aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/cxxsupp/openmp/kmp_barrier.cpp
blob: ee05bb3587ca1f60dd2dc1a33477f2ca40b2d679 (plain) (blame)
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
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
/*
 * kmp_barrier.cpp
 */

//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//

#include "kmp_wait_release.h"
#include "kmp_barrier.h"
#include "kmp_itt.h"
#include "kmp_os.h"
#include "kmp_stats.h"
#include "ompt-specific.h"
// for distributed barrier
#include "kmp_affinity.h"

#if KMP_MIC
#include <immintrin.h>
#define USE_NGO_STORES 1
#endif // KMP_MIC

#if KMP_MIC && USE_NGO_STORES
// ICV copying
#define ngo_load(src) __m512d Vt = _mm512_load_pd((void *)(src))
#define ngo_store_icvs(dst, src) _mm512_storenrngo_pd((void *)(dst), Vt)
#define ngo_store_go(dst, src) _mm512_storenrngo_pd((void *)(dst), Vt)
#define ngo_sync() __asm__ volatile("lock; addl $0,0(%%rsp)" ::: "memory")
#else
#define ngo_load(src) ((void)0)
#define ngo_store_icvs(dst, src) copy_icvs((dst), (src))
#define ngo_store_go(dst, src) KMP_MEMCPY((dst), (src), CACHE_LINE)
#define ngo_sync() ((void)0)
#endif /* KMP_MIC && USE_NGO_STORES */

void __kmp_print_structure(void); // Forward declaration

// ---------------------------- Barrier Algorithms ----------------------------
// Distributed barrier

// Compute how many threads to have polling each cache-line.
// We want to limit the number of writes to IDEAL_GO_RESOLUTION.
void distributedBarrier::computeVarsForN(size_t n) {
  int nsockets = 1;
  if (__kmp_topology) {
    int socket_level = __kmp_topology->get_level(KMP_HW_SOCKET);
    int core_level = __kmp_topology->get_level(KMP_HW_CORE);
    int ncores_per_socket =
        __kmp_topology->calculate_ratio(core_level, socket_level);
    nsockets = __kmp_topology->get_count(socket_level);

    if (nsockets <= 0)
      nsockets = 1;
    if (ncores_per_socket <= 0)
      ncores_per_socket = 1;

    threads_per_go = ncores_per_socket >> 1;
    if (!fix_threads_per_go) {
      // Minimize num_gos
      if (threads_per_go > 4) {
        if (KMP_OPTIMIZE_FOR_REDUCTIONS) {
          threads_per_go = threads_per_go >> 1;
        }
        if (threads_per_go > 4 && nsockets == 1)
          threads_per_go = threads_per_go >> 1;
      }
    }
    if (threads_per_go == 0)
      threads_per_go = 1;
    fix_threads_per_go = true;
    num_gos = n / threads_per_go;
    if (n % threads_per_go)
      num_gos++;
    if (nsockets == 1 || num_gos == 1)
      num_groups = 1;
    else {
      num_groups = num_gos / nsockets;
      if (num_gos % nsockets)
        num_groups++;
    }
    if (num_groups <= 0)
      num_groups = 1;
    gos_per_group = num_gos / num_groups;
    if (num_gos % num_groups)
      gos_per_group++;
    threads_per_group = threads_per_go * gos_per_group;
  } else {
    num_gos = n / threads_per_go;
    if (n % threads_per_go)
      num_gos++;
    if (num_gos == 1)
      num_groups = 1;
    else {
      num_groups = num_gos / 2;
      if (num_gos % 2)
        num_groups++;
    }
    gos_per_group = num_gos / num_groups;
    if (num_gos % num_groups)
      gos_per_group++;
    threads_per_group = threads_per_go * gos_per_group;
  }
}

void distributedBarrier::computeGo(size_t n) {
  // Minimize num_gos
  for (num_gos = 1;; num_gos++)
    if (IDEAL_CONTENTION * num_gos >= n)
      break;
  threads_per_go = n / num_gos;
  if (n % num_gos)
    threads_per_go++;
  while (num_gos > MAX_GOS) {
    threads_per_go++;
    num_gos = n / threads_per_go;
    if (n % threads_per_go)
      num_gos++;
  }
  computeVarsForN(n);
}

// This function is to resize the barrier arrays when the new number of threads
// exceeds max_threads, which is the current size of all the arrays
void distributedBarrier::resize(size_t nthr) {
  KMP_DEBUG_ASSERT(nthr > max_threads);

  // expand to requested size * 2
  max_threads = nthr * 2;

  // allocate arrays to new max threads
  for (int i = 0; i < MAX_ITERS; ++i) {
    if (flags[i])
      flags[i] = (flags_s *)KMP_INTERNAL_REALLOC(flags[i],
                                                 max_threads * sizeof(flags_s));
    else
      flags[i] = (flags_s *)KMP_INTERNAL_MALLOC(max_threads * sizeof(flags_s));
  }

  if (go)
    go = (go_s *)KMP_INTERNAL_REALLOC(go, max_threads * sizeof(go_s));
  else
    go = (go_s *)KMP_INTERNAL_MALLOC(max_threads * sizeof(go_s));

  if (iter)
    iter = (iter_s *)KMP_INTERNAL_REALLOC(iter, max_threads * sizeof(iter_s));
  else
    iter = (iter_s *)KMP_INTERNAL_MALLOC(max_threads * sizeof(iter_s));

  if (sleep)
    sleep =
        (sleep_s *)KMP_INTERNAL_REALLOC(sleep, max_threads * sizeof(sleep_s));
  else
    sleep = (sleep_s *)KMP_INTERNAL_MALLOC(max_threads * sizeof(sleep_s));
}

// This function is to set all the go flags that threads might be waiting
// on, and when blocktime is not infinite, it should be followed by a wake-up
// call to each thread
kmp_uint64 distributedBarrier::go_release() {
  kmp_uint64 next_go = iter[0].iter + distributedBarrier::MAX_ITERS;
  for (size_t j = 0; j < num_gos; j++) {
    go[j].go.store(next_go);
  }
  return next_go;
}

void distributedBarrier::go_reset() {
  for (size_t j = 0; j < max_threads; ++j) {
    for (size_t i = 0; i < distributedBarrier::MAX_ITERS; ++i) {
      flags[i][j].stillNeed = 1;
    }
    go[j].go.store(0);
    iter[j].iter = 0;
  }
}

// This function inits/re-inits the distributed barrier for a particular number
// of threads. If a resize of arrays is needed, it calls the resize function.
void distributedBarrier::init(size_t nthr) {
  size_t old_max = max_threads;
  if (nthr > max_threads) { // need more space in arrays
    resize(nthr);
  }

  for (size_t i = 0; i < max_threads; i++) {
    for (size_t j = 0; j < distributedBarrier::MAX_ITERS; j++) {
      flags[j][i].stillNeed = 1;
    }
    go[i].go.store(0);
    iter[i].iter = 0;
    if (i >= old_max)
      sleep[i].sleep = false;
  }

  // Recalculate num_gos, etc. based on new nthr
  computeVarsForN(nthr);

  num_threads = nthr;

  if (team_icvs == NULL)
    team_icvs = __kmp_allocate(sizeof(kmp_internal_control_t));
}

// This function is used only when KMP_BLOCKTIME is not infinite.
// static
void __kmp_dist_barrier_wakeup(enum barrier_type bt, kmp_team_t *team,
                               size_t start, size_t stop, size_t inc,
                               size_t tid) {
  KMP_DEBUG_ASSERT(__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME);
  if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
    return;

  kmp_info_t **other_threads = team->t.t_threads;
  for (size_t thr = start; thr < stop; thr += inc) {
    KMP_DEBUG_ASSERT(other_threads[thr]);
    int gtid = other_threads[thr]->th.th_info.ds.ds_gtid;
    // Wake up worker regardless of if it appears to be sleeping or not
    __kmp_atomic_resume_64(gtid, (kmp_atomic_flag_64<> *)NULL);
  }
}

static void __kmp_dist_barrier_gather(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    void (*reduce)(void *, void *) USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_dist_gather);
  kmp_team_t *team;
  distributedBarrier *b;
  kmp_info_t **other_threads;
  kmp_uint64 my_current_iter, my_next_iter;
  kmp_uint32 nproc;
  bool group_leader;

  team = this_thr->th.th_team;
  nproc = this_thr->th.th_team_nproc;
  other_threads = team->t.t_threads;
  b = team->t.b;
  my_current_iter = b->iter[tid].iter;
  my_next_iter = (my_current_iter + 1) % distributedBarrier::MAX_ITERS;
  group_leader = ((tid % b->threads_per_group) == 0);

  KA_TRACE(20,
           ("__kmp_dist_barrier_gather: T#%d(%d:%d) enter; barrier type %d\n",
            gtid, team->t.t_id, tid, bt));

#if USE_ITT_BUILD && USE_ITT_NOTIFY
  // Barrier imbalance - save arrive time to the thread
  if (__kmp_forkjoin_frames_mode == 3 || __kmp_forkjoin_frames_mode == 2) {
    this_thr->th.th_bar_arrive_time = this_thr->th.th_bar_min_time =
        __itt_get_timestamp();
  }
#endif

  if (group_leader) {
    // Start from the thread after the group leader
    size_t group_start = tid + 1;
    size_t group_end = tid + b->threads_per_group;
    size_t threads_pending = 0;

    if (group_end > nproc)
      group_end = nproc;
    do { // wait for threads in my group
      threads_pending = 0;
      // Check all the flags every time to avoid branch misspredict
      for (size_t thr = group_start; thr < group_end; thr++) {
        // Each thread uses a different cache line
        threads_pending += b->flags[my_current_iter][thr].stillNeed;
      }
      // Execute tasks here
      if (__kmp_tasking_mode != tskm_immediate_exec) {
        kmp_task_team_t *task_team = this_thr->th.th_task_team;
        if (task_team != NULL) {
          if (TCR_SYNC_4(task_team->tt.tt_active)) {
            if (KMP_TASKING_ENABLED(task_team)) {
              int tasks_completed = FALSE;
              __kmp_atomic_execute_tasks_64(
                  this_thr, gtid, (kmp_atomic_flag_64<> *)NULL, FALSE,
                  &tasks_completed USE_ITT_BUILD_ARG(itt_sync_obj), 0);
            } else
              this_thr->th.th_reap_state = KMP_SAFE_TO_REAP;
          }
        } else {
          this_thr->th.th_reap_state = KMP_SAFE_TO_REAP;
        } // if
      }
      if (TCR_4(__kmp_global.g.g_done)) {
        if (__kmp_global.g.g_abort)
          __kmp_abort_thread();
        break;
      } else if (__kmp_tasking_mode != tskm_immediate_exec &&
                 this_thr->th.th_reap_state == KMP_SAFE_TO_REAP) {
        this_thr->th.th_reap_state = KMP_NOT_SAFE_TO_REAP;
      }
    } while (threads_pending > 0);

    if (reduce) { // Perform reduction if needed
      OMPT_REDUCTION_DECL(this_thr, gtid);
      OMPT_REDUCTION_BEGIN;
      // Group leader reduces all threads in group
      for (size_t thr = group_start; thr < group_end; thr++) {
        (*reduce)(this_thr->th.th_local.reduce_data,
                  other_threads[thr]->th.th_local.reduce_data);
      }
      OMPT_REDUCTION_END;
    }

    // Set flag for next iteration
    b->flags[my_next_iter][tid].stillNeed = 1;
    // Each thread uses a different cache line; resets stillNeed to 0 to
    // indicate it has reached the barrier
    b->flags[my_current_iter][tid].stillNeed = 0;

    do { // wait for all group leaders
      threads_pending = 0;
      for (size_t thr = 0; thr < nproc; thr += b->threads_per_group) {
        threads_pending += b->flags[my_current_iter][thr].stillNeed;
      }
      // Execute tasks here
      if (__kmp_tasking_mode != tskm_immediate_exec) {
        kmp_task_team_t *task_team = this_thr->th.th_task_team;
        if (task_team != NULL) {
          if (TCR_SYNC_4(task_team->tt.tt_active)) {
            if (KMP_TASKING_ENABLED(task_team)) {
              int tasks_completed = FALSE;
              __kmp_atomic_execute_tasks_64(
                  this_thr, gtid, (kmp_atomic_flag_64<> *)NULL, FALSE,
                  &tasks_completed USE_ITT_BUILD_ARG(itt_sync_obj), 0);
            } else
              this_thr->th.th_reap_state = KMP_SAFE_TO_REAP;
          }
        } else {
          this_thr->th.th_reap_state = KMP_SAFE_TO_REAP;
        } // if
      }
      if (TCR_4(__kmp_global.g.g_done)) {
        if (__kmp_global.g.g_abort)
          __kmp_abort_thread();
        break;
      } else if (__kmp_tasking_mode != tskm_immediate_exec &&
                 this_thr->th.th_reap_state == KMP_SAFE_TO_REAP) {
        this_thr->th.th_reap_state = KMP_NOT_SAFE_TO_REAP;
      }
    } while (threads_pending > 0);

    if (reduce) { // Perform reduction if needed
      if (KMP_MASTER_TID(tid)) { // Master reduces over group leaders
        OMPT_REDUCTION_DECL(this_thr, gtid);
        OMPT_REDUCTION_BEGIN;
        for (size_t thr = b->threads_per_group; thr < nproc;
             thr += b->threads_per_group) {
          (*reduce)(this_thr->th.th_local.reduce_data,
                    other_threads[thr]->th.th_local.reduce_data);
        }
        OMPT_REDUCTION_END;
      }
    }
  } else {
    // Set flag for next iteration
    b->flags[my_next_iter][tid].stillNeed = 1;
    // Each thread uses a different cache line; resets stillNeed to 0 to
    // indicate it has reached the barrier
    b->flags[my_current_iter][tid].stillNeed = 0;
  }

  KMP_MFENCE();

  KA_TRACE(20,
           ("__kmp_dist_barrier_gather: T#%d(%d:%d) exit for barrier type %d\n",
            gtid, team->t.t_id, tid, bt));
}

static void __kmp_dist_barrier_release(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    int propagate_icvs USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_dist_release);
  kmp_team_t *team;
  distributedBarrier *b;
  kmp_bstate_t *thr_bar;
  kmp_uint64 my_current_iter, next_go;
  size_t my_go_index;
  bool group_leader;

  KA_TRACE(20, ("__kmp_dist_barrier_release: T#%d(%d) enter; barrier type %d\n",
                gtid, tid, bt));

  thr_bar = &this_thr->th.th_bar[bt].bb;

  if (!KMP_MASTER_TID(tid)) {
    // workers and non-master group leaders need to check their presence in team
    do {
      if (this_thr->th.th_used_in_team.load() != 1 &&
          this_thr->th.th_used_in_team.load() != 3) {
        // Thread is not in use in a team. Wait on location in tid's thread
        // struct. The 0 value tells anyone looking that this thread is spinning
        // or sleeping until this location becomes 3 again; 3 is the transition
        // state to get to 1 which is waiting on go and being in the team
        kmp_flag_32<false, false> my_flag(&(this_thr->th.th_used_in_team), 3);
        if (KMP_COMPARE_AND_STORE_ACQ32(&(this_thr->th.th_used_in_team), 2,
                                        0) ||
            this_thr->th.th_used_in_team.load() == 0) {
          my_flag.wait(this_thr, true USE_ITT_BUILD_ARG(itt_sync_obj));
        }
#if USE_ITT_BUILD && USE_ITT_NOTIFY
        if ((__itt_sync_create_ptr && itt_sync_obj == NULL) || KMP_ITT_DEBUG) {
          // In fork barrier where we could not get the object reliably
          itt_sync_obj =
              __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier, 0, -1);
          // Cancel wait on previous parallel region...
          __kmp_itt_task_starting(itt_sync_obj);

          if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
            return;

          itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
          if (itt_sync_obj != NULL)
            // Call prepare as early as possible for "new" barrier
            __kmp_itt_task_finished(itt_sync_obj);
        } else
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY */
            if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
          return;
      }
      if (this_thr->th.th_used_in_team.load() != 1 &&
          this_thr->th.th_used_in_team.load() != 3) // spurious wake-up?
        continue;
      if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
        return;

      // At this point, the thread thinks it is in use in a team, or in
      // transition to be used in a team, but it might have reached this barrier
      // before it was marked unused by the team. Unused threads are awoken and
      // shifted to wait on local thread struct elsewhere. It also might reach
      // this point by being picked up for use by a different team. Either way,
      // we need to update the tid.
      tid = __kmp_tid_from_gtid(gtid);
      team = this_thr->th.th_team;
      KMP_DEBUG_ASSERT(tid >= 0);
      KMP_DEBUG_ASSERT(team);
      b = team->t.b;
      my_current_iter = b->iter[tid].iter;
      next_go = my_current_iter + distributedBarrier::MAX_ITERS;
      my_go_index = tid / b->threads_per_go;
      if (this_thr->th.th_used_in_team.load() == 3) {
        KMP_COMPARE_AND_STORE_ACQ32(&(this_thr->th.th_used_in_team), 3, 1);
      }
      // Check if go flag is set
      if (b->go[my_go_index].go.load() != next_go) {
        // Wait on go flag on team
        kmp_atomic_flag_64<false, true> my_flag(
            &(b->go[my_go_index].go), next_go, &(b->sleep[tid].sleep));
        my_flag.wait(this_thr, true USE_ITT_BUILD_ARG(itt_sync_obj));
        KMP_DEBUG_ASSERT(my_current_iter == b->iter[tid].iter ||
                         b->iter[tid].iter == 0);
        KMP_DEBUG_ASSERT(b->sleep[tid].sleep == false);
      }

      if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
        return;
      // At this point, the thread's go location was set. This means the primary
      // thread is safely in the barrier, and so this thread's data is
      // up-to-date, but we should check again that this thread is really in
      // use in the team, as it could have been woken up for the purpose of
      // changing team size, or reaping threads at shutdown.
      if (this_thr->th.th_used_in_team.load() == 1)
        break;
    } while (1);

    if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
      return;

    group_leader = ((tid % b->threads_per_group) == 0);
    if (group_leader) {
      // Tell all the threads in my group they can go!
      for (size_t go_idx = my_go_index + 1;
           go_idx < my_go_index + b->gos_per_group; go_idx++) {
        b->go[go_idx].go.store(next_go);
      }
      // Fence added so that workers can see changes to go. sfence inadequate.
      KMP_MFENCE();
    }

#if KMP_BARRIER_ICV_PUSH
    if (propagate_icvs) { // copy ICVs to final dest
      __kmp_init_implicit_task(team->t.t_ident, team->t.t_threads[tid], team,
                               tid, FALSE);
      copy_icvs(&team->t.t_implicit_task_taskdata[tid].td_icvs,
                (kmp_internal_control_t *)team->t.b->team_icvs);
      copy_icvs(&thr_bar->th_fixed_icvs,
                &team->t.t_implicit_task_taskdata[tid].td_icvs);
    }
#endif
    if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME && group_leader) {
      // This thread is now awake and participating in the barrier;
      // wake up the other threads in the group
      size_t nproc = this_thr->th.th_team_nproc;
      size_t group_end = tid + b->threads_per_group;
      if (nproc < group_end)
        group_end = nproc;
      __kmp_dist_barrier_wakeup(bt, team, tid + 1, group_end, 1, tid);
    }
  } else { //  Primary thread
    team = this_thr->th.th_team;
    b = team->t.b;
    my_current_iter = b->iter[tid].iter;
    next_go = my_current_iter + distributedBarrier::MAX_ITERS;
#if KMP_BARRIER_ICV_PUSH
    if (propagate_icvs) {
      // primary thread has ICVs in final destination; copy
      copy_icvs(&thr_bar->th_fixed_icvs,
                &team->t.t_implicit_task_taskdata[tid].td_icvs);
    }
#endif
    // Tell all the group leaders they can go!
    for (size_t go_idx = 0; go_idx < b->num_gos; go_idx += b->gos_per_group) {
      b->go[go_idx].go.store(next_go);
    }

    if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME) {
      // Wake-up the group leaders
      size_t nproc = this_thr->th.th_team_nproc;
      __kmp_dist_barrier_wakeup(bt, team, tid + b->threads_per_group, nproc,
                                b->threads_per_group, tid);
    }

    // Tell all the threads in my group they can go!
    for (size_t go_idx = 1; go_idx < b->gos_per_group; go_idx++) {
      b->go[go_idx].go.store(next_go);
    }

    // Fence added so that workers can see changes to go. sfence inadequate.
    KMP_MFENCE();

    if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME) {
      // Wake-up the other threads in my group
      size_t nproc = this_thr->th.th_team_nproc;
      size_t group_end = tid + b->threads_per_group;
      if (nproc < group_end)
        group_end = nproc;
      __kmp_dist_barrier_wakeup(bt, team, tid + 1, group_end, 1, tid);
    }
  }
  // Update to next iteration
  KMP_ASSERT(my_current_iter == b->iter[tid].iter);
  b->iter[tid].iter = (b->iter[tid].iter + 1) % distributedBarrier::MAX_ITERS;

  KA_TRACE(
      20, ("__kmp_dist_barrier_release: T#%d(%d:%d) exit for barrier type %d\n",
           gtid, team->t.t_id, tid, bt));
}

// Linear Barrier
template <bool cancellable = false>
static bool __kmp_linear_barrier_gather_template(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    void (*reduce)(void *, void *) USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_linear_gather);
  kmp_team_t *team = this_thr->th.th_team;
  kmp_bstate_t *thr_bar = &this_thr->th.th_bar[bt].bb;
  kmp_info_t **other_threads = team->t.t_threads;

  KA_TRACE(
      20,
      ("__kmp_linear_barrier_gather: T#%d(%d:%d) enter for barrier type %d\n",
       gtid, team->t.t_id, tid, bt));
  KMP_DEBUG_ASSERT(this_thr == other_threads[this_thr->th.th_info.ds.ds_tid]);

#if USE_ITT_BUILD && USE_ITT_NOTIFY
  // Barrier imbalance - save arrive time to the thread
  if (__kmp_forkjoin_frames_mode == 3 || __kmp_forkjoin_frames_mode == 2) {
    this_thr->th.th_bar_arrive_time = this_thr->th.th_bar_min_time =
        __itt_get_timestamp();
  }
#endif
  // We now perform a linear reduction to signal that all of the threads have
  // arrived.
  if (!KMP_MASTER_TID(tid)) {
    KA_TRACE(20,
             ("__kmp_linear_barrier_gather: T#%d(%d:%d) releasing T#%d(%d:%d)"
              "arrived(%p): %llu => %llu\n",
              gtid, team->t.t_id, tid, __kmp_gtid_from_tid(0, team),
              team->t.t_id, 0, &thr_bar->b_arrived, thr_bar->b_arrived,
              thr_bar->b_arrived + KMP_BARRIER_STATE_BUMP));
    // Mark arrival to primary thread
    /* After performing this write, a worker thread may not assume that the team
       is valid any more - it could be deallocated by the primary thread at any
       time. */
    kmp_flag_64<> flag(&thr_bar->b_arrived, other_threads[0]);
    flag.release();
  } else {
    kmp_balign_team_t *team_bar = &team->t.t_bar[bt];
    int nproc = this_thr->th.th_team_nproc;
    int i;
    // Don't have to worry about sleep bit here or atomic since team setting
    kmp_uint64 new_state = team_bar->b_arrived + KMP_BARRIER_STATE_BUMP;

    // Collect all the worker team member threads.
    for (i = 1; i < nproc; ++i) {
#if KMP_CACHE_MANAGE
      // Prefetch next thread's arrived count
      if (i + 1 < nproc)
        KMP_CACHE_PREFETCH(&other_threads[i + 1]->th.th_bar[bt].bb.b_arrived);
#endif /* KMP_CACHE_MANAGE */
      KA_TRACE(20, ("__kmp_linear_barrier_gather: T#%d(%d:%d) wait T#%d(%d:%d) "
                    "arrived(%p) == %llu\n",
                    gtid, team->t.t_id, tid, __kmp_gtid_from_tid(i, team),
                    team->t.t_id, i,
                    &other_threads[i]->th.th_bar[bt].bb.b_arrived, new_state));

      // Wait for worker thread to arrive
      if (cancellable) {
        kmp_flag_64<true, false> flag(
            &other_threads[i]->th.th_bar[bt].bb.b_arrived, new_state);
        if (flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj)))
          return true;
      } else {
        kmp_flag_64<> flag(&other_threads[i]->th.th_bar[bt].bb.b_arrived,
                           new_state);
        flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
      }
#if USE_ITT_BUILD && USE_ITT_NOTIFY
      // Barrier imbalance - write min of the thread time and the other thread
      // time to the thread.
      if (__kmp_forkjoin_frames_mode == 2) {
        this_thr->th.th_bar_min_time = KMP_MIN(
            this_thr->th.th_bar_min_time, other_threads[i]->th.th_bar_min_time);
      }
#endif
      if (reduce) {
        KA_TRACE(100,
                 ("__kmp_linear_barrier_gather: T#%d(%d:%d) += T#%d(%d:%d)\n",
                  gtid, team->t.t_id, tid, __kmp_gtid_from_tid(i, team),
                  team->t.t_id, i));
        OMPT_REDUCTION_DECL(this_thr, gtid);
        OMPT_REDUCTION_BEGIN;
        (*reduce)(this_thr->th.th_local.reduce_data,
                  other_threads[i]->th.th_local.reduce_data);
        OMPT_REDUCTION_END;
      }
    }
    // Don't have to worry about sleep bit here or atomic since team setting
    team_bar->b_arrived = new_state;
    KA_TRACE(20, ("__kmp_linear_barrier_gather: T#%d(%d:%d) set team %d "
                  "arrived(%p) = %llu\n",
                  gtid, team->t.t_id, tid, team->t.t_id, &team_bar->b_arrived,
                  new_state));
  }
  KA_TRACE(
      20,
      ("__kmp_linear_barrier_gather: T#%d(%d:%d) exit for barrier type %d\n",
       gtid, team->t.t_id, tid, bt));
  return false;
}

template <bool cancellable = false>
static bool __kmp_linear_barrier_release_template(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    int propagate_icvs USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_linear_release);
  kmp_bstate_t *thr_bar = &this_thr->th.th_bar[bt].bb;
  kmp_team_t *team;

  if (KMP_MASTER_TID(tid)) {
    unsigned int i;
    kmp_uint32 nproc = this_thr->th.th_team_nproc;
    kmp_info_t **other_threads;

    team = __kmp_threads[gtid]->th.th_team;
    KMP_DEBUG_ASSERT(team != NULL);
    other_threads = team->t.t_threads;

    KA_TRACE(20, ("__kmp_linear_barrier_release: T#%d(%d:%d) primary enter for "
                  "barrier type %d\n",
                  gtid, team->t.t_id, tid, bt));

    if (nproc > 1) {
#if KMP_BARRIER_ICV_PUSH
      {
        KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(USER_icv_copy);
        if (propagate_icvs) {
          ngo_load(&team->t.t_implicit_task_taskdata[0].td_icvs);
          for (i = 1; i < nproc; ++i) {
            __kmp_init_implicit_task(team->t.t_ident, team->t.t_threads[i],
                                     team, i, FALSE);
            ngo_store_icvs(&team->t.t_implicit_task_taskdata[i].td_icvs,
                           &team->t.t_implicit_task_taskdata[0].td_icvs);
          }
          ngo_sync();
        }
      }
#endif // KMP_BARRIER_ICV_PUSH

      // Now, release all of the worker threads
      for (i = 1; i < nproc; ++i) {
#if KMP_CACHE_MANAGE
        // Prefetch next thread's go flag
        if (i + 1 < nproc)
          KMP_CACHE_PREFETCH(&other_threads[i + 1]->th.th_bar[bt].bb.b_go);
#endif /* KMP_CACHE_MANAGE */
        KA_TRACE(
            20,
            ("__kmp_linear_barrier_release: T#%d(%d:%d) releasing T#%d(%d:%d) "
             "go(%p): %u => %u\n",
             gtid, team->t.t_id, tid, other_threads[i]->th.th_info.ds.ds_gtid,
             team->t.t_id, i, &other_threads[i]->th.th_bar[bt].bb.b_go,
             other_threads[i]->th.th_bar[bt].bb.b_go,
             other_threads[i]->th.th_bar[bt].bb.b_go + KMP_BARRIER_STATE_BUMP));
        kmp_flag_64<> flag(&other_threads[i]->th.th_bar[bt].bb.b_go,
                           other_threads[i]);
        flag.release();
      }
    }
  } else { // Wait for the PRIMARY thread to release us
    KA_TRACE(20, ("__kmp_linear_barrier_release: T#%d wait go(%p) == %u\n",
                  gtid, &thr_bar->b_go, KMP_BARRIER_STATE_BUMP));
    if (cancellable) {
      kmp_flag_64<true, false> flag(&thr_bar->b_go, KMP_BARRIER_STATE_BUMP);
      if (flag.wait(this_thr, TRUE USE_ITT_BUILD_ARG(itt_sync_obj)))
        return true;
    } else {
      kmp_flag_64<> flag(&thr_bar->b_go, KMP_BARRIER_STATE_BUMP);
      flag.wait(this_thr, TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
    }
#if USE_ITT_BUILD && USE_ITT_NOTIFY
    if ((__itt_sync_create_ptr && itt_sync_obj == NULL) || KMP_ITT_DEBUG) {
      // In a fork barrier; cannot get the object reliably (or ITTNOTIFY is
      // disabled)
      itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier, 0, -1);
      // Cancel wait on previous parallel region...
      __kmp_itt_task_starting(itt_sync_obj);

      if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
        return false;

      itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
      if (itt_sync_obj != NULL)
        // Call prepare as early as possible for "new" barrier
        __kmp_itt_task_finished(itt_sync_obj);
    } else
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY */
        // Early exit for reaping threads releasing forkjoin barrier
        if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
      return false;
// The worker thread may now assume that the team is valid.
#ifdef KMP_DEBUG
    tid = __kmp_tid_from_gtid(gtid);
    team = __kmp_threads[gtid]->th.th_team;
#endif
    KMP_DEBUG_ASSERT(team != NULL);
    TCW_4(thr_bar->b_go, KMP_INIT_BARRIER_STATE);
    KA_TRACE(20,
             ("__kmp_linear_barrier_release: T#%d(%d:%d) set go(%p) = %u\n",
              gtid, team->t.t_id, tid, &thr_bar->b_go, KMP_INIT_BARRIER_STATE));
    KMP_MB(); // Flush all pending memory write invalidates.
  }
  KA_TRACE(
      20,
      ("__kmp_linear_barrier_release: T#%d(%d:%d) exit for barrier type %d\n",
       gtid, team->t.t_id, tid, bt));
  return false;
}

static void __kmp_linear_barrier_gather(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    void (*reduce)(void *, void *) USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  __kmp_linear_barrier_gather_template<false>(
      bt, this_thr, gtid, tid, reduce USE_ITT_BUILD_ARG(itt_sync_obj));
}

static bool __kmp_linear_barrier_gather_cancellable(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    void (*reduce)(void *, void *) USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  return __kmp_linear_barrier_gather_template<true>(
      bt, this_thr, gtid, tid, reduce USE_ITT_BUILD_ARG(itt_sync_obj));
}

static void __kmp_linear_barrier_release(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    int propagate_icvs USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  __kmp_linear_barrier_release_template<false>(
      bt, this_thr, gtid, tid, propagate_icvs USE_ITT_BUILD_ARG(itt_sync_obj));
}

static bool __kmp_linear_barrier_release_cancellable(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    int propagate_icvs USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  return __kmp_linear_barrier_release_template<true>(
      bt, this_thr, gtid, tid, propagate_icvs USE_ITT_BUILD_ARG(itt_sync_obj));
}

// Tree barrier
static void __kmp_tree_barrier_gather(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    void (*reduce)(void *, void *) USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_tree_gather);
  kmp_team_t *team = this_thr->th.th_team;
  kmp_bstate_t *thr_bar = &this_thr->th.th_bar[bt].bb;
  kmp_info_t **other_threads = team->t.t_threads;
  kmp_uint32 nproc = this_thr->th.th_team_nproc;
  kmp_uint32 branch_bits = __kmp_barrier_gather_branch_bits[bt];
  kmp_uint32 branch_factor = 1 << branch_bits;
  kmp_uint32 child;
  kmp_uint32 child_tid;
  kmp_uint64 new_state = 0;

  KA_TRACE(
      20, ("__kmp_tree_barrier_gather: T#%d(%d:%d) enter for barrier type %d\n",
           gtid, team->t.t_id, tid, bt));
  KMP_DEBUG_ASSERT(this_thr == other_threads[this_thr->th.th_info.ds.ds_tid]);

#if USE_ITT_BUILD && USE_ITT_NOTIFY
  // Barrier imbalance - save arrive time to the thread
  if (__kmp_forkjoin_frames_mode == 3 || __kmp_forkjoin_frames_mode == 2) {
    this_thr->th.th_bar_arrive_time = this_thr->th.th_bar_min_time =
        __itt_get_timestamp();
  }
#endif
  // Perform tree gather to wait until all threads have arrived; reduce any
  // required data as we go
  child_tid = (tid << branch_bits) + 1;
  if (child_tid < nproc) {
    // Parent threads wait for all their children to arrive
    new_state = team->t.t_bar[bt].b_arrived + KMP_BARRIER_STATE_BUMP;
    child = 1;
    do {
      kmp_info_t *child_thr = other_threads[child_tid];
      kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
#if KMP_CACHE_MANAGE
      // Prefetch next thread's arrived count
      if (child + 1 <= branch_factor && child_tid + 1 < nproc)
        KMP_CACHE_PREFETCH(
            &other_threads[child_tid + 1]->th.th_bar[bt].bb.b_arrived);
#endif /* KMP_CACHE_MANAGE */
      KA_TRACE(20,
               ("__kmp_tree_barrier_gather: T#%d(%d:%d) wait T#%d(%d:%u) "
                "arrived(%p) == %llu\n",
                gtid, team->t.t_id, tid, __kmp_gtid_from_tid(child_tid, team),
                team->t.t_id, child_tid, &child_bar->b_arrived, new_state));
      // Wait for child to arrive
      kmp_flag_64<> flag(&child_bar->b_arrived, new_state);
      flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
#if USE_ITT_BUILD && USE_ITT_NOTIFY
      // Barrier imbalance - write min of the thread time and a child time to
      // the thread.
      if (__kmp_forkjoin_frames_mode == 2) {
        this_thr->th.th_bar_min_time = KMP_MIN(this_thr->th.th_bar_min_time,
                                               child_thr->th.th_bar_min_time);
      }
#endif
      if (reduce) {
        KA_TRACE(100,
                 ("__kmp_tree_barrier_gather: T#%d(%d:%d) += T#%d(%d:%u)\n",
                  gtid, team->t.t_id, tid, __kmp_gtid_from_tid(child_tid, team),
                  team->t.t_id, child_tid));
        OMPT_REDUCTION_DECL(this_thr, gtid);
        OMPT_REDUCTION_BEGIN;
        (*reduce)(this_thr->th.th_local.reduce_data,
                  child_thr->th.th_local.reduce_data);
        OMPT_REDUCTION_END;
      }
      child++;
      child_tid++;
    } while (child <= branch_factor && child_tid < nproc);
  }

  if (!KMP_MASTER_TID(tid)) { // Worker threads
    kmp_int32 parent_tid = (tid - 1) >> branch_bits;

    KA_TRACE(20,
             ("__kmp_tree_barrier_gather: T#%d(%d:%d) releasing T#%d(%d:%d) "
              "arrived(%p): %llu => %llu\n",
              gtid, team->t.t_id, tid, __kmp_gtid_from_tid(parent_tid, team),
              team->t.t_id, parent_tid, &thr_bar->b_arrived, thr_bar->b_arrived,
              thr_bar->b_arrived + KMP_BARRIER_STATE_BUMP));

    // Mark arrival to parent thread
    /* After performing this write, a worker thread may not assume that the team
       is valid any more - it could be deallocated by the primary thread at any
       time.  */
    kmp_flag_64<> flag(&thr_bar->b_arrived, other_threads[parent_tid]);
    flag.release();
  } else {
    // Need to update the team arrived pointer if we are the primary thread
    if (nproc > 1) // New value was already computed above
      team->t.t_bar[bt].b_arrived = new_state;
    else
      team->t.t_bar[bt].b_arrived += KMP_BARRIER_STATE_BUMP;
    KA_TRACE(20, ("__kmp_tree_barrier_gather: T#%d(%d:%d) set team %d "
                  "arrived(%p) = %llu\n",
                  gtid, team->t.t_id, tid, team->t.t_id,
                  &team->t.t_bar[bt].b_arrived, team->t.t_bar[bt].b_arrived));
  }
  KA_TRACE(20,
           ("__kmp_tree_barrier_gather: T#%d(%d:%d) exit for barrier type %d\n",
            gtid, team->t.t_id, tid, bt));
}

static void __kmp_tree_barrier_release(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    int propagate_icvs USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_tree_release);
  kmp_team_t *team;
  kmp_bstate_t *thr_bar = &this_thr->th.th_bar[bt].bb;
  kmp_uint32 nproc;
  kmp_uint32 branch_bits = __kmp_barrier_release_branch_bits[bt];
  kmp_uint32 branch_factor = 1 << branch_bits;
  kmp_uint32 child;
  kmp_uint32 child_tid;

  // Perform a tree release for all of the threads that have been gathered
  if (!KMP_MASTER_TID(
          tid)) { // Handle fork barrier workers who aren't part of a team yet
    KA_TRACE(20, ("__kmp_tree_barrier_release: T#%d wait go(%p) == %u\n", gtid,
                  &thr_bar->b_go, KMP_BARRIER_STATE_BUMP));
    // Wait for parent thread to release us
    kmp_flag_64<> flag(&thr_bar->b_go, KMP_BARRIER_STATE_BUMP);
    flag.wait(this_thr, TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
#if USE_ITT_BUILD && USE_ITT_NOTIFY
    if ((__itt_sync_create_ptr && itt_sync_obj == NULL) || KMP_ITT_DEBUG) {
      // In fork barrier where we could not get the object reliably (or
      // ITTNOTIFY is disabled)
      itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier, 0, -1);
      // Cancel wait on previous parallel region...
      __kmp_itt_task_starting(itt_sync_obj);

      if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
        return;

      itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
      if (itt_sync_obj != NULL)
        // Call prepare as early as possible for "new" barrier
        __kmp_itt_task_finished(itt_sync_obj);
    } else
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY */
        // Early exit for reaping threads releasing forkjoin barrier
        if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
      return;

    // The worker thread may now assume that the team is valid.
    team = __kmp_threads[gtid]->th.th_team;
    KMP_DEBUG_ASSERT(team != NULL);
    tid = __kmp_tid_from_gtid(gtid);

    TCW_4(thr_bar->b_go, KMP_INIT_BARRIER_STATE);
    KA_TRACE(20,
             ("__kmp_tree_barrier_release: T#%d(%d:%d) set go(%p) = %u\n", gtid,
              team->t.t_id, tid, &thr_bar->b_go, KMP_INIT_BARRIER_STATE));
    KMP_MB(); // Flush all pending memory write invalidates.
  } else {
    team = __kmp_threads[gtid]->th.th_team;
    KMP_DEBUG_ASSERT(team != NULL);
    KA_TRACE(20, ("__kmp_tree_barrier_release: T#%d(%d:%d) primary enter for "
                  "barrier type %d\n",
                  gtid, team->t.t_id, tid, bt));
  }
  nproc = this_thr->th.th_team_nproc;
  child_tid = (tid << branch_bits) + 1;

  if (child_tid < nproc) {
    kmp_info_t **other_threads = team->t.t_threads;
    child = 1;
    // Parent threads release all their children
    do {
      kmp_info_t *child_thr = other_threads[child_tid];
      kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
#if KMP_CACHE_MANAGE
      // Prefetch next thread's go count
      if (child + 1 <= branch_factor && child_tid + 1 < nproc)
        KMP_CACHE_PREFETCH(
            &other_threads[child_tid + 1]->th.th_bar[bt].bb.b_go);
#endif /* KMP_CACHE_MANAGE */

#if KMP_BARRIER_ICV_PUSH
      {
        KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(USER_icv_copy);
        if (propagate_icvs) {
          __kmp_init_implicit_task(team->t.t_ident,
                                   team->t.t_threads[child_tid], team,
                                   child_tid, FALSE);
          copy_icvs(&team->t.t_implicit_task_taskdata[child_tid].td_icvs,
                    &team->t.t_implicit_task_taskdata[0].td_icvs);
        }
      }
#endif // KMP_BARRIER_ICV_PUSH
      KA_TRACE(20,
               ("__kmp_tree_barrier_release: T#%d(%d:%d) releasing T#%d(%d:%u)"
                "go(%p): %u => %u\n",
                gtid, team->t.t_id, tid, __kmp_gtid_from_tid(child_tid, team),
                team->t.t_id, child_tid, &child_bar->b_go, child_bar->b_go,
                child_bar->b_go + KMP_BARRIER_STATE_BUMP));
      // Release child from barrier
      kmp_flag_64<> flag(&child_bar->b_go, child_thr);
      flag.release();
      child++;
      child_tid++;
    } while (child <= branch_factor && child_tid < nproc);
  }
  KA_TRACE(
      20, ("__kmp_tree_barrier_release: T#%d(%d:%d) exit for barrier type %d\n",
           gtid, team->t.t_id, tid, bt));
}

// Hyper Barrier
static void __kmp_hyper_barrier_gather(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    void (*reduce)(void *, void *) USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_hyper_gather);
  kmp_team_t *team = this_thr->th.th_team;
  kmp_bstate_t *thr_bar = &this_thr->th.th_bar[bt].bb;
  kmp_info_t **other_threads = team->t.t_threads;
  kmp_uint64 new_state = KMP_BARRIER_UNUSED_STATE;
  kmp_uint32 num_threads = this_thr->th.th_team_nproc;
  kmp_uint32 branch_bits = __kmp_barrier_gather_branch_bits[bt];
  kmp_uint32 branch_factor = 1 << branch_bits;
  kmp_uint32 offset;
  kmp_uint32 level;

  KA_TRACE(
      20,
      ("__kmp_hyper_barrier_gather: T#%d(%d:%d) enter for barrier type %d\n",
       gtid, team->t.t_id, tid, bt));
  KMP_DEBUG_ASSERT(this_thr == other_threads[this_thr->th.th_info.ds.ds_tid]);

#if USE_ITT_BUILD && USE_ITT_NOTIFY
  // Barrier imbalance - save arrive time to the thread
  if (__kmp_forkjoin_frames_mode == 3 || __kmp_forkjoin_frames_mode == 2) {
    this_thr->th.th_bar_arrive_time = this_thr->th.th_bar_min_time =
        __itt_get_timestamp();
  }
#endif
  /* Perform a hypercube-embedded tree gather to wait until all of the threads
     have arrived, and reduce any required data as we go.  */
  kmp_flag_64<> p_flag(&thr_bar->b_arrived);
  for (level = 0, offset = 1; offset < num_threads;
       level += branch_bits, offset <<= branch_bits) {
    kmp_uint32 child;
    kmp_uint32 child_tid;

    if (((tid >> level) & (branch_factor - 1)) != 0) {
      kmp_int32 parent_tid = tid & ~((1 << (level + branch_bits)) - 1);

      KMP_MB(); // Synchronize parent and child threads.
      KA_TRACE(20,
               ("__kmp_hyper_barrier_gather: T#%d(%d:%d) releasing T#%d(%d:%d) "
                "arrived(%p): %llu => %llu\n",
                gtid, team->t.t_id, tid, __kmp_gtid_from_tid(parent_tid, team),
                team->t.t_id, parent_tid, &thr_bar->b_arrived,
                thr_bar->b_arrived,
                thr_bar->b_arrived + KMP_BARRIER_STATE_BUMP));
      // Mark arrival to parent thread
      /* After performing this write (in the last iteration of the enclosing for
         loop), a worker thread may not assume that the team is valid any more
         - it could be deallocated by the primary thread at any time.  */
      p_flag.set_waiter(other_threads[parent_tid]);
      p_flag.release();
      break;
    }

    // Parent threads wait for children to arrive
    if (new_state == KMP_BARRIER_UNUSED_STATE)
      new_state = team->t.t_bar[bt].b_arrived + KMP_BARRIER_STATE_BUMP;
    for (child = 1, child_tid = tid + (1 << level);
         child < branch_factor && child_tid < num_threads;
         child++, child_tid += (1 << level)) {
      kmp_info_t *child_thr = other_threads[child_tid];
      kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
#if KMP_CACHE_MANAGE
      kmp_uint32 next_child_tid = child_tid + (1 << level);
      // Prefetch next thread's arrived count
      if (child + 1 < branch_factor && next_child_tid < num_threads)
        KMP_CACHE_PREFETCH(
            &other_threads[next_child_tid]->th.th_bar[bt].bb.b_arrived);
#endif /* KMP_CACHE_MANAGE */
      KA_TRACE(20,
               ("__kmp_hyper_barrier_gather: T#%d(%d:%d) wait T#%d(%d:%u) "
                "arrived(%p) == %llu\n",
                gtid, team->t.t_id, tid, __kmp_gtid_from_tid(child_tid, team),
                team->t.t_id, child_tid, &child_bar->b_arrived, new_state));
      // Wait for child to arrive
      kmp_flag_64<> c_flag(&child_bar->b_arrived, new_state);
      c_flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
      KMP_MB(); // Synchronize parent and child threads.
#if USE_ITT_BUILD && USE_ITT_NOTIFY
      // Barrier imbalance - write min of the thread time and a child time to
      // the thread.
      if (__kmp_forkjoin_frames_mode == 2) {
        this_thr->th.th_bar_min_time = KMP_MIN(this_thr->th.th_bar_min_time,
                                               child_thr->th.th_bar_min_time);
      }
#endif
      if (reduce) {
        KA_TRACE(100,
                 ("__kmp_hyper_barrier_gather: T#%d(%d:%d) += T#%d(%d:%u)\n",
                  gtid, team->t.t_id, tid, __kmp_gtid_from_tid(child_tid, team),
                  team->t.t_id, child_tid));
        OMPT_REDUCTION_DECL(this_thr, gtid);
        OMPT_REDUCTION_BEGIN;
        (*reduce)(this_thr->th.th_local.reduce_data,
                  child_thr->th.th_local.reduce_data);
        OMPT_REDUCTION_END;
      }
    }
  }

  if (KMP_MASTER_TID(tid)) {
    // Need to update the team arrived pointer if we are the primary thread
    if (new_state == KMP_BARRIER_UNUSED_STATE)
      team->t.t_bar[bt].b_arrived += KMP_BARRIER_STATE_BUMP;
    else
      team->t.t_bar[bt].b_arrived = new_state;
    KA_TRACE(20, ("__kmp_hyper_barrier_gather: T#%d(%d:%d) set team %d "
                  "arrived(%p) = %llu\n",
                  gtid, team->t.t_id, tid, team->t.t_id,
                  &team->t.t_bar[bt].b_arrived, team->t.t_bar[bt].b_arrived));
  }
  KA_TRACE(
      20, ("__kmp_hyper_barrier_gather: T#%d(%d:%d) exit for barrier type %d\n",
           gtid, team->t.t_id, tid, bt));
}

// The reverse versions seem to beat the forward versions overall
#define KMP_REVERSE_HYPER_BAR
static void __kmp_hyper_barrier_release(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    int propagate_icvs USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_hyper_release);
  kmp_team_t *team;
  kmp_bstate_t *thr_bar = &this_thr->th.th_bar[bt].bb;
  kmp_info_t **other_threads;
  kmp_uint32 num_threads;
  kmp_uint32 branch_bits = __kmp_barrier_release_branch_bits[bt];
  kmp_uint32 branch_factor = 1 << branch_bits;
  kmp_uint32 child;
  kmp_uint32 child_tid;
  kmp_uint32 offset;
  kmp_uint32 level;

  /* Perform a hypercube-embedded tree release for all of the threads that have
     been gathered. If KMP_REVERSE_HYPER_BAR is defined (default) the threads
     are released in the reverse order of the corresponding gather, otherwise
     threads are released in the same order. */
  if (KMP_MASTER_TID(tid)) { // primary thread
    team = __kmp_threads[gtid]->th.th_team;
    KMP_DEBUG_ASSERT(team != NULL);
    KA_TRACE(20, ("__kmp_hyper_barrier_release: T#%d(%d:%d) primary enter for "
                  "barrier type %d\n",
                  gtid, team->t.t_id, tid, bt));
#if KMP_BARRIER_ICV_PUSH
    if (propagate_icvs) { // primary already has ICVs in final destination; copy
      copy_icvs(&thr_bar->th_fixed_icvs,
                &team->t.t_implicit_task_taskdata[tid].td_icvs);
    }
#endif
  } else { // Handle fork barrier workers who aren't part of a team yet
    KA_TRACE(20, ("__kmp_hyper_barrier_release: T#%d wait go(%p) == %u\n", gtid,
                  &thr_bar->b_go, KMP_BARRIER_STATE_BUMP));
    // Wait for parent thread to release us
    kmp_flag_64<> flag(&thr_bar->b_go, KMP_BARRIER_STATE_BUMP);
    flag.wait(this_thr, TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
#if USE_ITT_BUILD && USE_ITT_NOTIFY
    if ((__itt_sync_create_ptr && itt_sync_obj == NULL) || KMP_ITT_DEBUG) {
      // In fork barrier where we could not get the object reliably
      itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier, 0, -1);
      // Cancel wait on previous parallel region...
      __kmp_itt_task_starting(itt_sync_obj);

      if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
        return;

      itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
      if (itt_sync_obj != NULL)
        // Call prepare as early as possible for "new" barrier
        __kmp_itt_task_finished(itt_sync_obj);
    } else
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY */
        // Early exit for reaping threads releasing forkjoin barrier
        if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
      return;

    // The worker thread may now assume that the team is valid.
    team = __kmp_threads[gtid]->th.th_team;
    KMP_DEBUG_ASSERT(team != NULL);
    tid = __kmp_tid_from_gtid(gtid);

    TCW_4(thr_bar->b_go, KMP_INIT_BARRIER_STATE);
    KA_TRACE(20,
             ("__kmp_hyper_barrier_release: T#%d(%d:%d) set go(%p) = %u\n",
              gtid, team->t.t_id, tid, &thr_bar->b_go, KMP_INIT_BARRIER_STATE));
    KMP_MB(); // Flush all pending memory write invalidates.
  }
  num_threads = this_thr->th.th_team_nproc;
  other_threads = team->t.t_threads;

#ifdef KMP_REVERSE_HYPER_BAR
  // Count up to correct level for parent
  for (level = 0, offset = 1;
       offset < num_threads && (((tid >> level) & (branch_factor - 1)) == 0);
       level += branch_bits, offset <<= branch_bits)
    ;

  // Now go down from there
  for (level -= branch_bits, offset >>= branch_bits; offset != 0;
       level -= branch_bits, offset >>= branch_bits)
#else
  // Go down the tree, level by level
  for (level = 0, offset = 1; offset < num_threads;
       level += branch_bits, offset <<= branch_bits)
#endif // KMP_REVERSE_HYPER_BAR
  {
#ifdef KMP_REVERSE_HYPER_BAR
    /* Now go in reverse order through the children, highest to lowest.
       Initial setting of child is conservative here. */
    child = num_threads >> ((level == 0) ? level : level - 1);
    for (child = (child < branch_factor - 1) ? child : branch_factor - 1,
        child_tid = tid + (child << level);
         child >= 1; child--, child_tid -= (1 << level))
#else
    if (((tid >> level) & (branch_factor - 1)) != 0)
      // No need to go lower than this, since this is the level parent would be
      // notified
      break;
    // Iterate through children on this level of the tree
    for (child = 1, child_tid = tid + (1 << level);
         child < branch_factor && child_tid < num_threads;
         child++, child_tid += (1 << level))
#endif // KMP_REVERSE_HYPER_BAR
    {
      if (child_tid >= num_threads)
        continue; // Child doesn't exist so keep going
      else {
        kmp_info_t *child_thr = other_threads[child_tid];
        kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
#if KMP_CACHE_MANAGE
        kmp_uint32 next_child_tid = child_tid - (1 << level);
// Prefetch next thread's go count
#ifdef KMP_REVERSE_HYPER_BAR
        if (child - 1 >= 1 && next_child_tid < num_threads)
#else
        if (child + 1 < branch_factor && next_child_tid < num_threads)
#endif // KMP_REVERSE_HYPER_BAR
          KMP_CACHE_PREFETCH(
              &other_threads[next_child_tid]->th.th_bar[bt].bb.b_go);
#endif /* KMP_CACHE_MANAGE */

#if KMP_BARRIER_ICV_PUSH
        if (propagate_icvs) // push my fixed ICVs to my child
          copy_icvs(&child_bar->th_fixed_icvs, &thr_bar->th_fixed_icvs);
#endif // KMP_BARRIER_ICV_PUSH

        KA_TRACE(
            20,
            ("__kmp_hyper_barrier_release: T#%d(%d:%d) releasing T#%d(%d:%u)"
             "go(%p): %u => %u\n",
             gtid, team->t.t_id, tid, __kmp_gtid_from_tid(child_tid, team),
             team->t.t_id, child_tid, &child_bar->b_go, child_bar->b_go,
             child_bar->b_go + KMP_BARRIER_STATE_BUMP));
        // Release child from barrier
        kmp_flag_64<> flag(&child_bar->b_go, child_thr);
        flag.release();
      }
    }
  }
#if KMP_BARRIER_ICV_PUSH
  if (propagate_icvs &&
      !KMP_MASTER_TID(tid)) { // copy ICVs locally to final dest
    __kmp_init_implicit_task(team->t.t_ident, team->t.t_threads[tid], team, tid,
                             FALSE);
    copy_icvs(&team->t.t_implicit_task_taskdata[tid].td_icvs,
              &thr_bar->th_fixed_icvs);
  }
#endif
  KA_TRACE(
      20,
      ("__kmp_hyper_barrier_release: T#%d(%d:%d) exit for barrier type %d\n",
       gtid, team->t.t_id, tid, bt));
}

// Hierarchical Barrier

// Initialize thread barrier data
/* Initializes/re-initializes the hierarchical barrier data stored on a thread.
   Performs the minimum amount of initialization required based on how the team
   has changed. Returns true if leaf children will require both on-core and
   traditional wake-up mechanisms. For example, if the team size increases,
   threads already in the team will respond to on-core wakeup on their parent
   thread, but threads newly added to the team will only be listening on the
   their local b_go. */
static bool __kmp_init_hierarchical_barrier_thread(enum barrier_type bt,
                                                   kmp_bstate_t *thr_bar,
                                                   kmp_uint32 nproc, int gtid,
                                                   int tid, kmp_team_t *team) {
  // Checks to determine if (re-)initialization is needed
  bool uninitialized = thr_bar->team == NULL;
  bool team_changed = team != thr_bar->team;
  bool team_sz_changed = nproc != thr_bar->nproc;
  bool tid_changed = tid != thr_bar->old_tid;
  bool retval = false;

  if (uninitialized || team_sz_changed) {
    __kmp_get_hierarchy(nproc, thr_bar);
  }

  if (uninitialized || team_sz_changed || tid_changed) {
    thr_bar->my_level = thr_bar->depth - 1; // default for primary thread
    thr_bar->parent_tid = -1; // default for primary thread
    if (!KMP_MASTER_TID(tid)) {
      // if not primary thread, find parent thread in hierarchy
      kmp_uint32 d = 0;
      while (d < thr_bar->depth) { // find parent based on level of thread in
        // hierarchy, and note level
        kmp_uint32 rem;
        if (d == thr_bar->depth - 2) { // reached level right below the primary
          thr_bar->parent_tid = 0;
          thr_bar->my_level = d;
          break;
        } else if ((rem = tid % thr_bar->skip_per_level[d + 1]) != 0) {
          // TODO: can we make the above op faster?
          // thread is not a subtree root at next level, so this is max
          thr_bar->parent_tid = tid - rem;
          thr_bar->my_level = d;
          break;
        }
        ++d;
      }
    }
    __kmp_type_convert(7 - ((tid - thr_bar->parent_tid) /
                            (thr_bar->skip_per_level[thr_bar->my_level])),
                       &(thr_bar->offset));
    thr_bar->old_tid = tid;
    thr_bar->wait_flag = KMP_BARRIER_NOT_WAITING;
    thr_bar->team = team;
    thr_bar->parent_bar =
        &team->t.t_threads[thr_bar->parent_tid]->th.th_bar[bt].bb;
  }
  if (uninitialized || team_changed || tid_changed) {
    thr_bar->team = team;
    thr_bar->parent_bar =
        &team->t.t_threads[thr_bar->parent_tid]->th.th_bar[bt].bb;
    retval = true;
  }
  if (uninitialized || team_sz_changed || tid_changed) {
    thr_bar->nproc = nproc;
    thr_bar->leaf_kids = thr_bar->base_leaf_kids;
    if (thr_bar->my_level == 0)
      thr_bar->leaf_kids = 0;
    if (thr_bar->leaf_kids && (kmp_uint32)tid + thr_bar->leaf_kids + 1 > nproc)
      __kmp_type_convert(nproc - tid - 1, &(thr_bar->leaf_kids));
    thr_bar->leaf_state = 0;
    for (int i = 0; i < thr_bar->leaf_kids; ++i)
      ((char *)&(thr_bar->leaf_state))[7 - i] = 1;
  }
  return retval;
}

static void __kmp_hierarchical_barrier_gather(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    void (*reduce)(void *, void *) USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_hier_gather);
  kmp_team_t *team = this_thr->th.th_team;
  kmp_bstate_t *thr_bar = &this_thr->th.th_bar[bt].bb;
  kmp_uint32 nproc = this_thr->th.th_team_nproc;
  kmp_info_t **other_threads = team->t.t_threads;
  kmp_uint64 new_state = 0;

  int level = team->t.t_level;
  if (other_threads[0]
          ->th.th_teams_microtask) // are we inside the teams construct?
    if (this_thr->th.th_teams_size.nteams > 1)
      ++level; // level was not increased in teams construct for team_of_masters
  if (level == 1)
    thr_bar->use_oncore_barrier = 1;
  else
    thr_bar->use_oncore_barrier = 0; // Do not use oncore barrier when nested

  KA_TRACE(20, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) enter for "
                "barrier type %d\n",
                gtid, team->t.t_id, tid, bt));
  KMP_DEBUG_ASSERT(this_thr == other_threads[this_thr->th.th_info.ds.ds_tid]);

#if USE_ITT_BUILD && USE_ITT_NOTIFY
  // Barrier imbalance - save arrive time to the thread
  if (__kmp_forkjoin_frames_mode == 3 || __kmp_forkjoin_frames_mode == 2) {
    this_thr->th.th_bar_arrive_time = __itt_get_timestamp();
  }
#endif

  (void)__kmp_init_hierarchical_barrier_thread(bt, thr_bar, nproc, gtid, tid,
                                               team);

  if (thr_bar->my_level) { // not a leaf (my_level==0 means leaf)
    kmp_int32 child_tid;
    new_state =
        (kmp_uint64)team->t.t_bar[bt].b_arrived + KMP_BARRIER_STATE_BUMP;
    if (__kmp_dflt_blocktime == KMP_MAX_BLOCKTIME &&
        thr_bar->use_oncore_barrier) {
      if (thr_bar->leaf_kids) {
        // First, wait for leaf children to check-in on my b_arrived flag
        kmp_uint64 leaf_state =
            KMP_MASTER_TID(tid)
                ? thr_bar->b_arrived | thr_bar->leaf_state
                : team->t.t_bar[bt].b_arrived | thr_bar->leaf_state;
        KA_TRACE(20, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) waiting "
                      "for leaf kids\n",
                      gtid, team->t.t_id, tid));
        kmp_flag_64<> flag(&thr_bar->b_arrived, leaf_state);
        flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
        if (reduce) {
          OMPT_REDUCTION_DECL(this_thr, gtid);
          OMPT_REDUCTION_BEGIN;
          for (child_tid = tid + 1; child_tid <= tid + thr_bar->leaf_kids;
               ++child_tid) {
            KA_TRACE(100, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) += "
                           "T#%d(%d:%d)\n",
                           gtid, team->t.t_id, tid,
                           __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                           child_tid));
            (*reduce)(this_thr->th.th_local.reduce_data,
                      other_threads[child_tid]->th.th_local.reduce_data);
          }
          OMPT_REDUCTION_END;
        }
        // clear leaf_state bits
        KMP_TEST_THEN_AND64(&thr_bar->b_arrived, ~(thr_bar->leaf_state));
      }
      // Next, wait for higher level children on each child's b_arrived flag
      for (kmp_uint32 d = 1; d < thr_bar->my_level;
           ++d) { // gather lowest level threads first, but skip 0
        kmp_uint32 last = tid + thr_bar->skip_per_level[d + 1],
                   skip = thr_bar->skip_per_level[d];
        if (last > nproc)
          last = nproc;
        for (child_tid = tid + skip; child_tid < (int)last; child_tid += skip) {
          kmp_info_t *child_thr = other_threads[child_tid];
          kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
          KA_TRACE(20, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) wait "
                        "T#%d(%d:%d) "
                        "arrived(%p) == %llu\n",
                        gtid, team->t.t_id, tid,
                        __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                        child_tid, &child_bar->b_arrived, new_state));
          kmp_flag_64<> flag(&child_bar->b_arrived, new_state);
          flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
          if (reduce) {
            KA_TRACE(100, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) += "
                           "T#%d(%d:%d)\n",
                           gtid, team->t.t_id, tid,
                           __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                           child_tid));
            (*reduce)(this_thr->th.th_local.reduce_data,
                      child_thr->th.th_local.reduce_data);
          }
        }
      }
    } else { // Blocktime is not infinite
      for (kmp_uint32 d = 0; d < thr_bar->my_level;
           ++d) { // Gather lowest level threads first
        kmp_uint32 last = tid + thr_bar->skip_per_level[d + 1],
                   skip = thr_bar->skip_per_level[d];
        if (last > nproc)
          last = nproc;
        for (child_tid = tid + skip; child_tid < (int)last; child_tid += skip) {
          kmp_info_t *child_thr = other_threads[child_tid];
          kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
          KA_TRACE(20, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) wait "
                        "T#%d(%d:%d) "
                        "arrived(%p) == %llu\n",
                        gtid, team->t.t_id, tid,
                        __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                        child_tid, &child_bar->b_arrived, new_state));
          kmp_flag_64<> flag(&child_bar->b_arrived, new_state);
          flag.wait(this_thr, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
          if (reduce) {
            KA_TRACE(100, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) += "
                           "T#%d(%d:%d)\n",
                           gtid, team->t.t_id, tid,
                           __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                           child_tid));
            (*reduce)(this_thr->th.th_local.reduce_data,
                      child_thr->th.th_local.reduce_data);
          }
        }
      }
    }
  }
  // All subordinates are gathered; now release parent if not primary thread

  if (!KMP_MASTER_TID(tid)) { // worker threads release parent in hierarchy
    KA_TRACE(20, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) releasing"
                  " T#%d(%d:%d) arrived(%p): %llu => %llu\n",
                  gtid, team->t.t_id, tid,
                  __kmp_gtid_from_tid(thr_bar->parent_tid, team), team->t.t_id,
                  thr_bar->parent_tid, &thr_bar->b_arrived, thr_bar->b_arrived,
                  thr_bar->b_arrived + KMP_BARRIER_STATE_BUMP));
    /* Mark arrival to parent: After performing this write, a worker thread may
       not assume that the team is valid any more - it could be deallocated by
       the primary thread at any time. */
    if (thr_bar->my_level || __kmp_dflt_blocktime != KMP_MAX_BLOCKTIME ||
        !thr_bar->use_oncore_barrier) { // Parent is waiting on my b_arrived
      // flag; release it
      kmp_flag_64<> flag(&thr_bar->b_arrived,
                         other_threads[thr_bar->parent_tid]);
      flag.release();
    } else {
      // Leaf does special release on "offset" bits of parent's b_arrived flag
      thr_bar->b_arrived = team->t.t_bar[bt].b_arrived + KMP_BARRIER_STATE_BUMP;
      kmp_flag_oncore flag(&thr_bar->parent_bar->b_arrived,
                           thr_bar->offset + 1);
      flag.set_waiter(other_threads[thr_bar->parent_tid]);
      flag.release();
    }
  } else { // Primary thread needs to update the team's b_arrived value
    team->t.t_bar[bt].b_arrived = new_state;
    KA_TRACE(20, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) set team %d "
                  "arrived(%p) = %llu\n",
                  gtid, team->t.t_id, tid, team->t.t_id,
                  &team->t.t_bar[bt].b_arrived, team->t.t_bar[bt].b_arrived));
  }
  // Is the team access below unsafe or just technically invalid?
  KA_TRACE(20, ("__kmp_hierarchical_barrier_gather: T#%d(%d:%d) exit for "
                "barrier type %d\n",
                gtid, team->t.t_id, tid, bt));
}

static void __kmp_hierarchical_barrier_release(
    enum barrier_type bt, kmp_info_t *this_thr, int gtid, int tid,
    int propagate_icvs USE_ITT_BUILD_ARG(void *itt_sync_obj)) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_hier_release);
  kmp_team_t *team;
  kmp_bstate_t *thr_bar = &this_thr->th.th_bar[bt].bb;
  kmp_uint32 nproc;
  bool team_change = false; // indicates on-core barrier shouldn't be used

  if (KMP_MASTER_TID(tid)) {
    team = __kmp_threads[gtid]->th.th_team;
    KMP_DEBUG_ASSERT(team != NULL);
    KA_TRACE(20, ("__kmp_hierarchical_barrier_release: T#%d(%d:%d) primary "
                  "entered barrier type %d\n",
                  gtid, team->t.t_id, tid, bt));
  } else { // Worker threads
    // Wait for parent thread to release me
    if (!thr_bar->use_oncore_barrier ||
        __kmp_dflt_blocktime != KMP_MAX_BLOCKTIME || thr_bar->my_level != 0 ||
        thr_bar->team == NULL) {
      // Use traditional method of waiting on my own b_go flag
      thr_bar->wait_flag = KMP_BARRIER_OWN_FLAG;
      kmp_flag_64<> flag(&thr_bar->b_go, KMP_BARRIER_STATE_BUMP);
      flag.wait(this_thr, TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
      TCW_8(thr_bar->b_go,
            KMP_INIT_BARRIER_STATE); // Reset my b_go flag for next time
    } else { // Thread barrier data is initialized, this is a leaf, blocktime is
      // infinite, not nested
      // Wait on my "offset" bits on parent's b_go flag
      thr_bar->wait_flag = KMP_BARRIER_PARENT_FLAG;
      kmp_flag_oncore flag(&thr_bar->parent_bar->b_go, KMP_BARRIER_STATE_BUMP,
                           thr_bar->offset + 1, bt,
                           this_thr USE_ITT_BUILD_ARG(itt_sync_obj));
      flag.wait(this_thr, TRUE);
      if (thr_bar->wait_flag ==
          KMP_BARRIER_SWITCHING) { // Thread was switched to own b_go
        TCW_8(thr_bar->b_go,
              KMP_INIT_BARRIER_STATE); // Reset my b_go flag for next time
      } else { // Reset my bits on parent's b_go flag
        (RCAST(volatile char *,
               &(thr_bar->parent_bar->b_go)))[thr_bar->offset + 1] = 0;
      }
    }
    thr_bar->wait_flag = KMP_BARRIER_NOT_WAITING;
    // Early exit for reaping threads releasing forkjoin barrier
    if (bt == bs_forkjoin_barrier && TCR_4(__kmp_global.g.g_done))
      return;
    // The worker thread may now assume that the team is valid.
    team = __kmp_threads[gtid]->th.th_team;
    KMP_DEBUG_ASSERT(team != NULL);
    tid = __kmp_tid_from_gtid(gtid);

    KA_TRACE(
        20,
        ("__kmp_hierarchical_barrier_release: T#%d(%d:%d) set go(%p) = %u\n",
         gtid, team->t.t_id, tid, &thr_bar->b_go, KMP_INIT_BARRIER_STATE));
    KMP_MB(); // Flush all pending memory write invalidates.
  }

  nproc = this_thr->th.th_team_nproc;
  int level = team->t.t_level;
  if (team->t.t_threads[0]
          ->th.th_teams_microtask) { // are we inside the teams construct?
    if (team->t.t_pkfn != (microtask_t)__kmp_teams_master &&
        this_thr->th.th_teams_level == level)
      ++level; // level was not increased in teams construct for team_of_workers
    if (this_thr->th.th_teams_size.nteams > 1)
      ++level; // level was not increased in teams construct for team_of_masters
  }
  if (level == 1)
    thr_bar->use_oncore_barrier = 1;
  else
    thr_bar->use_oncore_barrier = 0; // Do not use oncore barrier when nested

  // If the team size has increased, we still communicate with old leaves via
  // oncore barrier.
  unsigned short int old_leaf_kids = thr_bar->leaf_kids;
  kmp_uint64 old_leaf_state = thr_bar->leaf_state;
  team_change = __kmp_init_hierarchical_barrier_thread(bt, thr_bar, nproc, gtid,
                                                       tid, team);
  // But if the entire team changes, we won't use oncore barrier at all
  if (team_change)
    old_leaf_kids = 0;

#if KMP_BARRIER_ICV_PUSH
  if (propagate_icvs) {
    __kmp_init_implicit_task(team->t.t_ident, team->t.t_threads[tid], team, tid,
                             FALSE);
    if (KMP_MASTER_TID(
            tid)) { // primary already has copy in final destination; copy
      copy_icvs(&thr_bar->th_fixed_icvs,
                &team->t.t_implicit_task_taskdata[tid].td_icvs);
    } else if (__kmp_dflt_blocktime == KMP_MAX_BLOCKTIME &&
               thr_bar->use_oncore_barrier) { // optimization for inf blocktime
      if (!thr_bar->my_level) // I'm a leaf in the hierarchy (my_level==0)
        // leaves (on-core children) pull parent's fixed ICVs directly to local
        // ICV store
        copy_icvs(&team->t.t_implicit_task_taskdata[tid].td_icvs,
                  &thr_bar->parent_bar->th_fixed_icvs);
      // non-leaves will get ICVs piggybacked with b_go via NGO store
    } else { // blocktime is not infinite; pull ICVs from parent's fixed ICVs
      if (thr_bar->my_level) // not a leaf; copy ICVs to my fixed ICVs child can
        // access
        copy_icvs(&thr_bar->th_fixed_icvs, &thr_bar->parent_bar->th_fixed_icvs);
      else // leaves copy parent's fixed ICVs directly to local ICV store
        copy_icvs(&team->t.t_implicit_task_taskdata[tid].td_icvs,
                  &thr_bar->parent_bar->th_fixed_icvs);
    }
  }
#endif // KMP_BARRIER_ICV_PUSH

  // Now, release my children
  if (thr_bar->my_level) { // not a leaf
    kmp_int32 child_tid;
    kmp_uint32 last;
    if (__kmp_dflt_blocktime == KMP_MAX_BLOCKTIME &&
        thr_bar->use_oncore_barrier) {
      if (KMP_MASTER_TID(tid)) { // do a flat release
        // Set local b_go to bump children via NGO store of the cache line
        // containing IVCs and b_go.
        thr_bar->b_go = KMP_BARRIER_STATE_BUMP;
        // Use ngo stores if available; b_go piggybacks in the last 8 bytes of
        // the cache line
        ngo_load(&thr_bar->th_fixed_icvs);
        // This loops over all the threads skipping only the leaf nodes in the
        // hierarchy
        for (child_tid = thr_bar->skip_per_level[1]; child_tid < (int)nproc;
             child_tid += thr_bar->skip_per_level[1]) {
          kmp_bstate_t *child_bar =
              &team->t.t_threads[child_tid]->th.th_bar[bt].bb;
          KA_TRACE(20, ("__kmp_hierarchical_barrier_release: T#%d(%d:%d) "
                        "releasing T#%d(%d:%d)"
                        " go(%p): %u => %u\n",
                        gtid, team->t.t_id, tid,
                        __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                        child_tid, &child_bar->b_go, child_bar->b_go,
                        child_bar->b_go + KMP_BARRIER_STATE_BUMP));
          // Use ngo store (if available) to both store ICVs and release child
          // via child's b_go
          ngo_store_go(&child_bar->th_fixed_icvs, &thr_bar->th_fixed_icvs);
        }
        ngo_sync();
      }
      TCW_8(thr_bar->b_go,
            KMP_INIT_BARRIER_STATE); // Reset my b_go flag for next time
      // Now, release leaf children
      if (thr_bar->leaf_kids) { // if there are any
        // We test team_change on the off-chance that the level 1 team changed.
        if (team_change ||
            old_leaf_kids < thr_bar->leaf_kids) { // some old, some new
          if (old_leaf_kids) { // release old leaf kids
            thr_bar->b_go |= old_leaf_state;
          }
          // Release new leaf kids
          last = tid + thr_bar->skip_per_level[1];
          if (last > nproc)
            last = nproc;
          for (child_tid = tid + 1 + old_leaf_kids; child_tid < (int)last;
               ++child_tid) { // skip_per_level[0]=1
            kmp_info_t *child_thr = team->t.t_threads[child_tid];
            kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
            KA_TRACE(
                20,
                ("__kmp_hierarchical_barrier_release: T#%d(%d:%d) releasing"
                 " T#%d(%d:%d) go(%p): %u => %u\n",
                 gtid, team->t.t_id, tid, __kmp_gtid_from_tid(child_tid, team),
                 team->t.t_id, child_tid, &child_bar->b_go, child_bar->b_go,
                 child_bar->b_go + KMP_BARRIER_STATE_BUMP));
            // Release child using child's b_go flag
            kmp_flag_64<> flag(&child_bar->b_go, child_thr);
            flag.release();
          }
        } else { // Release all children at once with leaf_state bits on my own
          // b_go flag
          thr_bar->b_go |= thr_bar->leaf_state;
        }
      }
    } else { // Blocktime is not infinite; do a simple hierarchical release
      for (int d = thr_bar->my_level - 1; d >= 0;
           --d) { // Release highest level threads first
        last = tid + thr_bar->skip_per_level[d + 1];
        kmp_uint32 skip = thr_bar->skip_per_level[d];
        if (last > nproc)
          last = nproc;
        for (child_tid = tid + skip; child_tid < (int)last; child_tid += skip) {
          kmp_info_t *child_thr = team->t.t_threads[child_tid];
          kmp_bstate_t *child_bar = &child_thr->th.th_bar[bt].bb;
          KA_TRACE(20, ("__kmp_hierarchical_barrier_release: T#%d(%d:%d) "
                        "releasing T#%d(%d:%d) go(%p): %u => %u\n",
                        gtid, team->t.t_id, tid,
                        __kmp_gtid_from_tid(child_tid, team), team->t.t_id,
                        child_tid, &child_bar->b_go, child_bar->b_go,
                        child_bar->b_go + KMP_BARRIER_STATE_BUMP));
          // Release child using child's b_go flag
          kmp_flag_64<> flag(&child_bar->b_go, child_thr);
          flag.release();
        }
      }
    }
#if KMP_BARRIER_ICV_PUSH
    if (propagate_icvs && !KMP_MASTER_TID(tid))
      // non-leaves copy ICVs from fixed ICVs to local dest
      copy_icvs(&team->t.t_implicit_task_taskdata[tid].td_icvs,
                &thr_bar->th_fixed_icvs);
#endif // KMP_BARRIER_ICV_PUSH
  }
  KA_TRACE(20, ("__kmp_hierarchical_barrier_release: T#%d(%d:%d) exit for "
                "barrier type %d\n",
                gtid, team->t.t_id, tid, bt));
}

// End of Barrier Algorithms

// type traits for cancellable value
// if cancellable is true, then is_cancellable is a normal boolean variable
// if cancellable is false, then is_cancellable is a compile time constant
template <bool cancellable> struct is_cancellable {};
template <> struct is_cancellable<true> {
  bool value;
  is_cancellable() : value(false) {}
  is_cancellable(bool b) : value(b) {}
  is_cancellable &operator=(bool b) {
    value = b;
    return *this;
  }
  operator bool() const { return value; }
};
template <> struct is_cancellable<false> {
  is_cancellable &operator=(bool b) { return *this; }
  constexpr operator bool() const { return false; }
};

// Internal function to do a barrier.
/* If is_split is true, do a split barrier, otherwise, do a plain barrier
   If reduce is non-NULL, do a split reduction barrier, otherwise, do a split
   barrier
   When cancellable = false,
     Returns 0 if primary thread, 1 if worker thread.
   When cancellable = true
     Returns 0 if not cancelled, 1 if cancelled.  */
template <bool cancellable = false>
static int __kmp_barrier_template(enum barrier_type bt, int gtid, int is_split,
                                  size_t reduce_size, void *reduce_data,
                                  void (*reduce)(void *, void *)) {
  KMP_TIME_PARTITIONED_BLOCK(OMP_plain_barrier);
  KMP_SET_THREAD_STATE_BLOCK(PLAIN_BARRIER);
  int tid = __kmp_tid_from_gtid(gtid);
  kmp_info_t *this_thr = __kmp_threads[gtid];
  kmp_team_t *team = this_thr->th.th_team;
  int status = 0;
  is_cancellable<cancellable> cancelled;
#if OMPT_SUPPORT && OMPT_OPTIONAL
  ompt_data_t *my_task_data;
  ompt_data_t *my_parallel_data;
  void *return_address;
  ompt_sync_region_t barrier_kind;
#endif

  KA_TRACE(15, ("__kmp_barrier: T#%d(%d:%d) has arrived\n", gtid,
                __kmp_team_from_gtid(gtid)->t.t_id, __kmp_tid_from_gtid(gtid)));

#if OMPT_SUPPORT
  if (ompt_enabled.enabled) {
#if OMPT_OPTIONAL
    my_task_data = OMPT_CUR_TASK_DATA(this_thr);
    my_parallel_data = OMPT_CUR_TEAM_DATA(this_thr);
    return_address = OMPT_LOAD_RETURN_ADDRESS(gtid);
    barrier_kind = __ompt_get_barrier_kind(bt, this_thr);
    if (ompt_enabled.ompt_callback_sync_region) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
          barrier_kind, ompt_scope_begin, my_parallel_data, my_task_data,
          return_address);
    }
    if (ompt_enabled.ompt_callback_sync_region_wait) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
          barrier_kind, ompt_scope_begin, my_parallel_data, my_task_data,
          return_address);
    }
#endif
    // It is OK to report the barrier state after the barrier begin callback.
    // According to the OMPT specification, a compliant implementation may
    // even delay reporting this state until the barrier begins to wait.
    this_thr->th.ompt_thread_info.state = ompt_state_wait_barrier;
  }
#endif

  if (!team->t.t_serialized) {
#if USE_ITT_BUILD
    // This value will be used in itt notify events below.
    void *itt_sync_obj = NULL;
#if USE_ITT_NOTIFY
    if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
      itt_sync_obj = __kmp_itt_barrier_object(gtid, bt, 1);
#endif
#endif /* USE_ITT_BUILD */
    if (__kmp_tasking_mode == tskm_extra_barrier) {
      __kmp_tasking_barrier(team, this_thr, gtid);
      KA_TRACE(15,
               ("__kmp_barrier: T#%d(%d:%d) past tasking barrier\n", gtid,
                __kmp_team_from_gtid(gtid)->t.t_id, __kmp_tid_from_gtid(gtid)));
    }

    /* Copy the blocktime info to the thread, where __kmp_wait_template() can
       access it when the team struct is not guaranteed to exist. */
    // See note about the corresponding code in __kmp_join_barrier() being
    // performance-critical.
    if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME) {
#if KMP_USE_MONITOR
      this_thr->th.th_team_bt_intervals =
          team->t.t_implicit_task_taskdata[tid].td_icvs.bt_intervals;
      this_thr->th.th_team_bt_set =
          team->t.t_implicit_task_taskdata[tid].td_icvs.bt_set;
#else
      this_thr->th.th_team_bt_intervals = KMP_BLOCKTIME_INTERVAL(team, tid);
#endif
    }

#if USE_ITT_BUILD
    if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
      __kmp_itt_barrier_starting(gtid, itt_sync_obj);
#endif /* USE_ITT_BUILD */
#if USE_DEBUGGER
    // Let the debugger know: the thread arrived to the barrier and waiting.
    if (KMP_MASTER_TID(tid)) { // Primary thread counter stored in team struct
      team->t.t_bar[bt].b_master_arrived += 1;
    } else {
      this_thr->th.th_bar[bt].bb.b_worker_arrived += 1;
    } // if
#endif /* USE_DEBUGGER */
    if (reduce != NULL) {
      // KMP_DEBUG_ASSERT( is_split == TRUE );  // #C69956
      this_thr->th.th_local.reduce_data = reduce_data;
    }

    if (KMP_MASTER_TID(tid) && __kmp_tasking_mode != tskm_immediate_exec)
      // use 0 to only setup the current team if nthreads > 1
      __kmp_task_team_setup(this_thr, team, 0);

    if (cancellable) {
      cancelled = __kmp_linear_barrier_gather_cancellable(
          bt, this_thr, gtid, tid, reduce USE_ITT_BUILD_ARG(itt_sync_obj));
    } else {
      switch (__kmp_barrier_gather_pattern[bt]) {
      case bp_dist_bar: {
        __kmp_dist_barrier_gather(bt, this_thr, gtid, tid,
                                  reduce USE_ITT_BUILD_ARG(itt_sync_obj));
        break;
      }
      case bp_hyper_bar: {
        // don't set branch bits to 0; use linear
        KMP_ASSERT(__kmp_barrier_gather_branch_bits[bt]);
        __kmp_hyper_barrier_gather(bt, this_thr, gtid, tid,
                                   reduce USE_ITT_BUILD_ARG(itt_sync_obj));
        break;
      }
      case bp_hierarchical_bar: {
        __kmp_hierarchical_barrier_gather(
            bt, this_thr, gtid, tid, reduce USE_ITT_BUILD_ARG(itt_sync_obj));
        break;
      }
      case bp_tree_bar: {
        // don't set branch bits to 0; use linear
        KMP_ASSERT(__kmp_barrier_gather_branch_bits[bt]);
        __kmp_tree_barrier_gather(bt, this_thr, gtid, tid,
                                  reduce USE_ITT_BUILD_ARG(itt_sync_obj));
        break;
      }
      default: {
        __kmp_linear_barrier_gather(bt, this_thr, gtid, tid,
                                    reduce USE_ITT_BUILD_ARG(itt_sync_obj));
      }
      }
    }

    KMP_MB();

    if (KMP_MASTER_TID(tid)) {
      status = 0;
      if (__kmp_tasking_mode != tskm_immediate_exec && !cancelled) {
        __kmp_task_team_wait(this_thr, team USE_ITT_BUILD_ARG(itt_sync_obj));
      }
#if USE_DEBUGGER
      // Let the debugger know: All threads are arrived and starting leaving the
      // barrier.
      team->t.t_bar[bt].b_team_arrived += 1;
#endif

      if (__kmp_omp_cancellation) {
        kmp_int32 cancel_request = KMP_ATOMIC_LD_RLX(&team->t.t_cancel_request);
        // Reset cancellation flag for worksharing constructs
        if (cancel_request == cancel_loop ||
            cancel_request == cancel_sections) {
          KMP_ATOMIC_ST_RLX(&team->t.t_cancel_request, cancel_noreq);
        }
      }
#if USE_ITT_BUILD
      /* TODO: In case of split reduction barrier, primary thread may send
         acquired event early, before the final summation into the shared
         variable is done (final summation can be a long operation for array
         reductions).  */
      if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
        __kmp_itt_barrier_middle(gtid, itt_sync_obj);
#endif /* USE_ITT_BUILD */
#if USE_ITT_BUILD && USE_ITT_NOTIFY
      // Barrier - report frame end (only if active_level == 1)
      if ((__itt_frame_submit_v3_ptr || KMP_ITT_DEBUG) &&
          __kmp_forkjoin_frames_mode &&
          (this_thr->th.th_teams_microtask == NULL || // either not in teams
           this_thr->th.th_teams_size.nteams == 1) && // or inside single team
          team->t.t_active_level == 1) {
        ident_t *loc = __kmp_threads[gtid]->th.th_ident;
        kmp_uint64 cur_time = __itt_get_timestamp();
        kmp_info_t **other_threads = team->t.t_threads;
        int nproc = this_thr->th.th_team_nproc;
        int i;
        switch (__kmp_forkjoin_frames_mode) {
        case 1:
          __kmp_itt_frame_submit(gtid, this_thr->th.th_frame_time, cur_time, 0,
                                 loc, nproc);
          this_thr->th.th_frame_time = cur_time;
          break;
        case 2: // AC 2015-01-19: currently does not work for hierarchical (to
          // be fixed)
          __kmp_itt_frame_submit(gtid, this_thr->th.th_bar_min_time, cur_time,
                                 1, loc, nproc);
          break;
        case 3:
          if (__itt_metadata_add_ptr) {
            // Initialize with primary thread's wait time
            kmp_uint64 delta = cur_time - this_thr->th.th_bar_arrive_time;
            // Set arrive time to zero to be able to check it in
            // __kmp_invoke_task(); the same is done inside the loop below
            this_thr->th.th_bar_arrive_time = 0;
            for (i = 1; i < nproc; ++i) {
              delta += (cur_time - other_threads[i]->th.th_bar_arrive_time);
              other_threads[i]->th.th_bar_arrive_time = 0;
            }
            __kmp_itt_metadata_imbalance(gtid, this_thr->th.th_frame_time,
                                         cur_time, delta,
                                         (kmp_uint64)(reduce != NULL));
          }
          __kmp_itt_frame_submit(gtid, this_thr->th.th_frame_time, cur_time, 0,
                                 loc, nproc);
          this_thr->th.th_frame_time = cur_time;
          break;
        }
      }
#endif /* USE_ITT_BUILD */
    } else {
      status = 1;
#if USE_ITT_BUILD
      if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
        __kmp_itt_barrier_middle(gtid, itt_sync_obj);
#endif /* USE_ITT_BUILD */
    }
    if ((status == 1 || !is_split) && !cancelled) {
      if (cancellable) {
        cancelled = __kmp_linear_barrier_release_cancellable(
            bt, this_thr, gtid, tid, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
      } else {
        switch (__kmp_barrier_release_pattern[bt]) {
        case bp_dist_bar: {
          KMP_ASSERT(__kmp_barrier_release_branch_bits[bt]);
          __kmp_dist_barrier_release(bt, this_thr, gtid, tid,
                                     FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
          break;
        }
        case bp_hyper_bar: {
          KMP_ASSERT(__kmp_barrier_release_branch_bits[bt]);
          __kmp_hyper_barrier_release(bt, this_thr, gtid, tid,
                                      FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
          break;
        }
        case bp_hierarchical_bar: {
          __kmp_hierarchical_barrier_release(
              bt, this_thr, gtid, tid, FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
          break;
        }
        case bp_tree_bar: {
          KMP_ASSERT(__kmp_barrier_release_branch_bits[bt]);
          __kmp_tree_barrier_release(bt, this_thr, gtid, tid,
                                     FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
          break;
        }
        default: {
          __kmp_linear_barrier_release(bt, this_thr, gtid, tid,
                                       FALSE USE_ITT_BUILD_ARG(itt_sync_obj));
        }
        }
      }
      if (__kmp_tasking_mode != tskm_immediate_exec && !cancelled) {
        __kmp_task_team_sync(this_thr, team);
      }
    }

#if USE_ITT_BUILD
    /* GEH: TODO: Move this under if-condition above and also include in
       __kmp_end_split_barrier(). This will more accurately represent the actual
       release time of the threads for split barriers.  */
    if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
      __kmp_itt_barrier_finished(gtid, itt_sync_obj);
#endif /* USE_ITT_BUILD */
  } else { // Team is serialized.
    status = 0;
    if (__kmp_tasking_mode != tskm_immediate_exec) {
      if (this_thr->th.th_task_team != NULL) {
#if USE_ITT_NOTIFY
        void *itt_sync_obj = NULL;
        if (__itt_sync_create_ptr || KMP_ITT_DEBUG) {
          itt_sync_obj = __kmp_itt_barrier_object(gtid, bt, 1);
          __kmp_itt_barrier_starting(gtid, itt_sync_obj);
        }
#endif

        KMP_DEBUG_ASSERT(
            this_thr->th.th_task_team->tt.tt_found_proxy_tasks == TRUE ||
            this_thr->th.th_task_team->tt.tt_hidden_helper_task_encountered ==
                TRUE);
        __kmp_task_team_wait(this_thr, team USE_ITT_BUILD_ARG(itt_sync_obj));
        __kmp_task_team_setup(this_thr, team, 0);

#if USE_ITT_BUILD
        if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
          __kmp_itt_barrier_finished(gtid, itt_sync_obj);
#endif /* USE_ITT_BUILD */
      }
    }
  }
  KA_TRACE(15, ("__kmp_barrier: T#%d(%d:%d) is leaving with return value %d\n",
                gtid, __kmp_team_from_gtid(gtid)->t.t_id,
                __kmp_tid_from_gtid(gtid), status));

#if OMPT_SUPPORT
  if (ompt_enabled.enabled) {
#if OMPT_OPTIONAL
    if (ompt_enabled.ompt_callback_sync_region_wait) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
          barrier_kind, ompt_scope_end, my_parallel_data, my_task_data,
          return_address);
    }
    if (ompt_enabled.ompt_callback_sync_region) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
          barrier_kind, ompt_scope_end, my_parallel_data, my_task_data,
          return_address);
    }
#endif
    this_thr->th.ompt_thread_info.state = ompt_state_work_parallel;
  }
#endif

  if (cancellable)
    return (int)cancelled;
  return status;
}

// Returns 0 if primary thread, 1 if worker thread.
int __kmp_barrier(enum barrier_type bt, int gtid, int is_split,
                  size_t reduce_size, void *reduce_data,
                  void (*reduce)(void *, void *)) {
  return __kmp_barrier_template<>(bt, gtid, is_split, reduce_size, reduce_data,
                                  reduce);
}

#if defined(KMP_GOMP_COMPAT)
// Returns 1 if cancelled, 0 otherwise
int __kmp_barrier_gomp_cancel(int gtid) {
  if (__kmp_omp_cancellation) {
    int cancelled = __kmp_barrier_template<true>(bs_plain_barrier, gtid, FALSE,
                                                 0, NULL, NULL);
    if (cancelled) {
      int tid = __kmp_tid_from_gtid(gtid);
      kmp_info_t *this_thr = __kmp_threads[gtid];
      if (KMP_MASTER_TID(tid)) {
        // Primary thread does not need to revert anything
      } else {
        // Workers need to revert their private b_arrived flag
        this_thr->th.th_bar[bs_plain_barrier].bb.b_arrived -=
            KMP_BARRIER_STATE_BUMP;
      }
    }
    return cancelled;
  }
  __kmp_barrier(bs_plain_barrier, gtid, FALSE, 0, NULL, NULL);
  return FALSE;
}
#endif

void __kmp_end_split_barrier(enum barrier_type bt, int gtid) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_end_split_barrier);
  KMP_SET_THREAD_STATE_BLOCK(PLAIN_BARRIER);
  KMP_DEBUG_ASSERT(bt < bs_last_barrier);
  int tid = __kmp_tid_from_gtid(gtid);
  kmp_info_t *this_thr = __kmp_threads[gtid];
  kmp_team_t *team = this_thr->th.th_team;

  if (!team->t.t_serialized) {
    if (KMP_MASTER_GTID(gtid)) {
      switch (__kmp_barrier_release_pattern[bt]) {
      case bp_dist_bar: {
        __kmp_dist_barrier_release(bt, this_thr, gtid, tid,
                                   FALSE USE_ITT_BUILD_ARG(NULL));
        break;
      }
      case bp_hyper_bar: {
        KMP_ASSERT(__kmp_barrier_release_branch_bits[bt]);
        __kmp_hyper_barrier_release(bt, this_thr, gtid, tid,
                                    FALSE USE_ITT_BUILD_ARG(NULL));
        break;
      }
      case bp_hierarchical_bar: {
        __kmp_hierarchical_barrier_release(bt, this_thr, gtid, tid,
                                           FALSE USE_ITT_BUILD_ARG(NULL));
        break;
      }
      case bp_tree_bar: {
        KMP_ASSERT(__kmp_barrier_release_branch_bits[bt]);
        __kmp_tree_barrier_release(bt, this_thr, gtid, tid,
                                   FALSE USE_ITT_BUILD_ARG(NULL));
        break;
      }
      default: {
        __kmp_linear_barrier_release(bt, this_thr, gtid, tid,
                                     FALSE USE_ITT_BUILD_ARG(NULL));
      }
      }
      if (__kmp_tasking_mode != tskm_immediate_exec) {
        __kmp_task_team_sync(this_thr, team);
      } // if
    }
  }
}

void __kmp_join_barrier(int gtid) {
  KMP_TIME_PARTITIONED_BLOCK(OMP_join_barrier);
  KMP_SET_THREAD_STATE_BLOCK(FORK_JOIN_BARRIER);

  KMP_DEBUG_ASSERT(__kmp_threads && __kmp_threads[gtid]);

  kmp_info_t *this_thr = __kmp_threads[gtid];
  kmp_team_t *team;
  kmp_uint nproc;
  int tid;
#ifdef KMP_DEBUG
  int team_id;
#endif /* KMP_DEBUG */
#if USE_ITT_BUILD
  void *itt_sync_obj = NULL;
#if USE_ITT_NOTIFY
  if (__itt_sync_create_ptr || KMP_ITT_DEBUG) // Don't call routine without need
    // Get object created at fork_barrier
    itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
#endif
#endif /* USE_ITT_BUILD */
  KMP_MB();

  // Get current info
  team = this_thr->th.th_team;
  nproc = this_thr->th.th_team_nproc;
  KMP_DEBUG_ASSERT((int)nproc == team->t.t_nproc);
  tid = __kmp_tid_from_gtid(gtid);
#ifdef KMP_DEBUG
  team_id = team->t.t_id;
  kmp_info_t *master_thread = this_thr->th.th_team_master;
  if (master_thread != team->t.t_threads[0]) {
    __kmp_print_structure();
  }
#endif /* KMP_DEBUG */
  KMP_DEBUG_ASSERT(master_thread == team->t.t_threads[0]);
  KMP_MB();

  // Verify state
  KMP_DEBUG_ASSERT(TCR_PTR(this_thr->th.th_team));
  KMP_DEBUG_ASSERT(TCR_PTR(this_thr->th.th_root));
  KMP_DEBUG_ASSERT(this_thr == team->t.t_threads[tid]);
  KA_TRACE(10, ("__kmp_join_barrier: T#%d(%d:%d) arrived at join barrier\n",
                gtid, team_id, tid));

#if OMPT_SUPPORT
  if (ompt_enabled.enabled) {
#if OMPT_OPTIONAL
    ompt_data_t *my_task_data;
    ompt_data_t *my_parallel_data;
    void *codeptr = NULL;
    int ds_tid = this_thr->th.th_info.ds.ds_tid;
    if (KMP_MASTER_TID(ds_tid) &&
        (ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait) ||
         ompt_callbacks.ompt_callback(ompt_callback_sync_region)))
      codeptr = team->t.ompt_team_info.master_return_address;
    my_task_data = OMPT_CUR_TASK_DATA(this_thr);
    my_parallel_data = OMPT_CUR_TEAM_DATA(this_thr);
    if (ompt_enabled.ompt_callback_sync_region) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
          ompt_sync_region_barrier_implicit, ompt_scope_begin, my_parallel_data,
          my_task_data, codeptr);
    }
    if (ompt_enabled.ompt_callback_sync_region_wait) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
          ompt_sync_region_barrier_implicit, ompt_scope_begin, my_parallel_data,
          my_task_data, codeptr);
    }
    if (!KMP_MASTER_TID(ds_tid))
      this_thr->th.ompt_thread_info.task_data = *OMPT_CUR_TASK_DATA(this_thr);
#endif
    this_thr->th.ompt_thread_info.state = ompt_state_wait_barrier_implicit;
  }
#endif

  if (__kmp_tasking_mode == tskm_extra_barrier) {
    __kmp_tasking_barrier(team, this_thr, gtid);
    KA_TRACE(10, ("__kmp_join_barrier: T#%d(%d:%d) past tasking barrier\n",
                  gtid, team_id, tid));
  }
#ifdef KMP_DEBUG
  if (__kmp_tasking_mode != tskm_immediate_exec) {
    KA_TRACE(20, ("__kmp_join_barrier: T#%d, old team = %d, old task_team = "
                  "%p, th_task_team = %p\n",
                  __kmp_gtid_from_thread(this_thr), team_id,
                  team->t.t_task_team[this_thr->th.th_task_state],
                  this_thr->th.th_task_team));
    if (this_thr->th.th_task_team)
      KMP_DEBUG_ASSERT(this_thr->th.th_task_team ==
                       team->t.t_task_team[this_thr->th.th_task_state]);
  }
#endif /* KMP_DEBUG */

  /* Copy the blocktime info to the thread, where __kmp_wait_template() can
     access it when the team struct is not guaranteed to exist. Doing these
     loads causes a cache miss slows down EPCC parallel by 2x. As a workaround,
     we do not perform the copy if blocktime=infinite, since the values are not
     used by __kmp_wait_template() in that case. */
  if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME) {
#if KMP_USE_MONITOR
    this_thr->th.th_team_bt_intervals =
        team->t.t_implicit_task_taskdata[tid].td_icvs.bt_intervals;
    this_thr->th.th_team_bt_set =
        team->t.t_implicit_task_taskdata[tid].td_icvs.bt_set;
#else
    this_thr->th.th_team_bt_intervals = KMP_BLOCKTIME_INTERVAL(team, tid);
#endif
  }

#if USE_ITT_BUILD
  if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
    __kmp_itt_barrier_starting(gtid, itt_sync_obj);
#endif /* USE_ITT_BUILD */

  switch (__kmp_barrier_gather_pattern[bs_forkjoin_barrier]) {
  case bp_dist_bar: {
    __kmp_dist_barrier_gather(bs_forkjoin_barrier, this_thr, gtid, tid,
                              NULL USE_ITT_BUILD_ARG(itt_sync_obj));
    break;
  }
  case bp_hyper_bar: {
    KMP_ASSERT(__kmp_barrier_gather_branch_bits[bs_forkjoin_barrier]);
    __kmp_hyper_barrier_gather(bs_forkjoin_barrier, this_thr, gtid, tid,
                               NULL USE_ITT_BUILD_ARG(itt_sync_obj));
    break;
  }
  case bp_hierarchical_bar: {
    __kmp_hierarchical_barrier_gather(bs_forkjoin_barrier, this_thr, gtid, tid,
                                      NULL USE_ITT_BUILD_ARG(itt_sync_obj));
    break;
  }
  case bp_tree_bar: {
    KMP_ASSERT(__kmp_barrier_gather_branch_bits[bs_forkjoin_barrier]);
    __kmp_tree_barrier_gather(bs_forkjoin_barrier, this_thr, gtid, tid,
                              NULL USE_ITT_BUILD_ARG(itt_sync_obj));
    break;
  }
  default: {
    __kmp_linear_barrier_gather(bs_forkjoin_barrier, this_thr, gtid, tid,
                                NULL USE_ITT_BUILD_ARG(itt_sync_obj));
  }
  }

  /* From this point on, the team data structure may be deallocated at any time
     by the primary thread - it is unsafe to reference it in any of the worker
     threads. Any per-team data items that need to be referenced before the
     end of the barrier should be moved to the kmp_task_team_t structs.  */
  if (KMP_MASTER_TID(tid)) {
    if (__kmp_tasking_mode != tskm_immediate_exec) {
      __kmp_task_team_wait(this_thr, team USE_ITT_BUILD_ARG(itt_sync_obj));
    }
    if (__kmp_display_affinity) {
      KMP_CHECK_UPDATE(team->t.t_display_affinity, 0);
    }
#if KMP_STATS_ENABLED
    // Have primary thread flag the workers to indicate they are now waiting for
    // next parallel region, Also wake them up so they switch their timers to
    // idle.
    for (int i = 0; i < team->t.t_nproc; ++i) {
      kmp_info_t *team_thread = team->t.t_threads[i];
      if (team_thread == this_thr)
        continue;
      team_thread->th.th_stats->setIdleFlag();
      if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME &&
          team_thread->th.th_sleep_loc != NULL)
        __kmp_null_resume_wrapper(team_thread);
    }
#endif
#if USE_ITT_BUILD
    if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
      __kmp_itt_barrier_middle(gtid, itt_sync_obj);
#endif /* USE_ITT_BUILD */

#if USE_ITT_BUILD && USE_ITT_NOTIFY
    // Join barrier - report frame end
    if ((__itt_frame_submit_v3_ptr || KMP_ITT_DEBUG) &&
        __kmp_forkjoin_frames_mode &&
        (this_thr->th.th_teams_microtask == NULL || // either not in teams
         this_thr->th.th_teams_size.nteams == 1) && // or inside single team
        team->t.t_active_level == 1) {
      kmp_uint64 cur_time = __itt_get_timestamp();
      ident_t *loc = team->t.t_ident;
      kmp_info_t **other_threads = team->t.t_threads;
      switch (__kmp_forkjoin_frames_mode) {
      case 1:
        __kmp_itt_frame_submit(gtid, this_thr->th.th_frame_time, cur_time, 0,
                               loc, nproc);
        break;
      case 2:
        __kmp_itt_frame_submit(gtid, this_thr->th.th_bar_min_time, cur_time, 1,
                               loc, nproc);
        break;
      case 3:
        if (__itt_metadata_add_ptr) {
          // Initialize with primary thread's wait time
          kmp_uint64 delta = cur_time - this_thr->th.th_bar_arrive_time;
          // Set arrive time to zero to be able to check it in
          // __kmp_invoke_task(); the same is done inside the loop below
          this_thr->th.th_bar_arrive_time = 0;
          for (kmp_uint i = 1; i < nproc; ++i) {
            delta += (cur_time - other_threads[i]->th.th_bar_arrive_time);
            other_threads[i]->th.th_bar_arrive_time = 0;
          }
          __kmp_itt_metadata_imbalance(gtid, this_thr->th.th_frame_time,
                                       cur_time, delta, 0);
        }
        __kmp_itt_frame_submit(gtid, this_thr->th.th_frame_time, cur_time, 0,
                               loc, nproc);
        this_thr->th.th_frame_time = cur_time;
        break;
      }
    }
#endif /* USE_ITT_BUILD */
  }
#if USE_ITT_BUILD
  else {
    if (__itt_sync_create_ptr || KMP_ITT_DEBUG)
      __kmp_itt_barrier_middle(gtid, itt_sync_obj);
  }
#endif /* USE_ITT_BUILD */

#if KMP_DEBUG
  if (KMP_MASTER_TID(tid)) {
    KA_TRACE(
        15,
        ("__kmp_join_barrier: T#%d(%d:%d) says all %d team threads arrived\n",
         gtid, team_id, tid, nproc));
  }
#endif /* KMP_DEBUG */

  // TODO now, mark worker threads as done so they may be disbanded
  KMP_MB(); // Flush all pending memory write invalidates.
  KA_TRACE(10,
           ("__kmp_join_barrier: T#%d(%d:%d) leaving\n", gtid, team_id, tid));

}

// TODO release worker threads' fork barriers as we are ready instead of all at
// once
void __kmp_fork_barrier(int gtid, int tid) {
  KMP_TIME_PARTITIONED_BLOCK(OMP_fork_barrier);
  KMP_SET_THREAD_STATE_BLOCK(FORK_JOIN_BARRIER);
  kmp_info_t *this_thr = __kmp_threads[gtid];
  kmp_team_t *team = (tid == 0) ? this_thr->th.th_team : NULL;
#if USE_ITT_BUILD
  void *itt_sync_obj = NULL;
#endif /* USE_ITT_BUILD */
  if (team)

  KA_TRACE(10, ("__kmp_fork_barrier: T#%d(%d:%d) has arrived\n", gtid,
                (team != NULL) ? team->t.t_id : -1, tid));

  // th_team pointer only valid for primary thread here
  if (KMP_MASTER_TID(tid)) {
#if USE_ITT_BUILD && USE_ITT_NOTIFY
    if (__itt_sync_create_ptr || KMP_ITT_DEBUG) {
      // Create itt barrier object
      itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier, 1);
      __kmp_itt_barrier_middle(gtid, itt_sync_obj); // Call acquired/releasing
    }
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY */

#ifdef KMP_DEBUG
    KMP_DEBUG_ASSERT(team);
    kmp_info_t **other_threads = team->t.t_threads;
    int i;

    // Verify state
    KMP_MB();

    for (i = 1; i < team->t.t_nproc; ++i) {
      KA_TRACE(500,
               ("__kmp_fork_barrier: T#%d(%d:0) checking T#%d(%d:%d) fork go "
                "== %u.\n",
                gtid, team->t.t_id, other_threads[i]->th.th_info.ds.ds_gtid,
                team->t.t_id, other_threads[i]->th.th_info.ds.ds_tid,
                other_threads[i]->th.th_bar[bs_forkjoin_barrier].bb.b_go));
      KMP_DEBUG_ASSERT(
          (TCR_4(other_threads[i]->th.th_bar[bs_forkjoin_barrier].bb.b_go) &
           ~(KMP_BARRIER_SLEEP_STATE)) == KMP_INIT_BARRIER_STATE);
      KMP_DEBUG_ASSERT(other_threads[i]->th.th_team == team);
    }
#endif

    if (__kmp_tasking_mode != tskm_immediate_exec) {
      // 0 indicates setup current task team if nthreads > 1
      __kmp_task_team_setup(this_thr, team, 0);
    }

    /* The primary thread may have changed its blocktime between join barrier
       and fork barrier. Copy the blocktime info to the thread, where
       __kmp_wait_template() can access it when the team struct is not
       guaranteed to exist. */
    // See note about the corresponding code in __kmp_join_barrier() being
    // performance-critical
    if (__kmp_dflt_blocktime != KMP_MAX_BLOCKTIME) {
#if KMP_USE_MONITOR
      this_thr->th.th_team_bt_intervals =
          team->t.t_implicit_task_taskdata[tid].td_icvs.bt_intervals;
      this_thr->th.th_team_bt_set =
          team->t.t_implicit_task_taskdata[tid].td_icvs.bt_set;
#else
      this_thr->th.th_team_bt_intervals = KMP_BLOCKTIME_INTERVAL(team, tid);
#endif
    }
  } // primary thread

  switch (__kmp_barrier_release_pattern[bs_forkjoin_barrier]) {
  case bp_dist_bar: {
    __kmp_dist_barrier_release(bs_forkjoin_barrier, this_thr, gtid, tid,
                               TRUE USE_ITT_BUILD_ARG(NULL));
    break;
  }
  case bp_hyper_bar: {
    KMP_ASSERT(__kmp_barrier_release_branch_bits[bs_forkjoin_barrier]);
    __kmp_hyper_barrier_release(bs_forkjoin_barrier, this_thr, gtid, tid,
                                TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
    break;
  }
  case bp_hierarchical_bar: {
    __kmp_hierarchical_barrier_release(bs_forkjoin_barrier, this_thr, gtid, tid,
                                       TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
    break;
  }
  case bp_tree_bar: {
    KMP_ASSERT(__kmp_barrier_release_branch_bits[bs_forkjoin_barrier]);
    __kmp_tree_barrier_release(bs_forkjoin_barrier, this_thr, gtid, tid,
                               TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
    break;
  }
  default: {
    __kmp_linear_barrier_release(bs_forkjoin_barrier, this_thr, gtid, tid,
                                 TRUE USE_ITT_BUILD_ARG(itt_sync_obj));
  }
  }

#if OMPT_SUPPORT
  if (ompt_enabled.enabled &&
      this_thr->th.ompt_thread_info.state == ompt_state_wait_barrier_implicit) {
    int ds_tid = this_thr->th.th_info.ds.ds_tid;
    ompt_data_t *task_data = (team)
                                 ? OMPT_CUR_TASK_DATA(this_thr)
                                 : &(this_thr->th.ompt_thread_info.task_data);
    this_thr->th.ompt_thread_info.state = ompt_state_overhead;
#if OMPT_OPTIONAL
    void *codeptr = NULL;
    if (KMP_MASTER_TID(ds_tid) &&
        (ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait) ||
         ompt_callbacks.ompt_callback(ompt_callback_sync_region)))
      codeptr = team ? team->t.ompt_team_info.master_return_address : NULL;
    if (ompt_enabled.ompt_callback_sync_region_wait) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region_wait)(
          ompt_sync_region_barrier_implicit, ompt_scope_end, NULL, task_data,
          codeptr);
    }
    if (ompt_enabled.ompt_callback_sync_region) {
      ompt_callbacks.ompt_callback(ompt_callback_sync_region)(
          ompt_sync_region_barrier_implicit, ompt_scope_end, NULL, task_data,
          codeptr);
    }
#endif
    if (!KMP_MASTER_TID(ds_tid) && ompt_enabled.ompt_callback_implicit_task) {
      ompt_callbacks.ompt_callback(ompt_callback_implicit_task)(
          ompt_scope_end, NULL, task_data, 0, ds_tid,
          ompt_task_implicit); // TODO: Can this be ompt_task_initial?
    }
  }
#endif

  // Early exit for reaping threads releasing forkjoin barrier
  if (TCR_4(__kmp_global.g.g_done)) {
    this_thr->th.th_task_team = NULL;

#if USE_ITT_BUILD && USE_ITT_NOTIFY
    if (__itt_sync_create_ptr || KMP_ITT_DEBUG) {
      if (!KMP_MASTER_TID(tid)) {
        itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
        if (itt_sync_obj)
          __kmp_itt_barrier_finished(gtid, itt_sync_obj);
      }
    }
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY */
    KA_TRACE(10, ("__kmp_fork_barrier: T#%d is leaving early\n", gtid));
    return;
  }

  /* We can now assume that a valid team structure has been allocated by the
     primary thread and propagated to all worker threads. The current thread,
     however, may not be part of the team, so we can't blindly assume that the
     team pointer is non-null.  */
  team = (kmp_team_t *)TCR_PTR(this_thr->th.th_team);
  KMP_DEBUG_ASSERT(team != NULL);
  tid = __kmp_tid_from_gtid(gtid);

#if KMP_BARRIER_ICV_PULL
  /* Primary thread's copy of the ICVs was set up on the implicit taskdata in
     __kmp_reinitialize_team. __kmp_fork_call() assumes the primary thread's
     implicit task has this data before this function is called. We cannot
     modify __kmp_fork_call() to look at the fixed ICVs in the primary thread's
     thread struct, because it is not always the case that the threads arrays
     have been allocated when __kmp_fork_call() is executed. */
  {
    KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(USER_icv_copy);
    if (!KMP_MASTER_TID(tid)) { // primary thread already has ICVs
      // Copy the initial ICVs from the primary thread's thread struct to the
      // implicit task for this tid.
      KA_TRACE(10,
               ("__kmp_fork_barrier: T#%d(%d) is PULLing ICVs\n", gtid, tid));
      __kmp_init_implicit_task(team->t.t_ident, team->t.t_threads[tid], team,
                               tid, FALSE);
      copy_icvs(&team->t.t_implicit_task_taskdata[tid].td_icvs,
                &team->t.t_threads[0]
                     ->th.th_bar[bs_forkjoin_barrier]
                     .bb.th_fixed_icvs);
    }
  }
#endif // KMP_BARRIER_ICV_PULL

  if (__kmp_tasking_mode != tskm_immediate_exec) {
    __kmp_task_team_sync(this_thr, team);
  }

#if KMP_AFFINITY_SUPPORTED
  kmp_proc_bind_t proc_bind = team->t.t_proc_bind;
  if (proc_bind == proc_bind_intel) {
    // Call dynamic affinity settings
    if (__kmp_affinity_type == affinity_balanced && team->t.t_size_changed) {
      __kmp_balanced_affinity(this_thr, team->t.t_nproc);
    }
  } else if (proc_bind != proc_bind_false) {
    if (this_thr->th.th_new_place == this_thr->th.th_current_place) {
      KA_TRACE(100, ("__kmp_fork_barrier: T#%d already in correct place %d\n",
                     __kmp_gtid_from_thread(this_thr),
                     this_thr->th.th_current_place));
    } else {
      __kmp_affinity_set_place(gtid);
    }
  }
#endif // KMP_AFFINITY_SUPPORTED
  // Perform the display affinity functionality
  if (__kmp_display_affinity) {
    if (team->t.t_display_affinity
#if KMP_AFFINITY_SUPPORTED
        || (__kmp_affinity_type == affinity_balanced && team->t.t_size_changed)
#endif
    ) {
      // NULL means use the affinity-format-var ICV
      __kmp_aux_display_affinity(gtid, NULL);
      this_thr->th.th_prev_num_threads = team->t.t_nproc;
      this_thr->th.th_prev_level = team->t.t_level;
    }
  }
  if (!KMP_MASTER_TID(tid))
    KMP_CHECK_UPDATE(this_thr->th.th_def_allocator, team->t.t_def_allocator);

#if USE_ITT_BUILD && USE_ITT_NOTIFY
  if (__itt_sync_create_ptr || KMP_ITT_DEBUG) {
    if (!KMP_MASTER_TID(tid)) {
      // Get correct barrier object
      itt_sync_obj = __kmp_itt_barrier_object(gtid, bs_forkjoin_barrier);
      __kmp_itt_barrier_finished(gtid, itt_sync_obj); // Workers call acquired
    } // (prepare called inside barrier_release)
  }
#endif /* USE_ITT_BUILD && USE_ITT_NOTIFY */
  KA_TRACE(10, ("__kmp_fork_barrier: T#%d(%d:%d) is leaving\n", gtid,
                team->t.t_id, tid));
}

void __kmp_setup_icv_copy(kmp_team_t *team, int new_nproc,
                          kmp_internal_control_t *new_icvs, ident_t *loc) {
  KMP_TIME_DEVELOPER_PARTITIONED_BLOCK(KMP_setup_icv_copy);

  KMP_DEBUG_ASSERT(team && new_nproc && new_icvs);
  KMP_DEBUG_ASSERT((!TCR_4(__kmp_init_parallel)) || new_icvs->nproc);

/* Primary thread's copy of the ICVs was set up on the implicit taskdata in
   __kmp_reinitialize_team. __kmp_fork_call() assumes the primary thread's
   implicit task has this data before this function is called. */
#if KMP_BARRIER_ICV_PULL
  /* Copy ICVs to primary thread's thread structure into th_fixed_icvs (which
     remains untouched), where all of the worker threads can access them and
     make their own copies after the barrier. */
  KMP_DEBUG_ASSERT(team->t.t_threads[0]); // The threads arrays should be
  // allocated at this point
  copy_icvs(
      &team->t.t_threads[0]->th.th_bar[bs_forkjoin_barrier].bb.th_fixed_icvs,
      new_icvs);
  KF_TRACE(10, ("__kmp_setup_icv_copy: PULL: T#%d this_thread=%p team=%p\n", 0,
                team->t.t_threads[0], team));
#elif KMP_BARRIER_ICV_PUSH
  // The ICVs will be propagated in the fork barrier, so nothing needs to be
  // done here.
  KF_TRACE(10, ("__kmp_setup_icv_copy: PUSH: T#%d this_thread=%p team=%p\n", 0,
                team->t.t_threads[0], team));
#else
  // Copy the ICVs to each of the non-primary threads.  This takes O(nthreads)
  // time.
  ngo_load(new_icvs);
  KMP_DEBUG_ASSERT(team->t.t_threads[0]); // The threads arrays should be
  // allocated at this point
  for (int f = 1; f < new_nproc; ++f) { // Skip the primary thread
    // TODO: GEH - pass in better source location info since usually NULL here
    KF_TRACE(10, ("__kmp_setup_icv_copy: LINEAR: T#%d this_thread=%p team=%p\n",
                  f, team->t.t_threads[f], team));
    __kmp_init_implicit_task(loc, team->t.t_threads[f], team, f, FALSE);
    ngo_store_icvs(&team->t.t_implicit_task_taskdata[f].td_icvs, new_icvs);
    KF_TRACE(10, ("__kmp_setup_icv_copy: LINEAR: T#%d this_thread=%p team=%p\n",
                  f, team->t.t_threads[f], team));
  }
  ngo_sync();
#endif // KMP_BARRIER_ICV_PULL
}