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
|
<!DOCTYPE html>
<html>
<head>
<meta charset="utf-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>LLVM Atomic Instructions and Concurrency Guide — LLVM 13 documentation</title>
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="_static/llvm-theme.css" type="text/css" />
<script id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="LLVM Bitcode File Format" href="BitCodeFormat.html" />
<link rel="prev" title="Reference" href="Reference.html" />
<style type="text/css">
table.right { float: right; margin-left: 20px; }
table.right td { border: 1px solid #ccc; }
</style>
</head><body>
<div class="logo">
<a href="index.html">
<img src="_static/logo.png"
alt="LLVM Logo" width="250" height="88"/></a>
</div>
<div class="related" role="navigation" aria-label="related navigation">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="genindex.html" title="General Index"
accesskey="I">index</a></li>
<li class="right" >
<a href="BitCodeFormat.html" title="LLVM Bitcode File Format"
accesskey="N">next</a> |</li>
<li class="right" >
<a href="Reference.html" title="Reference"
accesskey="P">previous</a> |</li>
<li><a href="https://llvm.org/">LLVM Home</a> | </li>
<li><a href="index.html">Documentation</a>»</li>
<li class="nav-item nav-item-1"><a href="Reference.html" accesskey="U">Reference</a> »</li>
<li class="nav-item nav-item-this"><a href="">LLVM Atomic Instructions and Concurrency Guide</a></li>
</ul>
</div>
<div class="sphinxsidebar" role="navigation" aria-label="main navigation">
<div class="sphinxsidebarwrapper">
<h3>Documentation</h3>
<ul class="want-points">
<li><a href="https://llvm.org/docs/GettingStartedTutorials.html">Getting Started/Tutorials</a></li>
<li><a href="https://llvm.org/docs/UserGuides.html">User Guides</a></li>
<li><a href="https://llvm.org/docs/Reference.html">Reference</a></li>
</ul>
<h3>Getting Involved</h3>
<ul class="want-points">
<li><a href="https://llvm.org/docs/Contributing.html">Contributing to LLVM</a></li>
<li><a href="https://llvm.org/docs/HowToSubmitABug.html">Submitting Bug Reports</a></li>
<li><a href="https://llvm.org/docs/GettingInvolved.html#mailing-lists">Mailing Lists</a></li>
<li><a href="https://llvm.org/docs/GettingInvolved.html#irc">IRC</a></li>
<li><a href="https://llvm.org/docs/GettingInvolved.html#meetups-and-social-events">Meetups and Social Events</a></li>
</ul>
<h3>Additional Links</h3>
<ul class="want-points">
<li><a href="https://llvm.org/docs/FAQ.html">FAQ</a></li>
<li><a href="https://llvm.org/docs/Lexicon.html">Glossary</a></li>
<li><a href="https://llvm.org/pubs">Publications</a></li>
<li><a href="https://github.com/llvm/llvm-project//">Github Repository</a></li>
</ul>
<div role="note" aria-label="source link">
<h3>This Page</h3>
<ul class="this-page-menu">
<li><a href="_sources/Atomics.rst.txt"
rel="nofollow">Show Source</a></li>
</ul>
</div>
<div id="searchbox" style="display: none" role="search">
<h3 id="searchlabel">Quick search</h3>
<div class="searchformwrapper">
<form class="search" action="search.html" method="get">
<input type="text" name="q" aria-labelledby="searchlabel" />
<input type="submit" value="Go" />
</form>
</div>
</div>
<script>$('#searchbox').show(0);</script>
</div>
</div>
<div class="document">
<div class="documentwrapper">
<div class="bodywrapper">
<div class="body" role="main">
<div class="section" id="llvm-atomic-instructions-and-concurrency-guide">
<h1>LLVM Atomic Instructions and Concurrency Guide<a class="headerlink" href="#llvm-atomic-instructions-and-concurrency-guide" title="Permalink to this headline">¶</a></h1>
<div class="contents local topic" id="contents">
<ul class="simple">
<li><p><a class="reference internal" href="#introduction" id="id4">Introduction</a></p></li>
<li><p><a class="reference internal" href="#optimization-outside-atomic" id="id5">Optimization outside atomic</a></p></li>
<li><p><a class="reference internal" href="#atomic-instructions" id="id6">Atomic instructions</a></p></li>
<li><p><a class="reference internal" href="#atomic-orderings" id="id7">Atomic orderings</a></p>
<ul>
<li><p><a class="reference internal" href="#notatomic" id="id8">NotAtomic</a></p></li>
<li><p><a class="reference internal" href="#unordered" id="id9">Unordered</a></p></li>
<li><p><a class="reference internal" href="#monotonic" id="id10">Monotonic</a></p></li>
<li><p><a class="reference internal" href="#acquire" id="id11">Acquire</a></p></li>
<li><p><a class="reference internal" href="#release" id="id12">Release</a></p></li>
<li><p><a class="reference internal" href="#acquirerelease" id="id13">AcquireRelease</a></p></li>
<li><p><a class="reference internal" href="#sequentiallyconsistent" id="id14">SequentiallyConsistent</a></p></li>
</ul>
</li>
<li><p><a class="reference internal" href="#atomics-and-ir-optimization" id="id15">Atomics and IR optimization</a></p></li>
<li><p><a class="reference internal" href="#atomics-and-codegen" id="id16">Atomics and Codegen</a></p></li>
<li><p><a class="reference internal" href="#libcalls-atomic" id="id17">Libcalls: __atomic_*</a></p></li>
<li><p><a class="reference internal" href="#libcalls-sync" id="id18">Libcalls: __sync_*</a></p></li>
</ul>
</div>
<div class="section" id="introduction">
<h2><a class="toc-backref" href="#id4">Introduction</a><a class="headerlink" href="#introduction" title="Permalink to this headline">¶</a></h2>
<p>LLVM supports instructions which are well-defined in the presence of threads and
asynchronous signals.</p>
<p>The atomic instructions are designed specifically to provide readable IR and
optimized code generation for the following:</p>
<ul class="simple">
<li><p>The C++11 <code class="docutils literal notranslate"><span class="pre"><atomic></span></code> header. (<a class="reference external" href="http://www.open-std.org/jtc1/sc22/wg21/">C++11 draft available here</a>.) (<a class="reference external" href="http://www.open-std.org/jtc1/sc22/wg14/">C11 draft available here</a>.)</p></li>
<li><p>Proper semantics for Java-style memory, for both <code class="docutils literal notranslate"><span class="pre">volatile</span></code> and regular
shared variables. (<a class="reference external" href="http://docs.oracle.com/javase/specs/jls/se8/html/jls-17.html">Java Specification</a>)</p></li>
<li><p>gcc-compatible <code class="docutils literal notranslate"><span class="pre">__sync_*</span></code> builtins. (<a class="reference external" href="https://gcc.gnu.org/onlinedocs/gcc/_005f_005fsync-Builtins.html">Description</a>)</p></li>
<li><p>Other scenarios with atomic semantics, including <code class="docutils literal notranslate"><span class="pre">static</span></code> variables with
non-trivial constructors in C++.</p></li>
</ul>
<p>Atomic and volatile in the IR are orthogonal; “volatile” is the C/C++ volatile,
which ensures that every volatile load and store happens and is performed in the
stated order. A couple examples: if a SequentiallyConsistent store is
immediately followed by another SequentiallyConsistent store to the same
address, the first store can be erased. This transformation is not allowed for a
pair of volatile stores. On the other hand, a non-volatile non-atomic load can
be moved across a volatile load freely, but not an Acquire load.</p>
<p>This document is intended to provide a guide to anyone either writing a frontend
for LLVM or working on optimization passes for LLVM with a guide for how to deal
with instructions with special semantics in the presence of concurrency. This
is not intended to be a precise guide to the semantics; the details can get
extremely complicated and unreadable, and are not usually necessary.</p>
</div>
<div class="section" id="optimization-outside-atomic">
<span id="id1"></span><h2><a class="toc-backref" href="#id5">Optimization outside atomic</a><a class="headerlink" href="#optimization-outside-atomic" title="Permalink to this headline">¶</a></h2>
<p>The basic <code class="docutils literal notranslate"><span class="pre">'load'</span></code> and <code class="docutils literal notranslate"><span class="pre">'store'</span></code> allow a variety of optimizations, but can
lead to undefined results in a concurrent environment; see <a class="reference internal" href="#notatomic">NotAtomic</a>. This
section specifically goes into the one optimizer restriction which applies in
concurrent environments, which gets a bit more of an extended description
because any optimization dealing with stores needs to be aware of it.</p>
<p>From the optimizer’s point of view, the rule is that if there are not any
instructions with atomic ordering involved, concurrency does not matter, with
one exception: if a variable might be visible to another thread or signal
handler, a store cannot be inserted along a path where it might not execute
otherwise. Take the following example:</p>
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="cm">/* C code, for readability; run through clang -O2 -S -emit-llvm to get</span>
<span class="cm"> equivalent IR */</span>
<span class="kt">int</span> <span class="n">x</span><span class="p">;</span>
<span class="kt">void</span> <span class="nf">f</span><span class="p">(</span><span class="kt">int</span><span class="o">*</span> <span class="n">a</span><span class="p">)</span> <span class="p">{</span>
<span class="k">for</span> <span class="p">(</span><span class="kt">int</span> <span class="n">i</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span> <span class="n">i</span> <span class="o"><</span> <span class="mi">100</span><span class="p">;</span> <span class="n">i</span><span class="o">++</span><span class="p">)</span> <span class="p">{</span>
<span class="k">if</span> <span class="p">(</span><span class="n">a</span><span class="p">[</span><span class="n">i</span><span class="p">])</span>
<span class="n">x</span> <span class="o">+=</span> <span class="mi">1</span><span class="p">;</span>
<span class="p">}</span>
<span class="p">}</span>
</pre></div>
</div>
<p>The following is equivalent in non-concurrent situations:</p>
<div class="highlight-c notranslate"><div class="highlight"><pre><span></span><span class="kt">int</span> <span class="n">x</span><span class="p">;</span>
<span class="kt">void</span> <span class="nf">f</span><span class="p">(</span><span class="kt">int</span><span class="o">*</span> <span class="n">a</span><span class="p">)</span> <span class="p">{</span>
<span class="kt">int</span> <span class="n">xtemp</span> <span class="o">=</span> <span class="n">x</span><span class="p">;</span>
<span class="k">for</span> <span class="p">(</span><span class="kt">int</span> <span class="n">i</span> <span class="o">=</span> <span class="mi">0</span><span class="p">;</span> <span class="n">i</span> <span class="o"><</span> <span class="mi">100</span><span class="p">;</span> <span class="n">i</span><span class="o">++</span><span class="p">)</span> <span class="p">{</span>
<span class="k">if</span> <span class="p">(</span><span class="n">a</span><span class="p">[</span><span class="n">i</span><span class="p">])</span>
<span class="n">xtemp</span> <span class="o">+=</span> <span class="mi">1</span><span class="p">;</span>
<span class="p">}</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">xtemp</span><span class="p">;</span>
<span class="p">}</span>
</pre></div>
</div>
<p>However, LLVM is not allowed to transform the former to the latter: it could
indirectly introduce undefined behavior if another thread can access <code class="docutils literal notranslate"><span class="pre">x</span></code> at
the same time. That thread would read <cite>undef</cite> instead of the value it was
expecting, which can lead to undefined behavior down the line. (This example is
particularly of interest because before the concurrency model was implemented,
LLVM would perform this transformation.)</p>
<p>Note that speculative loads are allowed; a load which is part of a race returns
<code class="docutils literal notranslate"><span class="pre">undef</span></code>, but does not have undefined behavior.</p>
</div>
<div class="section" id="atomic-instructions">
<h2><a class="toc-backref" href="#id6">Atomic instructions</a><a class="headerlink" href="#atomic-instructions" title="Permalink to this headline">¶</a></h2>
<p>For cases where simple loads and stores are not sufficient, LLVM provides
various atomic instructions. The exact guarantees provided depend on the
ordering; see <a class="reference internal" href="#atomic-orderings">Atomic orderings</a>.</p>
<p><code class="docutils literal notranslate"><span class="pre">load</span> <span class="pre">atomic</span></code> and <code class="docutils literal notranslate"><span class="pre">store</span> <span class="pre">atomic</span></code> provide the same basic functionality as
non-atomic loads and stores, but provide additional guarantees in situations
where threads and signals are involved.</p>
<p><code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code> and <code class="docutils literal notranslate"><span class="pre">atomicrmw</span></code> are essentially like an atomic load followed by an
atomic store (where the store is conditional for <code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code>), but no other
memory operation can happen on any thread between the load and store.</p>
<p>A <code class="docutils literal notranslate"><span class="pre">fence</span></code> provides Acquire and/or Release ordering which is not part of
another operation; it is normally used along with Monotonic memory operations.
A Monotonic load followed by an Acquire fence is roughly equivalent to an
Acquire load, and a Monotonic store following a Release fence is roughly
equivalent to a Release store. SequentiallyConsistent fences behave as both
an Acquire and a Release fence, and offer some additional complicated
guarantees, see the C++11 standard for details.</p>
<p>Frontends generating atomic instructions generally need to be aware of the
target to some degree; atomic instructions are guaranteed to be lock-free, and
therefore an instruction which is wider than the target natively supports can be
impossible to generate.</p>
</div>
<div class="section" id="atomic-orderings">
<span id="id2"></span><h2><a class="toc-backref" href="#id7">Atomic orderings</a><a class="headerlink" href="#atomic-orderings" title="Permalink to this headline">¶</a></h2>
<p>In order to achieve a balance between performance and necessary guarantees,
there are six levels of atomicity. They are listed in order of strength; each
level includes all the guarantees of the previous level except for
Acquire/Release. (See also <a class="reference external" href="LangRef.html#ordering">LangRef Ordering</a>.)</p>
<div class="section" id="notatomic">
<span id="id3"></span><h3><a class="toc-backref" href="#id8">NotAtomic</a><a class="headerlink" href="#notatomic" title="Permalink to this headline">¶</a></h3>
<p>NotAtomic is the obvious, a load or store which is not atomic. (This isn’t
really a level of atomicity, but is listed here for comparison.) This is
essentially a regular load or store. If there is a race on a given memory
location, loads from that location return undef.</p>
<dl class="simple">
<dt>Relevant standard</dt><dd><p>This is intended to match shared variables in C/C++, and to be used in any
other context where memory access is necessary, and a race is impossible. (The
precise definition is in <a class="reference external" href="LangRef.html#memmodel">LangRef Memory Model</a>.)</p>
</dd>
<dt>Notes for frontends</dt><dd><p>The rule is essentially that all memory accessed with basic loads and stores
by multiple threads should be protected by a lock or other synchronization;
otherwise, you are likely to run into undefined behavior. If your frontend is
for a “safe” language like Java, use Unordered to load and store any shared
variable. Note that NotAtomic volatile loads and stores are not properly
atomic; do not try to use them as a substitute. (Per the C/C++ standards,
volatile does provide some limited guarantees around asynchronous signals, but
atomics are generally a better solution.)</p>
</dd>
<dt>Notes for optimizers</dt><dd><p>Introducing loads to shared variables along a codepath where they would not
otherwise exist is allowed; introducing stores to shared variables is not. See
<a class="reference internal" href="#optimization-outside-atomic">Optimization outside atomic</a>.</p>
</dd>
<dt>Notes for code generation</dt><dd><p>The one interesting restriction here is that it is not allowed to write to
bytes outside of the bytes relevant to a store. This is mostly relevant to
unaligned stores: it is not allowed in general to convert an unaligned store
into two aligned stores of the same width as the unaligned store. Backends are
also expected to generate an i8 store as an i8 store, and not an instruction
which writes to surrounding bytes. (If you are writing a backend for an
architecture which cannot satisfy these restrictions and cares about
concurrency, please send an email to llvm-dev.)</p>
</dd>
</dl>
</div>
<div class="section" id="unordered">
<h3><a class="toc-backref" href="#id9">Unordered</a><a class="headerlink" href="#unordered" title="Permalink to this headline">¶</a></h3>
<p>Unordered is the lowest level of atomicity. It essentially guarantees that races
produce somewhat sane results instead of having undefined behavior. It also
guarantees the operation to be lock-free, so it does not depend on the data
being part of a special atomic structure or depend on a separate per-process
global lock. Note that code generation will fail for unsupported atomic
operations; if you need such an operation, use explicit locking.</p>
<dl class="simple">
<dt>Relevant standard</dt><dd><p>This is intended to match the Java memory model for shared variables.</p>
</dd>
<dt>Notes for frontends</dt><dd><p>This cannot be used for synchronization, but is useful for Java and other
“safe” languages which need to guarantee that the generated code never
exhibits undefined behavior. Note that this guarantee is cheap on common
platforms for loads of a native width, but can be expensive or unavailable for
wider loads, like a 64-bit store on ARM. (A frontend for Java or other “safe”
languages would normally split a 64-bit store on ARM into two 32-bit unordered
stores.)</p>
</dd>
<dt>Notes for optimizers</dt><dd><p>In terms of the optimizer, this prohibits any transformation that transforms a
single load into multiple loads, transforms a store into multiple stores,
narrows a store, or stores a value which would not be stored otherwise. Some
examples of unsafe optimizations are narrowing an assignment into a bitfield,
rematerializing a load, and turning loads and stores into a memcpy
call. Reordering unordered operations is safe, though, and optimizers should
take advantage of that because unordered operations are common in languages
that need them.</p>
</dd>
<dt>Notes for code generation</dt><dd><p>These operations are required to be atomic in the sense that if you use
unordered loads and unordered stores, a load cannot see a value which was
never stored. A normal load or store instruction is usually sufficient, but
note that an unordered load or store cannot be split into multiple
instructions (or an instruction which does multiple memory operations, like
<code class="docutils literal notranslate"><span class="pre">LDRD</span></code> on ARM without LPAE, or not naturally-aligned <code class="docutils literal notranslate"><span class="pre">LDRD</span></code> on LPAE ARM).</p>
</dd>
</dl>
</div>
<div class="section" id="monotonic">
<h3><a class="toc-backref" href="#id10">Monotonic</a><a class="headerlink" href="#monotonic" title="Permalink to this headline">¶</a></h3>
<p>Monotonic is the weakest level of atomicity that can be used in synchronization
primitives, although it does not provide any general synchronization. It
essentially guarantees that if you take all the operations affecting a specific
address, a consistent ordering exists.</p>
<dl class="simple">
<dt>Relevant standard</dt><dd><p>This corresponds to the C++11/C11 <code class="docutils literal notranslate"><span class="pre">memory_order_relaxed</span></code>; see those
standards for the exact definition.</p>
</dd>
<dt>Notes for frontends</dt><dd><p>If you are writing a frontend which uses this directly, use with caution. The
guarantees in terms of synchronization are very weak, so make sure these are
only used in a pattern which you know is correct. Generally, these would
either be used for atomic operations which do not protect other memory (like
an atomic counter), or along with a <code class="docutils literal notranslate"><span class="pre">fence</span></code>.</p>
</dd>
<dt>Notes for optimizers</dt><dd><p>In terms of the optimizer, this can be treated as a read+write on the relevant
memory location (and alias analysis will take advantage of that). In addition,
it is legal to reorder non-atomic and Unordered loads around Monotonic
loads. CSE/DSE and a few other optimizations are allowed, but Monotonic
operations are unlikely to be used in ways which would make those
optimizations useful.</p>
</dd>
<dt>Notes for code generation</dt><dd><p>Code generation is essentially the same as that for unordered for loads and
stores. No fences are required. <code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code> and <code class="docutils literal notranslate"><span class="pre">atomicrmw</span></code> are required
to appear as a single operation.</p>
</dd>
</dl>
</div>
<div class="section" id="acquire">
<h3><a class="toc-backref" href="#id11">Acquire</a><a class="headerlink" href="#acquire" title="Permalink to this headline">¶</a></h3>
<p>Acquire provides a barrier of the sort necessary to acquire a lock to access
other memory with normal loads and stores.</p>
<dl class="simple">
<dt>Relevant standard</dt><dd><p>This corresponds to the C++11/C11 <code class="docutils literal notranslate"><span class="pre">memory_order_acquire</span></code>. It should also be
used for C++11/C11 <code class="docutils literal notranslate"><span class="pre">memory_order_consume</span></code>.</p>
</dd>
<dt>Notes for frontends</dt><dd><p>If you are writing a frontend which uses this directly, use with caution.
Acquire only provides a semantic guarantee when paired with a Release
operation.</p>
</dd>
<dt>Notes for optimizers</dt><dd><p>Optimizers not aware of atomics can treat this like a nothrow call. It is
also possible to move stores from before an Acquire load or read-modify-write
operation to after it, and move non-Acquire loads from before an Acquire
operation to after it.</p>
</dd>
<dt>Notes for code generation</dt><dd><p>Architectures with weak memory ordering (essentially everything relevant today
except x86 and SPARC) require some sort of fence to maintain the Acquire
semantics. The precise fences required varies widely by architecture, but for
a simple implementation, most architectures provide a barrier which is strong
enough for everything (<code class="docutils literal notranslate"><span class="pre">dmb</span></code> on ARM, <code class="docutils literal notranslate"><span class="pre">sync</span></code> on PowerPC, etc.). Putting
such a fence after the equivalent Monotonic operation is sufficient to
maintain Acquire semantics for a memory operation.</p>
</dd>
</dl>
</div>
<div class="section" id="release">
<h3><a class="toc-backref" href="#id12">Release</a><a class="headerlink" href="#release" title="Permalink to this headline">¶</a></h3>
<p>Release is similar to Acquire, but with a barrier of the sort necessary to
release a lock.</p>
<dl class="simple">
<dt>Relevant standard</dt><dd><p>This corresponds to the C++11/C11 <code class="docutils literal notranslate"><span class="pre">memory_order_release</span></code>.</p>
</dd>
<dt>Notes for frontends</dt><dd><p>If you are writing a frontend which uses this directly, use with caution.
Release only provides a semantic guarantee when paired with a Acquire
operation.</p>
</dd>
<dt>Notes for optimizers</dt><dd><p>Optimizers not aware of atomics can treat this like a nothrow call. It is
also possible to move loads from after a Release store or read-modify-write
operation to before it, and move non-Release stores from after a Release
operation to before it.</p>
</dd>
<dt>Notes for code generation</dt><dd><p>See the section on Acquire; a fence before the relevant operation is usually
sufficient for Release. Note that a store-store fence is not sufficient to
implement Release semantics; store-store fences are generally not exposed to
IR because they are extremely difficult to use correctly.</p>
</dd>
</dl>
</div>
<div class="section" id="acquirerelease">
<h3><a class="toc-backref" href="#id13">AcquireRelease</a><a class="headerlink" href="#acquirerelease" title="Permalink to this headline">¶</a></h3>
<p>AcquireRelease (<code class="docutils literal notranslate"><span class="pre">acq_rel</span></code> in IR) provides both an Acquire and a Release
barrier (for fences and operations which both read and write memory).</p>
<dl class="simple">
<dt>Relevant standard</dt><dd><p>This corresponds to the C++11/C11 <code class="docutils literal notranslate"><span class="pre">memory_order_acq_rel</span></code>.</p>
</dd>
<dt>Notes for frontends</dt><dd><p>If you are writing a frontend which uses this directly, use with caution.
Acquire only provides a semantic guarantee when paired with a Release
operation, and vice versa.</p>
</dd>
<dt>Notes for optimizers</dt><dd><p>In general, optimizers should treat this like a nothrow call; the possible
optimizations are usually not interesting.</p>
</dd>
<dt>Notes for code generation</dt><dd><p>This operation has Acquire and Release semantics; see the sections on Acquire
and Release.</p>
</dd>
</dl>
</div>
<div class="section" id="sequentiallyconsistent">
<h3><a class="toc-backref" href="#id14">SequentiallyConsistent</a><a class="headerlink" href="#sequentiallyconsistent" title="Permalink to this headline">¶</a></h3>
<p>SequentiallyConsistent (<code class="docutils literal notranslate"><span class="pre">seq_cst</span></code> in IR) provides Acquire semantics for loads
and Release semantics for stores. Additionally, it guarantees that a total
ordering exists between all SequentiallyConsistent operations.</p>
<dl class="simple">
<dt>Relevant standard</dt><dd><p>This corresponds to the C++11/C11 <code class="docutils literal notranslate"><span class="pre">memory_order_seq_cst</span></code>, Java volatile, and
the gcc-compatible <code class="docutils literal notranslate"><span class="pre">__sync_*</span></code> builtins which do not specify otherwise.</p>
</dd>
<dt>Notes for frontends</dt><dd><p>If a frontend is exposing atomic operations, these are much easier to reason
about for the programmer than other kinds of operations, and using them is
generally a practical performance tradeoff.</p>
</dd>
<dt>Notes for optimizers</dt><dd><p>Optimizers not aware of atomics can treat this like a nothrow call. For
SequentiallyConsistent loads and stores, the same reorderings are allowed as
for Acquire loads and Release stores, except that SequentiallyConsistent
operations may not be reordered.</p>
</dd>
<dt>Notes for code generation</dt><dd><p>SequentiallyConsistent loads minimally require the same barriers as Acquire
operations and SequentiallyConsistent stores require Release
barriers. Additionally, the code generator must enforce ordering between
SequentiallyConsistent stores followed by SequentiallyConsistent loads. This
is usually done by emitting either a full fence before the loads or a full
fence after the stores; which is preferred varies by architecture.</p>
</dd>
</dl>
</div>
</div>
<div class="section" id="atomics-and-ir-optimization">
<h2><a class="toc-backref" href="#id15">Atomics and IR optimization</a><a class="headerlink" href="#atomics-and-ir-optimization" title="Permalink to this headline">¶</a></h2>
<p>Predicates for optimizer writers to query:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">isSimple()</span></code>: A load or store which is not volatile or atomic. This is
what, for example, memcpyopt would check for operations it might transform.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">isUnordered()</span></code>: A load or store which is not volatile and at most
Unordered. This would be checked, for example, by LICM before hoisting an
operation.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">mayReadFromMemory()</span></code>/<code class="docutils literal notranslate"><span class="pre">mayWriteToMemory()</span></code>: Existing predicate, but note
that they return true for any operation which is volatile or at least
Monotonic.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">isStrongerThan</span></code> / <code class="docutils literal notranslate"><span class="pre">isAtLeastOrStrongerThan</span></code>: These are predicates on
orderings. They can be useful for passes that are aware of atomics, for
example to do DSE across a single atomic access, but not across a
release-acquire pair (see MemoryDependencyAnalysis for an example of this)</p></li>
<li><p>Alias analysis: Note that AA will return ModRef for anything Acquire or
Release, and for the address accessed by any Monotonic operation.</p></li>
</ul>
<p>To support optimizing around atomic operations, make sure you are using the
right predicates; everything should work if that is done. If your pass should
optimize some atomic operations (Unordered operations in particular), make sure
it doesn’t replace an atomic load or store with a non-atomic operation.</p>
<p>Some examples of how optimizations interact with various kinds of atomic
operations:</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">memcpyopt</span></code>: An atomic operation cannot be optimized into part of a
memcpy/memset, including unordered loads/stores. It can pull operations
across some atomic operations.</p></li>
<li><p>LICM: Unordered loads/stores can be moved out of a loop. It just treats
monotonic operations like a read+write to a memory location, and anything
stricter than that like a nothrow call.</p></li>
<li><p>DSE: Unordered stores can be DSE’ed like normal stores. Monotonic stores can
be DSE’ed in some cases, but it’s tricky to reason about, and not especially
important. It is possible in some case for DSE to operate across a stronger
atomic operation, but it is fairly tricky. DSE delegates this reasoning to
MemoryDependencyAnalysis (which is also used by other passes like GVN).</p></li>
<li><p>Folding a load: Any atomic load from a constant global can be constant-folded,
because it cannot be observed. Similar reasoning allows sroa with
atomic loads and stores.</p></li>
</ul>
</div>
<div class="section" id="atomics-and-codegen">
<h2><a class="toc-backref" href="#id16">Atomics and Codegen</a><a class="headerlink" href="#atomics-and-codegen" title="Permalink to this headline">¶</a></h2>
<p>Atomic operations are represented in the SelectionDAG with <code class="docutils literal notranslate"><span class="pre">ATOMIC_*</span></code> opcodes.
On architectures which use barrier instructions for all atomic ordering (like
ARM), appropriate fences can be emitted by the AtomicExpand Codegen pass if
<code class="docutils literal notranslate"><span class="pre">setInsertFencesForAtomic()</span></code> was used.</p>
<p>The MachineMemOperand for all atomic operations is currently marked as volatile;
this is not correct in the IR sense of volatile, but CodeGen handles anything
marked volatile very conservatively. This should get fixed at some point.</p>
<p>One very important property of the atomic operations is that if your backend
supports any inline lock-free atomic operations of a given size, you should
support <em>ALL</em> operations of that size in a lock-free manner.</p>
<p>When the target implements atomic <code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code> or LL/SC instructions (as most do)
this is trivial: all the other operations can be implemented on top of those
primitives. However, on many older CPUs (e.g. ARMv5, SparcV8, Intel 80386) there
are atomic load and store instructions, but no <code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code> or LL/SC. As it is
invalid to implement <code class="docutils literal notranslate"><span class="pre">atomic</span> <span class="pre">load</span></code> using the native instruction, but
<code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code> using a library call to a function that uses a mutex, <code class="docutils literal notranslate"><span class="pre">atomic</span>
<span class="pre">load</span></code> must <em>also</em> expand to a library call on such architectures, so that it
can remain atomic with regards to a simultaneous <code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code>, by using the same
mutex.</p>
<p>AtomicExpandPass can help with that: it will expand all atomic operations to the
proper <code class="docutils literal notranslate"><span class="pre">__atomic_*</span></code> libcalls for any size above the maximum set by
<code class="docutils literal notranslate"><span class="pre">setMaxAtomicSizeInBitsSupported</span></code> (which defaults to 0).</p>
<p>On x86, all atomic loads generate a <code class="docutils literal notranslate"><span class="pre">MOV</span></code>. SequentiallyConsistent stores
generate an <code class="docutils literal notranslate"><span class="pre">XCHG</span></code>, other stores generate a <code class="docutils literal notranslate"><span class="pre">MOV</span></code>. SequentiallyConsistent
fences generate an <code class="docutils literal notranslate"><span class="pre">MFENCE</span></code>, other fences do not cause any code to be
generated. <code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code> uses the <code class="docutils literal notranslate"><span class="pre">LOCK</span> <span class="pre">CMPXCHG</span></code> instruction. <code class="docutils literal notranslate"><span class="pre">atomicrmw</span> <span class="pre">xchg</span></code>
uses <code class="docutils literal notranslate"><span class="pre">XCHG</span></code>, <code class="docutils literal notranslate"><span class="pre">atomicrmw</span> <span class="pre">add</span></code> and <code class="docutils literal notranslate"><span class="pre">atomicrmw</span> <span class="pre">sub</span></code> use <code class="docutils literal notranslate"><span class="pre">XADD</span></code>, and all
other <code class="docutils literal notranslate"><span class="pre">atomicrmw</span></code> operations generate a loop with <code class="docutils literal notranslate"><span class="pre">LOCK</span> <span class="pre">CMPXCHG</span></code>. Depending
on the users of the result, some <code class="docutils literal notranslate"><span class="pre">atomicrmw</span></code> operations can be translated into
operations like <code class="docutils literal notranslate"><span class="pre">LOCK</span> <span class="pre">AND</span></code>, but that does not work in general.</p>
<p>On ARM (before v8), MIPS, and many other RISC architectures, Acquire, Release,
and SequentiallyConsistent semantics require barrier instructions for every such
operation. Loads and stores generate normal instructions. <code class="docutils literal notranslate"><span class="pre">cmpxchg</span></code> and
<code class="docutils literal notranslate"><span class="pre">atomicrmw</span></code> can be represented using a loop with LL/SC-style instructions
which take some sort of exclusive lock on a cache line (<code class="docutils literal notranslate"><span class="pre">LDREX</span></code> and <code class="docutils literal notranslate"><span class="pre">STREX</span></code>
on ARM, etc.).</p>
<p>It is often easiest for backends to use AtomicExpandPass to lower some of the
atomic constructs. Here are some lowerings it can do:</p>
<ul class="simple">
<li><p>cmpxchg -> loop with load-linked/store-conditional
by overriding <code class="docutils literal notranslate"><span class="pre">shouldExpandAtomicCmpXchgInIR()</span></code>, <code class="docutils literal notranslate"><span class="pre">emitLoadLinked()</span></code>,
<code class="docutils literal notranslate"><span class="pre">emitStoreConditional()</span></code></p></li>
<li><p>large loads/stores -> ll-sc/cmpxchg
by overriding <code class="docutils literal notranslate"><span class="pre">shouldExpandAtomicStoreInIR()</span></code>/<code class="docutils literal notranslate"><span class="pre">shouldExpandAtomicLoadInIR()</span></code></p></li>
<li><p>strong atomic accesses -> monotonic accesses + fences by overriding
<code class="docutils literal notranslate"><span class="pre">shouldInsertFencesForAtomic()</span></code>, <code class="docutils literal notranslate"><span class="pre">emitLeadingFence()</span></code>, and
<code class="docutils literal notranslate"><span class="pre">emitTrailingFence()</span></code></p></li>
<li><p>atomic rmw -> loop with cmpxchg or load-linked/store-conditional
by overriding <code class="docutils literal notranslate"><span class="pre">expandAtomicRMWInIR()</span></code></p></li>
<li><p>expansion to __atomic_* libcalls for unsupported sizes.</p></li>
<li><p>part-word atomicrmw/cmpxchg -> target-specific intrinsic by overriding
<code class="docutils literal notranslate"><span class="pre">shouldExpandAtomicRMWInIR</span></code>, <code class="docutils literal notranslate"><span class="pre">emitMaskedAtomicRMWIntrinsic</span></code>,
<code class="docutils literal notranslate"><span class="pre">shouldExpandAtomicCmpXchgInIR</span></code>, and <code class="docutils literal notranslate"><span class="pre">emitMaskedAtomicCmpXchgIntrinsic</span></code>.</p></li>
</ul>
<p>For an example of these look at the ARM (first five lowerings) or RISC-V (last
lowering) backend.</p>
<p>AtomicExpandPass supports two strategies for lowering atomicrmw/cmpxchg to
load-linked/store-conditional (LL/SC) loops. The first expands the LL/SC loop
in IR, calling target lowering hooks to emit intrinsics for the LL and SC
operations. However, many architectures have strict requirements for LL/SC
loops to ensure forward progress, such as restrictions on the number and type
of instructions in the loop. It isn’t possible to enforce these restrictions
when the loop is expanded in LLVM IR, and so affected targets may prefer to
expand to LL/SC loops at a very late stage (i.e. after register allocation).
AtomicExpandPass can help support lowering of part-word atomicrmw or cmpxchg
using this strategy by producing IR for any shifting and masking that can be
performed outside of the LL/SC loop.</p>
</div>
<div class="section" id="libcalls-atomic">
<h2><a class="toc-backref" href="#id17">Libcalls: __atomic_*</a><a class="headerlink" href="#libcalls-atomic" title="Permalink to this headline">¶</a></h2>
<p>There are two kinds of atomic library calls that are generated by LLVM. Please
note that both sets of library functions somewhat confusingly share the names of
builtin functions defined by clang. Despite this, the library functions are
not directly related to the builtins: it is <em>not</em> the case that <code class="docutils literal notranslate"><span class="pre">__atomic_*</span></code>
builtins lower to <code class="docutils literal notranslate"><span class="pre">__atomic_*</span></code> library calls and <code class="docutils literal notranslate"><span class="pre">__sync_*</span></code> builtins lower
to <code class="docutils literal notranslate"><span class="pre">__sync_*</span></code> library calls.</p>
<p>The first set of library functions are named <code class="docutils literal notranslate"><span class="pre">__atomic_*</span></code>. This set has been
“standardized” by GCC, and is described below. (See also <a class="reference external" href="https://gcc.gnu.org/wiki/Atomic/GCCMM/LIbrary">GCC’s documentation</a>)</p>
<p>LLVM’s AtomicExpandPass will translate atomic operations on data sizes above
<code class="docutils literal notranslate"><span class="pre">MaxAtomicSizeInBitsSupported</span></code> into calls to these functions.</p>
<p>There are four generic functions, which can be called with data of any size or
alignment:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">void</span> <span class="n">__atomic_load</span><span class="p">(</span><span class="n">size_t</span> <span class="n">size</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">ret</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">void</span> <span class="n">__atomic_store</span><span class="p">(</span><span class="n">size_t</span> <span class="n">size</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">void</span> <span class="n">__atomic_exchange</span><span class="p">(</span><span class="n">size_t</span> <span class="n">size</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">val</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">ret</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="nb">bool</span> <span class="n">__atomic_compare_exchange</span><span class="p">(</span><span class="n">size_t</span> <span class="n">size</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">expected</span><span class="p">,</span> <span class="n">void</span> <span class="o">*</span><span class="n">desired</span><span class="p">,</span> <span class="nb">int</span> <span class="n">success_order</span><span class="p">,</span> <span class="nb">int</span> <span class="n">failure_order</span><span class="p">)</span>
</pre></div>
</div>
<p>There are also size-specialized versions of the above functions, which can only
be used with <em>naturally-aligned</em> pointers of the appropriate size. In the
signatures below, “N” is one of 1, 2, 4, 8, and 16, and “iN” is the appropriate
integer type of that size; if no such integer type exists, the specialization
cannot be used:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">iN</span> <span class="n">__atomic_load_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">void</span> <span class="n">__atomic_store_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__atomic_exchange_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="nb">bool</span> <span class="n">__atomic_compare_exchange_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="o">*</span><span class="n">expected</span><span class="p">,</span> <span class="n">iN</span> <span class="n">desired</span><span class="p">,</span> <span class="nb">int</span> <span class="n">success_order</span><span class="p">,</span> <span class="nb">int</span> <span class="n">failure_order</span><span class="p">)</span>
</pre></div>
</div>
<p>Finally there are some read-modify-write functions, which are only available in
the size-specific variants (any other sizes use a <code class="docutils literal notranslate"><span class="pre">__atomic_compare_exchange</span></code>
loop):</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">iN</span> <span class="n">__atomic_fetch_add_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__atomic_fetch_sub_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__atomic_fetch_and_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__atomic_fetch_or_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__atomic_fetch_xor_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__atomic_fetch_nand_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="nb">int</span> <span class="n">ordering</span><span class="p">)</span>
</pre></div>
</div>
<p>This set of library functions have some interesting implementation requirements
to take note of:</p>
<ul class="simple">
<li><p>They support all sizes and alignments – including those which cannot be
implemented natively on any existing hardware. Therefore, they will certainly
use mutexes in for some sizes/alignments.</p></li>
<li><p>As a consequence, they cannot be shipped in a statically linked
compiler-support library, as they have state which must be shared amongst all
DSOs loaded in the program. They must be provided in a shared library used by
all objects.</p></li>
<li><p>The set of atomic sizes supported lock-free must be a superset of the sizes
any compiler can emit. That is: if a new compiler introduces support for
inline-lock-free atomics of size N, the <code class="docutils literal notranslate"><span class="pre">__atomic_*</span></code> functions must also have a
lock-free implementation for size N. This is a requirement so that code
produced by an old compiler (which will have called the <code class="docutils literal notranslate"><span class="pre">__atomic_*</span></code> function)
interoperates with code produced by the new compiler (which will use native
the atomic instruction).</p></li>
</ul>
<p>Note that it’s possible to write an entirely target-independent implementation
of these library functions by using the compiler atomic builtins themselves to
implement the operations on naturally-aligned pointers of supported sizes, and a
generic mutex implementation otherwise.</p>
</div>
<div class="section" id="libcalls-sync">
<h2><a class="toc-backref" href="#id18">Libcalls: __sync_*</a><a class="headerlink" href="#libcalls-sync" title="Permalink to this headline">¶</a></h2>
<p>Some targets or OS/target combinations can support lock-free atomics, but for
various reasons, it is not practical to emit the instructions inline.</p>
<p>There’s two typical examples of this.</p>
<p>Some CPUs support multiple instruction sets which can be switched back and forth
on function-call boundaries. For example, MIPS supports the MIPS16 ISA, which
has a smaller instruction encoding than the usual MIPS32 ISA. ARM, similarly,
has the Thumb ISA. In MIPS16 and earlier versions of Thumb, the atomic
instructions are not encodable. However, those instructions are available via a
function call to a function with the longer encoding.</p>
<p>Additionally, a few OS/target pairs provide kernel-supported lock-free
atomics. ARM/Linux is an example of this: the kernel <a class="reference external" href="https://www.kernel.org/doc/Documentation/arm/kernel_user_helpers.txt">provides</a> a
function which on older CPUs contains a “magically-restartable” atomic sequence
(which looks atomic so long as there’s only one CPU), and contains actual atomic
instructions on newer multicore models. This sort of functionality can typically
be provided on any architecture, if all CPUs which are missing atomic
compare-and-swap support are uniprocessor (no SMP). This is almost always the
case. The only common architecture without that property is SPARC – SPARCV8 SMP
systems were common, yet it doesn’t support any sort of compare-and-swap
operation.</p>
<p>In either of these cases, the Target in LLVM can claim support for atomics of an
appropriate size, and then implement some subset of the operations via libcalls
to a <code class="docutils literal notranslate"><span class="pre">__sync_*</span></code> function. Such functions <em>must</em> not use locks in their
implementation, because unlike the <code class="docutils literal notranslate"><span class="pre">__atomic_*</span></code> routines used by
AtomicExpandPass, these may be mixed-and-matched with native instructions by the
target lowering.</p>
<p>Further, these routines do not need to be shared, as they are stateless. So,
there is no issue with having multiple copies included in one binary. Thus,
typically these routines are implemented by the statically-linked compiler
runtime support library.</p>
<p>LLVM will emit a call to an appropriate <code class="docutils literal notranslate"><span class="pre">__sync_*</span></code> routine if the target
ISelLowering code has set the corresponding <code class="docutils literal notranslate"><span class="pre">ATOMIC_CMPXCHG</span></code>, <code class="docutils literal notranslate"><span class="pre">ATOMIC_SWAP</span></code>,
or <code class="docutils literal notranslate"><span class="pre">ATOMIC_LOAD_*</span></code> operation to “Expand”, and if it has opted-into the
availability of those library functions via a call to <code class="docutils literal notranslate"><span class="pre">initSyncLibcalls()</span></code>.</p>
<p>The full set of functions that may be called by LLVM is (for <code class="docutils literal notranslate"><span class="pre">N</span></code> being 1, 2,
4, 8, or 16):</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">iN</span> <span class="n">__sync_val_compare_and_swap_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">expected</span><span class="p">,</span> <span class="n">iN</span> <span class="n">desired</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_lock_test_and_set_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_add_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_sub_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_and_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_or_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_xor_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_nand_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_max_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_umax_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_min_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__sync_fetch_and_umin_N</span><span class="p">(</span><span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">,</span> <span class="n">iN</span> <span class="n">val</span><span class="p">)</span>
</pre></div>
</div>
<p>This list doesn’t include any function for atomic load or store; all known
architectures support atomic loads and stores directly (possibly by emitting a
fence on either side of a normal load or store.)</p>
<p>There’s also, somewhat separately, the possibility to lower <code class="docutils literal notranslate"><span class="pre">ATOMIC_FENCE</span></code> to
<code class="docutils literal notranslate"><span class="pre">__sync_synchronize()</span></code>. This may happen or not happen independent of all the
above, controlled purely by <code class="docutils literal notranslate"><span class="pre">setOperationAction(ISD::ATOMIC_FENCE,</span> <span class="pre">...)</span></code>.</p>
<p>On AArch64, a variant of the __sync_* routines is used which contain the memory
order as part of the function name. These routines may determine at runtime
whether the single-instruction atomic operations which were introduced as part
of AArch64 Large System Extensions “LSE” instruction set are available, or if
it needs to fall back to an LL/SC loop. The following helper functions are
implemented in both <code class="docutils literal notranslate"><span class="pre">compiler-rt</span></code> and <code class="docutils literal notranslate"><span class="pre">libgcc</span></code> libraries
(<code class="docutils literal notranslate"><span class="pre">N</span></code> is one of 1, 2, 4, 8, and <code class="docutils literal notranslate"><span class="pre">M</span></code> is one of 1, 2, 4, 8 and 16, and
<code class="docutils literal notranslate"><span class="pre">ORDER</span></code> is one of ‘relax’, ‘acq’, ‘rel’, ‘acq_rel’):</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">iM</span> <span class="n">__aarch64_casM_ORDER</span><span class="p">(</span><span class="n">iM</span> <span class="n">expected</span><span class="p">,</span> <span class="n">iM</span> <span class="n">desired</span><span class="p">,</span> <span class="n">iM</span> <span class="o">*</span><span class="n">ptr</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__aarch64_swpN_ORDER</span><span class="p">(</span><span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__aarch64_ldaddN_ORDER</span><span class="p">(</span><span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__aarch64_ldclrN_ORDER</span><span class="p">(</span><span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__aarch64_ldeorN_ORDER</span><span class="p">(</span><span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">)</span>
<span class="n">iN</span> <span class="n">__aarch64_ldsetN_ORDER</span><span class="p">(</span><span class="n">iN</span> <span class="n">val</span><span class="p">,</span> <span class="n">iN</span> <span class="o">*</span><span class="n">ptr</span><span class="p">)</span>
</pre></div>
</div>
<p>Please note, if LSE instruction set is specified for AArch64 target then
out-of-line atomics calls are not generated and single-instruction atomic
operations are used in place.</p>
</div>
</div>
<div class="clearer"></div>
</div>
</div>
</div>
<div class="clearer"></div>
</div>
<div class="related" role="navigation" aria-label="related navigation">
<h3>Navigation</h3>
<ul>
<li class="right" style="margin-right: 10px">
<a href="genindex.html" title="General Index"
>index</a></li>
<li class="right" >
<a href="BitCodeFormat.html" title="LLVM Bitcode File Format"
>next</a> |</li>
<li class="right" >
<a href="Reference.html" title="Reference"
>previous</a> |</li>
<li><a href="https://llvm.org/">LLVM Home</a> | </li>
<li><a href="index.html">Documentation</a>»</li>
<li class="nav-item nav-item-1"><a href="Reference.html" >Reference</a> »</li>
<li class="nav-item nav-item-this"><a href="">LLVM Atomic Instructions and Concurrency Guide</a></li>
</ul>
</div>
<div class="footer" role="contentinfo">
© Copyright 2003-2021, LLVM Project.
Last updated on 2021-09-18.
Created using <a href="https://www.sphinx-doc.org/">Sphinx</a> 3.5.4.
</div>
</body>
</html>
|
