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
|
/* Copyright (C) 2007-2019 Free Software Foundation, Inc.
Contributed by Richard Henderson <rth@redhat.com>.
This file is part of the GNU Offloading and Multi Processing Library
(libgomp).
Libgomp is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 3, or (at your option)
any later version.
Libgomp is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
FOR A PARTICULAR PURPOSE. See the GNU General Public License for
more details.
Under Section 7 of GPL version 3, you are granted additional
permissions described in the GCC Runtime Library Exception, version
3.1, as published by the Free Software Foundation.
You should have received a copy of the GNU General Public License and
a copy of the GCC Runtime Library Exception along with this program;
see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
<http://www.gnu.org/licenses/>. */
/* This file handles the maintainence of tasks in response to task
creation and termination. */
#include "libgomp.h"
#include <stdlib.h>
#include <string.h>
#include "gomp-constants.h"
typedef struct gomp_task_depend_entry *hash_entry_type;
static inline void *
htab_alloc (size_t size)
{
return gomp_malloc (size);
}
static inline void
htab_free (void *ptr)
{
free (ptr);
}
#include "hashtab.h"
static inline hashval_t
htab_hash (hash_entry_type element)
{
return hash_pointer (element->addr);
}
static inline bool
htab_eq (hash_entry_type x, hash_entry_type y)
{
return x->addr == y->addr;
}
/* Create a new task data structure. */
void
gomp_init_task (struct gomp_task *task, struct gomp_task *parent_task,
struct gomp_task_icv *prev_icv)
{
/* It would seem that using memset here would be a win, but it turns
out that partially filling gomp_task allows us to keep the
overhead of task creation low. In the nqueens-1.c test, for a
sufficiently large N, we drop the overhead from 5-6% to 1%.
Note, the nqueens-1.c test in serial mode is a good test to
benchmark the overhead of creating tasks as there are millions of
tiny tasks created that all run undeferred. */
task->parent = parent_task;
task->icv = *prev_icv;
task->kind = GOMP_TASK_IMPLICIT;
task->taskwait = NULL;
task->in_tied_task = false;
task->final_task = false;
task->copy_ctors_done = false;
task->parent_depends_on = false;
priority_queue_init (&task->children_queue);
task->taskgroup = NULL;
task->dependers = NULL;
task->depend_hash = NULL;
task->depend_count = 0;
}
/* Clean up a task, after completing it. */
void
gomp_end_task (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_task *task = thr->task;
gomp_finish_task (task);
thr->task = task->parent;
}
/* Clear the parent field of every task in LIST. */
static inline void
gomp_clear_parent_in_list (struct priority_list *list)
{
struct priority_node *p = list->tasks;
if (p)
do
{
priority_node_to_task (PQ_CHILDREN, p)->parent = NULL;
p = p->next;
}
while (p != list->tasks);
}
/* Splay tree version of gomp_clear_parent_in_list.
Clear the parent field of every task in NODE within SP, and free
the node when done. */
static void
gomp_clear_parent_in_tree (prio_splay_tree sp, prio_splay_tree_node node)
{
if (!node)
return;
prio_splay_tree_node left = node->left, right = node->right;
gomp_clear_parent_in_list (&node->key.l);
#if _LIBGOMP_CHECKING_
memset (node, 0xaf, sizeof (*node));
#endif
/* No need to remove the node from the tree. We're nuking
everything, so just free the nodes and our caller can clear the
entire splay tree. */
free (node);
gomp_clear_parent_in_tree (sp, left);
gomp_clear_parent_in_tree (sp, right);
}
/* Clear the parent field of every task in Q and remove every task
from Q. */
static inline void
gomp_clear_parent (struct priority_queue *q)
{
if (priority_queue_multi_p (q))
{
gomp_clear_parent_in_tree (&q->t, q->t.root);
/* All the nodes have been cleared in gomp_clear_parent_in_tree.
No need to remove anything. We can just nuke everything. */
q->t.root = NULL;
}
else
gomp_clear_parent_in_list (&q->l);
}
/* Helper function for GOMP_task and gomp_create_target_task.
For a TASK with in/out dependencies, fill in the various dependency
queues. PARENT is the parent of said task. DEPEND is as in
GOMP_task. */
static void
gomp_task_handle_depend (struct gomp_task *task, struct gomp_task *parent,
void **depend)
{
size_t ndepend = (uintptr_t) depend[0];
size_t i;
hash_entry_type ent;
if (ndepend)
{
/* depend[0] is total # */
size_t nout = (uintptr_t) depend[1]; /* # of out: and inout: */
/* ndepend - nout is # of in: */
for (i = 0; i < ndepend; i++)
{
task->depend[i].addr = depend[2 + i];
task->depend[i].is_in = i >= nout;
}
}
else
{
ndepend = (uintptr_t) depend[1]; /* total # */
size_t nout = (uintptr_t) depend[2]; /* # of out: and inout: */
size_t nmutexinoutset = (uintptr_t) depend[3]; /* # of mutexinoutset: */
/* For now we treat mutexinoutset like out, which is compliant, but
inefficient. */
size_t nin = (uintptr_t) depend[4]; /* # of in: */
/* ndepend - nout - nmutexinoutset - nin is # of depobjs */
size_t normal = nout + nmutexinoutset + nin;
size_t n = 0;
for (i = normal; i < ndepend; i++)
{
void **d = (void **) (uintptr_t) depend[5 + i];
switch ((uintptr_t) d[1])
{
case GOMP_DEPEND_OUT:
case GOMP_DEPEND_INOUT:
case GOMP_DEPEND_MUTEXINOUTSET:
break;
case GOMP_DEPEND_IN:
continue;
default:
gomp_fatal ("unknown omp_depend_t dependence type %d",
(int) (uintptr_t) d[1]);
}
task->depend[n].addr = d[0];
task->depend[n++].is_in = 0;
}
for (i = 0; i < normal; i++)
{
task->depend[n].addr = depend[5 + i];
task->depend[n++].is_in = i >= nout + nmutexinoutset;
}
for (i = normal; i < ndepend; i++)
{
void **d = (void **) (uintptr_t) depend[5 + i];
if ((uintptr_t) d[1] != GOMP_DEPEND_IN)
continue;
task->depend[n].addr = d[0];
task->depend[n++].is_in = 1;
}
}
task->depend_count = ndepend;
task->num_dependees = 0;
if (parent->depend_hash == NULL)
parent->depend_hash = htab_create (2 * ndepend > 12 ? 2 * ndepend : 12);
for (i = 0; i < ndepend; i++)
{
task->depend[i].next = NULL;
task->depend[i].prev = NULL;
task->depend[i].task = task;
task->depend[i].redundant = false;
task->depend[i].redundant_out = false;
hash_entry_type *slot = htab_find_slot (&parent->depend_hash,
&task->depend[i], INSERT);
hash_entry_type out = NULL, last = NULL;
if (*slot)
{
/* If multiple depends on the same task are the same, all but the
first one are redundant. As inout/out come first, if any of them
is inout/out, it will win, which is the right semantics. */
if ((*slot)->task == task)
{
task->depend[i].redundant = true;
continue;
}
for (ent = *slot; ent; ent = ent->next)
{
if (ent->redundant_out)
break;
last = ent;
/* depend(in:...) doesn't depend on earlier depend(in:...). */
if (task->depend[i].is_in && ent->is_in)
continue;
if (!ent->is_in)
out = ent;
struct gomp_task *tsk = ent->task;
if (tsk->dependers == NULL)
{
tsk->dependers
= gomp_malloc (sizeof (struct gomp_dependers_vec)
+ 6 * sizeof (struct gomp_task *));
tsk->dependers->n_elem = 1;
tsk->dependers->allocated = 6;
tsk->dependers->elem[0] = task;
task->num_dependees++;
continue;
}
/* We already have some other dependency on tsk from earlier
depend clause. */
else if (tsk->dependers->n_elem
&& (tsk->dependers->elem[tsk->dependers->n_elem - 1]
== task))
continue;
else if (tsk->dependers->n_elem == tsk->dependers->allocated)
{
tsk->dependers->allocated
= tsk->dependers->allocated * 2 + 2;
tsk->dependers
= gomp_realloc (tsk->dependers,
sizeof (struct gomp_dependers_vec)
+ (tsk->dependers->allocated
* sizeof (struct gomp_task *)));
}
tsk->dependers->elem[tsk->dependers->n_elem++] = task;
task->num_dependees++;
}
task->depend[i].next = *slot;
(*slot)->prev = &task->depend[i];
}
*slot = &task->depend[i];
/* There is no need to store more than one depend({,in}out:) task per
address in the hash table chain for the purpose of creation of
deferred tasks, because each out depends on all earlier outs, thus it
is enough to record just the last depend({,in}out:). For depend(in:),
we need to keep all of the previous ones not terminated yet, because
a later depend({,in}out:) might need to depend on all of them. So, if
the new task's clause is depend({,in}out:), we know there is at most
one other depend({,in}out:) clause in the list (out). For
non-deferred tasks we want to see all outs, so they are moved to the
end of the chain, after first redundant_out entry all following
entries should be redundant_out. */
if (!task->depend[i].is_in && out)
{
if (out != last)
{
out->next->prev = out->prev;
out->prev->next = out->next;
out->next = last->next;
out->prev = last;
last->next = out;
if (out->next)
out->next->prev = out;
}
out->redundant_out = true;
}
}
}
/* Called when encountering an explicit task directive. If IF_CLAUSE is
false, then we must not delay in executing the task. If UNTIED is true,
then the task may be executed by any member of the team.
DEPEND is an array containing:
if depend[0] is non-zero, then:
depend[0]: number of depend elements.
depend[1]: number of depend elements of type "out/inout".
depend[2..N+1]: address of [1..N]th depend element.
otherwise, when depend[0] is zero, then:
depend[1]: number of depend elements.
depend[2]: number of depend elements of type "out/inout".
depend[3]: number of depend elements of type "mutexinoutset".
depend[4]: number of depend elements of type "in".
depend[5..4+depend[2]+depend[3]+depend[4]]: address of depend elements
depend[5+depend[2]+depend[3]+depend[4]..4+depend[1]]: address of
omp_depend_t objects. */
void
GOMP_task (void (*fn) (void *), void *data, void (*cpyfn) (void *, void *),
long arg_size, long arg_align, bool if_clause, unsigned flags,
void **depend, int priority)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
#ifdef HAVE_BROKEN_POSIX_SEMAPHORES
/* If pthread_mutex_* is used for omp_*lock*, then each task must be
tied to one thread all the time. This means UNTIED tasks must be
tied and if CPYFN is non-NULL IF(0) must be forced, as CPYFN
might be running on different thread than FN. */
if (cpyfn)
if_clause = false;
flags &= ~GOMP_TASK_FLAG_UNTIED;
#endif
/* If parallel or taskgroup has been cancelled, don't start new tasks. */
if (__builtin_expect (gomp_cancel_var, 0) && team)
{
if (gomp_team_barrier_cancelled (&team->barrier))
return;
if (thr->task->taskgroup)
{
if (thr->task->taskgroup->cancelled)
return;
if (thr->task->taskgroup->workshare
&& thr->task->taskgroup->prev
&& thr->task->taskgroup->prev->cancelled)
return;
}
}
if ((flags & GOMP_TASK_FLAG_PRIORITY) == 0)
priority = 0;
else if (priority > gomp_max_task_priority_var)
priority = gomp_max_task_priority_var;
if (!if_clause || team == NULL
|| (thr->task && thr->task->final_task)
|| team->task_count > 64 * team->nthreads)
{
struct gomp_task task;
/* If there are depend clauses and earlier deferred sibling tasks
with depend clauses, check if there isn't a dependency. If there
is, we need to wait for them. There is no need to handle
depend clauses for non-deferred tasks other than this, because
the parent task is suspended until the child task finishes and thus
it can't start further child tasks. */
if ((flags & GOMP_TASK_FLAG_DEPEND)
&& thr->task && thr->task->depend_hash)
gomp_task_maybe_wait_for_dependencies (depend);
gomp_init_task (&task, thr->task, gomp_icv (false));
task.kind = GOMP_TASK_UNDEFERRED;
task.final_task = (thr->task && thr->task->final_task)
|| (flags & GOMP_TASK_FLAG_FINAL);
task.priority = priority;
if (thr->task)
{
task.in_tied_task = thr->task->in_tied_task;
task.taskgroup = thr->task->taskgroup;
}
thr->task = &task;
if (__builtin_expect (cpyfn != NULL, 0))
{
char buf[arg_size + arg_align - 1];
char *arg = (char *) (((uintptr_t) buf + arg_align - 1)
& ~(uintptr_t) (arg_align - 1));
cpyfn (arg, data);
fn (arg);
}
else
fn (data);
/* Access to "children" is normally done inside a task_lock
mutex region, but the only way this particular task.children
can be set is if this thread's task work function (fn)
creates children. So since the setter is *this* thread, we
need no barriers here when testing for non-NULL. We can have
task.children set by the current thread then changed by a
child thread, but seeing a stale non-NULL value is not a
problem. Once past the task_lock acquisition, this thread
will see the real value of task.children. */
if (!priority_queue_empty_p (&task.children_queue, MEMMODEL_RELAXED))
{
gomp_mutex_lock (&team->task_lock);
gomp_clear_parent (&task.children_queue);
gomp_mutex_unlock (&team->task_lock);
}
gomp_end_task ();
}
else
{
struct gomp_task *task;
struct gomp_task *parent = thr->task;
struct gomp_taskgroup *taskgroup = parent->taskgroup;
char *arg;
bool do_wake;
size_t depend_size = 0;
if (flags & GOMP_TASK_FLAG_DEPEND)
depend_size = ((uintptr_t) (depend[0] ? depend[0] : depend[1])
* sizeof (struct gomp_task_depend_entry));
task = gomp_malloc (sizeof (*task) + depend_size
+ arg_size + arg_align - 1);
arg = (char *) (((uintptr_t) (task + 1) + depend_size + arg_align - 1)
& ~(uintptr_t) (arg_align - 1));
gomp_init_task (task, parent, gomp_icv (false));
task->priority = priority;
task->kind = GOMP_TASK_UNDEFERRED;
task->in_tied_task = parent->in_tied_task;
task->taskgroup = taskgroup;
thr->task = task;
if (cpyfn)
{
cpyfn (arg, data);
task->copy_ctors_done = true;
}
else
memcpy (arg, data, arg_size);
thr->task = parent;
task->kind = GOMP_TASK_WAITING;
task->fn = fn;
task->fn_data = arg;
task->final_task = (flags & GOMP_TASK_FLAG_FINAL) >> 1;
gomp_mutex_lock (&team->task_lock);
/* If parallel or taskgroup has been cancelled, don't start new
tasks. */
if (__builtin_expect (gomp_cancel_var, 0)
&& !task->copy_ctors_done)
{
if (gomp_team_barrier_cancelled (&team->barrier))
{
do_cancel:
gomp_mutex_unlock (&team->task_lock);
gomp_finish_task (task);
free (task);
return;
}
if (taskgroup)
{
if (taskgroup->cancelled)
goto do_cancel;
if (taskgroup->workshare
&& taskgroup->prev
&& taskgroup->prev->cancelled)
goto do_cancel;
}
}
if (taskgroup)
taskgroup->num_children++;
if (depend_size)
{
gomp_task_handle_depend (task, parent, depend);
if (task->num_dependees)
{
/* Tasks that depend on other tasks are not put into the
various waiting queues, so we are done for now. Said
tasks are instead put into the queues via
gomp_task_run_post_handle_dependers() after their
dependencies have been satisfied. After which, they
can be picked up by the various scheduling
points. */
gomp_mutex_unlock (&team->task_lock);
return;
}
}
priority_queue_insert (PQ_CHILDREN, &parent->children_queue,
task, priority,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
if (taskgroup)
priority_queue_insert (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
task, priority,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
priority_queue_insert (PQ_TEAM, &team->task_queue,
task, priority,
PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
++team->task_count;
++team->task_queued_count;
gomp_team_barrier_set_task_pending (&team->barrier);
do_wake = team->task_running_count + !parent->in_tied_task
< team->nthreads;
gomp_mutex_unlock (&team->task_lock);
if (do_wake)
gomp_team_barrier_wake (&team->barrier, 1);
}
}
ialias (GOMP_taskgroup_start)
ialias (GOMP_taskgroup_end)
ialias (GOMP_taskgroup_reduction_register)
#define TYPE long
#define UTYPE unsigned long
#define TYPE_is_long 1
#include "taskloop.c"
#undef TYPE
#undef UTYPE
#undef TYPE_is_long
#define TYPE unsigned long long
#define UTYPE TYPE
#define GOMP_taskloop GOMP_taskloop_ull
#include "taskloop.c"
#undef TYPE
#undef UTYPE
#undef GOMP_taskloop
static void inline
priority_queue_move_task_first (enum priority_queue_type type,
struct priority_queue *head,
struct gomp_task *task)
{
#if _LIBGOMP_CHECKING_
if (!priority_queue_task_in_queue_p (type, head, task))
gomp_fatal ("Attempt to move first missing task %p", task);
#endif
struct priority_list *list;
if (priority_queue_multi_p (head))
{
list = priority_queue_lookup_priority (head, task->priority);
#if _LIBGOMP_CHECKING_
if (!list)
gomp_fatal ("Unable to find priority %d", task->priority);
#endif
}
else
list = &head->l;
priority_list_remove (list, task_to_priority_node (type, task), 0);
priority_list_insert (type, list, task, task->priority,
PRIORITY_INSERT_BEGIN, type == PQ_CHILDREN,
task->parent_depends_on);
}
/* Actual body of GOMP_PLUGIN_target_task_completion that is executed
with team->task_lock held, or is executed in the thread that called
gomp_target_task_fn if GOMP_PLUGIN_target_task_completion has been
run before it acquires team->task_lock. */
static void
gomp_target_task_completion (struct gomp_team *team, struct gomp_task *task)
{
struct gomp_task *parent = task->parent;
if (parent)
priority_queue_move_task_first (PQ_CHILDREN, &parent->children_queue,
task);
struct gomp_taskgroup *taskgroup = task->taskgroup;
if (taskgroup)
priority_queue_move_task_first (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
task);
priority_queue_insert (PQ_TEAM, &team->task_queue, task, task->priority,
PRIORITY_INSERT_BEGIN, false,
task->parent_depends_on);
task->kind = GOMP_TASK_WAITING;
if (parent && parent->taskwait)
{
if (parent->taskwait->in_taskwait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
parent->taskwait->in_taskwait = false;
gomp_sem_post (&parent->taskwait->taskwait_sem);
}
else if (parent->taskwait->in_depend_wait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
parent->taskwait->in_depend_wait = false;
gomp_sem_post (&parent->taskwait->taskwait_sem);
}
}
if (taskgroup && taskgroup->in_taskgroup_wait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
taskgroup->in_taskgroup_wait = false;
gomp_sem_post (&taskgroup->taskgroup_sem);
}
++team->task_queued_count;
gomp_team_barrier_set_task_pending (&team->barrier);
/* I'm afraid this can't be done after releasing team->task_lock,
as gomp_target_task_completion is run from unrelated thread and
therefore in between gomp_mutex_unlock and gomp_team_barrier_wake
the team could be gone already. */
if (team->nthreads > team->task_running_count)
gomp_team_barrier_wake (&team->barrier, 1);
}
/* Signal that a target task TTASK has completed the asynchronously
running phase and should be requeued as a task to handle the
variable unmapping. */
void
GOMP_PLUGIN_target_task_completion (void *data)
{
struct gomp_target_task *ttask = (struct gomp_target_task *) data;
struct gomp_task *task = ttask->task;
struct gomp_team *team = ttask->team;
gomp_mutex_lock (&team->task_lock);
if (ttask->state == GOMP_TARGET_TASK_READY_TO_RUN)
{
ttask->state = GOMP_TARGET_TASK_FINISHED;
gomp_mutex_unlock (&team->task_lock);
return;
}
ttask->state = GOMP_TARGET_TASK_FINISHED;
gomp_target_task_completion (team, task);
gomp_mutex_unlock (&team->task_lock);
}
static void gomp_task_run_post_handle_depend_hash (struct gomp_task *);
/* Called for nowait target tasks. */
bool
gomp_create_target_task (struct gomp_device_descr *devicep,
void (*fn) (void *), size_t mapnum, void **hostaddrs,
size_t *sizes, unsigned short *kinds,
unsigned int flags, void **depend, void **args,
enum gomp_target_task_state state)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
/* If parallel or taskgroup has been cancelled, don't start new tasks. */
if (__builtin_expect (gomp_cancel_var, 0) && team)
{
if (gomp_team_barrier_cancelled (&team->barrier))
return true;
if (thr->task->taskgroup)
{
if (thr->task->taskgroup->cancelled)
return true;
if (thr->task->taskgroup->workshare
&& thr->task->taskgroup->prev
&& thr->task->taskgroup->prev->cancelled)
return true;
}
}
struct gomp_target_task *ttask;
struct gomp_task *task;
struct gomp_task *parent = thr->task;
struct gomp_taskgroup *taskgroup = parent->taskgroup;
bool do_wake;
size_t depend_size = 0;
uintptr_t depend_cnt = 0;
size_t tgt_align = 0, tgt_size = 0;
if (depend != NULL)
{
depend_cnt = (uintptr_t) (depend[0] ? depend[0] : depend[1]);
depend_size = depend_cnt * sizeof (struct gomp_task_depend_entry);
}
if (fn)
{
/* GOMP_MAP_FIRSTPRIVATE need to be copied first, as they are
firstprivate on the target task. */
size_t i;
for (i = 0; i < mapnum; i++)
if ((kinds[i] & 0xff) == GOMP_MAP_FIRSTPRIVATE)
{
size_t align = (size_t) 1 << (kinds[i] >> 8);
if (tgt_align < align)
tgt_align = align;
tgt_size = (tgt_size + align - 1) & ~(align - 1);
tgt_size += sizes[i];
}
if (tgt_align)
tgt_size += tgt_align - 1;
else
tgt_size = 0;
}
task = gomp_malloc (sizeof (*task) + depend_size
+ sizeof (*ttask)
+ mapnum * (sizeof (void *) + sizeof (size_t)
+ sizeof (unsigned short))
+ tgt_size);
gomp_init_task (task, parent, gomp_icv (false));
task->priority = 0;
task->kind = GOMP_TASK_WAITING;
task->in_tied_task = parent->in_tied_task;
task->taskgroup = taskgroup;
ttask = (struct gomp_target_task *) &task->depend[depend_cnt];
ttask->devicep = devicep;
ttask->fn = fn;
ttask->mapnum = mapnum;
ttask->args = args;
memcpy (ttask->hostaddrs, hostaddrs, mapnum * sizeof (void *));
ttask->sizes = (size_t *) &ttask->hostaddrs[mapnum];
memcpy (ttask->sizes, sizes, mapnum * sizeof (size_t));
ttask->kinds = (unsigned short *) &ttask->sizes[mapnum];
memcpy (ttask->kinds, kinds, mapnum * sizeof (unsigned short));
if (tgt_align)
{
char *tgt = (char *) &ttask->kinds[mapnum];
size_t i;
uintptr_t al = (uintptr_t) tgt & (tgt_align - 1);
if (al)
tgt += tgt_align - al;
tgt_size = 0;
for (i = 0; i < mapnum; i++)
if ((kinds[i] & 0xff) == GOMP_MAP_FIRSTPRIVATE)
{
size_t align = (size_t) 1 << (kinds[i] >> 8);
tgt_size = (tgt_size + align - 1) & ~(align - 1);
memcpy (tgt + tgt_size, hostaddrs[i], sizes[i]);
ttask->hostaddrs[i] = tgt + tgt_size;
tgt_size = tgt_size + sizes[i];
}
}
ttask->flags = flags;
ttask->state = state;
ttask->task = task;
ttask->team = team;
task->fn = NULL;
task->fn_data = ttask;
task->final_task = 0;
gomp_mutex_lock (&team->task_lock);
/* If parallel or taskgroup has been cancelled, don't start new tasks. */
if (__builtin_expect (gomp_cancel_var, 0))
{
if (gomp_team_barrier_cancelled (&team->barrier))
{
do_cancel:
gomp_mutex_unlock (&team->task_lock);
gomp_finish_task (task);
free (task);
return true;
}
if (taskgroup)
{
if (taskgroup->cancelled)
goto do_cancel;
if (taskgroup->workshare
&& taskgroup->prev
&& taskgroup->prev->cancelled)
goto do_cancel;
}
}
if (depend_size)
{
gomp_task_handle_depend (task, parent, depend);
if (task->num_dependees)
{
if (taskgroup)
taskgroup->num_children++;
gomp_mutex_unlock (&team->task_lock);
return true;
}
}
if (state == GOMP_TARGET_TASK_DATA)
{
gomp_task_run_post_handle_depend_hash (task);
gomp_mutex_unlock (&team->task_lock);
gomp_finish_task (task);
free (task);
return false;
}
if (taskgroup)
taskgroup->num_children++;
/* For async offloading, if we don't need to wait for dependencies,
run the gomp_target_task_fn right away, essentially schedule the
mapping part of the task in the current thread. */
if (devicep != NULL
&& (devicep->capabilities & GOMP_OFFLOAD_CAP_OPENMP_400))
{
priority_queue_insert (PQ_CHILDREN, &parent->children_queue, task, 0,
PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
if (taskgroup)
priority_queue_insert (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
task, 0, PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
task->pnode[PQ_TEAM].next = NULL;
task->pnode[PQ_TEAM].prev = NULL;
task->kind = GOMP_TASK_TIED;
++team->task_count;
gomp_mutex_unlock (&team->task_lock);
thr->task = task;
gomp_target_task_fn (task->fn_data);
thr->task = parent;
gomp_mutex_lock (&team->task_lock);
task->kind = GOMP_TASK_ASYNC_RUNNING;
/* If GOMP_PLUGIN_target_task_completion has run already
in between gomp_target_task_fn and the mutex lock,
perform the requeuing here. */
if (ttask->state == GOMP_TARGET_TASK_FINISHED)
gomp_target_task_completion (team, task);
else
ttask->state = GOMP_TARGET_TASK_RUNNING;
gomp_mutex_unlock (&team->task_lock);
return true;
}
priority_queue_insert (PQ_CHILDREN, &parent->children_queue, task, 0,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
if (taskgroup)
priority_queue_insert (PQ_TASKGROUP, &taskgroup->taskgroup_queue, task, 0,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
priority_queue_insert (PQ_TEAM, &team->task_queue, task, 0,
PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
++team->task_count;
++team->task_queued_count;
gomp_team_barrier_set_task_pending (&team->barrier);
do_wake = team->task_running_count + !parent->in_tied_task
< team->nthreads;
gomp_mutex_unlock (&team->task_lock);
if (do_wake)
gomp_team_barrier_wake (&team->barrier, 1);
return true;
}
/* Given a parent_depends_on task in LIST, move it to the front of its
priority so it is run as soon as possible.
Care is taken to update the list's LAST_PARENT_DEPENDS_ON field.
We rearrange the queue such that all parent_depends_on tasks are
first, and last_parent_depends_on points to the last such task we
rearranged. For example, given the following tasks in a queue
where PD[123] are the parent_depends_on tasks:
task->children
|
V
C1 -> C2 -> C3 -> PD1 -> PD2 -> PD3 -> C4
We rearrange such that:
task->children
| +--- last_parent_depends_on
| |
V V
PD1 -> PD2 -> PD3 -> C1 -> C2 -> C3 -> C4. */
static void inline
priority_list_upgrade_task (struct priority_list *list,
struct priority_node *node)
{
struct priority_node *last_parent_depends_on
= list->last_parent_depends_on;
if (last_parent_depends_on)
{
node->prev->next = node->next;
node->next->prev = node->prev;
node->prev = last_parent_depends_on;
node->next = last_parent_depends_on->next;
node->prev->next = node;
node->next->prev = node;
}
else if (node != list->tasks)
{
node->prev->next = node->next;
node->next->prev = node->prev;
node->prev = list->tasks->prev;
node->next = list->tasks;
list->tasks = node;
node->prev->next = node;
node->next->prev = node;
}
list->last_parent_depends_on = node;
}
/* Given a parent_depends_on TASK in its parent's children_queue, move
it to the front of its priority so it is run as soon as possible.
PARENT is passed as an optimization.
(This function could be defined in priority_queue.c, but we want it
inlined, and putting it in priority_queue.h is not an option, given
that gomp_task has not been properly defined at that point). */
static void inline
priority_queue_upgrade_task (struct gomp_task *task,
struct gomp_task *parent)
{
struct priority_queue *head = &parent->children_queue;
struct priority_node *node = &task->pnode[PQ_CHILDREN];
#if _LIBGOMP_CHECKING_
if (!task->parent_depends_on)
gomp_fatal ("priority_queue_upgrade_task: task must be a "
"parent_depends_on task");
if (!priority_queue_task_in_queue_p (PQ_CHILDREN, head, task))
gomp_fatal ("priority_queue_upgrade_task: cannot find task=%p", task);
#endif
if (priority_queue_multi_p (head))
{
struct priority_list *list
= priority_queue_lookup_priority (head, task->priority);
priority_list_upgrade_task (list, node);
}
else
priority_list_upgrade_task (&head->l, node);
}
/* Given a CHILD_TASK in LIST that is about to be executed, move it out of
the way in LIST so that other tasks can be considered for
execution. LIST contains tasks of type TYPE.
Care is taken to update the queue's LAST_PARENT_DEPENDS_ON field
if applicable. */
static void inline
priority_list_downgrade_task (enum priority_queue_type type,
struct priority_list *list,
struct gomp_task *child_task)
{
struct priority_node *node = task_to_priority_node (type, child_task);
if (list->tasks == node)
list->tasks = node->next;
else if (node->next != list->tasks)
{
/* The task in NODE is about to become TIED and TIED tasks
cannot come before WAITING tasks. If we're about to
leave the queue in such an indeterminate state, rewire
things appropriately. However, a TIED task at the end is
perfectly fine. */
struct gomp_task *next_task = priority_node_to_task (type, node->next);
if (next_task->kind == GOMP_TASK_WAITING)
{
/* Remove from list. */
node->prev->next = node->next;
node->next->prev = node->prev;
/* Rewire at the end. */
node->next = list->tasks;
node->prev = list->tasks->prev;
list->tasks->prev->next = node;
list->tasks->prev = node;
}
}
/* If the current task is the last_parent_depends_on for its
priority, adjust last_parent_depends_on appropriately. */
if (__builtin_expect (child_task->parent_depends_on, 0)
&& list->last_parent_depends_on == node)
{
struct gomp_task *prev_child = priority_node_to_task (type, node->prev);
if (node->prev != node
&& prev_child->kind == GOMP_TASK_WAITING
&& prev_child->parent_depends_on)
list->last_parent_depends_on = node->prev;
else
{
/* There are no more parent_depends_on entries waiting
to run, clear the list. */
list->last_parent_depends_on = NULL;
}
}
}
/* Given a TASK in HEAD that is about to be executed, move it out of
the way so that other tasks can be considered for execution. HEAD
contains tasks of type TYPE.
Care is taken to update the queue's LAST_PARENT_DEPENDS_ON field
if applicable.
(This function could be defined in priority_queue.c, but we want it
inlined, and putting it in priority_queue.h is not an option, given
that gomp_task has not been properly defined at that point). */
static void inline
priority_queue_downgrade_task (enum priority_queue_type type,
struct priority_queue *head,
struct gomp_task *task)
{
#if _LIBGOMP_CHECKING_
if (!priority_queue_task_in_queue_p (type, head, task))
gomp_fatal ("Attempt to downgrade missing task %p", task);
#endif
if (priority_queue_multi_p (head))
{
struct priority_list *list
= priority_queue_lookup_priority (head, task->priority);
priority_list_downgrade_task (type, list, task);
}
else
priority_list_downgrade_task (type, &head->l, task);
}
/* Setup CHILD_TASK to execute. This is done by setting the task to
TIED, and updating all relevant queues so that CHILD_TASK is no
longer chosen for scheduling. Also, remove CHILD_TASK from the
overall team task queue entirely.
Return TRUE if task or its containing taskgroup has been
cancelled. */
static inline bool
gomp_task_run_pre (struct gomp_task *child_task, struct gomp_task *parent,
struct gomp_team *team)
{
#if _LIBGOMP_CHECKING_
if (child_task->parent)
priority_queue_verify (PQ_CHILDREN,
&child_task->parent->children_queue, true);
if (child_task->taskgroup)
priority_queue_verify (PQ_TASKGROUP,
&child_task->taskgroup->taskgroup_queue, false);
priority_queue_verify (PQ_TEAM, &team->task_queue, false);
#endif
/* Task is about to go tied, move it out of the way. */
if (parent)
priority_queue_downgrade_task (PQ_CHILDREN, &parent->children_queue,
child_task);
/* Task is about to go tied, move it out of the way. */
struct gomp_taskgroup *taskgroup = child_task->taskgroup;
if (taskgroup)
priority_queue_downgrade_task (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
child_task);
priority_queue_remove (PQ_TEAM, &team->task_queue, child_task,
MEMMODEL_RELAXED);
child_task->pnode[PQ_TEAM].next = NULL;
child_task->pnode[PQ_TEAM].prev = NULL;
child_task->kind = GOMP_TASK_TIED;
if (--team->task_queued_count == 0)
gomp_team_barrier_clear_task_pending (&team->barrier);
if (__builtin_expect (gomp_cancel_var, 0)
&& !child_task->copy_ctors_done)
{
if (gomp_team_barrier_cancelled (&team->barrier))
return true;
if (taskgroup)
{
if (taskgroup->cancelled)
return true;
if (taskgroup->workshare
&& taskgroup->prev
&& taskgroup->prev->cancelled)
return true;
}
}
return false;
}
static void
gomp_task_run_post_handle_depend_hash (struct gomp_task *child_task)
{
struct gomp_task *parent = child_task->parent;
size_t i;
for (i = 0; i < child_task->depend_count; i++)
if (!child_task->depend[i].redundant)
{
if (child_task->depend[i].next)
child_task->depend[i].next->prev = child_task->depend[i].prev;
if (child_task->depend[i].prev)
child_task->depend[i].prev->next = child_task->depend[i].next;
else
{
hash_entry_type *slot
= htab_find_slot (&parent->depend_hash, &child_task->depend[i],
NO_INSERT);
if (*slot != &child_task->depend[i])
abort ();
if (child_task->depend[i].next)
*slot = child_task->depend[i].next;
else
htab_clear_slot (parent->depend_hash, slot);
}
}
}
/* After a CHILD_TASK has been run, adjust the dependency queue for
each task that depends on CHILD_TASK, to record the fact that there
is one less dependency to worry about. If a task that depended on
CHILD_TASK now has no dependencies, place it in the various queues
so it gets scheduled to run.
TEAM is the team to which CHILD_TASK belongs to. */
static size_t
gomp_task_run_post_handle_dependers (struct gomp_task *child_task,
struct gomp_team *team)
{
struct gomp_task *parent = child_task->parent;
size_t i, count = child_task->dependers->n_elem, ret = 0;
for (i = 0; i < count; i++)
{
struct gomp_task *task = child_task->dependers->elem[i];
/* CHILD_TASK satisfies a dependency for TASK. Keep track of
TASK's remaining dependencies. Once TASK has no other
depenencies, put it into the various queues so it will get
scheduled for execution. */
if (--task->num_dependees != 0)
continue;
struct gomp_taskgroup *taskgroup = task->taskgroup;
if (parent)
{
priority_queue_insert (PQ_CHILDREN, &parent->children_queue,
task, task->priority,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/true,
task->parent_depends_on);
if (parent->taskwait)
{
if (parent->taskwait->in_taskwait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
parent->taskwait->in_taskwait = false;
gomp_sem_post (&parent->taskwait->taskwait_sem);
}
else if (parent->taskwait->in_depend_wait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
parent->taskwait->in_depend_wait = false;
gomp_sem_post (&parent->taskwait->taskwait_sem);
}
}
}
if (taskgroup)
{
priority_queue_insert (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
task, task->priority,
PRIORITY_INSERT_BEGIN,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
if (taskgroup->in_taskgroup_wait)
{
/* One more task has had its dependencies met.
Inform any waiters. */
taskgroup->in_taskgroup_wait = false;
gomp_sem_post (&taskgroup->taskgroup_sem);
}
}
priority_queue_insert (PQ_TEAM, &team->task_queue,
task, task->priority,
PRIORITY_INSERT_END,
/*adjust_parent_depends_on=*/false,
task->parent_depends_on);
++team->task_count;
++team->task_queued_count;
++ret;
}
free (child_task->dependers);
child_task->dependers = NULL;
if (ret > 1)
gomp_team_barrier_set_task_pending (&team->barrier);
return ret;
}
static inline size_t
gomp_task_run_post_handle_depend (struct gomp_task *child_task,
struct gomp_team *team)
{
if (child_task->depend_count == 0)
return 0;
/* If parent is gone already, the hash table is freed and nothing
will use the hash table anymore, no need to remove anything from it. */
if (child_task->parent != NULL)
gomp_task_run_post_handle_depend_hash (child_task);
if (child_task->dependers == NULL)
return 0;
return gomp_task_run_post_handle_dependers (child_task, team);
}
/* Remove CHILD_TASK from its parent. */
static inline void
gomp_task_run_post_remove_parent (struct gomp_task *child_task)
{
struct gomp_task *parent = child_task->parent;
if (parent == NULL)
return;
/* If this was the last task the parent was depending on,
synchronize with gomp_task_maybe_wait_for_dependencies so it can
clean up and return. */
if (__builtin_expect (child_task->parent_depends_on, 0)
&& --parent->taskwait->n_depend == 0
&& parent->taskwait->in_depend_wait)
{
parent->taskwait->in_depend_wait = false;
gomp_sem_post (&parent->taskwait->taskwait_sem);
}
if (priority_queue_remove (PQ_CHILDREN, &parent->children_queue,
child_task, MEMMODEL_RELEASE)
&& parent->taskwait && parent->taskwait->in_taskwait)
{
parent->taskwait->in_taskwait = false;
gomp_sem_post (&parent->taskwait->taskwait_sem);
}
child_task->pnode[PQ_CHILDREN].next = NULL;
child_task->pnode[PQ_CHILDREN].prev = NULL;
}
/* Remove CHILD_TASK from its taskgroup. */
static inline void
gomp_task_run_post_remove_taskgroup (struct gomp_task *child_task)
{
struct gomp_taskgroup *taskgroup = child_task->taskgroup;
if (taskgroup == NULL)
return;
bool empty = priority_queue_remove (PQ_TASKGROUP,
&taskgroup->taskgroup_queue,
child_task, MEMMODEL_RELAXED);
child_task->pnode[PQ_TASKGROUP].next = NULL;
child_task->pnode[PQ_TASKGROUP].prev = NULL;
if (taskgroup->num_children > 1)
--taskgroup->num_children;
else
{
/* We access taskgroup->num_children in GOMP_taskgroup_end
outside of the task lock mutex region, so
need a release barrier here to ensure memory
written by child_task->fn above is flushed
before the NULL is written. */
__atomic_store_n (&taskgroup->num_children, 0, MEMMODEL_RELEASE);
}
if (empty && taskgroup->in_taskgroup_wait)
{
taskgroup->in_taskgroup_wait = false;
gomp_sem_post (&taskgroup->taskgroup_sem);
}
}
void
gomp_barrier_handle_tasks (gomp_barrier_state_t state)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
struct gomp_task *task = thr->task;
struct gomp_task *child_task = NULL;
struct gomp_task *to_free = NULL;
int do_wake = 0;
gomp_mutex_lock (&team->task_lock);
if (gomp_barrier_last_thread (state))
{
if (team->task_count == 0)
{
gomp_team_barrier_done (&team->barrier, state);
gomp_mutex_unlock (&team->task_lock);
gomp_team_barrier_wake (&team->barrier, 0);
return;
}
gomp_team_barrier_set_waiting_for_tasks (&team->barrier);
}
while (1)
{
bool cancelled = false;
if (!priority_queue_empty_p (&team->task_queue, MEMMODEL_RELAXED))
{
bool ignored;
child_task
= priority_queue_next_task (PQ_TEAM, &team->task_queue,
PQ_IGNORED, NULL,
&ignored);
cancelled = gomp_task_run_pre (child_task, child_task->parent,
team);
if (__builtin_expect (cancelled, 0))
{
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
to_free = NULL;
}
goto finish_cancelled;
}
team->task_running_count++;
child_task->in_tied_task = true;
}
gomp_mutex_unlock (&team->task_lock);
if (do_wake)
{
gomp_team_barrier_wake (&team->barrier, do_wake);
do_wake = 0;
}
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
to_free = NULL;
}
if (child_task)
{
thr->task = child_task;
if (__builtin_expect (child_task->fn == NULL, 0))
{
if (gomp_target_task_fn (child_task->fn_data))
{
thr->task = task;
gomp_mutex_lock (&team->task_lock);
child_task->kind = GOMP_TASK_ASYNC_RUNNING;
team->task_running_count--;
struct gomp_target_task *ttask
= (struct gomp_target_task *) child_task->fn_data;
/* If GOMP_PLUGIN_target_task_completion has run already
in between gomp_target_task_fn and the mutex lock,
perform the requeuing here. */
if (ttask->state == GOMP_TARGET_TASK_FINISHED)
gomp_target_task_completion (team, child_task);
else
ttask->state = GOMP_TARGET_TASK_RUNNING;
child_task = NULL;
continue;
}
}
else
child_task->fn (child_task->fn_data);
thr->task = task;
}
else
return;
gomp_mutex_lock (&team->task_lock);
if (child_task)
{
finish_cancelled:;
size_t new_tasks
= gomp_task_run_post_handle_depend (child_task, team);
gomp_task_run_post_remove_parent (child_task);
gomp_clear_parent (&child_task->children_queue);
gomp_task_run_post_remove_taskgroup (child_task);
to_free = child_task;
child_task = NULL;
if (!cancelled)
team->task_running_count--;
if (new_tasks > 1)
{
do_wake = team->nthreads - team->task_running_count;
if (do_wake > new_tasks)
do_wake = new_tasks;
}
if (--team->task_count == 0
&& gomp_team_barrier_waiting_for_tasks (&team->barrier))
{
gomp_team_barrier_done (&team->barrier, state);
gomp_mutex_unlock (&team->task_lock);
gomp_team_barrier_wake (&team->barrier, 0);
gomp_mutex_lock (&team->task_lock);
}
}
}
}
/* Called when encountering a taskwait directive.
Wait for all children of the current task. */
void
GOMP_taskwait (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
struct gomp_task *task = thr->task;
struct gomp_task *child_task = NULL;
struct gomp_task *to_free = NULL;
struct gomp_taskwait taskwait;
int do_wake = 0;
/* The acquire barrier on load of task->children here synchronizes
with the write of a NULL in gomp_task_run_post_remove_parent. It is
not necessary that we synchronize with other non-NULL writes at
this point, but we must ensure that all writes to memory by a
child thread task work function are seen before we exit from
GOMP_taskwait. */
if (task == NULL
|| priority_queue_empty_p (&task->children_queue, MEMMODEL_ACQUIRE))
return;
memset (&taskwait, 0, sizeof (taskwait));
bool child_q = false;
gomp_mutex_lock (&team->task_lock);
while (1)
{
bool cancelled = false;
if (priority_queue_empty_p (&task->children_queue, MEMMODEL_RELAXED))
{
bool destroy_taskwait = task->taskwait != NULL;
task->taskwait = NULL;
gomp_mutex_unlock (&team->task_lock);
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
}
if (destroy_taskwait)
gomp_sem_destroy (&taskwait.taskwait_sem);
return;
}
struct gomp_task *next_task
= priority_queue_next_task (PQ_CHILDREN, &task->children_queue,
PQ_TEAM, &team->task_queue, &child_q);
if (next_task->kind == GOMP_TASK_WAITING)
{
child_task = next_task;
cancelled
= gomp_task_run_pre (child_task, task, team);
if (__builtin_expect (cancelled, 0))
{
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
to_free = NULL;
}
goto finish_cancelled;
}
}
else
{
/* All tasks we are waiting for are either running in other
threads, or they are tasks that have not had their
dependencies met (so they're not even in the queue). Wait
for them. */
if (task->taskwait == NULL)
{
taskwait.in_depend_wait = false;
gomp_sem_init (&taskwait.taskwait_sem, 0);
task->taskwait = &taskwait;
}
taskwait.in_taskwait = true;
}
gomp_mutex_unlock (&team->task_lock);
if (do_wake)
{
gomp_team_barrier_wake (&team->barrier, do_wake);
do_wake = 0;
}
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
to_free = NULL;
}
if (child_task)
{
thr->task = child_task;
if (__builtin_expect (child_task->fn == NULL, 0))
{
if (gomp_target_task_fn (child_task->fn_data))
{
thr->task = task;
gomp_mutex_lock (&team->task_lock);
child_task->kind = GOMP_TASK_ASYNC_RUNNING;
struct gomp_target_task *ttask
= (struct gomp_target_task *) child_task->fn_data;
/* If GOMP_PLUGIN_target_task_completion has run already
in between gomp_target_task_fn and the mutex lock,
perform the requeuing here. */
if (ttask->state == GOMP_TARGET_TASK_FINISHED)
gomp_target_task_completion (team, child_task);
else
ttask->state = GOMP_TARGET_TASK_RUNNING;
child_task = NULL;
continue;
}
}
else
child_task->fn (child_task->fn_data);
thr->task = task;
}
else
gomp_sem_wait (&taskwait.taskwait_sem);
gomp_mutex_lock (&team->task_lock);
if (child_task)
{
finish_cancelled:;
size_t new_tasks
= gomp_task_run_post_handle_depend (child_task, team);
if (child_q)
{
priority_queue_remove (PQ_CHILDREN, &task->children_queue,
child_task, MEMMODEL_RELAXED);
child_task->pnode[PQ_CHILDREN].next = NULL;
child_task->pnode[PQ_CHILDREN].prev = NULL;
}
gomp_clear_parent (&child_task->children_queue);
gomp_task_run_post_remove_taskgroup (child_task);
to_free = child_task;
child_task = NULL;
team->task_count--;
if (new_tasks > 1)
{
do_wake = team->nthreads - team->task_running_count
- !task->in_tied_task;
if (do_wake > new_tasks)
do_wake = new_tasks;
}
}
}
}
/* Called when encountering a taskwait directive with depend clause(s).
Wait as if it was an mergeable included task construct with empty body. */
void
GOMP_taskwait_depend (void **depend)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
/* If parallel or taskgroup has been cancelled, return early. */
if (__builtin_expect (gomp_cancel_var, 0) && team)
{
if (gomp_team_barrier_cancelled (&team->barrier))
return;
if (thr->task->taskgroup)
{
if (thr->task->taskgroup->cancelled)
return;
if (thr->task->taskgroup->workshare
&& thr->task->taskgroup->prev
&& thr->task->taskgroup->prev->cancelled)
return;
}
}
if (thr->task && thr->task->depend_hash)
gomp_task_maybe_wait_for_dependencies (depend);
}
/* An undeferred task is about to run. Wait for all tasks that this
undeferred task depends on.
This is done by first putting all known ready dependencies
(dependencies that have their own dependencies met) at the top of
the scheduling queues. Then we iterate through these imminently
ready tasks (and possibly other high priority tasks), and run them.
If we run out of ready dependencies to execute, we either wait for
the remaining dependencies to finish, or wait for them to get
scheduled so we can run them.
DEPEND is as in GOMP_task. */
void
gomp_task_maybe_wait_for_dependencies (void **depend)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_task *task = thr->task;
struct gomp_team *team = thr->ts.team;
struct gomp_task_depend_entry elem, *ent = NULL;
struct gomp_taskwait taskwait;
size_t orig_ndepend = (uintptr_t) depend[0];
size_t nout = (uintptr_t) depend[1];
size_t ndepend = orig_ndepend;
size_t normal = ndepend;
size_t n = 2;
size_t i;
size_t num_awaited = 0;
struct gomp_task *child_task = NULL;
struct gomp_task *to_free = NULL;
int do_wake = 0;
if (ndepend == 0)
{
ndepend = nout;
nout = (uintptr_t) depend[2] + (uintptr_t) depend[3];
normal = nout + (uintptr_t) depend[4];
n = 5;
}
gomp_mutex_lock (&team->task_lock);
for (i = 0; i < ndepend; i++)
{
elem.addr = depend[i + n];
elem.is_in = i >= nout;
if (__builtin_expect (i >= normal, 0))
{
void **d = (void **) elem.addr;
switch ((uintptr_t) d[1])
{
case GOMP_DEPEND_IN:
break;
case GOMP_DEPEND_OUT:
case GOMP_DEPEND_INOUT:
case GOMP_DEPEND_MUTEXINOUTSET:
elem.is_in = 0;
break;
default:
gomp_fatal ("unknown omp_depend_t dependence type %d",
(int) (uintptr_t) d[1]);
}
elem.addr = d[0];
}
ent = htab_find (task->depend_hash, &elem);
for (; ent; ent = ent->next)
if (elem.is_in && ent->is_in)
continue;
else
{
struct gomp_task *tsk = ent->task;
if (!tsk->parent_depends_on)
{
tsk->parent_depends_on = true;
++num_awaited;
/* If depenency TSK itself has no dependencies and is
ready to run, move it up front so that we run it as
soon as possible. */
if (tsk->num_dependees == 0 && tsk->kind == GOMP_TASK_WAITING)
priority_queue_upgrade_task (tsk, task);
}
}
}
if (num_awaited == 0)
{
gomp_mutex_unlock (&team->task_lock);
return;
}
memset (&taskwait, 0, sizeof (taskwait));
taskwait.n_depend = num_awaited;
gomp_sem_init (&taskwait.taskwait_sem, 0);
task->taskwait = &taskwait;
while (1)
{
bool cancelled = false;
if (taskwait.n_depend == 0)
{
task->taskwait = NULL;
gomp_mutex_unlock (&team->task_lock);
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
}
gomp_sem_destroy (&taskwait.taskwait_sem);
return;
}
/* Theoretically when we have multiple priorities, we should
chose between the highest priority item in
task->children_queue and team->task_queue here, so we should
use priority_queue_next_task(). However, since we are
running an undeferred task, perhaps that makes all tasks it
depends on undeferred, thus a priority of INF? This would
make it unnecessary to take anything into account here,
but the dependencies.
On the other hand, if we want to use priority_queue_next_task(),
care should be taken to only use priority_queue_remove()
below if the task was actually removed from the children
queue. */
bool ignored;
struct gomp_task *next_task
= priority_queue_next_task (PQ_CHILDREN, &task->children_queue,
PQ_IGNORED, NULL, &ignored);
if (next_task->kind == GOMP_TASK_WAITING)
{
child_task = next_task;
cancelled
= gomp_task_run_pre (child_task, task, team);
if (__builtin_expect (cancelled, 0))
{
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
to_free = NULL;
}
goto finish_cancelled;
}
}
else
/* All tasks we are waiting for are either running in other
threads, or they are tasks that have not had their
dependencies met (so they're not even in the queue). Wait
for them. */
taskwait.in_depend_wait = true;
gomp_mutex_unlock (&team->task_lock);
if (do_wake)
{
gomp_team_barrier_wake (&team->barrier, do_wake);
do_wake = 0;
}
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
to_free = NULL;
}
if (child_task)
{
thr->task = child_task;
if (__builtin_expect (child_task->fn == NULL, 0))
{
if (gomp_target_task_fn (child_task->fn_data))
{
thr->task = task;
gomp_mutex_lock (&team->task_lock);
child_task->kind = GOMP_TASK_ASYNC_RUNNING;
struct gomp_target_task *ttask
= (struct gomp_target_task *) child_task->fn_data;
/* If GOMP_PLUGIN_target_task_completion has run already
in between gomp_target_task_fn and the mutex lock,
perform the requeuing here. */
if (ttask->state == GOMP_TARGET_TASK_FINISHED)
gomp_target_task_completion (team, child_task);
else
ttask->state = GOMP_TARGET_TASK_RUNNING;
child_task = NULL;
continue;
}
}
else
child_task->fn (child_task->fn_data);
thr->task = task;
}
else
gomp_sem_wait (&taskwait.taskwait_sem);
gomp_mutex_lock (&team->task_lock);
if (child_task)
{
finish_cancelled:;
size_t new_tasks
= gomp_task_run_post_handle_depend (child_task, team);
if (child_task->parent_depends_on)
--taskwait.n_depend;
priority_queue_remove (PQ_CHILDREN, &task->children_queue,
child_task, MEMMODEL_RELAXED);
child_task->pnode[PQ_CHILDREN].next = NULL;
child_task->pnode[PQ_CHILDREN].prev = NULL;
gomp_clear_parent (&child_task->children_queue);
gomp_task_run_post_remove_taskgroup (child_task);
to_free = child_task;
child_task = NULL;
team->task_count--;
if (new_tasks > 1)
{
do_wake = team->nthreads - team->task_running_count
- !task->in_tied_task;
if (do_wake > new_tasks)
do_wake = new_tasks;
}
}
}
}
/* Called when encountering a taskyield directive. */
void
GOMP_taskyield (void)
{
/* Nothing at the moment. */
}
static inline struct gomp_taskgroup *
gomp_taskgroup_init (struct gomp_taskgroup *prev)
{
struct gomp_taskgroup *taskgroup
= gomp_malloc (sizeof (struct gomp_taskgroup));
taskgroup->prev = prev;
priority_queue_init (&taskgroup->taskgroup_queue);
taskgroup->reductions = prev ? prev->reductions : NULL;
taskgroup->in_taskgroup_wait = false;
taskgroup->cancelled = false;
taskgroup->workshare = false;
taskgroup->num_children = 0;
gomp_sem_init (&taskgroup->taskgroup_sem, 0);
return taskgroup;
}
void
GOMP_taskgroup_start (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
struct gomp_task *task = thr->task;
/* If team is NULL, all tasks are executed as
GOMP_TASK_UNDEFERRED tasks and thus all children tasks of
taskgroup and their descendant tasks will be finished
by the time GOMP_taskgroup_end is called. */
if (team == NULL)
return;
task->taskgroup = gomp_taskgroup_init (task->taskgroup);
}
void
GOMP_taskgroup_end (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
struct gomp_task *task = thr->task;
struct gomp_taskgroup *taskgroup;
struct gomp_task *child_task = NULL;
struct gomp_task *to_free = NULL;
int do_wake = 0;
if (team == NULL)
return;
taskgroup = task->taskgroup;
if (__builtin_expect (taskgroup == NULL, 0)
&& thr->ts.level == 0)
{
/* This can happen if GOMP_taskgroup_start is called when
thr->ts.team == NULL, but inside of the taskgroup there
is #pragma omp target nowait that creates an implicit
team with a single thread. In this case, we want to wait
for all outstanding tasks in this team. */
gomp_team_barrier_wait (&team->barrier);
return;
}
/* The acquire barrier on load of taskgroup->num_children here
synchronizes with the write of 0 in gomp_task_run_post_remove_taskgroup.
It is not necessary that we synchronize with other non-0 writes at
this point, but we must ensure that all writes to memory by a
child thread task work function are seen before we exit from
GOMP_taskgroup_end. */
if (__atomic_load_n (&taskgroup->num_children, MEMMODEL_ACQUIRE) == 0)
goto finish;
bool unused;
gomp_mutex_lock (&team->task_lock);
while (1)
{
bool cancelled = false;
if (priority_queue_empty_p (&taskgroup->taskgroup_queue,
MEMMODEL_RELAXED))
{
if (taskgroup->num_children)
{
if (priority_queue_empty_p (&task->children_queue,
MEMMODEL_RELAXED))
goto do_wait;
child_task
= priority_queue_next_task (PQ_CHILDREN, &task->children_queue,
PQ_TEAM, &team->task_queue,
&unused);
}
else
{
gomp_mutex_unlock (&team->task_lock);
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
}
goto finish;
}
}
else
child_task
= priority_queue_next_task (PQ_TASKGROUP, &taskgroup->taskgroup_queue,
PQ_TEAM, &team->task_queue, &unused);
if (child_task->kind == GOMP_TASK_WAITING)
{
cancelled
= gomp_task_run_pre (child_task, child_task->parent, team);
if (__builtin_expect (cancelled, 0))
{
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
to_free = NULL;
}
goto finish_cancelled;
}
}
else
{
child_task = NULL;
do_wait:
/* All tasks we are waiting for are either running in other
threads, or they are tasks that have not had their
dependencies met (so they're not even in the queue). Wait
for them. */
taskgroup->in_taskgroup_wait = true;
}
gomp_mutex_unlock (&team->task_lock);
if (do_wake)
{
gomp_team_barrier_wake (&team->barrier, do_wake);
do_wake = 0;
}
if (to_free)
{
gomp_finish_task (to_free);
free (to_free);
to_free = NULL;
}
if (child_task)
{
thr->task = child_task;
if (__builtin_expect (child_task->fn == NULL, 0))
{
if (gomp_target_task_fn (child_task->fn_data))
{
thr->task = task;
gomp_mutex_lock (&team->task_lock);
child_task->kind = GOMP_TASK_ASYNC_RUNNING;
struct gomp_target_task *ttask
= (struct gomp_target_task *) child_task->fn_data;
/* If GOMP_PLUGIN_target_task_completion has run already
in between gomp_target_task_fn and the mutex lock,
perform the requeuing here. */
if (ttask->state == GOMP_TARGET_TASK_FINISHED)
gomp_target_task_completion (team, child_task);
else
ttask->state = GOMP_TARGET_TASK_RUNNING;
child_task = NULL;
continue;
}
}
else
child_task->fn (child_task->fn_data);
thr->task = task;
}
else
gomp_sem_wait (&taskgroup->taskgroup_sem);
gomp_mutex_lock (&team->task_lock);
if (child_task)
{
finish_cancelled:;
size_t new_tasks
= gomp_task_run_post_handle_depend (child_task, team);
gomp_task_run_post_remove_parent (child_task);
gomp_clear_parent (&child_task->children_queue);
gomp_task_run_post_remove_taskgroup (child_task);
to_free = child_task;
child_task = NULL;
team->task_count--;
if (new_tasks > 1)
{
do_wake = team->nthreads - team->task_running_count
- !task->in_tied_task;
if (do_wake > new_tasks)
do_wake = new_tasks;
}
}
}
finish:
task->taskgroup = taskgroup->prev;
gomp_sem_destroy (&taskgroup->taskgroup_sem);
free (taskgroup);
}
static inline __attribute__((always_inline)) void
gomp_reduction_register (uintptr_t *data, uintptr_t *old, uintptr_t *orig,
unsigned nthreads)
{
size_t total_cnt = 0;
uintptr_t *d = data;
struct htab *old_htab = NULL, *new_htab;
do
{
if (__builtin_expect (orig != NULL, 0))
{
/* For worksharing task reductions, memory has been allocated
already by some other thread that encountered the construct
earlier. */
d[2] = orig[2];
d[6] = orig[6];
orig = (uintptr_t *) orig[4];
}
else
{
size_t sz = d[1] * nthreads;
/* Should use omp_alloc if d[3] is not -1. */
void *ptr = gomp_aligned_alloc (d[2], sz);
memset (ptr, '\0', sz);
d[2] = (uintptr_t) ptr;
d[6] = d[2] + sz;
}
d[5] = 0;
total_cnt += d[0];
if (d[4] == 0)
{
d[4] = (uintptr_t) old;
break;
}
else
d = (uintptr_t *) d[4];
}
while (1);
if (old && old[5])
{
old_htab = (struct htab *) old[5];
total_cnt += htab_elements (old_htab);
}
new_htab = htab_create (total_cnt);
if (old_htab)
{
/* Copy old hash table, like in htab_expand. */
hash_entry_type *p, *olimit;
new_htab->n_elements = htab_elements (old_htab);
olimit = old_htab->entries + old_htab->size;
p = old_htab->entries;
do
{
hash_entry_type x = *p;
if (x != HTAB_EMPTY_ENTRY && x != HTAB_DELETED_ENTRY)
*find_empty_slot_for_expand (new_htab, htab_hash (x)) = x;
p++;
}
while (p < olimit);
}
d = data;
do
{
size_t j;
for (j = 0; j < d[0]; ++j)
{
uintptr_t *p = d + 7 + j * 3;
p[2] = (uintptr_t) d;
/* Ugly hack, hash_entry_type is defined for the task dependencies,
which hash on the first element which is a pointer. We need
to hash also on the first sizeof (uintptr_t) bytes which contain
a pointer. Hide the cast from the compiler. */
hash_entry_type n;
__asm ("" : "=g" (n) : "0" (p));
*htab_find_slot (&new_htab, n, INSERT) = n;
}
if (d[4] == (uintptr_t) old)
break;
else
d = (uintptr_t *) d[4];
}
while (1);
d[5] = (uintptr_t) new_htab;
}
static void
gomp_create_artificial_team (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_task_icv *icv;
struct gomp_team *team = gomp_new_team (1);
struct gomp_task *task = thr->task;
icv = task ? &task->icv : &gomp_global_icv;
team->prev_ts = thr->ts;
thr->ts.team = team;
thr->ts.team_id = 0;
thr->ts.work_share = &team->work_shares[0];
thr->ts.last_work_share = NULL;
#ifdef HAVE_SYNC_BUILTINS
thr->ts.single_count = 0;
#endif
thr->ts.static_trip = 0;
thr->task = &team->implicit_task[0];
gomp_init_task (thr->task, NULL, icv);
if (task)
{
thr->task = task;
gomp_end_task ();
free (task);
thr->task = &team->implicit_task[0];
}
#ifdef LIBGOMP_USE_PTHREADS
else
pthread_setspecific (gomp_thread_destructor, thr);
#endif
}
/* The format of data is:
data[0] cnt
data[1] size
data[2] alignment (on output array pointer)
data[3] allocator (-1 if malloc allocator)
data[4] next pointer
data[5] used internally (htab pointer)
data[6] used internally (end of array)
cnt times
ent[0] address
ent[1] offset
ent[2] used internally (pointer to data[0])
The entries are sorted by increasing offset, so that a binary
search can be performed. Normally, data[8] is 0, exception is
for worksharing construct task reductions in cancellable parallel,
where at offset 0 there should be space for a pointer and an integer
which are used internally. */
void
GOMP_taskgroup_reduction_register (uintptr_t *data)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
struct gomp_task *task;
unsigned nthreads;
if (__builtin_expect (team == NULL, 0))
{
/* The task reduction code needs a team and task, so for
orphaned taskgroups just create the implicit team. */
gomp_create_artificial_team ();
ialias_call (GOMP_taskgroup_start) ();
team = thr->ts.team;
}
nthreads = team->nthreads;
task = thr->task;
gomp_reduction_register (data, task->taskgroup->reductions, NULL, nthreads);
task->taskgroup->reductions = data;
}
void
GOMP_taskgroup_reduction_unregister (uintptr_t *data)
{
uintptr_t *d = data;
htab_free ((struct htab *) data[5]);
do
{
gomp_aligned_free ((void *) d[2]);
d = (uintptr_t *) d[4];
}
while (d && !d[5]);
}
ialias (GOMP_taskgroup_reduction_unregister)
/* For i = 0 to cnt-1, remap ptrs[i] which is either address of the
original list item or address of previously remapped original list
item to address of the private copy, store that to ptrs[i].
For i < cntorig, additionally set ptrs[cnt+i] to the address of
the original list item. */
void
GOMP_task_reduction_remap (size_t cnt, size_t cntorig, void **ptrs)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_task *task = thr->task;
unsigned id = thr->ts.team_id;
uintptr_t *data = task->taskgroup->reductions;
uintptr_t *d;
struct htab *reduction_htab = (struct htab *) data[5];
size_t i;
for (i = 0; i < cnt; ++i)
{
hash_entry_type ent, n;
__asm ("" : "=g" (ent) : "0" (ptrs + i));
n = htab_find (reduction_htab, ent);
if (n)
{
uintptr_t *p;
__asm ("" : "=g" (p) : "0" (n));
/* At this point, p[0] should be equal to (uintptr_t) ptrs[i],
p[1] is the offset within the allocated chunk for each
thread, p[2] is the array registered with
GOMP_taskgroup_reduction_register, d[2] is the base of the
allocated memory and d[1] is the size of the allocated chunk
for one thread. */
d = (uintptr_t *) p[2];
ptrs[i] = (void *) (d[2] + id * d[1] + p[1]);
if (__builtin_expect (i < cntorig, 0))
ptrs[cnt + i] = (void *) p[0];
continue;
}
d = data;
while (d != NULL)
{
if ((uintptr_t) ptrs[i] >= d[2] && (uintptr_t) ptrs[i] < d[6])
break;
d = (uintptr_t *) d[4];
}
if (d == NULL)
gomp_fatal ("couldn't find matching task_reduction or reduction with "
"task modifier for %p", ptrs[i]);
uintptr_t off = ((uintptr_t) ptrs[i] - d[2]) % d[1];
ptrs[i] = (void *) (d[2] + id * d[1] + off);
if (__builtin_expect (i < cntorig, 0))
{
size_t lo = 0, hi = d[0] - 1;
while (lo <= hi)
{
size_t m = (lo + hi) / 2;
if (d[7 + 3 * m + 1] < off)
lo = m + 1;
else if (d[7 + 3 * m + 1] == off)
{
ptrs[cnt + i] = (void *) d[7 + 3 * m];
break;
}
else
hi = m - 1;
}
if (lo > hi)
gomp_fatal ("couldn't find matching task_reduction or reduction "
"with task modifier for %p", ptrs[i]);
}
}
}
struct gomp_taskgroup *
gomp_parallel_reduction_register (uintptr_t *data, unsigned nthreads)
{
struct gomp_taskgroup *taskgroup = gomp_taskgroup_init (NULL);
gomp_reduction_register (data, NULL, NULL, nthreads);
taskgroup->reductions = data;
return taskgroup;
}
void
gomp_workshare_task_reduction_register (uintptr_t *data, uintptr_t *orig)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
struct gomp_task *task = thr->task;
unsigned nthreads = team->nthreads;
gomp_reduction_register (data, task->taskgroup->reductions, orig, nthreads);
task->taskgroup->reductions = data;
}
void
gomp_workshare_taskgroup_start (void)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_team *team = thr->ts.team;
struct gomp_task *task;
if (team == NULL)
{
gomp_create_artificial_team ();
team = thr->ts.team;
}
task = thr->task;
task->taskgroup = gomp_taskgroup_init (task->taskgroup);
task->taskgroup->workshare = true;
}
void
GOMP_workshare_task_reduction_unregister (bool cancelled)
{
struct gomp_thread *thr = gomp_thread ();
struct gomp_task *task = thr->task;
struct gomp_team *team = thr->ts.team;
uintptr_t *data = task->taskgroup->reductions;
ialias_call (GOMP_taskgroup_end) ();
if (thr->ts.team_id == 0)
ialias_call (GOMP_taskgroup_reduction_unregister) (data);
else
htab_free ((struct htab *) data[5]);
if (!cancelled)
gomp_team_barrier_wait (&team->barrier);
}
int
omp_in_final (void)
{
struct gomp_thread *thr = gomp_thread ();
return thr->task && thr->task->final_task;
}
ialias (omp_in_final)
|