This file is indexed.

/usr/include/d/std/parallelism.d is in libphobos2-ldc-dev 1:0.17.1-1ubuntu1.

This file is owned by root:root, with mode 0o644.

The actual contents of the file can be viewed below.

   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
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527
3528
3529
3530
3531
3532
3533
3534
3535
3536
3537
3538
3539
3540
3541
3542
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571
3572
3573
3574
3575
3576
3577
3578
3579
3580
3581
3582
3583
3584
3585
3586
3587
3588
3589
3590
3591
3592
3593
3594
3595
3596
3597
3598
3599
3600
3601
3602
3603
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616
3617
3618
3619
3620
3621
3622
3623
3624
3625
3626
3627
3628
3629
3630
3631
3632
3633
3634
3635
3636
3637
3638
3639
3640
3641
3642
3643
3644
3645
3646
3647
3648
3649
3650
3651
3652
3653
3654
3655
3656
3657
3658
3659
3660
3661
3662
3663
3664
3665
3666
3667
3668
3669
3670
3671
3672
3673
3674
3675
3676
3677
3678
3679
3680
3681
3682
3683
3684
3685
3686
3687
3688
3689
3690
3691
3692
3693
3694
3695
3696
3697
3698
3699
3700
3701
3702
3703
3704
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719
3720
3721
3722
3723
3724
3725
3726
3727
3728
3729
3730
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741
3742
3743
3744
3745
3746
3747
3748
3749
3750
3751
3752
3753
3754
3755
3756
3757
3758
3759
3760
3761
3762
3763
3764
3765
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779
3780
3781
3782
3783
3784
3785
3786
3787
3788
3789
3790
3791
3792
3793
3794
3795
3796
3797
3798
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808
3809
3810
3811
3812
3813
3814
3815
3816
3817
3818
3819
3820
3821
3822
3823
3824
3825
3826
3827
3828
3829
3830
3831
3832
3833
3834
3835
3836
3837
3838
3839
3840
3841
3842
3843
3844
3845
3846
3847
3848
3849
3850
3851
3852
3853
3854
3855
3856
3857
3858
3859
3860
3861
3862
3863
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877
3878
3879
3880
3881
3882
3883
3884
3885
3886
3887
3888
3889
3890
3891
3892
3893
3894
3895
3896
3897
3898
3899
3900
3901
3902
3903
3904
3905
3906
3907
3908
3909
3910
3911
3912
3913
3914
3915
3916
3917
3918
3919
3920
3921
3922
3923
3924
3925
3926
3927
3928
3929
3930
3931
3932
3933
3934
3935
3936
3937
3938
3939
3940
3941
3942
3943
3944
3945
3946
3947
3948
3949
3950
3951
3952
3953
3954
3955
3956
3957
3958
3959
3960
3961
3962
3963
3964
3965
3966
3967
3968
3969
3970
3971
3972
3973
3974
3975
3976
3977
3978
3979
3980
3981
3982
3983
3984
3985
3986
3987
3988
3989
3990
3991
3992
3993
3994
3995
3996
3997
3998
3999
4000
4001
4002
4003
4004
4005
4006
4007
4008
4009
4010
4011
4012
4013
4014
4015
4016
4017
4018
4019
4020
4021
4022
4023
4024
4025
4026
4027
4028
4029
4030
4031
4032
4033
4034
4035
4036
4037
4038
4039
4040
4041
4042
4043
4044
4045
4046
4047
4048
4049
4050
4051
4052
4053
4054
4055
4056
4057
4058
4059
4060
4061
4062
4063
4064
4065
4066
4067
4068
4069
4070
4071
4072
4073
4074
4075
4076
4077
4078
4079
4080
4081
4082
4083
4084
4085
4086
4087
4088
4089
4090
4091
4092
4093
4094
4095
4096
4097
4098
4099
4100
4101
4102
4103
4104
4105
4106
4107
4108
4109
4110
4111
4112
4113
4114
4115
4116
4117
4118
4119
4120
4121
4122
4123
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138
4139
4140
4141
4142
4143
4144
4145
4146
4147
4148
4149
4150
4151
4152
4153
4154
4155
4156
4157
4158
4159
4160
4161
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178
4179
4180
4181
4182
4183
4184
4185
4186
4187
4188
4189
4190
4191
4192
4193
4194
4195
4196
4197
4198
4199
4200
4201
4202
4203
4204
4205
4206
4207
4208
4209
4210
4211
4212
4213
4214
4215
4216
4217
4218
4219
4220
4221
4222
4223
4224
4225
4226
4227
4228
4229
4230
4231
4232
4233
4234
4235
4236
4237
4238
4239
4240
4241
4242
4243
4244
4245
4246
4247
4248
4249
4250
4251
4252
4253
4254
4255
4256
4257
4258
4259
4260
4261
4262
4263
4264
4265
4266
4267
4268
4269
4270
4271
4272
4273
4274
4275
4276
4277
4278
4279
4280
4281
4282
4283
4284
4285
4286
4287
4288
4289
4290
4291
4292
4293
4294
4295
4296
4297
4298
4299
4300
4301
4302
4303
4304
4305
4306
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321
4322
4323
4324
4325
4326
4327
4328
4329
4330
4331
4332
4333
4334
4335
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350
4351
4352
4353
4354
4355
4356
4357
4358
4359
4360
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381
4382
4383
4384
4385
4386
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407
4408
4409
4410
4411
4412
4413
4414
4415
4416
4417
4418
4419
4420
4421
4422
4423
4424
4425
4426
4427
4428
4429
4430
4431
4432
4433
4434
4435
4436
4437
4438
4439
4440
4441
4442
4443
4444
4445
4446
4447
4448
4449
4450
4451
4452
4453
4454
4455
4456
4457
4458
4459
4460
4461
4462
4463
4464
4465
4466
4467
4468
4469
4470
4471
4472
4473
4474
4475
4476
4477
4478
4479
4480
4481
4482
4483
4484
4485
4486
4487
4488
4489
4490
4491
4492
4493
4494
4495
4496
4497
4498
4499
4500
4501
4502
4503
4504
4505
4506
4507
4508
4509
4510
4511
4512
4513
4514
4515
4516
4517
4518
4519
4520
4521
4522
4523
4524
4525
4526
4527
4528
4529
4530
4531
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544
4545
4546
4547
4548
4549
4550
4551
4552
4553
4554
4555
4556
4557
4558
4559
4560
4561
4562
4563
4564
4565
4566
4567
4568
4569
4570
4571
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586
/**
$(D std._parallelism) implements high-level primitives for SMP _parallelism.
These include parallel foreach, parallel reduce, parallel eager map, pipelining
and future/promise _parallelism.  $(D std._parallelism) is recommended when the
same operation is to be executed in parallel on different data, or when a
function is to be executed in a background thread and its result returned to a
well-defined main thread.  For communication between arbitrary threads, see
$(D std.concurrency).

$(D std._parallelism) is based on the concept of a $(D Task).  A $(D Task) is an
object that represents the fundamental unit of work in this library and may be
executed in parallel with any other $(D Task).  Using $(D Task)
directly allows programming with a future/promise paradigm.  All other
supported _parallelism paradigms (parallel foreach, map, reduce, pipelining)
represent an additional level of abstraction over $(D Task).  They
automatically create one or more $(D Task) objects, or closely related types
that are conceptually identical but not part of the public API.

After creation, a $(D Task) may be executed in a new thread, or submitted
to a $(D TaskPool) for execution.  A $(D TaskPool) encapsulates a task queue
and its worker threads.  Its purpose is to efficiently map a large
number of $(D Task)s onto a smaller number of threads.  A task queue is a
FIFO queue of $(D Task) objects that have been submitted to the
$(D TaskPool) and are awaiting execution.  A worker thread is a thread that
is associated with exactly one task queue.  It executes the $(D Task) at the
front of its queue when the queue has work available, or sleeps when
no work is available.  Each task queue is associated with zero or
more worker threads.  If the result of a $(D Task) is needed before execution
by a worker thread has begun, the $(D Task) can be removed from the task queue
and executed immediately in the thread where the result is needed.

Warning:  Unless marked as $(D @trusted) or $(D @safe), artifacts in
          this module allow implicit data sharing between threads and cannot
          guarantee that client code is free from low level data races.

Synopsis:

---
import std.algorithm, std.parallelism, std.range;

void main() {
    // Parallel reduce can be combined with
    // std.algorithm.map to interesting effect.
    // The following example (thanks to Russel Winder)
    // calculates pi by quadrature  using
    // std.algorithm.map and TaskPool.reduce.
    // getTerm is evaluated in parallel as needed by
    // TaskPool.reduce.
    //
    // Timings on an Athlon 64 X2 dual core machine:
    //
    // TaskPool.reduce:       12.170 s
    // std.algorithm.reduce:  24.065 s

    immutable n = 1_000_000_000;
    immutable delta = 1.0 / n;

    real getTerm(int i)
    {
        immutable x = ( i - 0.5 ) * delta;
        return delta / ( 1.0 + x * x ) ;
    }

    immutable pi = 4.0 * taskPool.reduce!"a + b"(
        std.algorithm.map!getTerm(iota(n))
    );
}
---

Source:    $(PHOBOSSRC std/_parallelism.d)
Author:  David Simcha
Copyright:  Copyright (c) 2009-2011, David Simcha.
License:    $(WEB boost.org/LICENSE_1_0.txt, Boost License 1.0)
*/
module std.parallelism;

import core.atomic;
import core.cpuid;
import core.exception;
import core.memory;
import core.sync.condition;
import core.thread;

import std.algorithm;
import std.conv;
import std.exception;
import std.functional;
import std.math;
import std.range;
import std.traits;
import std.typecons;
import std.typetuple;

version(OSX)
{
    version = useSysctlbyname;
}
else version(FreeBSD)
{
    version = useSysctlbyname;
}

version(Windows)
{
    // BUGS:  Only works on Windows 2000 and above.

    import core.sys.windows.windows;

    struct SYSTEM_INFO
    {
        union
        {
            DWORD  dwOemId;
            struct
            {
                WORD wProcessorArchitecture;
                WORD wReserved;
            }
        }
        DWORD     dwPageSize;
        LPVOID    lpMinimumApplicationAddress;
        LPVOID    lpMaximumApplicationAddress;
        LPVOID    dwActiveProcessorMask;
        DWORD     dwNumberOfProcessors;
        DWORD     dwProcessorType;
        DWORD     dwAllocationGranularity;
        WORD      wProcessorLevel;
        WORD      wProcessorRevision;
    }

    private extern(Windows) void GetSystemInfo(void*);

    shared static this()
    {
        SYSTEM_INFO si;
        GetSystemInfo(&si);
        totalCPUs = max(1, cast(uint) si.dwNumberOfProcessors);
    }

}
else version(linux)
{
    import core.sys.posix.unistd;

    shared static this()
    {
        totalCPUs = cast(uint) sysconf(_SC_NPROCESSORS_ONLN);
    }
}
else version(Solaris)
{
    import core.sys.posix.unistd;

    shared static this()
    {
        totalCPUs = cast(uint) sysconf(_SC_NPROCESSORS_ONLN);
    }
}
else version(useSysctlbyname)
{
    extern(C) int sysctlbyname(
        const char *, void *, size_t *, void *, size_t
    );

    shared static this()
    {
        version(OSX)
        {
            auto nameStr = "machdep.cpu.core_count\0".ptr;
        }
        else version(FreeBSD)
        {
            auto nameStr = "hw.ncpu\0".ptr;
        }

        uint ans;
        size_t len = uint.sizeof;
        sysctlbyname(nameStr, &ans, &len, null, 0);
        totalCPUs = ans;
    }

}
else
{
    static assert(0, "Don't know how to get N CPUs on this OS.");
}

/* Atomics code.  These forward to core.atomic, but are written like this
   for two reasons:

   1.  They used to actually contain ASM code and I don' want to have to change
       to directly calling core.atomic in a zillion different places.

   2.  core.atomic has some misc. issues that make my use cases difficult
       without wrapping it.  If I didn't wrap it, casts would be required
       basically everywhere.
*/
private void atomicSetUbyte(T)(ref T stuff, T newVal)
if (__traits(isIntegral, T) && is(T : ubyte))
{
    //core.atomic.cas(cast(shared) &stuff, stuff, newVal);
    atomicStore(*(cast(shared) &stuff), newVal);
}

private ubyte atomicReadUbyte(T)(ref T val)
if (__traits(isIntegral, T) && is(T : ubyte))
{
    return atomicLoad(*(cast(shared) &val));
}

// This gets rid of the need for a lot of annoying casts in other parts of the
// code, when enums are involved.
private bool atomicCasUbyte(T)(ref T stuff, T testVal, T newVal)
if (__traits(isIntegral, T) && is(T : ubyte))
{
    return core.atomic.cas(cast(shared) &stuff, testVal, newVal);
}

/*--------------------- Generic helper functions, etc.------------------------*/
private template MapType(R, functions...)
{
    static assert(functions.length);

    ElementType!R e = void;
    alias MapType =
        typeof(adjoin!(staticMap!(unaryFun, functions))(e));
}

private template ReduceType(alias fun, R, E)
{
    alias ReduceType = typeof(binaryFun!fun(E.init, ElementType!R.init));
}

private template noUnsharedAliasing(T)
{
    enum bool noUnsharedAliasing = !hasUnsharedAliasing!T;
}

// This template tests whether a function may be executed in parallel from
// @safe code via Task.executeInNewThread().  There is an additional
// requirement for executing it via a TaskPool.  (See isSafeReturn).
private template isSafeTask(F)
{
    enum bool isSafeTask =
        (functionAttributes!F & (FunctionAttribute.safe | FunctionAttribute.trusted)) != 0 &&
        (functionAttributes!F & FunctionAttribute.ref_) == 0 &&
        (isFunctionPointer!F || !hasUnsharedAliasing!F) &&
        allSatisfy!(noUnsharedAliasing, ParameterTypeTuple!F);
}

unittest
{
    alias F1 = void function() @safe;
    alias F2 = void function();
    alias F3 = void function(uint, string) @trusted;
    alias F4 = void function(uint, char[]);

    static assert( isSafeTask!F1);
    static assert(!isSafeTask!F2);
    static assert( isSafeTask!F3);
    static assert(!isSafeTask!F4);

    alias F5 = uint[] function(uint, string) pure @trusted;
    static assert( isSafeTask!F5);
}

// This function decides whether Tasks that meet all of the other requirements
// for being executed from @safe code can be executed on a TaskPool.
// When executing via TaskPool, it's theoretically possible
// to return a value that is also pointed to by a worker thread's thread local
// storage.  When executing from executeInNewThread(), the thread that executed
// the Task is terminated by the time the return value is visible in the calling
// thread, so this is a non-issue.  It's also a non-issue for pure functions
// since they can't read global state.
private template isSafeReturn(T)
{
    static if(!hasUnsharedAliasing!(T.ReturnType))
    {
        enum isSafeReturn = true;
    }
    else static if(T.isPure)
    {
        enum isSafeReturn = true;
    }
    else
    {
        enum isSafeReturn = false;
    }
}

private template randAssignable(R)
{
    enum randAssignable = isRandomAccessRange!R && hasAssignableElements!R;
}

// Work around syntactic ambiguity w.r.t. address of function return vals.
private T* addressOf(T)(ref T val) pure nothrow
{
    return &val;
}

private enum TaskStatus : ubyte
{
    notStarted,
    inProgress,
    done
}

private template AliasReturn(alias fun, T...)
{
    alias AliasReturn = typeof({ T args; return fun(args); });
}

// Should be private, but std.algorithm.reduce is used in the zero-thread case
// and won't work w/ private.
template reduceAdjoin(functions...)
{
    static if(functions.length == 1)
    {
        alias reduceAdjoin = binaryFun!(functions[0]);
    }
    else
    {
        T reduceAdjoin(T, U)(T lhs, U rhs)
        {
            alias funs = staticMap!(binaryFun, functions);

            foreach(i, Unused; typeof(lhs.expand))
            {
                lhs.expand[i] = funs[i](lhs.expand[i], rhs);
            }

            return lhs;
        }
    }
}

private template reduceFinish(functions...)
{
    static if(functions.length == 1)
    {
        alias reduceFinish = binaryFun!(functions[0]);
    }
    else
    {
        T reduceFinish(T)(T lhs, T rhs)
        {
            alias funs = staticMap!(binaryFun, functions);

            foreach(i, Unused; typeof(lhs.expand))
            {
                lhs.expand[i] = funs[i](lhs.expand[i], rhs.expand[i]);
            }

            return lhs;
        }
    }
}

private template isRoundRobin(R : RoundRobinBuffer!(C1, C2), C1, C2)
{
    enum isRoundRobin = true;
}

private template isRoundRobin(T)
{
    enum isRoundRobin = false;
}

unittest
{
    static assert( isRoundRobin!(RoundRobinBuffer!(void delegate(char[]), bool delegate())));
    static assert(!isRoundRobin!(uint));
}

// This is the base "class" for all of the other tasks.  Using C-style
// polymorphism to allow more direct control over memory allocation, etc.
private struct AbstractTask
{
    AbstractTask* prev;
    AbstractTask* next;

    // Pointer to a function that executes this task.
    void function(void*) runTask;

    Throwable exception;
    ubyte taskStatus = TaskStatus.notStarted;

    bool done() @property
    {
        if(atomicReadUbyte(taskStatus) == TaskStatus.done)
        {
            if(exception)
            {
                throw exception;
            }

            return true;
        }

        return false;
    }

    void job()
    {
        runTask(&this);
    }
}

/**
$(D Task) represents the fundamental unit of work.  A $(D Task) may be
executed in parallel with any other $(D Task).  Using this struct directly
allows future/promise _parallelism.  In this paradigm, a function (or delegate
or other callable) is executed in a thread other than the one it was called
from.  The calling thread does not block while the function is being executed.
A call to $(D workForce), $(D yieldForce), or $(D spinForce) is used to
ensure that the $(D Task) has finished executing and to obtain the return
value, if any.  These functions and $(D done) also act as full memory barriers,
meaning that any memory writes made in the thread that executed the $(D Task)
are guaranteed to be visible in the calling thread after one of these functions
returns.

The $(XREF parallelism, task) and $(XREF parallelism, scopedTask) functions can
be used to create an instance of this struct.  See $(D task) for usage examples.

Function results are returned from $(D yieldForce), $(D spinForce) and
$(D workForce) by ref.  If $(D fun) returns by ref, the reference will point
to the returned reference of $(D fun).  Otherwise it will point to a
field in this struct.

Copying of this struct is disabled, since it would provide no useful semantics.
If you want to pass this struct around, you should do so by reference or
pointer.

Bugs:  Changes to $(D ref) and $(D out) arguments are not propagated to the
       call site, only to $(D args) in this struct.
*/
struct Task(alias fun, Args...)
{
    AbstractTask base = {runTask : &impl};
    alias base this;

    private @property AbstractTask* basePtr()
    {
        return &base;
    }

    private static void impl(void* myTask)
    {
        Task* myCastedTask = cast(typeof(this)*) myTask;
        static if(is(ReturnType == void))
        {
            fun(myCastedTask._args);
        }
        else static if(is(typeof(addressOf(fun(myCastedTask._args)))))
        {
            myCastedTask.returnVal = addressOf(fun(myCastedTask._args));
        }
        else
        {
            myCastedTask.returnVal = fun(myCastedTask._args);
        }
    }

    private TaskPool pool;
    private bool isScoped;  // True if created with scopedTask.

    Args _args;

    /**
    The arguments the function was called with.  Changes to $(D out) and
    $(D ref) arguments will be visible here.
    */
    static if(__traits(isSame, fun, run))
    {
        alias args = _args[1..$];
    }
    else
    {
        alias args = _args;
    }


    // The purpose of this code is to decide whether functions whose
    // return values have unshared aliasing can be executed via
    // TaskPool from @safe code.  See isSafeReturn.
    static if(__traits(isSame, fun, run))
    {
        static if(isFunctionPointer!(_args[0]))
        {
            private enum bool isPure =
            functionAttributes!(Args[0]) & FunctionAttribute.pure_;
        }
        else
        {
            // BUG:  Should check this for delegates too, but std.traits
            //       apparently doesn't allow this.  isPure is irrelevant
            //       for delegates, at least for now since shared delegates
            //       don't work.
            private enum bool isPure = false;
        }

    }
    else
    {
        // We already know that we can't execute aliases in @safe code, so
        // just put a dummy value here.
        private enum bool isPure = false;
    }


    /**
    The return type of the function called by this $(D Task).  This can be
    $(D void).
    */
    alias ReturnType = typeof(fun(_args));

    static if(!is(ReturnType == void))
    {
        static if(is(typeof(&fun(_args))))
        {
            // Ref return.
            ReturnType* returnVal;

            ref ReturnType fixRef(ReturnType* val)
            {
                return *val;
            }

        }
        else
        {
            ReturnType returnVal;

            ref ReturnType fixRef(ref ReturnType val)
            {
                return val;
            }
        }
    }

    private void enforcePool()
    {
        enforce(this.pool !is null, "Job not submitted yet.");
    }

    static if(Args.length > 0)
    {
        private this(Args args)
        {
            _args = args;
        }
    }

    // Work around DMD bug 6588, allow immutable elements.
    static if(allSatisfy!(isAssignable, Args))
    {
        typeof(this) opAssign(typeof(this) rhs)
        {
            foreach(i, Type; typeof(this.tupleof))
            {
                this.tupleof[i] = rhs.tupleof[i];
            }
            return this;
        }
    }
    else
    {
        @disable typeof(this) opAssign(typeof(this) rhs)
        {
            assert(0);
        }
    }

    /**
    If the $(D Task) isn't started yet, execute it in the current thread.
    If it's done, return its return value, if any.  If it's in progress,
    busy spin until it's done, then return the return value.  If it threw
    an exception, rethrow that exception.

    This function should be used when you expect the result of the
    $(D Task) to be available on a timescale shorter than that of an OS
    context switch.
     */
    @property ref ReturnType spinForce() @trusted
    {
        enforcePool();

        this.pool.tryDeleteExecute(basePtr);

        while(atomicReadUbyte(this.taskStatus) != TaskStatus.done) {}

        if(exception)
        {
            throw exception;
        }

        static if(!is(ReturnType == void))
        {
            return fixRef(this.returnVal);
        }
    }

    /**
    If the $(D Task) isn't started yet, execute it in the current thread.
    If it's done, return its return value, if any.  If it's in progress,
    wait on a condition variable.  If it threw an exception, rethrow that
    exception.

    This function should be used for expensive functions, as waiting on a
    condition variable introduces latency, but avoids wasted CPU cycles.
     */
    @property ref ReturnType yieldForce() @trusted
    {
        enforcePool();
        this.pool.tryDeleteExecute(basePtr);

        if(done)
        {
            static if(is(ReturnType == void))
            {
                return;
            }
            else
            {
                return fixRef(this.returnVal);
            }
        }

        pool.waiterLock();
        scope(exit) pool.waiterUnlock();

        while(atomicReadUbyte(this.taskStatus) != TaskStatus.done)
        {
            pool.waitUntilCompletion();
        }

        if(exception)
        {
            throw exception;
        }

        static if(!is(ReturnType == void))
        {
            return fixRef(this.returnVal);
        }
    }

    /**
    If this $(D Task) was not started yet, execute it in the current
    thread.  If it is finished, return its result.  If it is in progress,
    execute any other $(D Task) from the $(D TaskPool) instance that
    this $(D Task) was submitted to until this one
    is finished.  If it threw an exception, rethrow that exception.
    If no other tasks are available or this $(D Task) was executed using
    $(D executeInNewThread), wait on a condition variable.
     */
    @property ref ReturnType workForce() @trusted
    {
        enforcePool();
        this.pool.tryDeleteExecute(basePtr);

        while(true)
        {
            if(done)    // done() implicitly checks for exceptions.
            {
                static if(is(ReturnType == void))
                {
                    return;
                }
                else
                {
                    return fixRef(this.returnVal);
                }
            }

            AbstractTask* job;
            {
                // Locking explicitly and calling popNoSync() because
                // pop() waits on a condition variable if there are no Tasks
                // in the queue.

                pool.queueLock();
                scope(exit) pool.queueUnlock();
                job = pool.popNoSync();
            }


            if(job !is null)
            {

                version(verboseUnittest)
                {
                    stderr.writeln("Doing workForce work.");
                }

                pool.doJob(job);

                if(done)
                {
                    static if(is(ReturnType == void))
                    {
                        return;
                    }
                    else
                    {
                        return fixRef(this.returnVal);
                    }
                }
            }
            else
            {
                version(verboseUnittest)
                {
                    stderr.writeln("Yield from workForce.");
                }

                return yieldForce;
            }
        }
    }

    /**
    Returns $(D true) if the $(D Task) is finished executing.

    Throws:  Rethrows any exception thrown during the execution of the
             $(D Task).
    */
    @property bool done() @trusted
    {
        // Explicitly forwarded for documentation purposes.
        return base.done;
    }

    /**
    Create a new thread for executing this $(D Task), execute it in the
    newly created thread, then terminate the thread.  This can be used for
    future/promise parallelism.  An explicit priority may be given
    to the $(D Task).  If one is provided, its value is forwarded to
    $(D core.thread.Thread.priority). See $(XREF parallelism, task) for
    usage example.
    */
    void executeInNewThread() @trusted
    {
        pool = new TaskPool(basePtr);
    }

    /// Ditto
    void executeInNewThread(int priority) @trusted
    {
        pool = new TaskPool(basePtr, priority);
    }

    @safe ~this()
    {
        if(isScoped && pool !is null && taskStatus != TaskStatus.done)
        {
            yieldForce;
        }
    }

    // When this is uncommented, it somehow gets called on returning from
    // scopedTask even though the struct shouldn't be getting copied.
    //@disable this(this) {}
}

// Calls $(D fpOrDelegate) with $(D args).  This is an
// adapter that makes $(D Task) work with delegates, function pointers and
// functors instead of just aliases.
ReturnType!F run(F, Args...)(F fpOrDelegate, ref Args args)
{
    return fpOrDelegate(args);
}

/**
Creates a $(D Task) on the GC heap that calls an alias.  This may be executed
via $(D Task.executeInNewThread) or by submitting to a
$(XREF parallelism, TaskPool).  A globally accessible instance of
$(D TaskPool) is provided by $(XREF parallelism, taskPool).

Returns:  A pointer to the $(D Task).

Examples:
---
// Read two files into memory at the same time.
import std.file;

void main()
{
    // Create and execute a Task for reading
    // foo.txt.
    auto file1Task = task!read("foo.txt");
    file1Task.executeInNewThread();

    // Read bar.txt in parallel.
    auto file2Data = read("bar.txt");

    // Get the results of reading foo.txt.
    auto file1Data = file1Task.yieldForce;
}
---

---
// Sorts an array using a parallel quick sort algorithm.
// The first partition is done serially.  Both recursion
// branches are then executed in parallel.
//
// Timings for sorting an array of 1,000,000 doubles on
// an Athlon 64 X2 dual core machine:
//
// This implementation:               176 milliseconds.
// Equivalent serial implementation:  280 milliseconds
void parallelSort(T)(T[] data)
{
    // Sort small subarrays serially.
    if(data.length < 100)
    {
         std.algorithm.sort(data);
         return;
    }

    // Partition the array.
    swap(data[$ / 2], data[$ - 1]);
    auto pivot = data[$ - 1];
    bool lessThanPivot(T elem) { return elem < pivot; }

    auto greaterEqual = partition!lessThanPivot(data[0..$ - 1]);
    swap(data[$ - greaterEqual.length - 1], data[$ - 1]);

    auto less = data[0..$ - greaterEqual.length - 1];
    greaterEqual = data[$ - greaterEqual.length..$];

    // Execute both recursion branches in parallel.
    auto recurseTask = task!parallelSort(greaterEqual);
    taskPool.put(recurseTask);
    parallelSort(less);
    recurseTask.yieldForce;
}
---
*/
auto task(alias fun, Args...)(Args args)
{
    return new Task!(fun, Args)(args);
}

/**
Creates a $(D Task) on the GC heap that calls a function pointer, delegate, or
class/struct with overloaded opCall.

Examples:
---
// Read two files in at the same time again,
// but this time use a function pointer instead
// of an alias to represent std.file.read.
import std.file;

void main()
{
    // Create and execute a Task for reading
    // foo.txt.
    auto file1Task = task(&read, "foo.txt");
    file1Task.executeInNewThread();

    // Read bar.txt in parallel.
    auto file2Data = read("bar.txt");

    // Get the results of reading foo.txt.
    auto file1Data = file1Task.yieldForce;
}
---

Notes: This function takes a non-scope delegate, meaning it can be
       used with closures.  If you can't allocate a closure due to objects
       on the stack that have scoped destruction, see $(D scopedTask), which
       takes a scope delegate.
 */
auto task(F, Args...)(F delegateOrFp, Args args)
if(is(typeof(delegateOrFp(args))) && !isSafeTask!F)
{
    return new Task!(run, F, Args)(delegateOrFp, args);
}

/**
Version of $(D task) usable from $(D @safe) code.  Usage mechanics are
identical to the non-@safe case, but safety introduces some restrictions:

1.  $(D fun) must be @safe or @trusted.

2.  $(D F) must not have any unshared aliasing as defined by
    $(XREF traits, hasUnsharedAliasing).  This means it
    may not be an unshared delegate or a non-shared class or struct
    with overloaded $(D opCall).  This also precludes accepting template
    alias parameters.

3.  $(D Args) must not have unshared aliasing.

4.  $(D fun) must not return by reference.

5.  The return type must not have unshared aliasing unless $(D fun) is
    $(D pure) or the $(D Task) is executed via $(D executeInNewThread) instead
    of using a $(D TaskPool).

*/
@trusted auto task(F, Args...)(F fun, Args args)
if(is(typeof(fun(args))) && isSafeTask!F)
{
    return new Task!(run, F, Args)(fun, args);
}

/**
These functions allow the creation of $(D Task) objects on the stack rather
than the GC heap.  The lifetime of a $(D Task) created by $(D scopedTask)
cannot exceed the lifetime of the scope it was created in.

$(D scopedTask) might be preferred over $(D task):

1.  When a $(D Task) that calls a delegate is being created and a closure
    cannot be allocated due to objects on the stack that have scoped
    destruction.  The delegate overload of $(D scopedTask) takes a $(D scope)
    delegate.

2.  As a micro-optimization, to avoid the heap allocation associated with
    $(D task) or with the creation of a closure.

Usage is otherwise identical to $(D task).

Notes:  $(D Task) objects created using $(D scopedTask) will automatically
call $(D Task.yieldForce) in their destructor if necessary to ensure
the $(D Task) is complete before the stack frame they reside on is destroyed.
*/
auto scopedTask(alias fun, Args...)(Args args)
{
    auto ret = Task!(fun, Args)(args);
    ret.isScoped = true;
    return ret;
}

/// Ditto
auto scopedTask(F, Args...)(scope F delegateOrFp, Args args)
if(is(typeof(delegateOrFp(args))) && !isSafeTask!F)
{
    auto ret = Task!(run, F, Args)(delegateOrFp, args);
    ret.isScoped = true;
    return ret;
}

/// Ditto
@trusted auto scopedTask(F, Args...)(F fun, Args args)
if(is(typeof(fun(args))) && isSafeTask!F)
{
    auto ret = Task!(run, F, Args)(fun, args);
    ret.isScoped = true;
    return ret;
}

/**
The total number of CPU cores available on the current machine, as reported by
the operating system.
*/
immutable uint totalCPUs;

/*
This class serves two purposes:

1.  It distinguishes std.parallelism threads from other threads so that
    the std.parallelism daemon threads can be terminated.

2.  It adds a reference to the pool that the thread is a member of,
    which is also necessary to allow the daemon threads to be properly
    terminated.
*/
private final class ParallelismThread : Thread
{
    this(void delegate() dg)
    {
        super(dg);
    }

    TaskPool pool;
}

// Kill daemon threads.
shared static ~this()
{
    auto allThreads = Thread.getAll();

    foreach(thread; allThreads)
    {
        auto pthread = cast(ParallelismThread) thread;
        if(pthread is null) continue;
        auto pool = pthread.pool;
        if(!pool.isDaemon) continue;
        pool.stop();
        pthread.join();
    }
}

/**
This class encapsulates a task queue and a set of worker threads.  Its purpose
is to efficiently map a large number of $(D Task)s onto a smaller number of
threads.  A task queue is a FIFO queue of $(D Task) objects that have been
submitted to the $(D TaskPool) and are awaiting execution.  A worker thread is a
thread that executes the $(D Task) at the front of the queue when one is
available and sleeps when the queue is empty.

This class should usually be used via the global instantiation
available via the $(XREF parallelism, taskPool) property.
Occasionally it is useful to explicitly instantiate a $(D TaskPool):

1.  When you want $(D TaskPool) instances with multiple priorities, for example
    a low priority pool and a high priority pool.

2.  When the threads in the global task pool are waiting on a synchronization
    primitive (for example a mutex), and you want to parallelize the code that
    needs to run before these threads can be resumed.
 */
final class TaskPool
{
private:

    // A pool can either be a regular pool or a single-task pool.  A
    // single-task pool is a dummy pool that's fired up for
    // Task.executeInNewThread().
    bool isSingleTask;

    ParallelismThread[] pool;
    Thread singleTaskThread;

    AbstractTask* head;
    AbstractTask* tail;
    PoolState status = PoolState.running;
    Condition workerCondition;
    Condition waiterCondition;
    Mutex queueMutex;
    Mutex waiterMutex;  // For waiterCondition

    // The instanceStartIndex of the next instance that will be created.
    __gshared static size_t nextInstanceIndex = 1;

    // The index of the current thread.
    static size_t threadIndex;

    // The index of the first thread in this instance.
    immutable size_t instanceStartIndex;

    // The index that the next thread to be initialized in this pool will have.
    size_t nextThreadIndex;

    enum PoolState : ubyte
    {
        running,
        finishing,
        stopNow
    }

    void doJob(AbstractTask* job)
    {
        assert(job.taskStatus == TaskStatus.inProgress);
        assert(job.next is null);
        assert(job.prev is null);

        scope(exit)
        {
            if(!isSingleTask)
            {
                waiterLock();
                scope(exit) waiterUnlock();
                notifyWaiters();
            }
        }

        try
        {
            job.job();
        }
        catch(Throwable e)
        {
            job.exception = e;
        }

        atomicSetUbyte(job.taskStatus, TaskStatus.done);
    }

    // This function is used for dummy pools created by Task.executeInNewThread().
    void doSingleTask()
    {
        // No synchronization.  Pool is guaranteed to only have one thread,
        // and the queue is submitted to before this thread is created.
        assert(head);
        auto t = head;
        t.next = t.prev = head = null;
        doJob(t);
    }

    // This function performs initialization for each thread that affects
    // thread local storage and therefore must be done from within the
    // worker thread.  It then calls executeWorkLoop().
    void startWorkLoop()
    {
        // Initialize thread index.
        {
            queueLock();
            scope(exit) queueUnlock();
            threadIndex = nextThreadIndex;
            nextThreadIndex++;
        }

        executeWorkLoop();
    }

    // This is the main work loop that worker threads spend their time in
    // until they terminate.  It's also entered by non-worker threads when
    // finish() is called with the blocking variable set to true.
    void executeWorkLoop()
    {
        while(atomicReadUbyte(status) != PoolState.stopNow)
        {
            AbstractTask* task = pop();
            if (task is null)
            {
                if(atomicReadUbyte(status) == PoolState.finishing)
                {
                    atomicSetUbyte(status, PoolState.stopNow);
                    return;
                }
            }
            else
            {
                doJob(task);
            }
        }
    }

    // Pop a task off the queue.
    AbstractTask* pop()
    {
        queueLock();
        scope(exit) queueUnlock();
        auto ret = popNoSync();
        while(ret is null && status == PoolState.running)
        {
            wait();
            ret = popNoSync();
        }
        return ret;
    }

    AbstractTask* popNoSync()
    out(returned)
    {
        /* If task.prev and task.next aren't null, then another thread
         * can try to delete this task from the pool after it's
         * alreadly been deleted/popped.
         */
        if(returned !is null)
        {
            assert(returned.next is null);
            assert(returned.prev is null);
        }
    }
    body
    {
        if(isSingleTask) return null;

        AbstractTask* returned = head;
        if (head !is null)
        {
            head = head.next;
            returned.prev = null;
            returned.next = null;
            returned.taskStatus = TaskStatus.inProgress;
        }
        if(head !is null)
        {
            head.prev = null;
        }

        return returned;
    }

    // Push a task onto the queue.
    void abstractPut(AbstractTask* task)
    {
        queueLock();
        scope(exit) queueUnlock();
        abstractPutNoSync(task);
    }

    void abstractPutNoSync(AbstractTask* task)
    in
    {
        assert(task);
    }
    out
    {
        assert(tail.prev !is tail);
        assert(tail.next is null, text(tail.prev, '\t', tail.next));
        if(tail.prev !is null)
        {
            assert(tail.prev.next is tail, text(tail.prev, '\t', tail.next));
        }
    }
    body
    {
        // Not using enforce() to save on function call overhead since this
        // is a performance critical function.
        if(status != PoolState.running)
        {
            throw new Error(
                "Cannot submit a new task to a pool after calling " ~
                "finish() or stop()."
            );
        }

        task.next = null;
        if (head is null)   //Queue is empty.
        {
            head = task;
            tail = task;
            tail.prev = null;
        }
        else {
            assert(tail);
            task.prev = tail;
            tail.next = task;
            tail = task;
        }
        notify();
    }

    void abstractPutGroupNoSync(AbstractTask* h, AbstractTask* t)
    {
        if(status != PoolState.running)
        {
            throw new Error(
                "Cannot submit a new task to a pool after calling " ~
                "finish() or stop()."
            );
        }

        if(head is null)
        {
            head = h;
            tail = t;
        }
        else
        {
            h.prev = tail;
            tail.next = h;
            tail = t;
        }

        notifyAll();
    }

    void tryDeleteExecute(AbstractTask* toExecute)
    {
        if(isSingleTask) return;

        if( !deleteItem(toExecute) )
        {
            return;
        }

        try
        {
            toExecute.job();
        }
        catch(Exception e)
        {
            toExecute.exception = e;
        }

        atomicSetUbyte(toExecute.taskStatus, TaskStatus.done);
    }

    bool deleteItem(AbstractTask* item)
    {
        queueLock();
        scope(exit) queueUnlock();
        return deleteItemNoSync(item);
    }

    bool deleteItemNoSync(AbstractTask* item)
    {
        if(item.taskStatus != TaskStatus.notStarted)
        {
            return false;
        }
        item.taskStatus = TaskStatus.inProgress;

        if(item is head)
        {
            // Make sure head gets set properly.
            popNoSync();
            return true;
        }
        if(item is tail)
        {
            tail = tail.prev;
            if(tail !is null)
            {
                tail.next = null;
            }
            item.next = null;
            item.prev = null;
            return true;
        }
        if(item.next !is null)
        {
            assert(item.next.prev is item);  // Check queue consistency.
            item.next.prev = item.prev;
        }
        if(item.prev !is null)
        {
            assert(item.prev.next is item);  // Check queue consistency.
            item.prev.next = item.next;
        }
        item.next = null;
        item.prev = null;
        return true;
    }

    void queueLock()
    {
        assert(queueMutex);
        if(!isSingleTask) queueMutex.lock();
    }

    void queueUnlock()
    {
        assert(queueMutex);
        if(!isSingleTask) queueMutex.unlock();
    }

    void waiterLock()
    {
        if(!isSingleTask) waiterMutex.lock();
    }

    void waiterUnlock()
    {
        if(!isSingleTask) waiterMutex.unlock();
    }

    void wait()
    {
        if(!isSingleTask) workerCondition.wait();
    }

    void notify()
    {
        if(!isSingleTask) workerCondition.notify();
    }

    void notifyAll()
    {
        if(!isSingleTask) workerCondition.notifyAll();
    }

    void waitUntilCompletion()
    {
        if(isSingleTask)
        {
            singleTaskThread.join();
        }
        else
        {
            waiterCondition.wait();
        }
    }

    void notifyWaiters()
    {
        if(!isSingleTask) waiterCondition.notifyAll();
    }

    // Private constructor for creating dummy pools that only have one thread,
    // only execute one Task, and then terminate.  This is used for
    // Task.executeInNewThread().
    this(AbstractTask* task, int priority = int.max)
    {
        assert(task);

        // Dummy value, not used.
        instanceStartIndex = 0;

        this.isSingleTask = true;
        task.taskStatus = TaskStatus.inProgress;
        this.head = task;
        singleTaskThread = new Thread(&doSingleTask);
        singleTaskThread.start();

        // Disabled until writing code to support
        // running thread with specified priority
        // See https://d.puremagic.com/issues/show_bug.cgi?id=8960

        /*if(priority != int.max)
        {
            singleTaskThread.priority = priority;
        }*/
    }

public:
    // This is used in parallel_algorithm but is too unstable to document
    // as public API.
    size_t defaultWorkUnitSize(size_t rangeLen) const @safe pure nothrow
    {
        if(this.size == 0)
        {
            return rangeLen;
        }

        immutable size_t eightSize = 4 * (this.size + 1);
        auto ret = (rangeLen / eightSize) + ((rangeLen % eightSize == 0) ? 0 : 1);
        return max(ret, 1);
    }

    /**
    Default constructor that initializes a $(D TaskPool) with
    $(D totalCPUs) - 1 worker threads.  The minus 1 is included because the
    main thread will also be available to do work.

    Note:  On single-core machines, the primitives provided by $(D TaskPool)
           operate transparently in single-threaded mode.
     */
    this() @trusted
    {
        this(totalCPUs - 1);
    }

    /**
    Allows for custom number of worker threads.
    */
    this(size_t nWorkers) @trusted
    {
        synchronized(typeid(TaskPool))
        {
            instanceStartIndex = nextInstanceIndex;

            // The first worker thread to be initialized will have this index,
            // and will increment it.  The second worker to be initialized will
            // have this index plus 1.
            nextThreadIndex = instanceStartIndex;
            nextInstanceIndex += nWorkers;
        }

        queueMutex = new Mutex(this);
        waiterMutex = new Mutex();
        workerCondition = new Condition(queueMutex);
        waiterCondition = new Condition(waiterMutex);

        pool = new ParallelismThread[nWorkers];
        foreach(ref poolThread; pool)
        {
            poolThread = new ParallelismThread(&startWorkLoop);
            poolThread.pool = this;
            poolThread.start();
        }
    }

    /**
    Implements a parallel foreach loop over a range.  This works by implicitly
    creating and submitting one $(D Task) to the $(D TaskPool) for each worker
    thread.  A work unit is a set of consecutive elements of $(D range) to
    be processed by a worker thread between communication with any other
    thread.  The number of elements processed per work unit is controlled by the
    $(D workUnitSize) parameter.  Smaller work units provide better load
    balancing, but larger work units avoid the overhead of communicating
    with other threads frequently to fetch the next work unit.  Large work
    units also avoid false sharing in cases where the range is being modified.
    The less time a single iteration of the loop takes, the larger
    $(D workUnitSize) should be.  For very expensive loop bodies,
    $(D workUnitSize) should  be 1.  An overload that chooses a default work
    unit size is also available.

    Examples:
    ---
    // Find the logarithm of every number from 1 to
    // 10_000_000 in parallel.
    auto logs = new double[10_000_000];

    // Parallel foreach works with or without an index
    // variable.  It can be iterate by ref if range.front
    // returns by ref.

    // Iterate over logs using work units of size 100.
    foreach(i, ref elem; taskPool.parallel(logs, 100))
    {
        elem = log(i + 1.0);
    }

    // Same thing, but use the default work unit size.
    //
    // Timings on an Athlon 64 X2 dual core machine:
    //
    // Parallel foreach:  388 milliseconds
    // Regular foreach:   619 milliseconds
    foreach(i, ref elem; taskPool.parallel(logs))
    {
        elem = log(i + 1.0);
    }
    ---

    Notes:

    The memory usage of this implementation is guaranteed to be constant
    in $(D range.length).

    Breaking from a parallel foreach loop via a break, labeled break,
    labeled continue, return or goto statement throws a
    $(D ParallelForeachError).

    In the case of non-random access ranges, parallel foreach buffers lazily
    to an array of size $(D workUnitSize) before executing the parallel portion
    of the loop.  The exception is that, if a parallel foreach is executed
    over a range returned by $(D asyncBuf) or $(D map), the copying is elided
    and the buffers are simply swapped.  In this case $(D workUnitSize) is
    ignored and the work unit size is set to the  buffer size of $(D range).

    A memory barrier is guaranteed to be executed on exit from the loop,
    so that results produced by all threads are visible in the calling thread.

    $(B Exception Handling):

    When at least one exception is thrown from inside a parallel foreach loop,
    the submission of additional $(D Task) objects is terminated as soon as
    possible, in a non-deterministic manner.  All executing or
    enqueued work units are allowed to complete.  Then, all exceptions that
    were thrown by any work unit are chained using $(D Throwable.next) and
    rethrown.  The order of the exception chaining is non-deterministic.
    */
    ParallelForeach!R parallel(R)(R range, size_t workUnitSize)
    {
        enforce(workUnitSize > 0, "workUnitSize must be > 0.");
        alias RetType = ParallelForeach!R;
        return RetType(this, range, workUnitSize);
    }


    /// Ditto
    ParallelForeach!R parallel(R)(R range)
    {
        static if(hasLength!R)
        {
            // Default work unit size is such that we would use 4x as many
            // slots as are in this thread pool.
            size_t workUnitSize = defaultWorkUnitSize(range.length);
            return parallel(range, workUnitSize);
        }
        else
        {
            // Just use a really, really dumb guess if the user is too lazy to
            // specify.
            return parallel(range, 512);
        }
    }

    ///
    template amap(functions...)
    {
        /**
        Eager parallel map.  The eagerness of this function means it has less
        overhead than the lazily evaluated $(D TaskPool.map) and should be
        preferred where the memory requirements of eagerness are acceptable.
        $(D functions) are the functions to be evaluated, passed as template
        alias parameters in a style similar to
        $(XREF_PACK algorithm,iteration,map).
        The first argument must be a random access range. For performance
        reasons, amap will assume the range elements have not yet been
        initialized. Elements will be overwritten without calling a destructor
        nor doing an assignment. As such, the range must not contain meaningful
        data: either un-initialized objects, or objects in their $(D .init)
        state.

        ---
        auto numbers = iota(100_000_000.0);

        // Find the square roots of numbers.
        //
        // Timings on an Athlon 64 X2 dual core machine:
        //
        // Parallel eager map:                   0.802 s
        // Equivalent serial implementation:     1.768 s
        auto squareRoots = taskPool.amap!sqrt(numbers);
        ---

        Immediately after the range argument, an optional work unit size argument
        may be provided.  Work units as used by $(D amap) are identical to those
        defined for parallel foreach.  If no work unit size is provided, the
        default work unit size is used.

        ---
        // Same thing, but make work unit size 100.
        auto squareRoots = taskPool.amap!sqrt(numbers, 100);
        ---

        An output range for returning the results may be provided as the last
        argument.  If one is not provided, an array of the proper type will be
        allocated on the garbage collected heap.  If one is provided, it must be a
        random access range with assignable elements, must have reference
        semantics with respect to assignment to its elements, and must have the
        same length as the input range.  Writing to adjacent elements from
        different threads must be safe.

        ---
        // Same thing, but explicitly allocate an array
        // to return the results in.  The element type
        // of the array may be either the exact type
        // returned by functions or an implicit conversion
        // target.
        auto squareRoots = new float[numbers.length];
        taskPool.amap!sqrt(numbers, squareRoots);

        // Multiple functions, explicit output range, and
        // explicit work unit size.
        auto results = new Tuple!(float, real)[numbers.length];
        taskPool.amap!(sqrt, log)(numbers, 100, results);
        ---

        Note:

        A memory barrier is guaranteed to be executed after all results are written
        but before returning so that results produced by all threads are visible
        in the calling thread.

        Tips:

        To perform the mapping operation in place, provide the same range for the
        input and output range.

        To parallelize the copying of a range with expensive to evaluate elements
        to an array, pass an identity function (a function that just returns
        whatever argument is provided to it) to $(D amap).

        $(B Exception Handling):

        When at least one exception is thrown from inside the map functions,
        the submission of additional $(D Task) objects is terminated as soon as
        possible, in a non-deterministic manner.  All currently executing or
        enqueued work units are allowed to complete.  Then, all exceptions that
        were thrown from any work unit are chained using $(D Throwable.next) and
        rethrown.  The order of the exception chaining is non-deterministic.
        */
        auto amap(Args...)(Args args)
        if(isRandomAccessRange!(Args[0]))
        {
            alias fun = adjoin!(staticMap!(unaryFun, functions));

            alias range = args[0];
            immutable len = range.length;

            static if(
                Args.length > 1 &&
                randAssignable!(Args[$ - 1]) &&
                is(MapType!(Args[0], functions) : ElementType!(Args[$ - 1]))
                )
            {
                alias buf = args[$ - 1];
                alias args2 = args[0..$ - 1];
                alias Args2 = Args[0..$ - 1];
                enforce(buf.length == len,
                        text("Can't use a user supplied buffer that's the wrong ",
                             "size.  (Expected  :", len, " Got:  ", buf.length));
            }
            else static if(randAssignable!(Args[$ - 1]) && Args.length > 1)
            {
                static assert(0, "Wrong buffer type.");
            }
            else
            {
                auto buf = uninitializedArray!(MapType!(Args[0], functions)[])(len);
                alias args2 = args;
                alias Args2 = Args;
            }

            if(!len) return buf;

            static if(isIntegral!(Args2[$ - 1]))
            {
                static assert(args2.length == 2);
                auto workUnitSize = cast(size_t) args2[1];
            }
            else
            {
                static assert(args2.length == 1, Args);
                auto workUnitSize = defaultWorkUnitSize(range.length);
            }

            alias R = typeof(range);

            if(workUnitSize > len)
            {
                workUnitSize = len;
            }

            // Handle as a special case:
            if(size == 0)
            {
                size_t index = 0;
                foreach(elem; range)
                {
                    emplaceRef(buf[index++], fun(elem));
                }
                return buf;
            }

            // Effectively -1:  chunkIndex + 1 == 0:
            shared size_t workUnitIndex = size_t.max;
            shared bool shouldContinue = true;

            void doIt()
            {
                scope(failure)
                {
                    // If an exception is thrown, all threads should bail.
                    atomicStore(shouldContinue, false);
                }

                while(atomicLoad(shouldContinue))
                {
                    immutable myUnitIndex = atomicOp!"+="(workUnitIndex, 1);
                    immutable start = workUnitSize * myUnitIndex;
                    if(start >= len)
                    {
                        atomicStore(shouldContinue, false);
                        break;
                    }

                    immutable end = min(len, start + workUnitSize);

                    static if (hasSlicing!R)
                    {
                        auto subrange = range[start..end];
                        foreach(i; start..end)
                        {
                            emplaceRef(buf[i], fun(subrange.front));
                            subrange.popFront();
                        }
                    }
                    else
                    {
                        foreach(i; start..end)
                        {
                            emplaceRef(buf[i], fun(range[i]));
                        }
                    }
                }
            }

            submitAndExecute(this, &doIt);
            return buf;
        }
    }

    ///
    template map(functions...)
    {
        /**
        A semi-lazy parallel map that can be used for pipelining.  The map
        functions are evaluated for the first $(D bufSize) elements and stored in a
        buffer and made available to $(D popFront).  Meanwhile, in the
        background a second buffer of the same size is filled.  When the first
        buffer is exhausted, it is swapped with the second buffer and filled while
        the values from what was originally the second buffer are read.  This
        implementation allows for elements to be written to the buffer without
        the need for atomic operations or synchronization for each write, and
        enables the mapping function to be evaluated efficiently in parallel.

        $(D map) has more overhead than the simpler procedure used by $(D amap)
        but avoids the need to keep all results in memory simultaneously and works
        with non-random access ranges.

        Params:

        source = The input range to be mapped.  If $(D source) is not random
        access it will be lazily buffered to an array of size $(D bufSize) before
        the map function is evaluated.  (For an exception to this rule, see Notes.)

        bufSize = The size of the buffer to store the evaluated elements.

        workUnitSize = The number of elements to evaluate in a single
        $(D Task).  Must be less than or equal to $(D bufSize), and
        should be a fraction of $(D bufSize) such that all worker threads can be
        used.  If the default of size_t.max is used, workUnitSize will be set to
        the pool-wide default.

        Returns:  An input range representing the results of the map.  This range
                  has a length iff $(D source) has a length.

        Notes:

        If a range returned by $(D map) or $(D asyncBuf) is used as an input to
        $(D map), then as an optimization the copying from the output buffer
        of the first range to the input buffer of the second range is elided, even
        though the ranges returned by $(D map) and $(D asyncBuf) are non-random
        access ranges.  This means that the $(D bufSize) parameter passed to the
        current call to $(D map) will be ignored and the size of the buffer
        will be the buffer size of $(D source).

        Examples:
        ---
        // Pipeline reading a file, converting each line
        // to a number, taking the logarithms of the numbers,
        // and performing the additions necessary to find
        // the sum of the logarithms.

        auto lineRange = File("numberList.txt").byLine();
        auto dupedLines = std.algorithm.map!"a.idup"(lineRange);
        auto nums = taskPool.map!(to!double)(dupedLines);
        auto logs = taskPool.map!log10(nums);

        double sum = 0;
        foreach(elem; logs)
        {
            sum += elem;
        }
        ---

        $(B Exception Handling):

        Any exceptions thrown while iterating over $(D source)
        or computing the map function are re-thrown on a call to $(D popFront) or,
        if thrown during construction, are simply allowed to propagate to the
        caller.  In the case of exceptions thrown while computing the map function,
        the exceptions are chained as in $(D TaskPool.amap).
        */
        auto
        map(S)(S source, size_t bufSize = 100, size_t workUnitSize = size_t.max)
        if(isInputRange!S)
        {
            enforce(workUnitSize == size_t.max || workUnitSize <= bufSize,
                    "Work unit size must be smaller than buffer size.");
            alias fun = adjoin!(staticMap!(unaryFun, functions));

            static final class Map
            {
                // This is a class because the task needs to be located on the
                // heap and in the non-random access case source needs to be on
                // the heap, too.

            private:
                enum bufferTrick = is(typeof(source.buf1)) &&
                is(typeof(source.bufPos)) &&
                is(typeof(source.doBufSwap()));

                alias E = MapType!(S, functions);
                E[] buf1, buf2;
                S source;
                TaskPool pool;
                Task!(run, E[] delegate(E[]), E[]) nextBufTask;
                size_t workUnitSize;
                size_t bufPos;
                bool lastTaskWaited;

            static if(isRandomAccessRange!S)
            {
                alias FromType = S;

                void popSource()
                {
                    static if(__traits(compiles, source[0..source.length]))
                    {
                        source = source[min(buf1.length, source.length)..source.length];
                    }
                    else static if(__traits(compiles, source[0..$]))
                    {
                        source = source[min(buf1.length, source.length)..$];
                    }
                    else
                    {
                        static assert(0, "S must have slicing for Map."
                                      ~ "  " ~ S.stringof ~ " doesn't.");
                    }
                }
            }
            else static if(bufferTrick)
            {
                // Make sure we don't have the buffer recycling overload of
                // asyncBuf.
                static if(
                    is(typeof(source.source)) &&
                    isRoundRobin!(typeof(source.source))
                )
                {
                    static assert(0, "Cannot execute a parallel map on " ~
                                  "the buffer recycling overload of asyncBuf."
                                 );
                }

                alias FromType = typeof(source.buf1);
                FromType from;

                // Just swap our input buffer with source's output buffer.
                // No need to copy element by element.
                FromType dumpToFrom()
                {
                    assert(source.buf1.length <= from.length);
                    from.length = source.buf1.length;
                    swap(source.buf1, from);

                    // Just in case this source has been popped before
                    // being sent to map:
                    from = from[source.bufPos..$];

                    static if(is(typeof(source._length)))
                    {
                        source._length -= (from.length - source.bufPos);
                    }

                    source.doBufSwap();

                    return from;
                }
            }
            else
            {
                alias FromType = ElementType!S[];

                // The temporary array that data is copied to before being
                // mapped.
                FromType from;

                FromType dumpToFrom()
                {
                    assert(from !is null);

                    size_t i;
                    for(; !source.empty && i < from.length; source.popFront())
                    {
                        from[i++] = source.front;
                    }

                    from = from[0..i];
                    return from;
                }
            }

            static if(hasLength!S)
            {
                size_t _length;

                public @property size_t length() const @safe pure nothrow
                {
                    return _length;
                }
            }

                this(S source, size_t bufSize, size_t workUnitSize, TaskPool pool)
                {
                    static if(bufferTrick)
                    {
                        bufSize = source.buf1.length;
                    }

                    buf1.length = bufSize;
                    buf2.length = bufSize;

                    static if(!isRandomAccessRange!S)
                    {
                        from.length = bufSize;
                    }

                    this.workUnitSize = (workUnitSize == size_t.max) ?
                            pool.defaultWorkUnitSize(bufSize) : workUnitSize;
                    this.source = source;
                    this.pool = pool;

                    static if(hasLength!S)
                    {
                        _length = source.length;
                    }

                    buf1 = fillBuf(buf1);
                    submitBuf2();
                }

                // The from parameter is a dummy and ignored in the random access
                // case.
                E[] fillBuf(E[] buf)
                {
                    static if(isRandomAccessRange!S)
                    {
                        auto toMap = take(source, buf.length);
                        scope(success) popSource();
                    }
                    else
                    {
                        auto toMap = dumpToFrom();
                    }

                    buf = buf[0..min(buf.length, toMap.length)];

                    // Handle as a special case:
                    if(pool.size == 0)
                    {
                        size_t index = 0;
                        foreach(elem; toMap)
                        {
                            buf[index++] = fun(elem);
                        }
                        return buf;
                    }

                    pool.amap!functions(toMap, workUnitSize, buf);

                    return buf;
                }

                void submitBuf2()
                in
                {
                    assert(nextBufTask.prev is null);
                    assert(nextBufTask.next is null);
                } body
                {
                    // Hack to reuse the task object.

                    nextBufTask = typeof(nextBufTask).init;
                    nextBufTask._args[0] = &fillBuf;
                    nextBufTask._args[1] = buf2;
                    pool.put(nextBufTask);
                }

                void doBufSwap()
                {
                    if(lastTaskWaited)
                    {
                        // Then the source is empty.  Signal it here.
                        buf1 = null;
                        buf2 = null;

                        static if(!isRandomAccessRange!S)
                        {
                            from = null;
                        }

                        return;
                    }

                    buf2 = buf1;
                    buf1 = nextBufTask.yieldForce;
                    bufPos = 0;

                    if(source.empty)
                    {
                        lastTaskWaited = true;
                    }
                    else
                    {
                        submitBuf2();
                    }
                }

            public:
                @property auto front()
                {
                    return buf1[bufPos];
                }

                void popFront()
                {
                    static if(hasLength!S)
                    {
                        _length--;
                    }

                    bufPos++;
                    if(bufPos >= buf1.length)
                    {
                        doBufSwap();
                    }
                }

                static if(std.range.isInfinite!S)
                {
                    enum bool empty = false;
                }
                else
                {

                    bool empty() @property
                    {
                        // popFront() sets this when source is empty
                        return buf1.length == 0;
                    }
                }
            }
            return new Map(source, bufSize, workUnitSize, this);
        }
    }

    /**
    Given a $(D source) range that is expensive to iterate over, returns an
    input range that asynchronously buffers the contents of
    $(D source) into a buffer of $(D bufSize) elements in a worker thread,
    while making previously buffered elements from a second buffer, also of size
    $(D bufSize), available via the range interface of the returned
    object.  The returned range has a length iff $(D hasLength!S).
    $(D asyncBuf) is useful, for example, when performing expensive operations
    on the elements of ranges that represent data on a disk or network.

    Examples:
    ---
    import std.conv, std.stdio;

    void main()
    {
        // Fetch lines of a file in a background thread
        // while processing previously fetched lines,
        // dealing with byLine's buffer recycling by
        // eagerly duplicating every line.
        auto lines = File("foo.txt").byLine();
        auto duped = std.algorithm.map!"a.idup"(lines);

        // Fetch more lines in the background while we
        // process the lines already read into memory
        // into a matrix of doubles.
        double[][] matrix;
        auto asyncReader = taskPool.asyncBuf(duped);

        foreach(line; asyncReader)
        {
            auto ls = line.split("\t");
            matrix ~= to!(double[])(ls);
        }
    }
    ---

    $(B Exception Handling):

    Any exceptions thrown while iterating over $(D source) are re-thrown on a
    call to $(D popFront) or, if thrown during construction, simply
    allowed to propagate to the caller.
    */
    auto asyncBuf(S)(S source, size_t bufSize = 100) if(isInputRange!S)
    {
        static final class AsyncBuf
        {
            // This is a class because the task and source both need to be on
            // the heap.

            // The element type of S.
            alias E = ElementType!S;  // Needs to be here b/c of forward ref bugs.

        private:
            E[] buf1, buf2;
            S source;
            TaskPool pool;
            Task!(run, E[] delegate(E[]), E[]) nextBufTask;
            size_t bufPos;
            bool lastTaskWaited;

            static if(hasLength!S)
            {
                size_t _length;

                // Available if hasLength!S.
                public @property size_t length() const @safe pure nothrow
                {
                    return _length;
                }
            }

            this(S source, size_t bufSize, TaskPool pool)
            {
                buf1.length = bufSize;
                buf2.length = bufSize;

                this.source = source;
                this.pool = pool;

                static if(hasLength!S)
                {
                    _length = source.length;
                }

                buf1 = fillBuf(buf1);
                submitBuf2();
            }

            E[] fillBuf(E[] buf)
            {
                assert(buf !is null);

                size_t i;
                for(; !source.empty && i < buf.length; source.popFront())
                {
                    buf[i++] = source.front;
                }

                buf = buf[0..i];
                return buf;
            }

            void submitBuf2()
            in
            {
                assert(nextBufTask.prev is null);
                assert(nextBufTask.next is null);
            } body
            {
                // Hack to reuse the task object.

                nextBufTask = typeof(nextBufTask).init;
                nextBufTask._args[0] = &fillBuf;
                nextBufTask._args[1] = buf2;
                pool.put(nextBufTask);
            }

            void doBufSwap()
            {
                if(lastTaskWaited)
                {
                    // Then source is empty.  Signal it here.
                    buf1 = null;
                    buf2 = null;
                    return;
                }

                buf2 = buf1;
                buf1 = nextBufTask.yieldForce;
                bufPos = 0;

                if(source.empty)
                {
                    lastTaskWaited = true;
                }
                else
                {
                    submitBuf2();
                }
            }

        public:
            E front() @property
            {
                return buf1[bufPos];
            }

            void popFront()
            {
                static if(hasLength!S)
                {
                    _length--;
                }

                bufPos++;
                if(bufPos >= buf1.length)
                {
                    doBufSwap();
                }
            }

            static if(std.range.isInfinite!S)
            {
                enum bool empty = false;
            }

            else
            {
                ///
                bool empty() @property
                {
                    // popFront() sets this when source is empty:
                    return buf1.length == 0;
                }
            }
        }
        return new AsyncBuf(source, bufSize, this);
    }

    /**
    Given a callable object $(D next) that writes to a user-provided buffer and
    a second callable object $(D empty) that determines whether more data is
    available to write via $(D next), returns an input range that
    asynchronously calls $(D next) with a set of size $(D nBuffers) of buffers
    and makes the results available in the order they were obtained via the
    input range interface of the returned object.  Similarly to the
    input range overload of $(D asyncBuf), the first half of the buffers
    are made available via the range interface while the second half are
    filled and vice-versa.

    Params:

    next = A callable object that takes a single argument that must be an array
           with mutable elements.  When called, $(D next) writes data to
           the array provided by the caller.

    empty = A callable object that takes no arguments and returns a type
            implicitly convertible to $(D bool).  This is used to signify
            that no more data is available to be obtained by calling $(D next).

    initialBufSize = The initial size of each buffer.  If $(D next) takes its
                     array by reference, it may resize the buffers.

    nBuffers = The number of buffers to cycle through when calling $(D next).

    Examples:
    ---
    // Fetch lines of a file in a background
    // thread while processing previously fetched
    // lines, without duplicating any lines.
    auto file = File("foo.txt");

    void next(ref char[] buf)
    {
        file.readln(buf);
    }

    // Fetch more lines in the background while we
    // process the lines already read into memory
    // into a matrix of doubles.
    double[][] matrix;
    auto asyncReader = taskPool.asyncBuf(&next, &file.eof);

    foreach(line; asyncReader)
    {
        auto ls = line.split("\t");
        matrix ~= to!(double[])(ls);
    }
    ---

    $(B Exception Handling):

    Any exceptions thrown while iterating over $(D range) are re-thrown on a
    call to $(D popFront).

    Warning:

    Using the range returned by this function in a parallel foreach loop
    will not work because buffers may be overwritten while the task that
    processes them is in queue.  This is checked for at compile time
    and will result in a static assertion failure.
    */
    auto asyncBuf(C1, C2)(C1 next, C2 empty, size_t initialBufSize = 0, size_t nBuffers = 100)
    if(is(typeof(C2.init()) : bool) &&
        ParameterTypeTuple!C1.length == 1 &&
        ParameterTypeTuple!C2.length == 0 &&
        isArray!(ParameterTypeTuple!C1[0])
    ) {
        auto roundRobin = RoundRobinBuffer!(C1, C2)(next, empty, initialBufSize, nBuffers);
        return asyncBuf(roundRobin, nBuffers / 2);
    }

    ///
    template reduce(functions...)
    {
        /**
        Parallel reduce on a random access range.  Except as otherwise noted,
        usage is similar to $(XREF_PACK algorithm,iteration,_reduce).  This
        function works by splitting the range to be reduced into work units,
        which are slices to be reduced in parallel.  Once the results from all
        work units are computed, a final serial reduction is performed on these
        results to compute the final answer. Therefore, care must be taken to
        choose the seed value appropriately.

        Because the reduction is being performed in parallel, $(D functions)
        must be associative.  For notational simplicity, let # be an
        infix operator representing $(D functions).  Then, (a # b) # c must equal
        a # (b # c).  Floating point addition is not associative
        even though addition in exact arithmetic is.  Summing floating
        point numbers using this function may give different results than summing
        serially.  However, for many practical purposes floating point addition
        can be treated as associative.

        Note that, since $(D functions) are assumed to be associative,
        additional optimizations are made to the serial portion of the reduction
        algorithm. These take advantage of the instruction level parallelism of
        modern CPUs, in addition to the thread-level parallelism that the rest
        of this module exploits.  This can lead to better than linear speedups
        relative to $(XREF_PACK algorithm,iteration,_reduce), especially for
        fine-grained benchmarks like dot products.

        An explicit seed may be provided as the first argument.  If
        provided, it is used as the seed for all work units and for the final
        reduction of results from all work units.  Therefore, if it is not the
        identity value for the operation being performed, results may differ
        from those generated by $(XREF_PACK algorithm,iteration,_reduce) or
        depending on how many work units are used.  The next argument must be
        the range to be reduced.
        ---
        // Find the sum of squares of a range in parallel, using
        // an explicit seed.
        //
        // Timings on an Athlon 64 X2 dual core machine:
        //
        // Parallel reduce:                     72 milliseconds
        // Using std.algorithm.reduce instead:  181 milliseconds
        auto nums = iota(10_000_000.0f);
        auto sumSquares = taskPool.reduce!"a + b"(
            0.0, std.algorithm.map!"a * a"(nums)
        );
        ---

        If no explicit seed is provided, the first element of each work unit
        is used as a seed.  For the final reduction, the result from the first
        work unit is used as the seed.
        ---
        // Find the sum of a range in parallel, using the first
        // element of each work unit as the seed.
        auto sum = taskPool.reduce!"a + b"(nums);
        ---

        An explicit work unit size may be specified as the last argument.
        Specifying too small a work unit size will effectively serialize the
        reduction, as the final reduction of the result of each work unit will
        dominate computation time.  If $(D TaskPool.size) for this instance
        is zero, this parameter is ignored and one work unit is used.
        ---
        // Use a work unit size of 100.
        auto sum2 = taskPool.reduce!"a + b"(nums, 100);

        // Work unit size of 100 and explicit seed.
        auto sum3 = taskPool.reduce!"a + b"(0.0, nums, 100);
        ---

        Parallel reduce supports multiple functions, like
        $(D std.algorithm.reduce).
        ---
        // Find both the min and max of nums.
        auto minMax = taskPool.reduce!(min, max)(nums);
        assert(minMax[0] == reduce!min(nums));
        assert(minMax[1] == reduce!max(nums));
        ---

        $(B Exception Handling):

        After this function is finished executing, any exceptions thrown
        are chained together via $(D Throwable.next) and rethrown.  The chaining
        order is non-deterministic.
         */
        auto reduce(Args...)(Args args)
        {
            alias fun = reduceAdjoin!functions;
            alias finishFun = reduceFinish!functions;

            static if(isIntegral!(Args[$ - 1]))
            {
                size_t workUnitSize = cast(size_t) args[$ - 1];
                alias args2 = args[0..$ - 1];
                alias Args2 = Args[0..$ - 1];
            }
            else
            {
                alias args2 = args;
                alias Args2 = Args;
            }

            auto makeStartValue(Type)(Type e)
            {
                static if(functions.length == 1)
                {
                    return e;
                }
                else
                {
                    typeof(adjoin!(staticMap!(binaryFun, functions))(e, e)) seed = void;
                    foreach (i, T; seed.Types)
                    {
                        emplaceRef(seed.expand[i], e);
                    }

                    return seed;
                }
            }

            static if(args2.length == 2)
            {
                static assert(isInputRange!(Args2[1]));
                alias range = args2[1];
                alias seed = args2[0];
                enum explicitSeed = true;

                static if(!is(typeof(workUnitSize)))
                {
                    size_t workUnitSize = defaultWorkUnitSize(range.length);
                }
            }
            else
            {
                static assert(args2.length == 1);
                alias range = args2[0];

                static if(!is(typeof(workUnitSize)))
                {
                    size_t workUnitSize = defaultWorkUnitSize(range.length);
                }

                enforce(!range.empty,
                    "Cannot reduce an empty range with first element as start value.");

                auto seed = makeStartValue(range.front);
                enum explicitSeed = false;
                range.popFront();
            }

            alias E = typeof(seed);
            alias R = typeof(range);

            E reduceOnRange(R range, size_t lowerBound, size_t upperBound)
            {
                // This is for exploiting instruction level parallelism by
                // using multiple accumulator variables within each thread,
                // since we're assuming functions are associative anyhow.

                // This is so that loops can be unrolled automatically.
                enum ilpTuple = TypeTuple!(0, 1, 2, 3, 4, 5);
                enum nILP = ilpTuple.length;
                immutable subSize = (upperBound - lowerBound) / nILP;

                if(subSize <= 1)
                {
                    // Handle as a special case.
                    static if(explicitSeed)
                    {
                        E result = seed;
                    }
                    else
                    {
                        E result = makeStartValue(range[lowerBound]);
                        lowerBound++;
                    }

                    foreach(i; lowerBound..upperBound)
                    {
                        result = fun(result, range[i]);
                    }

                    return result;
                }

                assert(subSize > 1);
                E[nILP] results;
                size_t[nILP] offsets;

                foreach(i; ilpTuple)
                {
                    offsets[i] = lowerBound + subSize * i;

                    static if(explicitSeed)
                    {
                        results[i] = seed;
                    }
                    else
                    {
                        results[i] = makeStartValue(range[offsets[i]]);
                        offsets[i]++;
                    }
                }

                immutable nLoop = subSize - (!explicitSeed);
                foreach(i; 0..nLoop)
                {
                    foreach(j; ilpTuple)
                    {
                        results[j] = fun(results[j], range[offsets[j]]);
                        offsets[j]++;
                    }
                }

                // Finish the remainder.
                foreach(i; nILP * subSize + lowerBound..upperBound)
                {
                    results[$ - 1] = fun(results[$ - 1], range[i]);
                }

                foreach(i; ilpTuple[1..$])
                {
                    results[0] = finishFun(results[0], results[i]);
                }

                return results[0];
            }

            immutable len = range.length;
            if(len == 0)
            {
                return seed;
            }

            if(this.size == 0)
            {
                return finishFun(seed, reduceOnRange(range, 0, len));
            }

            // Unlike the rest of the functions here, I can't use the Task object
            // recycling trick here because this has to work on non-commutative
            // operations.  After all the tasks are done executing, fun() has to
            // be applied on the results of these to get a final result, but
            // it can't be evaluated out of order.

            if(workUnitSize > len)
            {
                workUnitSize = len;
            }

            immutable size_t nWorkUnits = (len / workUnitSize) + ((len % workUnitSize == 0) ? 0 : 1);
            assert(nWorkUnits * workUnitSize >= len);

            alias RTask = Task!(run, typeof(&reduceOnRange), R, size_t, size_t);
            RTask[] tasks;

            // Can't use alloca() due to Bug 3753.  Use a fixed buffer
            // backed by malloc().
            enum maxStack = 2_048;
            byte[maxStack] buf = void;
            immutable size_t nBytesNeeded = nWorkUnits * RTask.sizeof;

            import core.stdc.stdlib;
            if(nBytesNeeded < maxStack)
            {
                tasks = (cast(RTask*) buf.ptr)[0..nWorkUnits];
            }
            else
            {
                auto ptr = cast(RTask*) malloc(nBytesNeeded);
                if(!ptr)
                {
                    throw new OutOfMemoryError(
                        "Out of memory in std.parallelism."
                    );
                }

                tasks = ptr[0..nWorkUnits];
            }

            scope(exit)
            {
                if(nBytesNeeded > maxStack)
                {
                    free(tasks.ptr);
                }
            }

            foreach (ref t; tasks[])
                emplaceRef(t, RTask());

            // Hack to take the address of a nested function w/o
            // making a closure.
            static auto scopedAddress(D)(scope D del)
            {
                return del;
            }

            size_t curPos = 0;
            void useTask(ref RTask task)
            {
                task.pool = this;
                task._args[0] = scopedAddress(&reduceOnRange);
                task._args[3] = min(len, curPos + workUnitSize);  // upper bound.
                task._args[1] = range;  // range
                task._args[2] = curPos; // lower bound.

                curPos += workUnitSize;
            }

            foreach(ref task; tasks)
            {
                useTask(task);
            }

            foreach(i; 1..tasks.length - 1)
            {
                tasks[i].next = tasks[i + 1].basePtr;
                tasks[i + 1].prev = tasks[i].basePtr;
            }

            if(tasks.length > 1)
            {
                queueLock();
                scope(exit) queueUnlock();

                abstractPutGroupNoSync(
                    tasks[1].basePtr,
                    tasks[$ - 1].basePtr
                );
            }

            if(tasks.length > 0)
            {
                try
                {
                    tasks[0].job();
                }
                catch(Throwable e)
                {
                    tasks[0].exception = e;
                }
                tasks[0].taskStatus = TaskStatus.done;

                // Try to execute each of these in the current thread
                foreach(ref task; tasks[1..$])
                {
                    tryDeleteExecute(task.basePtr);
                }
            }

            // Now that we've tried to execute every task, they're all either
            // done or in progress.  Force all of them.
            E result = seed;

            Throwable firstException, lastException;

            foreach(ref task; tasks)
            {
                try
                {
                    task.yieldForce;
                }
                catch(Throwable e)
                {
                    addToChain(e, firstException, lastException);
                    continue;
                }

                if(!firstException) result = finishFun(result, task.returnVal);
            }

            if(firstException) throw firstException;

            return result;
        }
    }

    /**
    Gets the index of the current thread relative to this $(D TaskPool).  Any
    thread not in this pool will receive an index of 0.  The worker threads in
    this pool receive unique indices of 1 through $(D this.size).

    This function is useful for maintaining worker-local resources.

    Examples:
    ---
    // Execute a loop that computes the greatest common
    // divisor of every number from 0 through 999 with
    // 42 in parallel.  Write the results out to
    // a set of files, one for each thread.  This allows
    // results to be written out without any synchronization.

    import std.conv, std.range, std.numeric, std.stdio;

    void main()
    {
        auto filesHandles = new File[taskPool.size + 1];
        scope(exit) {
            foreach(ref handle; fileHandles) {
                handle.close();
            }
        }

        foreach(i, ref handle; fileHandles)
        {
            handle = File("workerResults" ~ to!string(i) ~ ".txt");
        }

        foreach(num; parallel(iota(1_000)))
        {
            auto outHandle = fileHandles[taskPool.workerIndex];
            outHandle.writeln(num, '\t', gcd(num, 42));
        }
    }
    ---
    */
    size_t workerIndex() @property @safe const nothrow
    {
        immutable rawInd = threadIndex;
        return (rawInd >= instanceStartIndex && rawInd < instanceStartIndex + size) ?
                (rawInd - instanceStartIndex + 1) : 0;
    }

    /**
    Struct for creating worker-local storage.  Worker-local storage is
    thread-local storage that exists only for worker threads in a given
    $(D TaskPool) plus a single thread outside the pool.  It is allocated on the
    garbage collected heap in a way that avoids _false sharing, and doesn't
    necessarily have global scope within any thread.  It can be accessed from
    any worker thread in the $(D TaskPool) that created it, and one thread
    outside this $(D TaskPool).  All threads outside the pool that created a
    given instance of worker-local storage share a single slot.

    Since the underlying data for this struct is heap-allocated, this struct
    has reference semantics when passed between functions.

    The main uses cases for $(D WorkerLocalStorageStorage) are:

    1.  Performing parallel reductions with an imperative, as opposed to
    functional, programming style.  In this case, it's useful to treat
    $(D WorkerLocalStorageStorage) as local to each thread for only the parallel
    portion of an algorithm.

    2.  Recycling temporary buffers across iterations of a parallel foreach loop.

    Examples:
    ---
    // Calculate pi as in our synopsis example, but
    // use an imperative instead of a functional style.
    immutable n = 1_000_000_000;
    immutable delta = 1.0L / n;

    auto sums = taskPool.workerLocalStorage(0.0L);
    foreach(i; parallel(iota(n)))
    {
        immutable x = ( i - 0.5L ) * delta;
        immutable toAdd = delta / ( 1.0 + x * x );
        sums.get += toAdd;
    }

    // Add up the results from each worker thread.
    real pi = 0;
    foreach(threadResult; sums.toRange)
    {
        pi += 4.0L * threadResult;
    }
    ---
     */
    static struct WorkerLocalStorage(T)
    {
    private:
        TaskPool pool;
        size_t size;

        static immutable size_t cacheLineSize;
        size_t elemSize;
        bool* stillThreadLocal;

        shared static this()
        {
            size_t lineSize = 0;
            foreach(cachelevel; datacache)
            {
                if(cachelevel.lineSize > lineSize && cachelevel.lineSize < uint.max)
                {
                    lineSize = cachelevel.lineSize;
                }
            }

            cacheLineSize = lineSize;
        }

        static size_t roundToLine(size_t num) pure nothrow
        {
            if(num % cacheLineSize == 0)
            {
                return num;
            }
            else {
                return ((num / cacheLineSize) + 1) * cacheLineSize;
            }
        }

        void* data;

        void initialize(TaskPool pool)
        {
            this.pool = pool;
            size = pool.size + 1;
            stillThreadLocal = new bool;
            *stillThreadLocal = true;

            // Determines whether the GC should scan the array.
            auto blkInfo = (typeid(T).flags & 1) ?
                           cast(GC.BlkAttr) 0 :
                           GC.BlkAttr.NO_SCAN;

            immutable nElem = pool.size + 1;
            elemSize = roundToLine(T.sizeof);

            // The + 3 is to pad one full cache line worth of space on either side
            // of the data structure to make sure false sharing with completely
            // unrelated heap data is prevented, and to provide enough padding to
            // make sure that data is cache line-aligned.
            data = GC.malloc(elemSize * (nElem + 3), blkInfo) + elemSize;

            // Cache line align data ptr.
            data = cast(void*) roundToLine(cast(size_t) data);

            foreach(i; 0..nElem)
            {
                this.opIndex(i) = T.init;
            }
        }

        ref T opIndex(size_t index)
        {
            assert(index < size, text(index, '\t', uint.max));
            return *(cast(T*) (data + elemSize * index));
        }

        void opIndexAssign(T val, size_t index)
        {
            assert(index < size);
            *(cast(T*) (data + elemSize * index)) = val;
        }

    public:
        /**
        Get the current thread's instance.  Returns by ref.
        Note that calling $(D get) from any thread
        outside the $(D TaskPool) that created this instance will return the
        same reference, so an instance of worker-local storage should only be
        accessed from one thread outside the pool that created it.  If this
        rule is violated, undefined behavior will result.

        If assertions are enabled and $(D toRange) has been called, then this
        WorkerLocalStorage instance is no longer worker-local and an assertion
        failure will result when calling this method.  This is not checked
        when assertions are disabled for performance reasons.
         */
        ref T get() @property
        {
            assert(*stillThreadLocal,
                "Cannot call get() on this instance of WorkerLocalStorage " ~
                "because it is no longer worker-local."
            );
            return opIndex(pool.workerIndex);
        }

        /**
        Assign a value to the current thread's instance.  This function has
        the same caveats as its overload.
        */
        void get(T val) @property
        {
            assert(*stillThreadLocal,
                "Cannot call get() on this instance of WorkerLocalStorage " ~
                "because it is no longer worker-local."
            );

            opIndexAssign(val, pool.workerIndex);
        }

        /**
        Returns a range view of the values for all threads, which can be used
        to further process the results of each thread after running the parallel
        part of your algorithm.  Do not use this method in the parallel portion
        of your algorithm.

        Calling this function sets a flag indicating that this struct is no
        longer worker-local, and attempting to use the $(D get) method again
        will result in an assertion failure if assertions are enabled.
         */
        WorkerLocalStorageRange!T toRange() @property
        {
            if(*stillThreadLocal)
            {
                *stillThreadLocal = false;

                // Make absolutely sure results are visible to all threads.
                // This is probably not necessary since some other
                // synchronization primitive will be used to signal that the
                // parallel part of the algorithm is done, but the
                // performance impact should be negligible, so it's better
                // to be safe.
                ubyte barrierDummy;
                atomicSetUbyte(barrierDummy, 1);
            }

            return WorkerLocalStorageRange!T(this);
        }
    }

    /**
    Range primitives for worker-local storage.  The purpose of this is to
    access results produced by each worker thread from a single thread once you
    are no longer using the worker-local storage from multiple threads.
    Do not use this struct in the parallel portion of your algorithm.

    The proper way to instantiate this object is to call
    $(D WorkerLocalStorage.toRange).  Once instantiated, this object behaves
    as a finite random-access range with assignable, lvalue elements and
    a length equal to the number of worker threads in the $(D TaskPool) that
    created it plus 1.
     */
    static struct WorkerLocalStorageRange(T)
    {
    private:
        WorkerLocalStorage!T workerLocalStorage;

        size_t _length;
        size_t beginOffset;

        this(WorkerLocalStorage!T wl)
        {
            this.workerLocalStorage = wl;
            _length = wl.size;
        }

    public:
        ref T front() @property
        {
            return this[0];
        }

        ref T back() @property
        {
            return this[_length - 1];
        }

        void popFront()
        {
            if(_length > 0)
            {
                beginOffset++;
                _length--;
            }
        }

        void popBack()
        {
            if(_length > 0)
            {
                _length--;
            }
        }

        typeof(this) save() @property
        {
            return this;
        }

        ref T opIndex(size_t index)
        {
            assert(index < _length);
            return workerLocalStorage[index + beginOffset];
        }

        void opIndexAssign(T val, size_t index)
        {
            assert(index < _length);
            workerLocalStorage[index] = val;
        }

        typeof(this) opSlice(size_t lower, size_t upper)
        {
            assert(upper <= _length);
            auto newWl = this.workerLocalStorage;
            newWl.data += lower * newWl.elemSize;
            newWl.size = upper - lower;
            return typeof(this)(newWl);
        }

        bool empty() @property
        {
            return length == 0;
        }

        size_t length() @property
        {
            return _length;
        }
    }

    /**
    Creates an instance of worker-local storage, initialized with a given
    value.  The value is $(D lazy) so that you can, for example, easily
    create one instance of a class for each worker.  For usage example,
    see the $(D WorkerLocalStorage) struct.
     */
    WorkerLocalStorage!T workerLocalStorage(T)(lazy T initialVal = T.init)
    {
        WorkerLocalStorage!T ret;
        ret.initialize(this);
        foreach(i; 0..size + 1)
        {
            ret[i] = initialVal;
        }

        // Memory barrier to make absolutely sure that what we wrote is
        // visible to worker threads.
        ubyte barrierDummy;
        atomicSetUbyte(barrierDummy, 0);

        return ret;
    }

    /**
    Signals to all worker threads to terminate as soon as they are finished
    with their current $(D Task), or immediately if they are not executing a
    $(D Task).  $(D Task)s that were in queue will not be executed unless
    a call to $(D Task.workForce), $(D Task.yieldForce) or $(D Task.spinForce)
    causes them to be executed.

    Use only if you have waited on every $(D Task) and therefore know the
    queue is empty, or if you speculatively executed some tasks and no longer
    need the results.
     */
    void stop() @trusted
    {
        queueLock();
        scope(exit) queueUnlock();
        atomicSetUbyte(status, PoolState.stopNow);
        notifyAll();
    }

    /**
    Signals worker threads to terminate when the queue becomes empty.

    If blocking argument is true, wait for all worker threads to terminate
    before returning.  This option might be used in applications where
    task results are never consumed-- e.g. when $(D TaskPool) is employed as a
    rudimentary scheduler for tasks which communicate by means other than
    return values.

    Warning:  Calling this function with $(D blocking = true) from a worker
              thread that is a member of the same $(D TaskPool) that
              $(D finish) is being called on will result in a deadlock.
     */
    void finish(bool blocking = false) @trusted
    {
        {
            queueLock();
            scope(exit) queueUnlock();
            atomicCasUbyte(status, PoolState.running, PoolState.finishing);
            notifyAll();
        }
        if (blocking)
        {
            // Use this thread as a worker until everything is finished.
            executeWorkLoop();

            foreach(t; pool)
            {
                // Maybe there should be something here to prevent a thread
                // from calling join() on itself if this function is called
                // from a worker thread in the same pool, but:
                //
                // 1.  Using an if statement to skip join() would result in
                //     finish() returning without all tasks being finished.
                //
                // 2.  If an exception were thrown, it would bubble up to the
                //     Task from which finish() was called and likely be
                //     swallowed.
                t.join();
            }
        }
    }

    /// Returns the number of worker threads in the pool.
    @property size_t size() @safe const pure nothrow
    {
        return pool.length;
    }

    /**
    Put a $(D Task) object on the back of the task queue.  The $(D Task)
    object may be passed by pointer or reference.

    Example:
    ---
    import std.file;

    // Create a task.
    auto t = task!read("foo.txt");

    // Add it to the queue to be executed.
    taskPool.put(t);
    ---

    Notes:

    @trusted overloads of this function are called for $(D Task)s if
    $(XREF traits, hasUnsharedAliasing) is false for the $(D Task)'s
    return type or the function the $(D Task) executes is $(D pure).
    $(D Task) objects that meet all other requirements specified in the
    $(D @trusted) overloads of $(D task) and $(D scopedTask) may be created
    and executed from $(D @safe) code via $(D Task.executeInNewThread) but
    not via $(D TaskPool).

    While this function takes the address of variables that may
    be on the stack, some overloads are marked as @trusted.
    $(D Task) includes a destructor that waits for the task to complete
    before destroying the stack frame it is allocated on.  Therefore,
    it is impossible for the stack frame to be destroyed before the task is
    complete and no longer referenced by a $(D TaskPool).
    */
    void put(alias fun, Args...)(ref Task!(fun, Args) task)
    if(!isSafeReturn!(typeof(task)))
    {
        task.pool = this;
        abstractPut(task.basePtr);
    }

    /// Ditto
    void put(alias fun, Args...)(Task!(fun, Args)* task)
    if(!isSafeReturn!(typeof(*task)))
    {
        enforce(task !is null, "Cannot put a null Task on a TaskPool queue.");
        put(*task);
    }

    @trusted void put(alias fun, Args...)(ref Task!(fun, Args) task)
    if(isSafeReturn!(typeof(task)))
    {
        task.pool = this;
        abstractPut(task.basePtr);
    }

    @trusted void put(alias fun, Args...)(Task!(fun, Args)* task)
    if(isSafeReturn!(typeof(*task)))
    {
        enforce(task !is null, "Cannot put a null Task on a TaskPool queue.");
        put(*task);
    }

    /**
    These properties control whether the worker threads are daemon threads.
    A daemon thread is automatically terminated when all non-daemon threads
    have terminated.  A non-daemon thread will prevent a program from
    terminating as long as it has not terminated.

    If any $(D TaskPool) with non-daemon threads is active, either $(D stop)
    or $(D finish) must be called on it before the program can terminate.

    The worker treads in the $(D TaskPool) instance returned by the
    $(D taskPool) property are daemon by default.  The worker threads of
    manually instantiated task pools are non-daemon by default.

    Note:  For a size zero pool, the getter arbitrarily returns true and the
           setter has no effect.
    */
    bool isDaemon() @property @trusted
    {
        queueLock();
        scope(exit) queueUnlock();
        return (size == 0) ? true : pool[0].isDaemon;
    }

    /// Ditto
    void isDaemon(bool newVal) @property @trusted
    {
        queueLock();
        scope(exit) queueUnlock();
        foreach(thread; pool)
        {
            thread.isDaemon = newVal;
        }
    }

    /**
    These functions allow getting and setting the OS scheduling priority of
    the worker threads in this $(D TaskPool).  They forward to
    $(D core.thread.Thread.priority), so a given priority value here means the
    same thing as an identical priority value in $(D core.thread).

    Note:  For a size zero pool, the getter arbitrarily returns
           $(D core.thread.Thread.PRIORITY_MIN) and the setter has no effect.
    */
    int priority() @property @trusted
    {
        return (size == 0) ? core.thread.Thread.PRIORITY_MIN :
        pool[0].priority;
    }

    /// Ditto
    void priority(int newPriority) @property @trusted
    {
        if(size > 0)
        {
            foreach(t; pool)
            {
                t.priority = newPriority;
            }
        }
    }
}

/**
Returns a lazily initialized global instantiation of $(D TaskPool).
This function can safely be called concurrently from multiple non-worker
threads.  The worker threads in this pool are daemon threads, meaning that it
is not necessary to call $(D TaskPool.stop) or $(D TaskPool.finish) before
terminating the main thread.
*/
@property TaskPool taskPool() @trusted
{
    static bool initialized;
    __gshared static TaskPool pool;

    if(!initialized)
    {
        synchronized(typeid(TaskPool))
        {
            if(!pool)
            {
                pool = new TaskPool(defaultPoolThreads);
                pool.isDaemon = true;
            }
        }

        initialized = true;
    }

    return pool;
}

private shared uint _defaultPoolThreads;
shared static this()
{
    atomicStore(_defaultPoolThreads, totalCPUs - 1);
}

/**
These properties get and set the number of worker threads in the $(D TaskPool)
instance returned by $(D taskPool).  The default value is $(D totalCPUs) - 1.
Calling the setter after the first call to $(D taskPool) does not changes
number of worker threads in the instance returned by $(D taskPool).
*/
@property uint defaultPoolThreads() @trusted
{
    return atomicLoad(_defaultPoolThreads);
}

/// Ditto
@property void defaultPoolThreads(uint newVal) @trusted
{
    atomicStore(_defaultPoolThreads, newVal);
}

/**
Convenience functions that forwards to $(D taskPool.parallel).  The
purpose of these is to make parallel foreach less verbose and more
readable.

Example:
---
// Find the logarithm of every number from
// 1 to 1_000_000 in parallel, using the
// default TaskPool instance.
auto logs = new double[1_000_000];

foreach(i, ref elem; parallel(logs)) {
    elem = log(i + 1.0);
}
---

*/
ParallelForeach!R parallel(R)(R range)
{
    return taskPool.parallel(range);
}

/// Ditto
ParallelForeach!R parallel(R)(R range, size_t workUnitSize)
{
    return taskPool.parallel(range, workUnitSize);
}

// Thrown when a parallel foreach loop is broken from.
class ParallelForeachError : Error
{
    this()
    {
        super("Cannot break from a parallel foreach loop using break, return, "
              ~ "labeled break/continue or goto statements.");
    }
}

/*------Structs that implement opApply for parallel foreach.------------------*/
private template randLen(R)
{
    enum randLen = isRandomAccessRange!R && hasLength!R;
}

private void submitAndExecute(
    TaskPool pool,
    scope void delegate() doIt
)
{
    immutable nThreads = pool.size + 1;

    alias PTask = typeof(scopedTask(doIt));
    import core.stdc.stdlib;
    import core.stdc.string : memcpy;

    // The logical thing to do would be to just use alloca() here, but that
    // causes problems on Windows for reasons that I don't understand
    // (tentatively a compiler bug) and definitely doesn't work on Posix due
    // to Bug 3753.  Therefore, allocate a fixed buffer and fall back to
    // malloc() if someone's using a ridiculous amount of threads.  Also,
    // the using a byte array instead of a PTask array as the fixed buffer
    // is to prevent d'tors from being called on uninitialized excess PTask
    // instances.
    enum nBuf = 64;
    byte[nBuf * PTask.sizeof] buf = void;
    PTask[] tasks;
    if(nThreads <= nBuf)
    {
        tasks = (cast(PTask*) buf.ptr)[0..nThreads];
    }
    else
    {
        auto ptr = cast(PTask*) malloc(nThreads * PTask.sizeof);
        if(!ptr) throw new OutOfMemoryError("Out of memory in std.parallelism.");
        tasks = ptr[0..nThreads];
    }

    scope(exit)
    {
        if(nThreads > nBuf)
        {
            free(tasks.ptr);
        }
    }

    foreach(ref t; tasks)
    {
        // This silly looking code is necessary to prevent d'tors from being
        // called on uninitialized objects.
        auto temp = scopedTask(doIt);
        core.stdc.string.memcpy(&t, &temp, PTask.sizeof);

        // This has to be done to t after copying, not temp before copying.
        // Otherwise, temp's destructor will sit here and wait for the
        // task to finish.
        t.pool = pool;
    }

    foreach(i; 1..tasks.length - 1)
    {
        tasks[i].next = tasks[i + 1].basePtr;
        tasks[i + 1].prev = tasks[i].basePtr;
    }

    if(tasks.length > 1)
    {
        pool.queueLock();
        scope(exit) pool.queueUnlock();

        pool.abstractPutGroupNoSync(
            tasks[1].basePtr,
            tasks[$ - 1].basePtr
        );
    }

    if(tasks.length > 0)
    {
        try
        {
            tasks[0].job();
        }
        catch(Throwable e)
        {
            tasks[0].exception = e;
        }
        tasks[0].taskStatus = TaskStatus.done;

        // Try to execute each of these in the current thread
        foreach(ref task; tasks[1..$])
        {
            pool.tryDeleteExecute(task.basePtr);
        }
    }

    Throwable firstException, lastException;

    foreach(i, ref task; tasks)
    {
        try
        {
            task.yieldForce;
        }
        catch(Throwable e)
        {
            addToChain(e, firstException, lastException);
            continue;
        }
    }

    if(firstException) throw firstException;
}

void foreachErr()
{
    throw new ParallelForeachError();
}

int doSizeZeroCase(R, Delegate)(ref ParallelForeach!R p, Delegate dg)
{
    with(p)
    {
        int res = 0;
        size_t index = 0;

        // The explicit ElementType!R in the foreach loops is necessary for
        // correct behavior when iterating over strings.
        static if(hasLvalueElements!R)
        {
            foreach(ref ElementType!R elem; range)
            {
                static if(ParameterTypeTuple!dg.length == 2)
                {
                    res = dg(index, elem);
                }
                else
                {
                    res = dg(elem);
                }
                if(res) foreachErr();
                index++;
            }
        }
        else
        {
            foreach(ElementType!R elem; range)
            {
                static if(ParameterTypeTuple!dg.length == 2)
                {
                    res = dg(index, elem);
                }
                else
                {
                    res = dg(elem);
                }
                if(res) foreachErr();
                index++;
            }
        }
        return res;
    }
}

private enum string parallelApplyMixinRandomAccess = q{
    // Handle empty thread pool as special case.
    if(pool.size == 0)
    {
        return doSizeZeroCase(this, dg);
    }

    // Whether iteration is with or without an index variable.
    enum withIndex = ParameterTypeTuple!(typeof(dg)).length == 2;

    shared size_t workUnitIndex = size_t.max;  // Effectively -1:  chunkIndex + 1 == 0
    immutable len = range.length;
    if(!len) return 0;

    shared bool shouldContinue = true;

    void doIt()
    {
        scope(failure)
        {
            // If an exception is thrown, all threads should bail.
            atomicStore(shouldContinue, false);
        }

        while(atomicLoad(shouldContinue))
        {
            immutable myUnitIndex = atomicOp!"+="(workUnitIndex, 1);
            immutable start = workUnitSize * myUnitIndex;
            if(start >= len)
            {
                atomicStore(shouldContinue, false);
                break;
            }

            immutable end = min(len, start + workUnitSize);

            foreach(i; start..end)
            {
                static if(withIndex)
                {
                    if(dg(i, range[i])) foreachErr();
                }
                else
                {
                    if(dg(range[i])) foreachErr();
                }
            }
        }
    }

    submitAndExecute(pool, &doIt);

    return 0;
};

enum string parallelApplyMixinInputRange = q{
    // Handle empty thread pool as special case.
    if(pool.size == 0)
    {
        return doSizeZeroCase(this, dg);
    }

    // Whether iteration is with or without an index variable.
    enum withIndex = ParameterTypeTuple!(typeof(dg)).length == 2;

    // This protects the range while copying it.
    auto rangeMutex = new Mutex();

    shared bool shouldContinue = true;

    // The total number of elements that have been popped off range.
    // This is updated only while protected by rangeMutex;
    size_t nPopped = 0;

    static if(
        is(typeof(range.buf1)) &&
        is(typeof(range.bufPos)) &&
        is(typeof(range.doBufSwap()))
    )
    {
        // Make sure we don't have the buffer recycling overload of
        // asyncBuf.
        static if(
            is(typeof(range.source)) &&
            isRoundRobin!(typeof(range.source))
        )
        {
            static assert(0, "Cannot execute a parallel foreach loop on " ~
            "the buffer recycling overload of asyncBuf.");
        }

        enum bool bufferTrick = true;
    }
    else
    {
        enum bool bufferTrick = false;
    }

    void doIt()
    {
        scope(failure)
        {
            // If an exception is thrown, all threads should bail.
            atomicStore(shouldContinue, false);
        }

        static if(hasLvalueElements!R)
        {
            alias Temp = ElementType!R*[];
            Temp temp;

            // Returns:  The previous value of nPopped.
            size_t makeTemp()
            {
                if(temp is null)
                {
                    temp = uninitializedArray!Temp(workUnitSize);
                }

                rangeMutex.lock();
                scope(exit) rangeMutex.unlock();

                size_t i = 0;
                for(; i < workUnitSize && !range.empty; range.popFront(), i++)
                {
                    temp[i] = addressOf(range.front);
                }

                temp = temp[0..i];
                auto ret = nPopped;
                nPopped += temp.length;
                return ret;
            }

        }
        else
        {

            alias Temp = ElementType!R[];
            Temp temp;

            // Returns:  The previous value of nPopped.
            static if(!bufferTrick) size_t makeTemp()
            {
                if(temp is null)
                {
                    temp = uninitializedArray!Temp(workUnitSize);
                }

                rangeMutex.lock();
                scope(exit) rangeMutex.unlock();

                size_t i = 0;
                for(; i < workUnitSize && !range.empty; range.popFront(), i++)
                {
                    temp[i] = range.front;
                }

                temp = temp[0..i];
                auto ret = nPopped;
                nPopped += temp.length;
                return ret;
            }

            static if(bufferTrick) size_t makeTemp()
            {
                rangeMutex.lock();
                scope(exit) rangeMutex.unlock();

                // Elide copying by just swapping buffers.
                temp.length = range.buf1.length;
                swap(range.buf1, temp);

                // This is necessary in case popFront() has been called on
                // range before entering the parallel foreach loop.
                temp = temp[range.bufPos..$];

                static if(is(typeof(range._length)))
                {
                    range._length -= (temp.length - range.bufPos);
                }

                range.doBufSwap();
                auto ret = nPopped;
                nPopped += temp.length;
                return ret;
            }
        }

        while(atomicLoad(shouldContinue))
        {
            auto overallIndex = makeTemp();
            if(temp.empty)
            {
                atomicStore(shouldContinue, false);
                break;
            }

            foreach(i; 0..temp.length)
            {
                scope(success) overallIndex++;

                static if(hasLvalueElements!R)
                {
                    static if(withIndex)
                    {
                        if(dg(overallIndex, *temp[i])) foreachErr();
                    }
                    else
                    {
                        if(dg(*temp[i])) foreachErr();
                    }
                }
                else
                {
                    static if(withIndex)
                    {
                        if(dg(overallIndex, temp[i])) foreachErr();
                    }
                    else
                    {
                        if(dg(temp[i])) foreachErr();
                    }
                }
            }
        }
    }

    submitAndExecute(pool, &doIt);

    return 0;
};

// Calls e.next until the end of the chain is found.
private Throwable findLastException(Throwable e) pure nothrow
{
    if(e is null) return null;

    while(e.next)
    {
        e = e.next;
    }

    return e;
}

// Adds e to the exception chain.
private void addToChain(
    Throwable e,
    ref Throwable firstException,
    ref Throwable lastException
) pure nothrow
{
    if(firstException)
    {
        assert(lastException);
        lastException.next = e;
        lastException = findLastException(e);
    }
    else
    {
        firstException = e;
        lastException = findLastException(e);
    }
}

private struct ParallelForeach(R)
{
    TaskPool pool;
    R range;
    size_t workUnitSize;
    alias E = ElementType!R;

    static if(hasLvalueElements!R)
    {
        alias NoIndexDg = int delegate(ref E);
        alias IndexDg = int delegate(size_t, ref E);
    }
    else
    {
        alias NoIndexDg = int delegate(E);
        alias IndexDg = int delegate(size_t, E);
    }

    int opApply(scope NoIndexDg dg)
    {
        static if(randLen!R)
        {
            mixin(parallelApplyMixinRandomAccess);
        }
        else
        {
            mixin(parallelApplyMixinInputRange);
        }
    }

    int opApply(scope IndexDg dg)
    {
        static if(randLen!R)
        {
            mixin(parallelApplyMixinRandomAccess);
        }
        else
        {
            mixin(parallelApplyMixinInputRange);
        }
    }
}

/*
This struct buffers the output of a callable that outputs data into a
user-supplied buffer into a set of buffers of some fixed size.  It allows these
buffers to be accessed with an input range interface.  This is used internally
in the buffer-recycling overload of TaskPool.asyncBuf, which creates an
instance and forwards it to the input range overload of asyncBuf.
*/
private struct RoundRobinBuffer(C1, C2)
{
    // No need for constraints because they're already checked for in asyncBuf.

    alias Array = ParameterTypeTuple!(C1.init)[0];
    alias T = typeof(Array.init[0]);

    T[][] bufs;
    size_t index;
    C1 nextDel;
    C2 emptyDel;
    bool _empty;
    bool primed;

    this(
        C1 nextDel,
        C2 emptyDel,
        size_t initialBufSize,
        size_t nBuffers
    ) {
        this.nextDel = nextDel;
        this.emptyDel = emptyDel;
        bufs.length = nBuffers;

        foreach(ref buf; bufs)
        {
            buf.length = initialBufSize;
        }
    }

    void prime()
    in
    {
        assert(!empty);
    }
    body
    {
        scope(success) primed = true;
        nextDel(bufs[index]);
    }


    T[] front() @property
    in
    {
        assert(!empty);
    }
    body
    {
        if(!primed) prime();
        return bufs[index];
    }

    void popFront()
    {
        if(empty || emptyDel())
        {
            _empty = true;
            return;
        }

        index = (index + 1) % bufs.length;
        primed = false;
    }

    bool empty() @property const @safe pure nothrow
    {
        return _empty;
    }
}

version(unittest)
{
    // This was the only way I could get nested maps to work.
    __gshared TaskPool poolInstance;

    import std.stdio;
}

// These test basic functionality but don't stress test for threading bugs.
// These are the tests that should be run every time Phobos is compiled.
unittest
{
    poolInstance = new TaskPool(2);
    scope(exit) poolInstance.stop();

    // The only way this can be verified is manually.
    debug(std_parallelism) stderr.writeln("totalCPUs = ", totalCPUs);

    auto oldPriority = poolInstance.priority;
    poolInstance.priority = Thread.PRIORITY_MAX;
    assert(poolInstance.priority == Thread.PRIORITY_MAX);

    poolInstance.priority = Thread.PRIORITY_MIN;
    assert(poolInstance.priority == Thread.PRIORITY_MIN);

    poolInstance.priority = oldPriority;
    assert(poolInstance.priority == oldPriority);

    static void refFun(ref uint num)
    {
        num++;
    }

    uint x;

    // Test task().
    auto t = task!refFun(x);
    poolInstance.put(t);
    t.yieldForce;
    assert(t.args[0] == 1);

    auto t2 = task(&refFun, x);
    poolInstance.put(t2);
    t2.yieldForce;
    assert(t2.args[0] == 1);

    // Test scopedTask().
    auto st = scopedTask!refFun(x);
    poolInstance.put(st);
    st.yieldForce;
    assert(st.args[0] == 1);

    auto st2 = scopedTask(&refFun, x);
    poolInstance.put(st2);
    st2.yieldForce;
    assert(st2.args[0] == 1);

    // Test executeInNewThread().
    auto ct = scopedTask!refFun(x);
    ct.executeInNewThread(Thread.PRIORITY_MAX);
    ct.yieldForce;
    assert(ct.args[0] == 1);

    // Test ref return.
    uint toInc = 0;
    static ref T makeRef(T)(ref T num)
    {
        return num;
    }

    auto t3 = task!makeRef(toInc);
    taskPool.put(t3);
    assert(t3.args[0] == 0);
    t3.spinForce++;
    assert(t3.args[0] == 1);

    static void testSafe() @safe {
        static int bump(int num)
        {
            return num + 1;
        }

        auto safePool = new TaskPool(0);
        auto t = task(&bump, 1);
        taskPool.put(t);
        assert(t.yieldForce == 2);

        auto st = scopedTask(&bump, 1);
        taskPool.put(st);
        assert(st.yieldForce == 2);
        safePool.stop();
    }

    auto arr = [1,2,3,4,5];
    auto nums = new uint[5];
    auto nums2 = new uint[5];

    foreach(i, ref elem; poolInstance.parallel(arr))
    {
        elem++;
        nums[i] = cast(uint) i + 2;
        nums2[i] = elem;
    }

    assert(nums == [2,3,4,5,6], text(nums));
    assert(nums2 == nums, text(nums2));
    assert(arr == nums, text(arr));

    // Test const/immutable arguments.
    static int add(int lhs, int rhs)
    {
        return lhs + rhs;
    }
    immutable addLhs = 1;
    immutable addRhs = 2;
    auto addTask = task(&add, addLhs, addRhs);
    auto addScopedTask = scopedTask(&add, addLhs, addRhs);
    poolInstance.put(addTask);
    poolInstance.put(addScopedTask);
    assert(addTask.yieldForce == 3);
    assert(addScopedTask.yieldForce == 3);

    // Test parallel foreach with non-random access range.
    auto range = filter!"a != 666"([0, 1, 2, 3, 4]);

    foreach(i, elem; poolInstance.parallel(range))
    {
        nums[i] = cast(uint) i;
    }

    assert(nums == [0,1,2,3,4]);

    auto logs = new double[1_000_000];
    foreach(i, ref elem; poolInstance.parallel(logs))
    {
        elem = log(i + 1.0);
    }

    foreach(i, elem; logs)
    {
        assert(approxEqual(elem, cast(double) log(i + 1)));
    }

    assert(poolInstance.amap!"a * a"([1,2,3,4,5]) == [1,4,9,16,25]);
    assert(poolInstance.amap!"a * a"([1,2,3,4,5], new long[5]) == [1,4,9,16,25]);
    assert(poolInstance.amap!("a * a", "-a")([1,2,3]) ==
           [tuple(1, -1), tuple(4, -2), tuple(9, -3)]);

    auto tupleBuf = new Tuple!(int, int)[3];
    poolInstance.amap!("a * a", "-a")([1,2,3], tupleBuf);
    assert(tupleBuf == [tuple(1, -1), tuple(4, -2), tuple(9, -3)]);
    poolInstance.amap!("a * a", "-a")([1,2,3], 5, tupleBuf);
    assert(tupleBuf == [tuple(1, -1), tuple(4, -2), tuple(9, -3)]);

    // Test amap with a non-array buffer.
    auto toIndex = new int[5];
    auto indexed = std.range.indexed(toIndex, [3, 1, 4, 0, 2]);
    poolInstance.amap!"a * 2"([1, 2, 3, 4, 5], indexed);
    assert(equal(indexed, [2, 4, 6, 8, 10]));
    assert(equal(toIndex, [8, 4, 10, 2, 6]));
    poolInstance.amap!"a / 2"(indexed, indexed);
    assert(equal(indexed, [1, 2, 3, 4, 5]));
    assert(equal(toIndex, [4, 2, 5, 1, 3]));

    auto buf = new int[5];
    poolInstance.amap!"a * a"([1,2,3,4,5], buf);
    assert(buf == [1,4,9,16,25]);
    poolInstance.amap!"a * a"([1,2,3,4,5], 4, buf);
    assert(buf == [1,4,9,16,25]);

    assert(poolInstance.reduce!"a + b"([1]) == 1);
    assert(poolInstance.reduce!"a + b"([1,2,3,4]) == 10);
    assert(poolInstance.reduce!"a + b"(0.0, [1,2,3,4]) == 10);
    assert(poolInstance.reduce!"a + b"(0.0, [1,2,3,4], 1) == 10);
    assert(poolInstance.reduce!(min, max)([1,2,3,4]) == tuple(1, 4));
    assert(poolInstance.reduce!("a + b", "a * b")(tuple(0, 1), [1,2,3,4]) ==
           tuple(10, 24));

    immutable serialAns = std.algorithm.reduce!"a + b"(iota(1000));
    assert(poolInstance.reduce!"a + b"(0, iota(1000)) == serialAns);
    assert(poolInstance.reduce!"a + b"(iota(1000)) == serialAns);

    // Test worker-local storage.
    auto wl = poolInstance.workerLocalStorage(0);
    foreach(i; poolInstance.parallel(iota(1000), 1))
    {
        wl.get = wl.get + i;
    }

    auto wlRange = wl.toRange;
    auto parallelSum = poolInstance.reduce!"a + b"(wlRange);
    assert(parallelSum == 499500);
    assert(wlRange[0..1][0] == wlRange[0]);
    assert(wlRange[1..2][0] == wlRange[1]);

    // Test finish()
    {
        static void slowFun() { Thread.sleep(dur!"msecs"(1)); }

        auto pool1 = new TaskPool();
        auto tSlow = task!slowFun();
        pool1.put(tSlow);
        pool1.finish();
        tSlow.yieldForce;
        // Can't assert that pool1.status == PoolState.stopNow because status
        // doesn't change until after the "done" flag is set and the waiting
        // thread is woken up.

        auto pool2 = new TaskPool();
        auto tSlow2 = task!slowFun();
        pool2.put(tSlow2);
        pool2.finish(true); // blocking
        assert(tSlow2.done);

        // Test fix for Bug 8582 by making pool size zero.
        auto pool3 = new TaskPool(0);
        auto tSlow3 = task!slowFun();
        pool3.put(tSlow3);
        pool3.finish(true); // blocking
        assert(tSlow3.done);

        // This is correct because no thread will terminate unless pool2.status
        // and pool3.status have already been set to stopNow.
        assert(pool2.status == TaskPool.PoolState.stopNow);
        assert(pool3.status == TaskPool.PoolState.stopNow);
    }

    // Test default pool stuff.
    assert(taskPool.size == totalCPUs - 1);

    nums = new uint[1000];
    foreach(i; parallel(iota(1000)))
    {
        nums[i] = cast(uint) i;
    }
    assert(equal(nums, iota(1000)));

    assert(equal(
               poolInstance.map!"a * a"(iota(30_000_001), 10_000),
               std.algorithm.map!"a * a"(iota(30_000_001))
           ));

    // The filter is to kill random access and test the non-random access
    // branch.
    assert(equal(
               poolInstance.map!"a * a"(
                   filter!"a == a"(iota(30_000_001)
                                  ), 10_000, 1000),
               std.algorithm.map!"a * a"(iota(30_000_001))
           ));

    assert(
        reduce!"a + b"(0UL,
                       poolInstance.map!"a * a"(iota(3_000_001), 10_000)
                      ) ==
        reduce!"a + b"(0UL,
                       std.algorithm.map!"a * a"(iota(3_000_001))
                      )
    );

    assert(equal(
               iota(1_000_002),
               poolInstance.asyncBuf(filter!"a == a"(iota(1_000_002)))
           ));

    {
        import std.file : deleteme;

        string temp_file = std.file.deleteme ~ "-tempDelMe.txt";
        auto file = File(temp_file, "wb");
        scope(exit)
        {
            file.close();
            import std.file;
            remove(temp_file);
        }

        auto written = [[1.0, 2, 3], [4.0, 5, 6], [7.0, 8, 9]];
        foreach(row; written)
        {
            file.writeln(join(to!(string[])(row), "\t"));
        }

        file = File(temp_file);

        void next(ref char[] buf)
        {
            file.readln(buf);
            import std.string : chomp;
            buf = chomp(buf);
        }

        double[][] read;
        auto asyncReader = taskPool.asyncBuf(&next, &file.eof);

        foreach(line; asyncReader)
        {
            if(line.length == 0) continue;
            auto ls = line.split("\t");
            read ~= to!(double[])(ls);
        }

        assert(read == written);
        file.close();
    }

    // Test Map/AsyncBuf chaining.

    auto abuf = poolInstance.asyncBuf(iota(-1.0, 3_000_000), 100);
    auto temp = poolInstance.map!sqrt(
                    abuf, 100, 5
                );
    auto lmchain = poolInstance.map!"a * a"(temp, 100, 5);
    lmchain.popFront();

    int ii;
    foreach( elem; (lmchain))
    {
        if(!approxEqual(elem, ii))
        {
            stderr.writeln(ii, '\t', elem);
        }
        ii++;
    }

    // Test buffer trick in parallel foreach.
    abuf = poolInstance.asyncBuf(iota(-1.0, 1_000_000), 100);
    abuf.popFront();
    auto bufTrickTest = new size_t[abuf.length];
    foreach(i, elem; parallel(abuf))
    {
        bufTrickTest[i] = i;
    }

    assert(equal(iota(1_000_000), bufTrickTest));

    auto myTask = task!(std.math.abs)(-1);
    taskPool.put(myTask);
    assert(myTask.spinForce == 1);

    // Test that worker local storage from one pool receives an index of 0
    // when the index is queried w.r.t. another pool.  The only way to do this
    // is non-deterministically.
    foreach(i; parallel(iota(1000), 1))
    {
        assert(poolInstance.workerIndex == 0);
    }

    foreach(i; poolInstance.parallel(iota(1000), 1))
    {
        assert(taskPool.workerIndex == 0);
    }

    // Test exception handling.
    static void parallelForeachThrow()
    {
        foreach(elem; parallel(iota(10)))
        {
            throw new Exception("");
        }
    }

    assertThrown!Exception(parallelForeachThrow());

    static int reduceException(int a, int b)
    {
        throw new Exception("");
    }

    assertThrown!Exception(
        poolInstance.reduce!reduceException(iota(3))
    );

    static int mapException(int a)
    {
        throw new Exception("");
    }

    assertThrown!Exception(
        poolInstance.amap!mapException(iota(3))
    );

    static void mapThrow()
    {
        auto m = poolInstance.map!mapException(iota(3));
        m.popFront();
    }

    assertThrown!Exception(mapThrow());

    struct ThrowingRange
    {
        @property int front()
        {
            return 1;
        }
        void popFront()
        {
            throw new Exception("");
        }
        enum bool empty = false;
    }

    assertThrown!Exception(poolInstance.asyncBuf(ThrowingRange.init));
}

//version = parallelismStressTest;

// These are more like stress tests than real unit tests.  They print out
// tons of stuff and should not be run every time make unittest is run.
version(parallelismStressTest)
{
    unittest
    {
        size_t attempt;
        for(; attempt < 10; attempt++)
            foreach(poolSize; [0, 4])
        {

            poolInstance = new TaskPool(poolSize);

            uint[] numbers = new uint[1_000];

            foreach(i; poolInstance.parallel( iota(0, numbers.length)) )
            {
                numbers[i] = cast(uint) i;
            }

            // Make sure it works.
            foreach(i; 0..numbers.length)
            {
                assert(numbers[i] == i);
            }

            stderr.writeln("Done creating nums.");


            auto myNumbers = filter!"a % 7 > 0"( iota(0, 1000));
            foreach(num; poolInstance.parallel(myNumbers))
            {
                assert(num % 7 > 0 && num < 1000);
            }
            stderr.writeln("Done modulus test.");

            uint[] squares = poolInstance.amap!"a * a"(numbers, 100);
            assert(squares.length == numbers.length);
            foreach(i, number; numbers)
            {
                assert(squares[i] == number * number);
            }
            stderr.writeln("Done squares.");

            auto sumFuture = task!( reduce!"a + b" )(numbers);
            poolInstance.put(sumFuture);

            ulong sumSquares = 0;
            foreach(elem; numbers)
            {
                sumSquares += elem * elem;
            }

            uint mySum = sumFuture.spinForce();
            assert(mySum == 999 * 1000 / 2);

            auto mySumParallel = poolInstance.reduce!"a + b"(numbers);
            assert(mySum == mySumParallel);
            stderr.writeln("Done sums.");

            auto myTask = task(
            {
                synchronized writeln("Our lives are parallel...Our lives are parallel.");
            });
            poolInstance.put(myTask);

            auto nestedOuter = "abcd";
            auto nestedInner =  iota(0, 10, 2);

            foreach(i, letter; poolInstance.parallel(nestedOuter, 1))
            {
                foreach(j, number; poolInstance.parallel(nestedInner, 1))
                {
                    synchronized writeln(i, ": ", letter, "  ", j, ": ", number);
                }
            }

            poolInstance.stop();
        }

        assert(attempt == 10);
        writeln("Press enter to go to next round of unittests.");
        readln();
    }

    // These unittests are intended more for actual testing and not so much
    // as examples.
    unittest
    {
        foreach(attempt; 0..10)
        foreach(poolSize; [0, 4])
        {
            poolInstance = new TaskPool(poolSize);

            // Test indexing.
            stderr.writeln("Creator Raw Index:  ", poolInstance.threadIndex);
            assert(poolInstance.workerIndex() == 0);

            // Test worker-local storage.
            auto workerLocalStorage = poolInstance.workerLocalStorage!uint(1);
            foreach(i; poolInstance.parallel(iota(0U, 1_000_000)))
            {
                workerLocalStorage.get++;
            }
            assert(reduce!"a + b"(workerLocalStorage.toRange) ==
            1_000_000 + poolInstance.size + 1);

            // Make sure work is reasonably balanced among threads.  This test is
            // non-deterministic and is more of a sanity check than something that
            // has an absolute pass/fail.
            shared(uint)[void*] nJobsByThread;
            foreach(thread; poolInstance.pool)
            {
                nJobsByThread[cast(void*) thread] = 0;
            }
            nJobsByThread[ cast(void*) Thread.getThis()] = 0;

            foreach(i; poolInstance.parallel( iota(0, 1_000_000), 100 ))
            {
                atomicOp!"+="( nJobsByThread[ cast(void*) Thread.getThis() ], 1);
            }

            stderr.writeln("\nCurrent thread is:  ",
            cast(void*) Thread.getThis());
            stderr.writeln("Workload distribution:  ");
            foreach(k, v; nJobsByThread)
            {
                stderr.writeln(k, '\t', v);
            }

            // Test whether amap can be nested.
            real[][] matrix = new real[][](1000, 1000);
            foreach(i; poolInstance.parallel( iota(0, matrix.length) ))
            {
                foreach(j; poolInstance.parallel( iota(0, matrix[0].length) ))
                {
                    matrix[i][j] = i * j;
                }
            }

            // Get around weird bugs having to do w/ sqrt being an intrinsic:
            static real mySqrt(real num)
            {
                return sqrt(num);
            }

            static real[] parallelSqrt(real[] nums)
            {
                return poolInstance.amap!mySqrt(nums);
            }

            real[][] sqrtMatrix = poolInstance.amap!parallelSqrt(matrix);

            foreach(i, row; sqrtMatrix)
            {
                foreach(j, elem; row)
                {
                    real shouldBe = sqrt( cast(real) i * j);
                    assert(approxEqual(shouldBe, elem));
                    sqrtMatrix[i][j] = shouldBe;
                }
            }

            auto saySuccess = task(
            {
                stderr.writeln(
                    "Success doing matrix stuff that involves nested pool use.");
            });
            poolInstance.put(saySuccess);
            saySuccess.workForce();

            // A more thorough test of amap, reduce:  Find the sum of the square roots of
            // matrix.

            static real parallelSum(real[] input)
            {
                return poolInstance.reduce!"a + b"(input);
            }

            auto sumSqrt = poolInstance.reduce!"a + b"(
                               poolInstance.amap!parallelSum(
                                   sqrtMatrix
                               )
                           );

            assert(approxEqual(sumSqrt, 4.437e8));
            stderr.writeln("Done sum of square roots.");

            // Test whether tasks work with function pointers.
            auto nanTask = task(&isNaN, 1.0L);
            poolInstance.put(nanTask);
            assert(nanTask.spinForce == false);

            if(poolInstance.size > 0)
            {
                // Test work waiting.
                static void uselessFun()
                {
                    foreach(i; 0..1_000_000) {}
                }

                auto uselessTasks = new typeof(task(&uselessFun))[1000];
                foreach(ref uselessTask; uselessTasks)
                {
                    uselessTask = task(&uselessFun);
                }
                foreach(ref uselessTask; uselessTasks)
                {
                    poolInstance.put(uselessTask);
                }
                foreach(ref uselessTask; uselessTasks)
                {
                    uselessTask.workForce();
                }
            }

            // Test the case of non-random access + ref returns.
            int[] nums = [1,2,3,4,5];
            static struct RemoveRandom
            {
                int[] arr;

                ref int front()
                {
                    return arr.front;
                }
                void popFront()
                {
                    arr.popFront();
                }
                bool empty()
                {
                    return arr.empty;
                }
            }

            auto refRange = RemoveRandom(nums);
            foreach(ref elem; poolInstance.parallel(refRange))
            {
                elem++;
            }
            assert(nums == [2,3,4,5,6], text(nums));
            stderr.writeln("Nums:  ", nums);

            poolInstance.stop();
        }
    }
}

version(unittest)
{
    struct __S_12733
    {
        invariant() { assert(checksum == 1234567890); }
        this(ulong u){n = u;}
        void opAssign(__S_12733 s){this.n = s.n;}
        ulong n;
        ulong checksum = 1234567890;
    }

    static auto __genPair_12733(ulong n) { return __S_12733(n); }
}

unittest
{
    immutable ulong[] data = [ 2UL^^59-1, 2UL^^59-1, 2UL^^59-1, 112_272_537_195_293UL ];

    auto result = taskPool.amap!__genPair_12733(data);
}