aboutsummaryrefslogtreecommitdiffhomepage
path: root/src/cubeb_aaudio.cpp
blob: 83ca85454ca34ea8fa3e4f7a4fed3a26a5e7a4d1 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
/* ex: set tabstop=2 shiftwidth=2 expandtab:
 * Copyright © 2019 Jan Kelling
 *
 * This program is made available under an ISC-style license.  See the
 * accompanying file LICENSE for details.
 */
#include "cubeb-internal.h"
#include "cubeb/cubeb.h"
#include "cubeb_android.h"
#include "cubeb_log.h"
#include "cubeb_resampler.h"
#include "cubeb_triple_buffer.h"
#include <aaudio/AAudio.h>
#include <atomic>
#include <cassert>
#include <chrono>
#include <condition_variable>
#include <cstdint>
#include <cstring>
#include <dlfcn.h>
#include <inttypes.h>
#include <limits>
#include <memory>
#include <mutex>
#include <thread>
#include <time.h>

#ifdef DISABLE_LIBAAUDIO_DLOPEN
#define WRAP(x) x
#else
#define WRAP(x) (*cubeb_##x)
#define LIBAAUDIO_API_VISIT(X)                                                 \
  X(AAudio_convertResultToText)                                                \
  X(AAudio_convertStreamStateToText)                                           \
  X(AAudio_createStreamBuilder)                                                \
  X(AAudioStreamBuilder_openStream)                                            \
  X(AAudioStreamBuilder_setChannelCount)                                       \
  X(AAudioStreamBuilder_setBufferCapacityInFrames)                             \
  X(AAudioStreamBuilder_setDirection)                                          \
  X(AAudioStreamBuilder_setFormat)                                             \
  X(AAudioStreamBuilder_setSharingMode)                                        \
  X(AAudioStreamBuilder_setPerformanceMode)                                    \
  X(AAudioStreamBuilder_setSampleRate)                                         \
  X(AAudioStreamBuilder_delete)                                                \
  X(AAudioStreamBuilder_setDataCallback)                                       \
  X(AAudioStreamBuilder_setErrorCallback)                                      \
  X(AAudioStream_close)                                                        \
  X(AAudioStream_read)                                                         \
  X(AAudioStream_requestStart)                                                 \
  X(AAudioStream_requestPause)                                                 \
  X(AAudioStream_setBufferSizeInFrames)                                        \
  X(AAudioStream_getTimestamp)                                                 \
  X(AAudioStream_requestFlush)                                                 \
  X(AAudioStream_requestStop)                                                  \
  X(AAudioStream_getPerformanceMode)                                           \
  X(AAudioStream_getSharingMode)                                               \
  X(AAudioStream_getBufferSizeInFrames)                                        \
  X(AAudioStream_getBufferCapacityInFrames)                                    \
  X(AAudioStream_getSampleRate)                                                \
  X(AAudioStream_waitForStateChange)                                           \
  X(AAudioStream_getFramesRead)                                                \
  X(AAudioStream_getState)                                                     \
  X(AAudioStream_getFramesWritten)                                             \
  X(AAudioStream_getFramesPerBurst)                                            \
  X(AAudioStreamBuilder_setInputPreset)                                        \
  X(AAudioStreamBuilder_setUsage)

// not needed or added later on
// X(AAudioStreamBuilder_setFramesPerDataCallback) \
  // X(AAudioStreamBuilder_setDeviceId)              \
  // X(AAudioStreamBuilder_setSamplesPerFrame)       \
  // X(AAudioStream_getSamplesPerFrame)              \
  // X(AAudioStream_getDeviceId)                     \
  // X(AAudioStream_write)                           \
  // X(AAudioStream_getChannelCount)                 \
  // X(AAudioStream_getFormat)                       \
  // X(AAudioStream_getXRunCount)                    \
  // X(AAudioStream_isMMapUsed)                      \
  // X(AAudioStreamBuilder_setContentType)           \
  // X(AAudioStreamBuilder_setSessionId)             \
  // X(AAudioStream_getUsage)                        \
  // X(AAudioStream_getContentType)                  \
  // X(AAudioStream_getInputPreset)                  \
  // X(AAudioStream_getSessionId)                    \
// END: not needed or added later on

#define MAKE_TYPEDEF(x) static decltype(x) * cubeb_##x;
LIBAAUDIO_API_VISIT(MAKE_TYPEDEF)
#undef MAKE_TYPEDEF
#endif

const uint8_t MAX_STREAMS = 16;
const int64_t NS_PER_S = static_cast<int64_t>(1e9);

using unique_lock = std::unique_lock<std::mutex>;
using lock_guard = std::lock_guard<std::mutex>;

enum class stream_state {
  INIT = 0,
  STOPPED,
  STOPPING,
  STARTED,
  STARTING,
  DRAINING,
  ERROR,
  SHUTDOWN,
};

struct AAudioTimingInfo {
  // The timestamp at which the audio engine last called the calback.
  uint64_t tstamp;
  // The number of output frames sent to the engine.
  uint64_t output_frame_index;
  // The current output latency in frames. 0 if there is no output stream.
  uint32_t output_latency;
  // The current input latency in frames. 0 if there is no input stream.
  uint32_t input_latency;
};

struct cubeb_stream {
  /* Note: Must match cubeb_stream layout in cubeb.c. */
  cubeb * context{};
  void * user_ptr{};

  std::atomic<bool> in_use{false};
  std::atomic<bool> latency_metrics_available{false};
  std::atomic<stream_state> state{stream_state::INIT};
  std::atomic<bool> in_data_callback{false};
  triple_buffer<AAudioTimingInfo> timing_info;

  AAudioStream * ostream{};
  AAudioStream * istream{};
  cubeb_data_callback data_callback{};
  cubeb_state_callback state_callback{};
  cubeb_resampler * resampler{};

  // mutex synchronizes access to the stream from the state thread
  // and user-called functions. Everything that is accessed in the
  // aaudio data (or error) callback is synchronized only via atomics.
  std::mutex mutex;

  std::unique_ptr<char[]> in_buf;
  unsigned in_frame_size{}; // size of one input frame

  cubeb_sample_format out_format{};
  uint32_t sample_rate{};
  std::atomic<float> volume{1.f};
  unsigned out_channels{};
  unsigned out_frame_size{};
  bool voice_input{};
  bool voice_output{};
  uint64_t previous_clock{};
};

struct cubeb {
  struct cubeb_ops const * ops{};
  void * libaaudio{};

  struct {
    // The state thread: it waits for state changes and stops
    // drained streams.
    std::thread thread;
    std::thread notifier;
    std::mutex mutex;
    std::condition_variable cond;
    std::atomic<bool> join{false};
    std::atomic<bool> waiting{false};
  } state;

  // streams[i].in_use signals whether a stream is used
  struct cubeb_stream streams[MAX_STREAMS];
};

struct AutoInCallback {
  AutoInCallback(cubeb_stream * stm) : stm(stm)
  {
    stm->in_data_callback.store(true);
  }
  ~AutoInCallback() { stm->in_data_callback.store(false); }
  cubeb_stream * stm;
};

// Returns when aaudio_stream's state is equal to desired_state.
// poll_frequency_ns is the duration that is slept in between asking for
// state updates and getting the new state.
// When waiting for a stream to stop, it is best to pick a value similar
// to the callback time because STOPPED will happen after
// draining.
static int
wait_for_state_change(AAudioStream * aaudio_stream,
                      aaudio_stream_state_t desired_state,
                      int64_t poll_frequency_ns)
{
  aaudio_stream_state_t new_state;
  do {
    aaudio_result_t res = WRAP(AAudioStream_waitForStateChange)(
        aaudio_stream, AAUDIO_STREAM_STATE_UNKNOWN, &new_state,
        poll_frequency_ns);
    if (res != AAUDIO_OK) {
      LOG("AAudioStream_waitForStateChanged: %s",
          WRAP(AAudio_convertResultToText)(res));
      return CUBEB_ERROR;
    }
  } while (new_state != desired_state);

  LOG("wait_for_state_change: current state now: %s",
      cubeb_AAudio_convertStreamStateToText(new_state));

  return CUBEB_OK;
}

// Only allowed from state thread, while mutex on stm is locked
static void
shutdown_with_error(cubeb_stream * stm)
{
  if (stm->istream) {
    WRAP(AAudioStream_requestStop)(stm->istream);
  }
  if (stm->ostream) {
    WRAP(AAudioStream_requestStop)(stm->ostream);
  }

  int64_t poll_frequency_ns = NS_PER_S * stm->out_frame_size / stm->sample_rate;
  if (stm->istream) {
    wait_for_state_change(stm->istream, AAUDIO_STREAM_STATE_STOPPED,
                          poll_frequency_ns);
  }
  if (stm->ostream) {
    wait_for_state_change(stm->ostream, AAUDIO_STREAM_STATE_STOPPED,
                          poll_frequency_ns);
  }

  assert(!stm->in_data_callback.load());
  stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_ERROR);
  stm->state.store(stream_state::SHUTDOWN);
}

// Returns whether the given state is one in which we wait for
// an asynchronous change
static bool
waiting_state(stream_state state)
{
  switch (state) {
  case stream_state::DRAINING:
  case stream_state::STARTING:
  case stream_state::STOPPING:
    return true;
  default:
    return false;
  }
}

static void
update_state(cubeb_stream * stm)
{
  // Fast path for streams that don't wait for state change or are invalid
  enum stream_state old_state = stm->state.load();
  if (old_state == stream_state::INIT || old_state == stream_state::STARTED ||
      old_state == stream_state::STOPPED ||
      old_state == stream_state::SHUTDOWN) {
    return;
  }

  // If the main thread currently operates on this thread, we don't
  // have to wait for it
  unique_lock lock(stm->mutex, std::try_to_lock);
  if (!lock.owns_lock()) {
    return;
  }

  // check again: if this is true now, the stream was destroyed or
  // changed between our fast path check and locking the mutex
  old_state = stm->state.load();
  if (old_state == stream_state::INIT || old_state == stream_state::STARTED ||
      old_state == stream_state::STOPPED ||
      old_state == stream_state::SHUTDOWN) {
    return;
  }

  // We compute the new state the stream has and then compare_exchange it
  // if it has changed. This way we will never just overwrite state
  // changes that were set from the audio thread in the meantime,
  // such as a DRAINING or error state.
  enum stream_state new_state;
  do {
    if (old_state == stream_state::SHUTDOWN) {
      return;
    }

    if (old_state == stream_state::ERROR) {
      shutdown_with_error(stm);
      return;
    }

    new_state = old_state;

    aaudio_stream_state_t istate = 0;
    aaudio_stream_state_t ostate = 0;

    // We use waitForStateChange (with zero timeout) instead of just
    // getState since only the former internally updates the state.
    // See the docs of aaudio getState/waitForStateChange for details,
    // why we are passing STATE_UNKNOWN.
    aaudio_result_t res;
    if (stm->istream) {
      res = WRAP(AAudioStream_waitForStateChange)(
          stm->istream, AAUDIO_STREAM_STATE_UNKNOWN, &istate, 0);
      if (res != AAUDIO_OK) {
        LOG("AAudioStream_waitForStateChanged: %s",
            WRAP(AAudio_convertResultToText)(res));
        return;
      }
      assert(istate);
    }

    if (stm->ostream) {
      res = WRAP(AAudioStream_waitForStateChange)(
          stm->ostream, AAUDIO_STREAM_STATE_UNKNOWN, &ostate, 0);
      if (res != AAUDIO_OK) {
        LOG("AAudioStream_waitForStateChanged: %s",
            WRAP(AAudio_convertResultToText)(res));
        return;
      }
      assert(ostate);
    }

    // handle invalid stream states
    if (istate == AAUDIO_STREAM_STATE_PAUSING ||
        istate == AAUDIO_STREAM_STATE_PAUSED ||
        istate == AAUDIO_STREAM_STATE_FLUSHING ||
        istate == AAUDIO_STREAM_STATE_FLUSHED ||
        istate == AAUDIO_STREAM_STATE_UNKNOWN ||
        istate == AAUDIO_STREAM_STATE_DISCONNECTED) {
      LOG("Unexpected android input stream state %s",
          WRAP(AAudio_convertStreamStateToText)(istate));
      shutdown_with_error(stm);
      return;
    }

    if (ostate == AAUDIO_STREAM_STATE_PAUSING ||
        ostate == AAUDIO_STREAM_STATE_PAUSED ||
        ostate == AAUDIO_STREAM_STATE_FLUSHING ||
        ostate == AAUDIO_STREAM_STATE_FLUSHED ||
        ostate == AAUDIO_STREAM_STATE_UNKNOWN ||
        ostate == AAUDIO_STREAM_STATE_DISCONNECTED) {
      LOG("Unexpected android output stream state %s",
          WRAP(AAudio_convertStreamStateToText)(istate));
      shutdown_with_error(stm);
      return;
    }

    switch (old_state) {
    case stream_state::STARTING:
      if ((!istate || istate == AAUDIO_STREAM_STATE_STARTED) &&
          (!ostate || ostate == AAUDIO_STREAM_STATE_STARTED)) {
        stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_STARTED);
        new_state = stream_state::STARTED;
      }
      break;
    case stream_state::DRAINING:
      // The DRAINING state means that we want to stop the streams but
      // may not have done so yet.
      // The aaudio docs state that returning STOP from the callback isn't
      // enough, the stream has to be stopped from another thread
      // afterwards.
      // No callbacks are triggered anymore when requestStop returns.
      // That is important as we otherwise might read from a closed istream
      // for a duplex stream.
      // Therefor it is important to close ostream first.
      if (ostate && ostate != AAUDIO_STREAM_STATE_STOPPING &&
          ostate != AAUDIO_STREAM_STATE_STOPPED) {
        res = WRAP(AAudioStream_requestStop)(stm->ostream);
        if (res != AAUDIO_OK) {
          LOG("AAudioStream_requestStop: %s",
              WRAP(AAudio_convertResultToText)(res));
          return;
        }
      }
      if (istate && istate != AAUDIO_STREAM_STATE_STOPPING &&
          istate != AAUDIO_STREAM_STATE_STOPPED) {
        res = WRAP(AAudioStream_requestStop)(stm->istream);
        if (res != AAUDIO_OK) {
          LOG("AAudioStream_requestStop: %s",
              WRAP(AAudio_convertResultToText)(res));
          return;
        }
      }

      // we always wait until both streams are stopped until we
      // send CUBEB_STATE_DRAINED. Then we can directly transition
      // our logical state to STOPPED, not triggering
      // an additional CUBEB_STATE_STOPPED callback (which might
      // be unexpected for the user).
      if ((!ostate || ostate == AAUDIO_STREAM_STATE_STOPPED) &&
          (!istate || istate == AAUDIO_STREAM_STATE_STOPPED)) {
        new_state = stream_state::STOPPED;
        stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_DRAINED);
      }
      break;
    case stream_state::STOPPING:
      assert(!istate || istate == AAUDIO_STREAM_STATE_STOPPING ||
             istate == AAUDIO_STREAM_STATE_STOPPED);
      assert(!ostate || ostate == AAUDIO_STREAM_STATE_STOPPING ||
             ostate == AAUDIO_STREAM_STATE_STOPPED);
      if ((!istate || istate == AAUDIO_STREAM_STATE_STOPPED) &&
          (!ostate || ostate == AAUDIO_STREAM_STATE_STOPPED)) {
        stm->state_callback(stm, stm->user_ptr, CUBEB_STATE_STOPPED);
        new_state = stream_state::STOPPED;
      }
      break;
    default:
      assert(false && "Unreachable: invalid state");
    }
  } while (old_state != new_state &&
           !stm->state.compare_exchange_strong(old_state, new_state));
}

// See https://nyorain.github.io/lock-free-wakeup.html for a note
// why this is needed. The audio thread notifies the state thread about
// state changes and must not block. The state thread on the other hand should
// sleep until there is work to be done. So we need a lockfree producer
// and blocking producer. This can only be achieved safely with a new thread
// that only serves as notifier backup (in case the notification happens
// right between the state thread checking and going to sleep in which case
// this thread will kick in and signal it right again).
static void
notifier_thread(cubeb * ctx)
{
  unique_lock lock(ctx->state.mutex);

  while (!ctx->state.join.load()) {
    ctx->state.cond.wait(lock);
    if (ctx->state.waiting.load()) {
      // This must signal our state thread since there is no other
      // thread currently waiting on the condition variable.
      // The state change thread is guaranteed to be waiting since
      // we hold the mutex it locks when awake.
      ctx->state.cond.notify_one();
    }
  }

  // make sure other thread joins as well
  ctx->state.cond.notify_one();
  LOG("Exiting notifier thread");
}

static void
state_thread(cubeb * ctx)
{
  unique_lock lock(ctx->state.mutex);

  bool waiting = false;
  while (!ctx->state.join.load()) {
    waiting |= ctx->state.waiting.load();
    if (waiting) {
      ctx->state.waiting.store(false);
      waiting = false;
      for (auto & stream : ctx->streams) {
        cubeb_stream * stm = &stream;
        update_state(stm);
        waiting |= waiting_state(atomic_load(&stm->state));
      }

      // state changed from another thread, update again immediately
      if (ctx->state.waiting.load()) {
        waiting = true;
        continue;
      }

      // Not waiting for any change anymore: we can wait on the
      // condition variable without timeout
      if (!waiting) {
        continue;
      }

      // while any stream is waiting for state change we sleep with regular
      // timeouts. But we wake up immediately if signaled.
      // This might seem like a poor man's implementation of state change
      // waiting but (as of october 2020), the implementation of
      // AAudioStream_waitForStateChange is just sleeping with regular
      // timeouts as well:
      // https://android.googlesource.com/platform/frameworks/av/+/refs/heads/master/media/libaaudio/src/core/AudioStream.cpp
      auto dur = std::chrono::milliseconds(5);
      ctx->state.cond.wait_for(lock, dur);
    } else {
      ctx->state.cond.wait(lock);
    }
  }

  // make sure other thread joins as well
  ctx->state.cond.notify_one();
  LOG("Exiting state thread");
}

static char const *
aaudio_get_backend_id(cubeb * /* ctx */)
{
  return "aaudio";
}

static int
aaudio_get_max_channel_count(cubeb * ctx, uint32_t * max_channels)
{
  assert(ctx && max_channels);
  // NOTE: we might get more, AAudio docs don't specify anything.
  *max_channels = 2;
  return CUBEB_OK;
}

static void
aaudio_destroy(cubeb * ctx)
{
  assert(ctx);

#ifndef NDEBUG
  // make sure all streams were destroyed
  for (auto & stream : ctx->streams) {
    assert(!stream.in_use.load());
  }
#endif

  // broadcast joining to both threads
  // they will additionally signal each other before joining
  ctx->state.join.store(true);
  ctx->state.cond.notify_all();

  if (ctx->state.thread.joinable()) {
    ctx->state.thread.join();
  }
  if (ctx->state.notifier.joinable()) {
    ctx->state.notifier.join();
  }
#ifndef DISABLE_LIBAAUDIO_DLOPEN
  if (ctx->libaaudio) {
    dlclose(ctx->libaaudio);
  }
#endif
  delete ctx;
}

static void
apply_volume(cubeb_stream * stm, void * audio_data, uint32_t num_frames)
{
  float volume = stm->volume.load();
  // optimization: we don't have to change anything in this case
  if (volume == 1.f) {
    return;
  }

  switch (stm->out_format) {
  case CUBEB_SAMPLE_S16NE: {
    int16_t* integer_data = static_cast<int16_t *>(audio_data);
    for (uint32_t i = 0u; i < num_frames * stm->out_channels; ++i) {
      integer_data[i] =
          static_cast<int16_t>(static_cast<float>(integer_data[i]) * volume);
    }
    break;
  }
  case CUBEB_SAMPLE_FLOAT32NE:
    for (uint32_t i = 0u; i < num_frames * stm->out_channels; ++i) {
      (static_cast<float *>(audio_data))[i] *= volume;
    }
    break;
  default:
    assert(false && "Unreachable: invalid stream out_format");
  }
}

uint64_t
now_ns()
{
  using namespace std::chrono;
  return duration_cast<nanoseconds>(steady_clock::now().time_since_epoch())
      .count();
}

// To be called from the real-time audio callback
uint64_t
aaudio_get_latency(cubeb_stream * stm, aaudio_direction_t direction,
                   uint64_t tstamp_ns)
{
  bool is_output = direction == AAUDIO_DIRECTION_OUTPUT;
  int64_t hw_frame_index;
  int64_t hw_tstamp;
  AAudioStream * stream = is_output ? stm->ostream : stm->istream;
  // For an output stream (resp. input stream), get the number of frames
  // written to (resp read from) the hardware.
  int64_t app_frame_index = is_output
                                ? WRAP(AAudioStream_getFramesWritten)(stream)
                                : WRAP(AAudioStream_getFramesRead)(stream);

  assert(tstamp_ns < std::numeric_limits<uint64_t>::max());
  int64_t signed_tstamp_ns = static_cast<int64_t>(tstamp_ns);

  // Get a timestamp for a particular frame index written to or read from the
  // hardware.
  auto result = WRAP(AAudioStream_getTimestamp)(stream, CLOCK_MONOTONIC,
                                                &hw_frame_index, &hw_tstamp);
  if (result != AAUDIO_OK) {
    LOG("AAudioStream_getTimestamp failure.");
    return 0;
  }

  // Compute the difference between the app and the hardware indices.
  int64_t frame_index_delta = app_frame_index - hw_frame_index;
  // Convert to ns
  int64_t frame_time_delta = (frame_index_delta * NS_PER_S) / stm->sample_rate;
  // Extrapolate from the known timestamp for a particular frame presented.
  int64_t app_frame_hw_time = hw_tstamp + frame_time_delta;
  // For an output stream, the latency is positive, for an input stream, it's
  // negative.
  int64_t latency_ns =
      is_output ? app_frame_hw_time - signed_tstamp_ns : signed_tstamp_ns - app_frame_hw_time;
  int64_t latency_frames = stm->sample_rate * latency_ns / NS_PER_S;

  LOGV("Latency in frames (%s): %d (%dms)", is_output ? "output" : "input",
       latency_frames, latency_ns / 1e6);

  return latency_frames;
}

void
compute_and_report_latency_metrics(cubeb_stream * stm)
{
  AAudioTimingInfo info = {};

  info.tstamp = now_ns();

  if (stm->ostream) {
    uint64_t latency_frames =
        aaudio_get_latency(stm, AAUDIO_DIRECTION_OUTPUT, info.tstamp);
    if (latency_frames) {
      info.output_latency = latency_frames;
      info.output_frame_index =
          WRAP(AAudioStream_getFramesWritten)(stm->ostream);
    }
  }
  if (stm->istream) {
    uint64_t latency_frames =
        aaudio_get_latency(stm, AAUDIO_DIRECTION_INPUT, info.tstamp);
    if (latency_frames) {
      info.input_latency = latency_frames;
    }
  }

  if (info.output_latency || info.input_latency) {
    stm->latency_metrics_available = true;
    stm->timing_info.write(info);
  }
}

// Returning AAUDIO_CALLBACK_RESULT_STOP seems to put the stream in
// an invalid state. Seems like an AAudio bug/bad documentation.
// We therefore only return it on error.

static aaudio_data_callback_result_t
aaudio_duplex_data_cb(AAudioStream * astream, void * user_data,
                      void * audio_data, int32_t num_frames)
{
  cubeb_stream * stm = (cubeb_stream *)user_data;
  AutoInCallback aic(stm);
  assert(stm->ostream == astream);
  assert(stm->istream);
  assert(num_frames >= 0);

  stream_state state = atomic_load(&stm->state);
  int istate = WRAP(AAudioStream_getState)(stm->istream);
  int ostate = WRAP(AAudioStream_getState)(stm->ostream);
  ALOGV(
      "aaudio duplex data cb on stream %p: state %ld (in: %d, out: %d), "
      "num_frames: %ld",
      (void*)stm, state, istate, ostate, num_frames);

  // all other states may happen since the callback might be called
  // from within requestStart
  assert(state != stream_state::SHUTDOWN);

  // This might happen when we started draining but not yet actually
  // stopped the stream from the state thread.
  if (state == stream_state::DRAINING) {
    std::memset(audio_data, 0x0, num_frames * stm->out_frame_size);
    return AAUDIO_CALLBACK_RESULT_CONTINUE;
  }

  // The aaudio docs state that AAudioStream_read must not be called on
  // the stream associated with a callback. But we call it on the input stream
  // while this callback is for the output stream so this is ok.
  // We also pass timeout 0, giving us strong non-blocking guarantees.
  // This is exactly how it's done in the aaudio duplex example code snippet.
  long in_num_frames =
      WRAP(AAudioStream_read)(stm->istream, stm->in_buf.get(), num_frames, 0);
  if (in_num_frames < 0) { // error
    stm->state.store(stream_state::ERROR);
    LOG("AAudioStream_read: %s",
        WRAP(AAudio_convertResultToText)(in_num_frames));
    return AAUDIO_CALLBACK_RESULT_STOP;
  }

  compute_and_report_latency_metrics(stm);

  // This can happen shortly after starting the stream. AAudio might immediately
  // begin to buffer output but not have any input ready yet. We could
  // block AAudioStream_read (passing a timeout > 0) but that leads to issues
  // since blocking in this callback is a bad idea in general and it might break
  // the stream when it is stopped by another thread shortly after being
  // started. We therefore simply send silent input to the application, as shown
  // in the AAudio duplex stream code example.
  if (in_num_frames < num_frames) {
    // LOG("AAudioStream_read returned not enough frames: %ld instead of %d",
    //   in_num_frames, num_frames);
    unsigned left = num_frames - in_num_frames;
    char * buf = stm->in_buf.get() + in_num_frames * stm->in_frame_size;
    std::memset(buf, 0x0, left * stm->in_frame_size);
    in_num_frames = num_frames;
  }

  long done_frames =
      cubeb_resampler_fill(stm->resampler, stm->in_buf.get(), &in_num_frames,
                           audio_data, num_frames);

  if (done_frames < 0 || done_frames > num_frames) {
    LOG("Error in data callback or resampler: %ld", done_frames);
    stm->state.store(stream_state::ERROR);
    return AAUDIO_CALLBACK_RESULT_STOP;
  }
  if (done_frames < num_frames) {
    stm->state.store(stream_state::DRAINING);
    stm->context->state.waiting.store(true);
    stm->context->state.cond.notify_one();

    char * begin =
        static_cast<char *>(audio_data) + done_frames * stm->out_frame_size;
    std::memset(begin, 0x0, (num_frames - done_frames) * stm->out_frame_size);
  }

  apply_volume(stm, audio_data, done_frames);
  return AAUDIO_CALLBACK_RESULT_CONTINUE;
}

static aaudio_data_callback_result_t
aaudio_output_data_cb(AAudioStream * astream, void * user_data,
                      void * audio_data, int32_t num_frames)
{
  cubeb_stream * stm = (cubeb_stream *)user_data;
  AutoInCallback aic(stm);
  assert(stm->ostream == astream);
  assert(!stm->istream);
  assert(num_frames >= 0);

  stream_state state = stm->state.load();
  int ostate = WRAP(AAudioStream_getState)(stm->ostream);
  ALOGV("aaudio output data cb on stream %p: state %ld (%d), num_frames: %ld",
        stm, state, ostate, num_frames);

  // all other states may happen since the callback might be called
  // from within requestStart
  assert(state != stream_state::SHUTDOWN);

  // This might happen when we started draining but not yet actually
  // stopped the stream from the state thread.
  if (state == stream_state::DRAINING) {
    std::memset(audio_data, 0x0, num_frames * stm->out_frame_size);
    return AAUDIO_CALLBACK_RESULT_CONTINUE;
  }

  compute_and_report_latency_metrics(stm);

  long done_frames =
      cubeb_resampler_fill(stm->resampler, nullptr, nullptr, audio_data, num_frames);
  if (done_frames < 0 || done_frames > num_frames) {
    LOG("Error in data callback or resampler: %ld", done_frames);
    stm->state.store(stream_state::ERROR);
    return AAUDIO_CALLBACK_RESULT_STOP;
  }

  if (done_frames < num_frames) {
    stm->state.store(stream_state::DRAINING);
    stm->context->state.waiting.store(true);
    stm->context->state.cond.notify_one();

    char * begin =
        static_cast<char *>(audio_data) + done_frames * stm->out_frame_size;
    std::memset(begin, 0x0, (num_frames - done_frames) * stm->out_frame_size);
  }

  apply_volume(stm, audio_data, done_frames);
  return AAUDIO_CALLBACK_RESULT_CONTINUE;
}

static aaudio_data_callback_result_t
aaudio_input_data_cb(AAudioStream * astream, void * user_data,
                     void * audio_data, int32_t num_frames)
{
  cubeb_stream * stm = (cubeb_stream *)user_data;
  AutoInCallback aic(stm);
  assert(stm->istream == astream);
  assert(!stm->ostream);
  assert(num_frames >= 0);

  stream_state state = stm->state.load();
  int istate = WRAP(AAudioStream_getState)(stm->istream);
  ALOGV("aaudio input data cb on stream %p: state %ld (%d), num_frames: %ld",
        stm, state, istate, num_frames);

  // all other states may happen since the callback might be called
  // from within requestStart
  assert(state != stream_state::SHUTDOWN);

  // This might happen when we started draining but not yet actually
  // STOPPED the stream from the state thread.
  if (state == stream_state::DRAINING) {
    return AAUDIO_CALLBACK_RESULT_CONTINUE;
  }

  compute_and_report_latency_metrics(stm);

  long input_frame_count = num_frames;
  long done_frames = cubeb_resampler_fill(stm->resampler, audio_data,
                                          &input_frame_count, nullptr, 0);

  if (done_frames < 0 || done_frames > num_frames) {
    LOG("Error in data callback or resampler: %ld", done_frames);
    stm->state.store(stream_state::ERROR);
    return AAUDIO_CALLBACK_RESULT_STOP;
  }

  if (done_frames < input_frame_count) {
    // we don't really drain an input stream, just have to
    // stop it from the state thread. That is signaled via the
    // DRAINING state.
    stm->state.store(stream_state::DRAINING);
    stm->context->state.waiting.store(true);
    stm->context->state.cond.notify_one();
  }

  return AAUDIO_CALLBACK_RESULT_CONTINUE;
}

static void
aaudio_error_cb(AAudioStream * astream, void * user_data, aaudio_result_t error)
{
  cubeb_stream * stm = static_cast<cubeb_stream *>(user_data);
  assert(stm->ostream == astream || stm->istream == astream);
  LOG("AAudio error callback: %s", WRAP(AAudio_convertResultToText)(error));
  stm->state.store(stream_state::ERROR);
}

static int
realize_stream(AAudioStreamBuilder * sb, const cubeb_stream_params * params,
               AAudioStream ** stream, unsigned * frame_size)
{
  aaudio_result_t res;
  assert(params->rate);
  assert(params->channels);

  WRAP(AAudioStreamBuilder_setSampleRate)(sb, static_cast<int32_t>(params->rate));
  WRAP(AAudioStreamBuilder_setChannelCount)(sb, static_cast<int32_t>(params->channels));

  aaudio_format_t fmt;
  switch (params->format) {
  case CUBEB_SAMPLE_S16NE:
    fmt = AAUDIO_FORMAT_PCM_I16;
    *frame_size = sizeof(int16_t) * params->channels;
    break;
  case CUBEB_SAMPLE_FLOAT32NE:
    fmt = AAUDIO_FORMAT_PCM_FLOAT;
    *frame_size = sizeof(float) * params->channels;
    break;
  default:
    return CUBEB_ERROR_INVALID_FORMAT;
  }

  WRAP(AAudioStreamBuilder_setFormat)(sb, fmt);
  res = WRAP(AAudioStreamBuilder_openStream)(sb, stream);
  if (res == AAUDIO_ERROR_INVALID_FORMAT) {
    LOG("AAudio device doesn't support output format %d", fmt);
    return CUBEB_ERROR_INVALID_FORMAT;
  }

  if (params->rate && res == AAUDIO_ERROR_INVALID_RATE) {
    // The requested rate is not supported.
    // Just try again with default rate, we create a resampler anyways
    WRAP(AAudioStreamBuilder_setSampleRate)(sb, AAUDIO_UNSPECIFIED);
    res = WRAP(AAudioStreamBuilder_openStream)(sb, stream);
    LOG("Requested rate of %u is not supported, inserting resampler",
        params->rate);
  }

  // When the app has no permission to record audio
  // (android.permission.RECORD_AUDIO) but requested and input stream, this will
  // return INVALID_ARGUMENT.
  if (res != AAUDIO_OK) {
    LOG("AAudioStreamBuilder_openStream: %s",
        WRAP(AAudio_convertResultToText)(res));
    return CUBEB_ERROR;
  }

  return CUBEB_OK;
}

static void
aaudio_stream_destroy(cubeb_stream * stm)
{
  lock_guard lock(stm->mutex);
  assert(stm->state == stream_state::STOPPED ||
         stm->state == stream_state::STOPPING ||
         stm->state == stream_state::INIT ||
         stm->state == stream_state::DRAINING ||
         stm->state == stream_state::ERROR ||
         stm->state == stream_state::SHUTDOWN);

  aaudio_result_t res;

  // No callbacks are triggered anymore when requestStop returns.
  // That is important as we otherwise might read from a closed istream
  // for a duplex stream.
  if (stm->ostream) {
    if (stm->state != stream_state::STOPPED &&
        stm->state != stream_state::STOPPING &&
        stm->state != stream_state::SHUTDOWN) {
      res = WRAP(AAudioStream_requestStop)(stm->ostream);
      if (res != AAUDIO_OK) {
        LOG("AAudioStreamBuilder_requestStop: %s",
            WRAP(AAudio_convertResultToText)(res));
      }
    }

    WRAP(AAudioStream_close)(stm->ostream);
    stm->ostream = nullptr;
  }

  if (stm->istream) {
    if (stm->state != stream_state::STOPPED &&
        stm->state != stream_state::STOPPING &&
        stm->state != stream_state::SHUTDOWN) {
      res = WRAP(AAudioStream_requestStop)(stm->istream);
      if (res != AAUDIO_OK) {
        LOG("AAudioStreamBuilder_requestStop: %s",
            WRAP(AAudio_convertResultToText)(res));
      }
    }

    WRAP(AAudioStream_close)(stm->istream);
    stm->istream = nullptr;
  }

  if (stm->resampler) {
    cubeb_resampler_destroy(stm->resampler);
    stm->resampler = nullptr;
  }

  stm->in_buf = {};
  stm->in_frame_size = {};
  stm->out_format = {};
  stm->out_channels = {};
  stm->out_frame_size = {};

  stm->state.store(stream_state::INIT);
  stm->in_use.store(false);
}

static int
aaudio_stream_init_impl(cubeb_stream * stm, cubeb_devid input_device,
                        cubeb_stream_params * input_stream_params,
                        cubeb_devid output_device,
                        cubeb_stream_params * output_stream_params,
                        unsigned int latency_frames)
{
  assert(stm->state.load() == stream_state::INIT);
  stm->in_use.store(true);

  aaudio_result_t res;
  AAudioStreamBuilder * sb;
  res = WRAP(AAudio_createStreamBuilder)(&sb);
  if (res != AAUDIO_OK) {
    LOG("AAudio_createStreamBuilder: %s",
        WRAP(AAudio_convertResultToText)(res));
    return CUBEB_ERROR;
  }

  // make sure the builder is always destroyed
  struct StreamBuilderDestructor {
    void operator()(AAudioStreamBuilder * sb)
    {
      WRAP(AAudioStreamBuilder_delete)(sb);
    }
  };

  std::unique_ptr<AAudioStreamBuilder, StreamBuilderDestructor> sbPtr(sb);

  WRAP(AAudioStreamBuilder_setErrorCallback)(sb, aaudio_error_cb, stm);
  WRAP(AAudioStreamBuilder_setBufferCapacityInFrames)(sb, static_cast<int32_t>(latency_frames));

  AAudioStream_dataCallback in_data_callback{};
  AAudioStream_dataCallback out_data_callback{};
  if (output_stream_params && input_stream_params) {
    out_data_callback = aaudio_duplex_data_cb;
    in_data_callback = nullptr;
  } else if (input_stream_params) {
    in_data_callback = aaudio_input_data_cb;
  } else if (output_stream_params) {
    out_data_callback = aaudio_output_data_cb;
  } else {
    LOG("Tried to open stream without input or output parameters");
    return CUBEB_ERROR;
  }

#ifdef CUBEB_AAUDIO_EXCLUSIVE_STREAM
  LOG("AAudio setting exclusive share mode for stream");
  WRAP(AAudioStreamBuilder_setSharingMode)(sb, AAUDIO_SHARING_MODE_EXCLUSIVE);
#endif

  if (latency_frames <= POWERSAVE_LATENCY_FRAMES_THRESHOLD) {
    LOG("AAudio setting low latency mode for stream");
    WRAP(AAudioStreamBuilder_setPerformanceMode)
    (sb, AAUDIO_PERFORMANCE_MODE_LOW_LATENCY);
  } else {
    LOG("AAudio setting power saving mode for stream");
    WRAP(AAudioStreamBuilder_setPerformanceMode)
    (sb, AAUDIO_PERFORMANCE_MODE_POWER_SAVING);
  }

  unsigned frame_size;

  // initialize streams
  // output
  cubeb_stream_params out_params;
  if (output_stream_params) {
    int output_preset = stm->voice_output ? AAUDIO_USAGE_VOICE_COMMUNICATION
                                          : AAUDIO_USAGE_MEDIA;
    WRAP(AAudioStreamBuilder_setUsage)(sb, output_preset);
    WRAP(AAudioStreamBuilder_setDirection)(sb, AAUDIO_DIRECTION_OUTPUT);
    WRAP(AAudioStreamBuilder_setDataCallback)(sb, out_data_callback, stm);
    int res_err =
        realize_stream(sb, output_stream_params, &stm->ostream, &frame_size);
    if (res_err) {
      return res_err;
    }

    int rate = WRAP(AAudioStream_getSampleRate)(stm->ostream);
    LOG("AAudio output stream sharing mode: %d",
        WRAP(AAudioStream_getSharingMode)(stm->ostream));
    LOG("AAudio output stream performance mode: %d",
        WRAP(AAudioStream_getPerformanceMode)(stm->ostream));
    LOG("AAudio output stream buffer capacity: %d",
        WRAP(AAudioStream_getBufferCapacityInFrames)(stm->ostream));
    LOG("AAudio output stream buffer size: %d",
        WRAP(AAudioStream_getBufferSizeInFrames)(stm->ostream));
    LOG("AAudio output stream sample-rate: %d", rate);

    stm->sample_rate = output_stream_params->rate;
    out_params = *output_stream_params;
    out_params.rate = rate;

    stm->out_channels = output_stream_params->channels;
    stm->out_format = output_stream_params->format;
    stm->out_frame_size = frame_size;
    stm->volume.store(1.f);
  }

  // input
  cubeb_stream_params in_params;
  if (input_stream_params) {
    // Match what the OpenSL backend does for now, we could use UNPROCESSED and
    // VOICE_COMMUNICATION here, but we'd need to make it clear that
    // application-level AEC and other voice processing should be disabled
    // there.
    int input_preset = stm->voice_input ? AAUDIO_INPUT_PRESET_VOICE_RECOGNITION
                                        : AAUDIO_INPUT_PRESET_CAMCORDER;
    WRAP(AAudioStreamBuilder_setInputPreset)(sb, input_preset);
    WRAP(AAudioStreamBuilder_setDirection)(sb, AAUDIO_DIRECTION_INPUT);
    WRAP(AAudioStreamBuilder_setDataCallback)(sb, in_data_callback, stm);
    int res_err =
        realize_stream(sb, input_stream_params, &stm->istream, &frame_size);
    if (res_err) {
      return res_err;
    }

    int bcap = WRAP(AAudioStream_getBufferCapacityInFrames)(stm->istream);
    int rate = WRAP(AAudioStream_getSampleRate)(stm->istream);
    LOG("AAudio input stream sharing mode: %d",
        WRAP(AAudioStream_getSharingMode)(stm->istream));
    LOG("AAudio input stream performance mode: %d",
        WRAP(AAudioStream_getPerformanceMode)(stm->istream));
    LOG("AAudio input stream buffer capacity: %d", bcap);
    LOG("AAudio input stream buffer size: %d",
        WRAP(AAudioStream_getBufferSizeInFrames)(stm->istream));
    LOG("AAudio input stream buffer rate: %d", rate);

    stm->in_buf.reset(new char[bcap * frame_size]());
    assert(!stm->sample_rate || stm->sample_rate == input_stream_params->rate);

    stm->sample_rate = input_stream_params->rate;
    in_params = *input_stream_params;
    in_params.rate = rate;
    stm->in_frame_size = frame_size;
  }

  // initialize resampler
  stm->resampler = cubeb_resampler_create(
      stm, input_stream_params ? &in_params : nullptr,
      output_stream_params ? &out_params : nullptr, stm->sample_rate,
      stm->data_callback, stm->user_ptr, CUBEB_RESAMPLER_QUALITY_DEFAULT,
      CUBEB_RESAMPLER_RECLOCK_NONE);

  if (!stm->resampler) {
    LOG("Failed to create resampler");
    return CUBEB_ERROR;
  }

  // the stream isn't started initially. We don't need to differentiate
  // between a stream that was just initialized and one that played
  // already but was stopped.
  stm->state.store(stream_state::STOPPED);
  LOG("Cubeb stream (%p) INIT success", (void *)stm);
  return CUBEB_OK;
}

static int
aaudio_stream_init(cubeb * ctx, cubeb_stream ** stream,
                   char const * /* stream_name */, cubeb_devid input_device,
                   cubeb_stream_params * input_stream_params,
                   cubeb_devid output_device,
                   cubeb_stream_params * output_stream_params,
                   unsigned int latency_frames,
                   cubeb_data_callback data_callback,
                   cubeb_state_callback state_callback, void * user_ptr)
{
  assert(!input_device);
  assert(!output_device);

  // atomically find a free stream.
  cubeb_stream * stm = nullptr;
  unique_lock lock;
  for (auto & stream : ctx->streams) {
    // This check is only an optimization, we don't strictly need it
    // since we check again after locking the mutex.
    if (stream.in_use.load()) {
      continue;
    }

    // if this fails, another thread initialized this stream
    // between our check of in_use and this.
    lock = unique_lock(stream.mutex, std::try_to_lock);
    if (!lock.owns_lock()) {
      continue;
    }

    if (stream.in_use.load()) {
      lock = {};
      continue;
    }

    stm = &stream;
    break;
  }

  if (!stm) {
    LOG("Error: maximum number of streams reached");
    return CUBEB_ERROR;
  }

  stm->context = ctx;
  stm->user_ptr = user_ptr;
  stm->data_callback = data_callback;
  stm->state_callback = state_callback;
  stm->voice_input = input_stream_params &&
                     !!(input_stream_params->prefs & CUBEB_STREAM_PREF_VOICE);
  stm->voice_output = output_stream_params &&
                      !!(output_stream_params->prefs & CUBEB_STREAM_PREF_VOICE);
  stm->previous_clock = 0;

  LOG("cubeb stream prefs: voice_input: %s voice_output: %s",
      stm->voice_input ? "true" : "false",
      stm->voice_output ? "true" : "false");

  int err = aaudio_stream_init_impl(stm, input_device, input_stream_params,
                                    output_device, output_stream_params,
                                    latency_frames);
  if (err != CUBEB_OK) {
    // This is needed since aaudio_stream_destroy will lock the mutex again.
    // It's no problem that there is a gap in between as the stream isn't
    // actually in u se.
    lock.unlock();
    aaudio_stream_destroy(stm);
    return err;
  }

  *stream = stm;
  return CUBEB_OK;
}

static int
aaudio_stream_start(cubeb_stream * stm)
{
  assert(stm && stm->in_use.load());
  lock_guard lock(stm->mutex);

  stream_state state = stm->state.load();
  int istate = stm->istream ? WRAP(AAudioStream_getState)(stm->istream) : 0;
  int ostate = stm->ostream ? WRAP(AAudioStream_getState)(stm->ostream) : 0;
  LOGV("STARTING stream %p: %d (%d %d)", (void *)stm, state, istate, ostate);

  switch (state) {
  case stream_state::STARTED:
  case stream_state::STARTING:
    LOG("cubeb stream %p already STARTING/STARTED", (void *)stm);
    return CUBEB_OK;
  case stream_state::ERROR:
  case stream_state::SHUTDOWN:
    return CUBEB_ERROR;
  case stream_state::INIT:
    assert(false && "Invalid stream");
    return CUBEB_ERROR;
  case stream_state::STOPPED:
  case stream_state::STOPPING:
  case stream_state::DRAINING:
    break;
  }

  aaudio_result_t res;

  // Important to start istream before ostream.
  // As soon as we start ostream, the callbacks might be triggered an we
  // might read from istream (on duplex). If istream wasn't started yet
  // this is a problem.
  if (stm->istream) {
    res = WRAP(AAudioStream_requestStart)(stm->istream);
    if (res != AAUDIO_OK) {
      LOG("AAudioStream_requestStart (istream): %s",
          WRAP(AAudio_convertResultToText)(res));
      stm->state.store(stream_state::ERROR);
      return CUBEB_ERROR;
    }
  }

  if (stm->ostream) {
    res = WRAP(AAudioStream_requestStart)(stm->ostream);
    if (res != AAUDIO_OK) {
      LOG("AAudioStream_requestStart (ostream): %s",
          WRAP(AAudio_convertResultToText)(res));
      stm->state.store(stream_state::ERROR);
      return CUBEB_ERROR;
    }
  }

  int ret = CUBEB_OK;
  bool success;

  while (!(success = stm->state.compare_exchange_strong(
               state, stream_state::STARTING))) {
    // we land here only if the state has changed in the meantime
    switch (state) {
    // If an error ocurred in the meantime, we can't change that.
    // The stream will be stopped when shut down.
    case stream_state::ERROR:
      ret = CUBEB_ERROR;
      break;
    // The only situation in which the state could have switched to draining
    // is if the callback was already fired and requested draining. Don't
    // overwrite that. It's not an error either though.
    case stream_state::DRAINING:
      break;

    // If the state switched [DRAINING -> STOPPING] or [DRAINING/STOPPING ->
    // STOPPED] in the meantime, we can simply overwrite that since we
    // restarted the stream.
    case stream_state::STOPPING:
    case stream_state::STOPPED:
      continue;

    // There is no situation in which the state could have been valid before
    // but now in shutdown mode, since we hold the streams mutex.
    // There is also no way that it switched *into* STARTING or
    // STARTED mode.
    default:
      assert(false && "Invalid state change");
      ret = CUBEB_ERROR;
      break;
    }

    break;
  }

  if (success) {
    stm->context->state.waiting.store(true);
    stm->context->state.cond.notify_one();
  }

  return ret;
}

static int
aaudio_stream_stop(cubeb_stream * stm)
{
  assert(stm && stm->in_use.load());
  lock_guard lock(stm->mutex);

  stream_state state = stm->state.load();
  int istate = stm->istream ? WRAP(AAudioStream_getState)(stm->istream) : 0;
  int ostate = stm->ostream ? WRAP(AAudioStream_getState)(stm->ostream) : 0;
  LOGV("STOPPING stream %p: %d (%d %d)", (void *)stm, state, istate, ostate);

  switch (state) {
  case stream_state::STOPPED:
  case stream_state::STOPPING:
  case stream_state::DRAINING:
    LOG("cubeb stream %p already STOPPING/STOPPED", (void *)stm);
    return CUBEB_OK;
  case stream_state::ERROR:
  case stream_state::SHUTDOWN:
    return CUBEB_ERROR;
  case stream_state::INIT:
    assert(false && "Invalid stream");
    return CUBEB_ERROR;
  case stream_state::STARTED:
  case stream_state::STARTING:
    break;
  }

  aaudio_result_t res;

  // No callbacks are triggered anymore when requestStop returns.
  // That is important as we otherwise might read from a closed istream
  // for a duplex stream.
  // Therefor it is important to close ostream first.
  if (stm->ostream) {
    // Could use pause + flush here as well, the public cubeb interface
    // doesn't state behavior.
    res = WRAP(AAudioStream_requestStop)(stm->ostream);
    if (res != AAUDIO_OK) {
      LOG("AAudioStream_requestStop (ostream): %s",
          WRAP(AAudio_convertResultToText)(res));
      stm->state.store(stream_state::ERROR);
      return CUBEB_ERROR;
    }
  }

  if (stm->istream) {
    res = WRAP(AAudioStream_requestStop)(stm->istream);
    if (res != AAUDIO_OK) {
      LOG("AAudioStream_requestStop (istream): %s",
          WRAP(AAudio_convertResultToText)(res));
      stm->state.store(stream_state::ERROR);
      return CUBEB_ERROR;
    }
  }

  int ret = CUBEB_OK;
  bool success;
  while (!(success = atomic_compare_exchange_strong(&stm->state, &state,
                                                    stream_state::STOPPING))) {
    // we land here only if the state has changed in the meantime
    switch (state) {
    // If an error ocurred in the meantime, we can't change that.
    // The stream will be STOPPED when shut down.
    case stream_state::ERROR:
      ret = CUBEB_ERROR;
      break;
    // If it was switched to DRAINING in the meantime, it was or
    // will be STOPPED soon anyways. We don't interfere with
    // the DRAINING process, no matter in which state.
    // Not an error
    case stream_state::DRAINING:
    case stream_state::STOPPING:
    case stream_state::STOPPED:
      break;

    // If the state switched from STARTING to STARTED in the meantime
    // we can simply overwrite that since we just STOPPED it.
    case stream_state::STARTED:
      continue;

    // There is no situation in which the state could have been valid before
    // but now in shutdown mode, since we hold the streams mutex.
    // There is also no way that it switched *into* STARTING mode.
    default:
      assert(false && "Invalid state change");
      ret = CUBEB_ERROR;
      break;
    }

    break;
  }

  if (success) {
    stm->context->state.waiting.store(true);
    stm->context->state.cond.notify_one();
  }

  return ret;
}

static int
aaudio_stream_get_position(cubeb_stream * stm, uint64_t * position)
{
  assert(stm && stm->in_use.load());
  lock_guard lock(stm->mutex);

  stream_state state = stm->state.load();
  AAudioStream * stream = stm->ostream ? stm->ostream : stm->istream;
  switch (state) {
  case stream_state::ERROR:
  case stream_state::SHUTDOWN:
    return CUBEB_ERROR;
  case stream_state::DRAINING:
  case stream_state::STOPPED:
  case stream_state::STOPPING:
    // getTimestamp is only valid when the stream is playing.
    // Simply return the number of frames passed to aaudio
    *position = WRAP(AAudioStream_getFramesRead)(stream);
    if (*position < stm->previous_clock) {
      *position = stm->previous_clock;
    } else {
      stm->previous_clock = *position;
    }
    return CUBEB_OK;
  case stream_state::INIT:
    assert(false && "Invalid stream");
    return CUBEB_ERROR;
  case stream_state::STARTED:
  case stream_state::STARTING:
    break;
  }

  // No callback yet, the stream hasn't really started.
  if (stm->previous_clock == 0 && !stm->timing_info.updated()) {
    LOG("Not timing info yet");
    *position = 0;
    return CUBEB_OK;
  }

  AAudioTimingInfo info = stm->timing_info.read();
  LOGV("AAudioTimingInfo idx:%lu tstamp:%lu latency:%u",
       info.output_frame_index, info.tstamp, info.output_latency);
  // Interpolate client side since the last callback.
  uint64_t interpolation = stm->sample_rate * (now_ns() - info.tstamp) / NS_PER_S;
  *position = info.output_frame_index + interpolation - info.output_latency;
  if (*position < stm->previous_clock) {
    *position = stm->previous_clock;
  } else {
    stm->previous_clock = *position;
  }

  LOG("aaudio_stream_get_position: %" PRIu64 " frames", *position);

  return CUBEB_OK;
}

static int
aaudio_stream_get_latency(cubeb_stream * stm, uint32_t * latency)
{
  if (!stm->ostream) {
    LOG("error: aaudio_stream_get_latency on input-only stream");
    return CUBEB_ERROR;
  }

  if (!stm->latency_metrics_available) {
    LOG("Not timing info yet (output)");
    return CUBEB_OK;
  }

  AAudioTimingInfo info = stm->timing_info.read();

  *latency = info.output_latency;
  LOG("aaudio_stream_get_latency, %u frames", *latency);

  return CUBEB_OK;
}

static int
aaudio_stream_get_input_latency(cubeb_stream * stm, uint32_t * latency)
{
  if (!stm->istream) {
    LOG("error: aaudio_stream_get_input_latency on an output-only stream");
    return CUBEB_ERROR;
  }

  if (!stm->latency_metrics_available) {
    LOG("Not timing info yet (input)");
    return CUBEB_OK;
  }

  AAudioTimingInfo info = stm->timing_info.read();

  *latency = info.input_latency;
  LOG("aaudio_stream_get_latency, %u frames", *latency);

  return CUBEB_OK;
}

static int
aaudio_stream_set_volume(cubeb_stream * stm, float volume)
{
  assert(stm && stm->in_use.load() && stm->ostream);
  stm->volume.store(volume);
  return CUBEB_OK;
}

aaudio_data_callback_result_t
dummy_callback(AAudioStream * stream, void * userData, void * audioData,
               int32_t numFrames)
{
  return AAUDIO_CALLBACK_RESULT_STOP;
}

// Returns a dummy stream with all default settings
static AAudioStream *
init_dummy_stream()
{
  AAudioStreamBuilder * streamBuilder;
  aaudio_result_t res;
  res = WRAP(AAudio_createStreamBuilder)(&streamBuilder);
  if (res != AAUDIO_OK) {
    LOG("init_dummy_stream: AAudio_createStreamBuilder: %s",
        WRAP(AAudio_convertResultToText)(res));
    return nullptr;
  }
  WRAP(AAudioStreamBuilder_setDataCallback)
  (streamBuilder, dummy_callback, nullptr);
  WRAP(AAudioStreamBuilder_setPerformanceMode)
  (streamBuilder, AAUDIO_PERFORMANCE_MODE_LOW_LATENCY);

  AAudioStream * stream;
  res = WRAP(AAudioStreamBuilder_openStream)(streamBuilder, &stream);
  if (res != AAUDIO_OK) {
    LOG("init_dummy_stream: AAudioStreamBuilder_openStream %s",
        WRAP(AAudio_convertResultToText)(res));
    return nullptr;
  }
  WRAP(AAudioStreamBuilder_delete)(streamBuilder);

  return stream;
}

static void
destroy_dummy_stream(AAudioStream * stream)
{
  WRAP(AAudioStream_close)(stream);
}

static int
aaudio_get_min_latency(cubeb * ctx, cubeb_stream_params params,
                       uint32_t * latency_frames)
{
  AAudioStream * stream = init_dummy_stream();

  if (!stream) {
    return CUBEB_ERROR;
  }

  // https://android.googlesource.com/platform/compatibility/cdd/+/refs/heads/master/5_multimedia/5_6_audio-latency.md
  *latency_frames = WRAP(AAudioStream_getFramesPerBurst)(stream);

  LOG("aaudio_get_min_latency: %u frames", *latency_frames);

  destroy_dummy_stream(stream);

  return CUBEB_OK;
}

int
aaudio_get_preferred_sample_rate(cubeb * ctx, uint32_t * rate)
{
  AAudioStream * stream = init_dummy_stream();

  if (!stream) {
    return CUBEB_ERROR;
  }

  *rate = WRAP(AAudioStream_getSampleRate)(stream);

  LOG("aaudio_get_preferred_sample_rate %uHz", *rate);

  destroy_dummy_stream(stream);

  return CUBEB_OK;
}

extern "C" int
aaudio_init(cubeb ** context, char const * context_name);

const static struct cubeb_ops aaudio_ops = {
    /*.init =*/aaudio_init,
    /*.get_backend_id =*/aaudio_get_backend_id,
    /*.get_max_channel_count =*/aaudio_get_max_channel_count,
    /* .get_min_latency =*/aaudio_get_min_latency,
    /*.get_preferred_sample_rate =*/aaudio_get_preferred_sample_rate,
    /*.enumerate_devices =*/nullptr,
    /*.device_collection_destroy =*/nullptr,
    /*.destroy =*/aaudio_destroy,
    /*.stream_init =*/aaudio_stream_init,
    /*.stream_destroy =*/aaudio_stream_destroy,
    /*.stream_start =*/aaudio_stream_start,
    /*.stream_stop =*/aaudio_stream_stop,
    /*.stream_get_position =*/aaudio_stream_get_position,
    /*.stream_get_latency =*/aaudio_stream_get_latency,
    /*.stream_get_input_latency =*/aaudio_stream_get_input_latency,
    /*.stream_set_volume =*/aaudio_stream_set_volume,
    /*.stream_set_name =*/nullptr,
    /*.stream_get_current_device =*/nullptr,
    /*.stream_device_destroy =*/nullptr,
    /*.stream_register_device_changed_callback =*/nullptr,
    /*.register_device_collection_changed =*/nullptr};

extern "C" /*static*/ int
aaudio_init(cubeb ** context, char const * /* context_name */)
{
  // load api
  void * libaaudio = nullptr;
#ifndef DISABLE_LIBAAUDIO_DLOPEN
  libaaudio = dlopen("libaaudio.so", RTLD_NOW);
  if (!libaaudio) {
    return CUBEB_ERROR;
  }

#define LOAD(x)                                                                \
  {                                                                            \
    cubeb_##x = (decltype(x) *)(dlsym(libaaudio, #x));                         \
    if (!WRAP(x)) {                                                            \
      LOG("AAudio: Failed to load %s", #x);                                    \
      dlclose(libaaudio);                                                      \
      return CUBEB_ERROR;                                                      \
    }                                                                          \
  }

  LIBAAUDIO_API_VISIT(LOAD);
#undef LOAD
#endif

  cubeb * ctx = new cubeb;
  ctx->ops = &aaudio_ops;
  ctx->libaaudio = libaaudio;

  ctx->state.thread = std::thread(state_thread, ctx);

  // NOTE: using platform-specific APIs we could set the priority of the
  // notifier thread lower than the priority of the state thread.
  // This way, it's more likely that the state thread will be woken up
  // by the condition variable signal when both are currently waiting
  ctx->state.notifier = std::thread(notifier_thread, ctx);

  *context = ctx;
  return CUBEB_OK;
}