets an out= parameter, so no temporary</span>
<span class="c1"># arrays are needed whatsoever.</span>
<span class="n">e</span> <span class="o">=</span> <span class="n">a</span><span class="o">+</span><span class="n">b</span><span class="o">+</span><span class="n">c</span><span class="o">*</span><span class="n">d</span>
<span class="c1"># Now e is a 'deferred array,' a, b, c, and d are marked READONLY</span>
<span class="n">d</span><span class="p">[:]</span> <span class="o">=</span> <span class="n">a</span>
<span class="c1"># d was marked readonly, but the assignment could see that</span>
<span class="c1"># this was due to it being a deferred expression operand.</span>
<span class="c1"># This triggered the deferred evaluation so it could assign</span>
<span class="c1"># the value of a to d.</span>
</pre></div>
</div>
<p>There may be some surprising behavior, though.:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">with</span> <span class="n">np</span><span class="o">.</span><span class="n">deferredstate</span><span class="p">(</span><span class="kc">True</span><span class="p">):</span>
<span class="n">d</span> <span class="o">=</span> <span class="n">a</span> <span class="o">+</span> <span class="n">b</span> <span class="o">+</span> <span class="n">c</span>
<span class="c1"># d is deferred</span>
<span class="n">e</span><span class="p">[:]</span> <span class="o">=</span> <span class="n">d</span>
<span class="n">f</span><span class="p">[:]</span> <span class="o">=</span> <span class="n">d</span>
<span class="n">g</span><span class="p">[:]</span> <span class="o">=</span> <span class="n">d</span>
<span class="c1"># d is still deferred, and its deferred expression</span>
<span class="c1"># was evaluated three times, once for each assignment.</span>
<span class="c1"># This could be detected, with d being converted to</span>
<span class="c1"># a regular ndarray the second time it is evaluated.</span>
</pre></div>
</div>
<p>I believe the usage that should be recommended in the documentation
is to leave the deferred state at its default, except when
evaluating a large expression that can benefit from it.:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># calculations</span>
<span class="k">with</span> <span class="n">np</span><span class="o">.</span><span class="n">deferredstate</span><span class="p">(</span><span class="kc">True</span><span class="p">):</span>
<span class="n">x</span> <span class="o">=</span> <span class="o"><</span><span class="n">big</span> <span class="n">expression</span><span class="o">></span>
<span class="c1"># more calculations</span>
</pre></div>
</div>
<p>This will avoid surprises which would be cause by always keeping
deferred usage True, like floating point warnings or exceptions
at surprising times when deferred expression are used later.
User questions like “Why does my print statement throw a
divide by zero error?” can hopefully be avoided by recommending
this approach.</p>
</section>
<section id="proposed-deferred-evaluation-api">
<h2>Proposed deferred evaluation API<a class="headerlink" href="#proposed-deferred-evaluation-api" title="Link to this heading">#</a></h2>
<p>For deferred evaluation to work, the C API needs to be aware of its
existence, and be able to trigger evaluation when necessary. The
ndarray would gain two new flag.</p>
<blockquote>
<div><p><code class="docutils literal notranslate"><span class="pre">NPY_ISDEFERRED</span></code></p>
<blockquote>
<div><p>Indicates the expression evaluation for this ndarray instance
has been deferred.</p>
</div></blockquote>
<p><code class="docutils literal notranslate"><span class="pre">NPY_DEFERRED_WASWRITEABLE</span></code></p>
<blockquote>
<div><p>Can only be set when <code class="docutils literal notranslate"><span class="pre">PyArray_GetDeferredUsageCount(arr)</span> <span class="pre">></span> <span class="pre">0</span></code>.
It indicates that when <code class="docutils literal notranslate"><span class="pre">arr</span></code> was first used in a deferred
expression, it was a writeable array. If this flag is set,
calling <code class="docutils literal notranslate"><span class="pre">PyArray_CalculateAllDeferred()</span></code> will make <code class="docutils literal notranslate"><span class="pre">arr</span></code>
writeable again.</p>
</div></blockquote>
</div></blockquote>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>QUESTION</p>
<p>Should NPY_DEFERRED and NPY_DEFERRED_WASWRITEABLE be visible
to Python, or should accessing the flags from python trigger
PyArray_CalculateAllDeferred if necessary?</p>
</div>
<p>The API would be expanded with a number of functions.</p>
<p><code class="docutils literal notranslate"><span class="pre">int</span> <span class="pre">PyArray_CalculateAllDeferred()</span></code></p>
<blockquote>
<div><p>This function forces all currently deferred calculations to occur.</p>
<p>For example, if the error state is set to ignore all, and
np.seterr({all=’raise’}), this would change what happens
to already deferred expressions. Thus, all the existing
deferred arrays should be evaluated before changing the
error state.</p>
</div></blockquote>
<p><code class="docutils literal notranslate"><span class="pre">int</span> <span class="pre">PyArray_CalculateDeferred(PyArrayObject*</span> <span class="pre">arr)</span></code></p>
<blockquote>
<div><p>If ‘arr’ is a deferred array, allocates memory for it and
evaluates the deferred expression. If ‘arr’ is not a deferred
array, simply returns success. Returns NPY_SUCCESS or NPY_FAILURE.</p>
</div></blockquote>
<p><code class="docutils literal notranslate"><span class="pre">int</span> <span class="pre">PyArray_CalculateDeferredAssignment(PyArrayObject*</span> <span class="pre">arr,</span> <span class="pre">PyArrayObject*</span> <span class="pre">out)</span></code></p>
<blockquote>
<div><p>If ‘arr’ is a deferred array, evaluates the deferred expression
into ‘out’, and ‘arr’ remains a deferred array. If ‘arr’ is not
a deferred array, copies its value into out. Returns NPY_SUCCESS
or NPY_FAILURE.</p>
</div></blockquote>
<p><code class="docutils literal notranslate"><span class="pre">int</span> <span class="pre">PyArray_GetDeferredUsageCount(PyArrayObject*</span> <span class="pre">arr)</span></code></p>
<blockquote>
<div><p>Returns a count of how many deferred expressions use this array
as an operand.</p>
</div></blockquote>
<p>The Python API would be expanded as follows.</p>
<blockquote>
<div><p><code class="docutils literal notranslate"><span class="pre">numpy.setdeferred(state)</span></code></p>
<blockquote>
<div><p>Enables or disables deferred evaluation. True means to always
use deferred evaluation. False means to never use deferred
evaluation. None means to use deferred evaluation if the error
handling state is set to ignore everything. At NumPy initialization,
the deferred state is None.</p>
<p>Returns the previous deferred state.</p>
</div></blockquote>
</div></blockquote>
<p><code class="docutils literal notranslate"><span class="pre">numpy.getdeferred()</span></code></p>
<blockquote>
<div><p>Returns the current deferred state.</p>
</div></blockquote>
<p><code class="docutils literal notranslate"><span class="pre">numpy.deferredstate(state)</span></code></p>
<blockquote>
<div><p>A context manager for deferred state handling, similar to
<code class="docutils literal notranslate"><span class="pre">numpy.errstate</span></code>.</p>
</div></blockquote>
<section id="error-handling">
<h3>Error handling<a class="headerlink" href="#error-handling" title="Link to this heading">#</a></h3>
<p>Error handling is a thorny issue for deferred evaluation. If the
NumPy error state is {all=’ignore’}, it might be reasonable to
introduce deferred evaluation as the default, however if a UFunc
can raise an error, it would be very strange for the later ‘print’
statement to throw the exception instead of the actual operation which
caused the error.</p>
<p>What may be a good approach is to by default enable deferred evaluation
only when the error state is set to ignore all, but allow user control with
‘setdeferred’ and ‘getdeferred’ functions. True would mean always
use deferred evaluation, False would mean never use it, and None would
mean use it only when safe (i.e. the error state is set to ignore all).</p>
</section>
<section id="interaction-with-updateifcopy">
<h3>Interaction with UPDATEIFCOPY<a class="headerlink" href="#interaction-with-updateifcopy" title="Link to this heading">#</a></h3>
<p>The <code class="docutils literal notranslate"><span class="pre">NPY_UPDATEIFCOPY</span></code> documentation states:</p>
<blockquote>
<div><p>The data area represents a (well-behaved) copy whose information
should be transferred back to the original when this array is deleted.</p>
<p>This is a special flag that is set if this array represents a copy
made because a user required certain flags in PyArray_FromAny and a
copy had to be made of some other array (and the user asked for this
flag to be set in such a situation). The base attribute then points
to the “misbehaved” array (which is set read_only). When the array
with this flag set is deallocated, it will copy its contents back to
the “misbehaved” array (casting if necessary) and will reset the
“misbehaved” array to NPY_WRITEABLE. If the “misbehaved” array was
not NPY_WRITEABLE to begin with then PyArray_FromAny would have
returned an error because NPY_UPDATEIFCOPY would not have been possible.</p>
</div></blockquote>
<p>The current implementation of UPDATEIFCOPY assumes that it is the only
mechanism mucking with the writeable flag in this manner. These mechanisms
must be aware of each other to work correctly. Here’s an example of how
they might go wrong:</p>
<ol class="arabic simple">
<li><p>Make a temporary copy of ‘arr’ with UPDATEIFCOPY (‘arr’ becomes read only)</p></li>
<li><p>Use ‘arr’ in a deferred expression (deferred usage count becomes one,
NPY_DEFERRED_WASWRITEABLE is <strong>not</strong> set, since ‘arr’ is read only)</p></li>
<li><p>Destroy the temporary copy, causing ‘arr’ to become writeable</p></li>
<li><p>Writing to ‘arr’ destroys the value of the deferred expression</p></li>
</ol>
<p>To deal with this issue, we make these two states mutually exclusive.</p>
<ul class="simple">
<li><p>Usage of UPDATEIFCOPY checks the <code class="docutils literal notranslate"><span class="pre">NPY_DEFERRED_WASWRITEABLE</span></code> flag,
and if it’s set, calls <code class="docutils literal notranslate"><span class="pre">PyArray_CalculateAllDeferred</span></code> to flush
all deferred calculation before proceeding.</p></li>
<li><p>The ndarray gets a new flag <code class="docutils literal notranslate"><span class="pre">NPY_UPDATEIFCOPY_TARGET</span></code> indicating
the array will be updated and made writeable at some point in the
future. If the deferred evaluation mechanism sees this flag in
any operand, it triggers immediate evaluation.</p></li>
</ul>
</section>
<section id="other-implementation-details">
<h3>Other implementation details<a class="headerlink" href="#other-implementation-details" title="Link to this heading">#</a></h3>
<p>When a deferred array is created, it gets references to all the
operands of the UFunc, along with the UFunc itself. The
‘DeferredUsageCount’ is incremented for each operand, and later
gets decremented when the deferred expression is calculated or
the deferred array is destroyed.</p>
<p>A global list of weak references to all the deferred arrays
is tracked, in order of creation. When <code class="docutils literal notranslate"><span class="pre">PyArray_CalculateAllDeferred</span></code>
gets called, the newest deferred array is calculated first.
This may release references to other deferred arrays contained
in the deferred expression tree, which then
never have to be calculated.</p>
</section>
<section id="further-optimization">
<h3>Further optimization<a class="headerlink" href="#further-optimization" title="Link to this heading">#</a></h3>
<p>Instead of conservatively disabling deferred evaluation when any
errors are not set to ‘ignore’, each UFunc could give a set
of possible errors it generates. Then, if all those errors
are set to ‘ignore’, deferred evaluation could be used even
if other errors are not set to ignore.</p>
<p>Once the expression tree is explicitly stored, it is possible to
do transformations on it. For example add(add(a,b),c) could
be transformed into add3(a,b,c), or add(multiply(a,b),c) could
become fma(a,b,c) using the CPU fused multiply-add instruction
where available.</p>
<p>While I’ve framed deferred evaluation as just for UFuncs, it could
be extended to other functions, such as dot(). For example, chained
matrix multiplications could be reordered to minimize the size
of intermediates, or peep-hole style optimizer passes could search
for patterns that match optimized BLAS/other high performance
library calls.</p>
<p>For operations on really large arrays, integrating a JIT like LLVM into
this system might be a big benefit. The UFuncs and other operations
would provide bitcode, which could be inlined together and optimized
by the LLVM optimizers, then executed. In fact, the iterator itself
could also be represented in bitcode, allowing LLVM to consider
the entire iteration while doing its optimization.</p>
</section>
</section>
</section>
</article>
</div>
<dialog id="pst-secondary-sidebar-modal"></dialog>
<div id="pst-secondary-sidebar" class="bd-sidebar-secondary bd-toc"><div class="sidebar-secondary-items sidebar-secondary__inner">
<div class="sidebar-secondary-item">
<div
id="pst-page-navigation-heading-2"
class="page-toc tocsection onthispage">
<i class="fa-solid fa-list"></i> On this page
</div>
<nav class="bd-toc-nav page-toc" aria-labelledby="pst-page-navigation-heading-2">
<ul class="visible nav section-nav flex-column">
<li class="toc-h2 nav-item toc-entry"><a class="reference internal nav-link" href="#abstract">Abstract</a></li>
<li class="toc-h2 nav-item toc-entry"><a class="reference internal nav-link" href="#motivation">Motivation</a></li>
<li class="toc-h2 nav-item toc-entry"><a class="reference internal nav-link" href="#example-python-code">Example Python code</a></li>
<li class="toc-h2 nav-item toc-entry"><a class="reference internal nav-link" href="#proposed-deferred-evaluation-api">Proposed deferred evaluation API</a><ul class="nav section-nav flex-column">
<li class="toc-h3 nav-item toc-entry"><a class="reference internal nav-link" href="#error-handling">Error handling</a></li>
<li class="toc-h3 nav-item toc-entry"><a class="reference internal nav-link" href="#interaction-with-updateifcopy">Interaction with UPDATEIFCOPY</a></li>
<li class="toc-h3 nav-item toc-entry"><a class="reference internal nav-link" href="#other-implementation-details">Other implementation details</a></li>
<li class="toc-h3 nav-item toc-entry"><a class="reference internal nav-link" href="#further-optimization">Further optimization</a></li>
</ul>
</li>
</ul>
</nav></div>
</div></div>
</div>
<footer class="bd-footer-content">
</footer>
</main>
</div>
</div>
<!-- Scripts loaded after <body> so the DOM is not blocked -->
<script defer src="_static/scripts/bootstrap.js?digest=8878045cc6db502f8baf"></script>
<script defer src="_static/scripts/pydata-sphinx-theme.js?digest=8878045cc6db502f8baf"></script>
<footer class="bd-footer">
<div class="bd-footer__inner bd-page-width">
<div class="footer-items__start">
<div class="footer-item">
<p class="copyright">
© Copyright 2017-2024, NumPy Developers.
<br/>
</p>
</div>
<div class="footer-item">
<p class="sphinx-version">
Created using <a href="https://www.sphinx-doc.org/">Sphinx</a> 7.2.6.
<br/>
</p>
</div>
</div>
<div class="footer-items__end">
<div class="footer-item">
<p class="theme-version">
<!-- # L10n: Setting the PST URL as an argument as this does not need to be localized -->
Built with the <a href="https://pydata-sphinx-theme.readthedocs.io/en/stable/index.html">PyData Sphinx Theme</a> 0.16.1.
</p></div>
</div>
</div>
</footer>
</body>
</html>