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
|
/* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * *
* Copyright by The HDF Group. *
* Copyright by the Board of Trustees of the University of Illinois. *
* All rights reserved. *
* *
* This file is part of HDF5. The full HDF5 copyright notice, including *
* terms governing use, modification, and redistribution, is contained in *
* the COPYING file, which can be found at the root of the source code *
* distribution tree, or in https://www.hdfgroup.org/licenses. *
* If you do not have access to either file, you may request a copy from *
* help@hdfgroup.org. *
* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * */
/*
* Programmer: Robb Matzke
* Thursday, July 29, 1999
*
* Purpose: This is the MPI-2 I/O driver.
*
*/
#include "H5FDdrvr_module.h" /* This source code file is part of the H5FD driver module */
#include "H5private.h" /* Generic Functions */
#include "H5CXprivate.h" /* API Contexts */
#include "H5Dprivate.h" /* Dataset functions */
#include "H5Eprivate.h" /* Error handling */
#include "H5Fprivate.h" /* File access */
#include "H5FDprivate.h" /* File drivers */
#include "H5FDmpi.h" /* MPI-based file drivers */
#include "H5Iprivate.h" /* IDs */
#include "H5MMprivate.h" /* Memory management */
#include "H5Pprivate.h" /* Property lists */
#ifdef H5_HAVE_PARALLEL
/*
* The driver identification number, initialized at runtime if H5_HAVE_PARALLEL
* is defined. This allows applications to still have the H5FD_MPIO
* "constants" in their source code.
*/
static hid_t H5FD_MPIO_g = 0;
/* Whether to allow collective I/O operations */
/* (Can be changed by setting "HDF5_MPI_OPT_TYPES" environment variable to '0' or '1') */
hbool_t H5FD_mpi_opt_types_g = TRUE;
/* Whether the driver initialized MPI on its own */
hbool_t H5FD_mpi_self_initialized = FALSE;
/*
* The view is set to this value
*/
static char H5FD_mpi_native_g[] = "native";
/*
* The description of a file belonging to this driver.
* The EOF value is only used just after the file is opened in order for the
* library to determine whether the file is empty, truncated, or okay. The MPIO
* driver doesn't bother to keep it updated since it's an expensive operation.
*/
typedef struct H5FD_mpio_t {
H5FD_t pub; /* Public stuff, must be first */
MPI_File f; /* MPIO file handle */
MPI_Comm comm; /* MPI Communicator */
MPI_Info info; /* MPI info object */
int mpi_rank; /* This process's rank */
int mpi_size; /* Total number of processes */
haddr_t eof; /* End-of-file marker */
haddr_t eoa; /* End-of-address marker */
haddr_t last_eoa; /* Last known end-of-address marker */
haddr_t local_eof; /* Local end-of-file address for each process */
} H5FD_mpio_t;
/* Private Prototypes */
/* Callbacks */
static herr_t H5FD__mpio_term(void);
static H5FD_t *H5FD__mpio_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t maxaddr);
static herr_t H5FD__mpio_close(H5FD_t *_file);
static herr_t H5FD__mpio_query(const H5FD_t *_f1, unsigned long *flags);
static haddr_t H5FD__mpio_get_eoa(const H5FD_t *_file, H5FD_mem_t type);
static herr_t H5FD__mpio_set_eoa(H5FD_t *_file, H5FD_mem_t type, haddr_t addr);
static haddr_t H5FD__mpio_get_eof(const H5FD_t *_file, H5FD_mem_t type);
static herr_t H5FD__mpio_get_handle(H5FD_t *_file, hid_t fapl, void **file_handle);
static herr_t H5FD__mpio_read(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr, size_t size,
void *buf);
static herr_t H5FD__mpio_write(H5FD_t *_file, H5FD_mem_t type, hid_t dxpl_id, haddr_t addr, size_t size,
const void *buf);
static herr_t H5FD__mpio_read_vector(H5FD_t *_file, hid_t H5_ATTR_UNUSED dxpl_id, uint32_t count,
H5FD_mem_t types[], haddr_t addrs[], size_t sizes[], void *bufs[]);
static herr_t H5FD__mpio_write_vector(H5FD_t *_file, hid_t H5_ATTR_UNUSED dxpl_id, uint32_t count,
H5FD_mem_t types[], haddr_t addrs[], size_t sizes[],
const void *bufs[]);
static herr_t H5FD__mpio_flush(H5FD_t *_file, hid_t dxpl_id, hbool_t closing);
static herr_t H5FD__mpio_truncate(H5FD_t *_file, hid_t dxpl_id, hbool_t closing);
static herr_t H5FD__mpio_delete(const char *filename, hid_t fapl_id);
static herr_t H5FD__mpio_ctl(H5FD_t *_file, uint64_t op_code, uint64_t flags, const void *input,
void **output);
/* Other functions */
static herr_t H5FD__mpio_vector_build_types(
uint32_t count, H5FD_mem_t types[], haddr_t addrs[], size_t sizes[], H5_flexible_const_ptr_t bufs[],
haddr_t *s_addrs[], size_t *s_sizes[], H5_flexible_const_ptr_t *s_bufs[], hbool_t *vector_was_sorted,
MPI_Offset *mpi_off, H5_flexible_const_ptr_t *mpi_bufs_base, int *size_i, MPI_Datatype *buf_type,
hbool_t *buf_type_created, MPI_Datatype *file_type, hbool_t *file_type_created, char *unused);
/* The MPIO file driver information */
static const H5FD_class_t H5FD_mpio_g = {
H5FD_CLASS_VERSION, /* struct version */
H5_VFD_MPIO, /* value */
"mpio", /* name */
HADDR_MAX, /* maxaddr */
H5F_CLOSE_SEMI, /* fc_degree */
H5FD__mpio_term, /* terminate */
NULL, /* sb_size */
NULL, /* sb_encode */
NULL, /* sb_decode */
0, /* fapl_size */
NULL, /* fapl_get */
NULL, /* fapl_copy */
NULL, /* fapl_free */
0, /* dxpl_size */
NULL, /* dxpl_copy */
NULL, /* dxpl_free */
H5FD__mpio_open, /* open */
H5FD__mpio_close, /* close */
NULL, /* cmp */
H5FD__mpio_query, /* query */
NULL, /* get_type_map */
NULL, /* alloc */
NULL, /* free */
H5FD__mpio_get_eoa, /* get_eoa */
H5FD__mpio_set_eoa, /* set_eoa */
H5FD__mpio_get_eof, /* get_eof */
H5FD__mpio_get_handle, /* get_handle */
H5FD__mpio_read, /* read */
H5FD__mpio_write, /* write */
H5FD__mpio_read_vector, /* read_vector */
H5FD__mpio_write_vector, /* write_vector */
NULL, /* read_selection */
NULL, /* write_selection */
H5FD__mpio_flush, /* flush */
H5FD__mpio_truncate, /* truncate */
NULL, /* lock */
NULL, /* unlock */
H5FD__mpio_delete, /* del */
H5FD__mpio_ctl, /* ctl */
H5FD_FLMAP_DICHOTOMY /* fl_map */
};
#ifdef H5FDmpio_DEBUG
/* Flags to control debug actions in the MPI-IO VFD.
* (Meant to be indexed by characters)
*
* These flags can be set with either (or both) the environment variable
* "H5FD_mpio_Debug" set to a string containing one or more characters
* (flags) or by setting them as a string value for the
* "H5F_mpio_debug_key" MPI Info key.
*
* Supported characters in 'H5FD_mpio_Debug' string:
* 't' trace function entry and exit
* 'r' show read offset and size
* 'w' show write offset and size
* '0'-'9' only show output from a single MPI rank (ranks 0-9 supported)
*/
static int H5FD_mpio_debug_flags_s[256];
static int H5FD_mpio_debug_rank_s = -1;
/* Indicate if this rank should output tracing info */
#define H5FD_MPIO_TRACE_THIS_RANK(file) \
(H5FD_mpio_debug_rank_s < 0 || H5FD_mpio_debug_rank_s == (file)->mpi_rank)
#endif
#ifdef H5FDmpio_DEBUG
/*---------------------------------------------------------------------------
* Function: H5FD__mpio_parse_debug_str
*
* Purpose: Parse a string for debugging flags
*
* Returns: N/A
*
* Programmer: Quincey Koziol
* Wednesday, Aug 12, 2020
*
*---------------------------------------------------------------------------
*/
static void
H5FD__mpio_parse_debug_str(const char *s)
{
FUNC_ENTER_PACKAGE_NOERR
/* Sanity check */
HDassert(s);
/* Set debug mask */
while (*s) {
if ((int)(*s) >= (int)'0' && (int)(*s) <= (int)'9')
H5FD_mpio_debug_rank_s = ((int)*s) - (int)'0';
else
H5FD_mpio_debug_flags_s[(int)*s]++;
s++;
} /* end while */
FUNC_LEAVE_NOAPI_VOID
} /* end H5FD__mpio_parse_debug_str() */
/*---------------------------------------------------------------------------
* Function: H5FD__mem_t_to_str
*
* Purpose: Returns a string representing the enum value in an H5FD_mem_t
* enum
*
* Returns: H5FD_mem_t enum value string
*
*---------------------------------------------------------------------------
*/
static const char *
H5FD__mem_t_to_str(H5FD_mem_t mem_type)
{
switch (mem_type) {
case H5FD_MEM_NOLIST:
return "H5FD_MEM_NOLIST";
case H5FD_MEM_DEFAULT:
return "H5FD_MEM_DEFAULT";
case H5FD_MEM_SUPER:
return "H5FD_MEM_SUPER";
case H5FD_MEM_BTREE:
return "H5FD_MEM_BTREE";
case H5FD_MEM_DRAW:
return "H5FD_MEM_DRAW";
case H5FD_MEM_GHEAP:
return "H5FD_MEM_GHEAP";
case H5FD_MEM_LHEAP:
return "H5FD_MEM_LHEAP";
case H5FD_MEM_OHDR:
return "H5FD_MEM_OHDR";
default:
return "(Unknown)";
}
}
#endif /* H5FDmpio_DEBUG */
/*-------------------------------------------------------------------------
* Function: H5FD_mpio_init
*
* Purpose: Initialize this driver by registering the driver with the
* library.
*
* Return: Success: The driver ID for the mpio driver
* Failure: H5I_INVALID_HID
*
* Programmer: Robb Matzke
* Thursday, August 5, 1999
*
*-------------------------------------------------------------------------
*/
hid_t
H5FD_mpio_init(void)
{
static int H5FD_mpio_Debug_inited = 0;
char * env = NULL;
hid_t ret_value = H5I_INVALID_HID; /* Return value */
FUNC_ENTER_NOAPI(H5I_INVALID_HID)
/* Register the MPI-IO VFD, if it isn't already */
if (H5I_VFL != H5I_get_type(H5FD_MPIO_g)) {
H5FD_MPIO_g = H5FD_register((const H5FD_class_t *)&H5FD_mpio_g, sizeof(H5FD_class_t), FALSE);
/* Check if MPI driver has been loaded dynamically */
env = HDgetenv(HDF5_DRIVER);
if (env && !HDstrcmp(env, "mpio")) {
int mpi_initialized = 0;
/* Initialize MPI if not already initialized */
if (MPI_SUCCESS != MPI_Initialized(&mpi_initialized))
HGOTO_ERROR(H5E_VFL, H5E_UNINITIALIZED, H5I_INVALID_HID, "can't check if MPI is initialized")
if (!mpi_initialized) {
if (MPI_SUCCESS != MPI_Init(NULL, NULL))
HGOTO_ERROR(H5E_VFL, H5E_CANTINIT, H5I_INVALID_HID, "can't initialize MPI")
H5FD_mpi_self_initialized = TRUE;
}
}
}
if (!H5FD_mpio_Debug_inited) {
const char *s; /* String for environment variables */
/* Allow MPI buf-and-file-type optimizations? */
s = HDgetenv("HDF5_MPI_OPT_TYPES");
if (s && HDisdigit(*s))
H5FD_mpi_opt_types_g = (0 == HDstrtol(s, NULL, 0)) ? FALSE : TRUE;
#ifdef H5FDmpio_DEBUG
/* Clear the flag buffer */
HDmemset(H5FD_mpio_debug_flags_s, 0, sizeof(H5FD_mpio_debug_flags_s));
/* Retrieve MPI-IO debugging environment variable */
s = HDgetenv("H5FD_mpio_Debug");
if (s)
H5FD__mpio_parse_debug_str(s);
#endif /* H5FDmpio_DEBUG */
H5FD_mpio_Debug_inited++;
} /* end if */
/* Set return value */
ret_value = H5FD_MPIO_g;
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD_mpio_init() */
/*---------------------------------------------------------------------------
* Function: H5FD__mpio_term
*
* Purpose: Shut down the VFD
*
* Returns: Non-negative on success or negative on failure
*
* Programmer: Quincey Koziol
* Friday, Jan 30, 2004
*
*---------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_term(void)
{
FUNC_ENTER_PACKAGE_NOERR
/* Terminate MPI if the driver initialized it */
if (H5FD_mpi_self_initialized) {
int mpi_finalized = 0;
MPI_Finalized(&mpi_finalized);
if (!mpi_finalized)
MPI_Finalize();
H5FD_mpi_self_initialized = FALSE;
}
/* Reset VFL ID */
H5FD_MPIO_g = 0;
FUNC_LEAVE_NOAPI(SUCCEED)
} /* end H5FD__mpio_term() */
/*-------------------------------------------------------------------------
* Function: H5Pset_fapl_mpio
*
* Purpose: Store the user supplied MPIO communicator comm and info in
* the file access property list FAPL_ID which can then be used
* to create and/or open the file. This function is available
* only in the parallel HDF5 library and is not collective.
*
* comm is the MPI communicator to be used for file open as
* defined in MPI_FILE_OPEN of MPI-2. This function makes a
* duplicate of comm. Any modification to comm after this function
* call returns has no effect on the access property list.
*
* info is the MPI Info object to be used for file open as
* defined in MPI_FILE_OPEN of MPI-2. This function makes a
* duplicate of info. Any modification to info after this
* function call returns has no effect on the access property
* list.
*
* If fapl_id has previously set comm and info values, they
* will be replaced and the old communicator and Info object
* are freed.
*
* Return: Success: Non-negative
* Failure: Negative
*
* Programmer: Albert Cheng
* Feb 3, 1998
*
*-------------------------------------------------------------------------
*/
herr_t
H5Pset_fapl_mpio(hid_t fapl_id, MPI_Comm comm, MPI_Info info)
{
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value;
FUNC_ENTER_API(FAIL)
H5TRACE3("e", "iMcMi", fapl_id, comm, info);
/* Check arguments */
if (fapl_id == H5P_DEFAULT)
HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "can't set values in default property list")
if (NULL == (plist = H5P_object_verify(fapl_id, H5P_FILE_ACCESS)))
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a file access list")
if (MPI_COMM_NULL == comm)
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "MPI_COMM_NULL is not a valid communicator")
/* Set the MPI communicator and info object */
if (H5P_set(plist, H5F_ACS_MPI_PARAMS_COMM_NAME, &comm) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "can't set MPI communicator")
if (H5P_set(plist, H5F_ACS_MPI_PARAMS_INFO_NAME, &info) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "can't set MPI info object")
/* duplication is done during driver setting. */
ret_value = H5P_set_driver(plist, H5FD_MPIO, NULL, NULL);
done:
FUNC_LEAVE_API(ret_value)
} /* H5Pset_fapl_mpio() */
/*-------------------------------------------------------------------------
* Function: H5Pget_fapl_mpio
*
* Purpose: If the file access property list is set to the H5FD_MPIO
* driver then this function returns duplicates of the MPI
* communicator and Info object stored through the comm and
* info pointers. It is the responsibility of the application
* to free the returned communicator and Info object.
*
* Return: Success: Non-negative with the communicator and
* Info object returned through the comm and
* info arguments if non-null. Since they are
* duplicates of the stored objects, future
* modifications to the access property list do
* not affect them and it is the responsibility
* of the application to free them.
* Failure: Negative
*
* Programmer: Robb Matzke
* Thursday, February 26, 1998
*
*-------------------------------------------------------------------------
*/
herr_t
H5Pget_fapl_mpio(hid_t fapl_id, MPI_Comm *comm /*out*/, MPI_Info *info /*out*/)
{
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_API(FAIL)
H5TRACE3("e", "ixx", fapl_id, comm, info);
/* Set comm and info in case we have problems */
if (comm)
*comm = MPI_COMM_NULL;
if (info)
*info = MPI_INFO_NULL;
/* Check arguments */
if (NULL == (plist = H5P_object_verify(fapl_id, H5P_FILE_ACCESS)))
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a file access list")
if (H5FD_MPIO != H5P_peek_driver(plist))
HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "VFL driver is not MPI-I/O")
/* Get the MPI communicator and info object */
if (comm)
if (H5P_get(plist, H5F_ACS_MPI_PARAMS_COMM_NAME, comm) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, FAIL, "can't get MPI communicator")
if (info)
if (H5P_get(plist, H5F_ACS_MPI_PARAMS_INFO_NAME, info) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTGET, FAIL, "can't get MPI info object")
done:
/* Clean up anything duplicated on errors. The free calls will set
* the output values to MPI_COMM|INFO_NULL.
*/
if (ret_value != SUCCEED) {
if (comm)
if (H5_mpi_comm_free(comm) < 0)
HDONE_ERROR(H5E_PLIST, H5E_CANTFREE, FAIL, "unable to free MPI communicator")
if (info)
if (H5_mpi_info_free(info) < 0)
HDONE_ERROR(H5E_PLIST, H5E_CANTFREE, FAIL, "unable to free MPI info object")
}
FUNC_LEAVE_API(ret_value)
} /* end H5Pget_fapl_mpio() */
/*-------------------------------------------------------------------------
* Function: H5Pset_dxpl_mpio
*
* Purpose: Set the data transfer property list DXPL_ID to use transfer
* mode XFER_MODE. The property list can then be used to control
* the I/O transfer mode during data I/O operations. The valid
* transfer modes are:
*
* H5FD_MPIO_INDEPENDENT:
* Use independent I/O access (the default).
*
* H5FD_MPIO_COLLECTIVE:
* Use collective I/O access.
*
* Return: Success: Non-negative
* Failure: Negative
*
* Programmer: Albert Cheng
* April 2, 1998
*
*-------------------------------------------------------------------------
*/
herr_t
H5Pset_dxpl_mpio(hid_t dxpl_id, H5FD_mpio_xfer_t xfer_mode)
{
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_API(FAIL)
H5TRACE2("e", "iDt", dxpl_id, xfer_mode);
/* Check arguments */
if (dxpl_id == H5P_DEFAULT)
HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "can't set values in default property list")
if (NULL == (plist = H5P_object_verify(dxpl_id, H5P_DATASET_XFER)))
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a dxpl")
if (H5FD_MPIO_INDEPENDENT != xfer_mode && H5FD_MPIO_COLLECTIVE != xfer_mode)
HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "incorrect xfer_mode")
/* Set the transfer mode */
if (H5P_set(plist, H5D_XFER_IO_XFER_MODE_NAME, &xfer_mode) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "unable to set value")
done:
FUNC_LEAVE_API(ret_value)
} /* end H5Pset_dxpl_mpio() */
/*-------------------------------------------------------------------------
* Function: H5Pget_dxpl_mpio
*
* Purpose: Queries the transfer mode current set in the data transfer
* property list DXPL_ID. This is not collective.
*
* Return: Success: Non-negative, with the transfer mode returned
* through the XFER_MODE argument if it is
* non-null.
* Failure: Negative
*
* Programmer: Albert Cheng
* April 2, 1998
*
*-------------------------------------------------------------------------
*/
herr_t
H5Pget_dxpl_mpio(hid_t dxpl_id, H5FD_mpio_xfer_t *xfer_mode /*out*/)
{
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_API(FAIL)
H5TRACE2("e", "ix", dxpl_id, xfer_mode);
/* Check arguments */
if (NULL == (plist = H5P_object_verify(dxpl_id, H5P_DATASET_XFER)))
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a dxpl")
/* Get the transfer mode */
if (xfer_mode)
if (H5P_get(plist, H5D_XFER_IO_XFER_MODE_NAME, xfer_mode) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "unable to get value")
done:
FUNC_LEAVE_API(ret_value)
} /* end H5Pget_dxpl_mpio() */
/*-------------------------------------------------------------------------
* Function: H5Pset_dxpl_mpio_collective_opt
*
* Purpose: To set a flag to choose linked chunk I/O or multi-chunk I/O
* without involving decision-making inside HDF5
*
* Note: The library will do linked chunk I/O or multi-chunk I/O without
* involving communications for decision-making process.
* The library won't behave as it asks for only when we find
* that the low-level MPI-IO package doesn't support this.
*
* Return: Success: Non-negative
* Failure: Negative
*
* Programmer: Kent Yang
*
*-------------------------------------------------------------------------
*/
herr_t
H5Pset_dxpl_mpio_collective_opt(hid_t dxpl_id, H5FD_mpio_collective_opt_t opt_mode)
{
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_API(FAIL)
H5TRACE2("e", "iDc", dxpl_id, opt_mode);
/* Check arguments */
if (dxpl_id == H5P_DEFAULT)
HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "can't set values in default property list")
if (NULL == (plist = H5P_object_verify(dxpl_id, H5P_DATASET_XFER)))
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a dxpl")
/* Set the transfer mode */
if (H5P_set(plist, H5D_XFER_MPIO_COLLECTIVE_OPT_NAME, &opt_mode) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "unable to set value")
done:
FUNC_LEAVE_API(ret_value)
} /* end H5Pset_dxpl_mpio_collective_opt() */
/*-------------------------------------------------------------------------
* Function: H5Pset_dxpl_mpio_chunk_opt
*
* Purpose: To set a flag to choose linked chunk I/O or multi-chunk I/O
* without involving decision-making inside HDF5
*
* Note: The library will do linked chunk I/O or multi-chunk I/O without
* involving communications for decision-making process.
* The library won't behave as it asks for only when we find
* that the low-level MPI-IO package doesn't support this.
*
* Return: Success: Non-negative
* Failure: Negative
*
* Programmer: Kent Yang
*
*-------------------------------------------------------------------------
*/
herr_t
H5Pset_dxpl_mpio_chunk_opt(hid_t dxpl_id, H5FD_mpio_chunk_opt_t opt_mode)
{
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_API(FAIL)
H5TRACE2("e", "iDh", dxpl_id, opt_mode);
/* Check arguments */
if (dxpl_id == H5P_DEFAULT)
HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "can't set values in default property list")
if (NULL == (plist = H5P_object_verify(dxpl_id, H5P_DATASET_XFER)))
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a dxpl")
/* Set the transfer mode */
if (H5P_set(plist, H5D_XFER_MPIO_CHUNK_OPT_HARD_NAME, &opt_mode) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "unable to set value")
done:
FUNC_LEAVE_API(ret_value)
} /* end H5Pset_dxpl_mpio_chunk_opt() */
/*-------------------------------------------------------------------------
* Function: H5Pset_dxpl_mpio_chunk_opt_num
*
* Purpose: To set a threshold for doing linked chunk IO
*
* Note: If the number is greater than the threshold set by the user,
* the library will do linked chunk I/O; otherwise, I/O will be
* done for every chunk.
*
* Return: Success: Non-negative
* Failure: Negative
*
* Programmer: Kent Yang
*
*-------------------------------------------------------------------------
*/
herr_t
H5Pset_dxpl_mpio_chunk_opt_num(hid_t dxpl_id, unsigned num_chunk_per_proc)
{
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_API(FAIL)
H5TRACE2("e", "iIu", dxpl_id, num_chunk_per_proc);
/* Check arguments */
if (dxpl_id == H5P_DEFAULT)
HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "can't set values in default property list")
if (NULL == (plist = H5P_object_verify(dxpl_id, H5P_DATASET_XFER)))
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a dxpl")
/* Set the transfer mode */
if (H5P_set(plist, H5D_XFER_MPIO_CHUNK_OPT_NUM_NAME, &num_chunk_per_proc) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "unable to set value")
done:
FUNC_LEAVE_API(ret_value)
} /* end H5Pset_dxpl_mpio_chunk_opt_num() */
/*-------------------------------------------------------------------------
* Function: H5Pset_dxpl_mpio_chunk_opt_ratio
*
* Purpose: To set a threshold for doing collective I/O for each chunk
*
* Note: The library will calculate the percentage of the number of
* process holding selections at each chunk. If that percentage
* of number of process in the individual chunk is greater than
* the threshold set by the user, the library will do collective
* chunk I/O for this chunk; otherwise, independent I/O will be
* done for this chunk.
*
* Return: Success: Non-negative
* Failure: Negative
*
* Programmer: Kent Yang
*
*-------------------------------------------------------------------------
*/
herr_t
H5Pset_dxpl_mpio_chunk_opt_ratio(hid_t dxpl_id, unsigned percent_num_proc_per_chunk)
{
H5P_genplist_t *plist; /* Property list pointer */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_API(FAIL)
H5TRACE2("e", "iIu", dxpl_id, percent_num_proc_per_chunk);
/* Check arguments */
if (dxpl_id == H5P_DEFAULT)
HGOTO_ERROR(H5E_PLIST, H5E_BADVALUE, FAIL, "can't set values in default property list")
if (NULL == (plist = H5P_object_verify(dxpl_id, H5P_DATASET_XFER)))
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL, "not a dxpl")
/* Set the transfer mode */
if (H5P_set(plist, H5D_XFER_MPIO_CHUNK_OPT_RATIO_NAME, &percent_num_proc_per_chunk) < 0)
HGOTO_ERROR(H5E_PLIST, H5E_CANTSET, FAIL, "unable to set value")
done:
FUNC_LEAVE_API(ret_value)
} /* end H5Pset_dxpl_mpio_chunk_opt_ratio() */
/*-------------------------------------------------------------------------
* Function: H5FD_set_mpio_atomicity
*
* Purpose: Sets the atomicity mode
*
* Return: SUCCEED/FAIL
*
* Programmer: Mohamad Chaarawi
* Feb 14, 2012
*
*-------------------------------------------------------------------------
*/
herr_t
H5FD_set_mpio_atomicity(H5FD_t *_file, hbool_t flag)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
#endif
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* set atomicity value */
if (MPI_SUCCESS != (mpi_code = MPI_File_set_atomicity(file->f, (int)(flag != FALSE))))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_atomicity", mpi_code)
done:
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Leaving\n", __func__, file->mpi_rank);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD_set_mpio_atomicity() */
/*-------------------------------------------------------------------------
* Function: H5FD_get_mpio_atomicity
*
* Purpose: Returns the atomicity mode
*
* Return: SUCCEED/FAIL
*
* Programmer: Mohamad Chaarawi
* Feb 14, 2012
*
*-------------------------------------------------------------------------
*/
herr_t
H5FD_get_mpio_atomicity(H5FD_t *_file, hbool_t *flag)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
int temp_flag;
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
#endif
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED;
FUNC_ENTER_NOAPI_NOINIT
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* Get atomicity value */
if (MPI_SUCCESS != (mpi_code = MPI_File_get_atomicity(file->f, &temp_flag)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_get_atomicity", mpi_code)
if (0 != temp_flag)
*flag = TRUE;
else
*flag = FALSE;
done:
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Leaving\n", __func__, file->mpi_rank);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD_get_mpio_atomicity() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_open
*
* Purpose: Opens a file with name NAME. The FLAGS are a bit field with
* purpose similar to the second argument of open(2) and which
* are defined in H5Fpublic.h. The file access property list
* FAPL_ID contains the properties driver properties and MAXADDR
* is the largest address which this file will be expected to
* access. This is collective.
*
* Return: Success: A new file pointer
* Failure: NULL
*
* Programmer: Robert Kim Yates
* January 30, 1998
*
*-------------------------------------------------------------------------
*/
static H5FD_t *
H5FD__mpio_open(const char *name, unsigned flags, hid_t fapl_id, haddr_t H5_ATTR_UNUSED maxaddr)
{
H5FD_mpio_t * file = NULL; /* VFD File struct for new file */
H5P_genplist_t *plist; /* Property list pointer */
MPI_Comm comm = MPI_COMM_NULL; /* MPI Communicator, from plist */
MPI_Info info = MPI_INFO_NULL; /* MPI Info, from plist */
MPI_Info info_used; /* MPI Info returned from MPI_File_open */
MPI_File fh; /* MPI file handle */
hbool_t file_opened = FALSE; /* Flag to indicate that the file was successfully opened */
int mpi_amode; /* MPI file access flags */
int mpi_rank = INT_MAX; /* MPI rank of this process */
int mpi_size; /* Total number of MPI processes */
MPI_Offset file_size; /* File size (of existing files) */
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = FALSE;
#endif
int mpi_code; /* MPI return code */
H5FD_t *ret_value = NULL; /* Return value */
FUNC_ENTER_PACKAGE
/* Get a pointer to the fapl */
if (NULL == (plist = H5P_object_verify(fapl_id, H5P_FILE_ACCESS)))
HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, NULL, "not a file access property list")
if (H5FD_mpi_self_initialized) {
comm = MPI_COMM_WORLD;
}
else {
/* Get the MPI communicator and info object from the property list */
if (H5P_get(plist, H5F_ACS_MPI_PARAMS_COMM_NAME, &comm) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, NULL, "can't get MPI communicator")
if (H5P_get(plist, H5F_ACS_MPI_PARAMS_INFO_NAME, &info) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, NULL, "can't get MPI info object")
}
/* Get the MPI rank of this process and the total number of processes */
if (MPI_SUCCESS != (mpi_code = MPI_Comm_rank(comm, &mpi_rank)))
HMPI_GOTO_ERROR(NULL, "MPI_Comm_rank failed", mpi_code)
if (MPI_SUCCESS != (mpi_code = MPI_Comm_size(comm, &mpi_size)))
HMPI_GOTO_ERROR(NULL, "MPI_Comm_size failed", mpi_code)
#ifdef H5FDmpio_DEBUG
H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] &&
(H5FD_mpio_debug_rank_s < 0 || H5FD_mpio_debug_rank_s == mpi_rank));
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering - name = \"%s\", flags = 0x%x, fapl_id = %d, maxaddr = %lu\n",
__func__, mpi_rank, name, flags, (int)fapl_id, (unsigned long)maxaddr);
#endif
/* Convert HDF5 flags to MPI-IO flags */
/* Some combinations are illegal; let MPI-IO figure it out */
mpi_amode = (flags & H5F_ACC_RDWR) ? MPI_MODE_RDWR : MPI_MODE_RDONLY;
if (flags & H5F_ACC_CREAT)
mpi_amode |= MPI_MODE_CREATE;
if (flags & H5F_ACC_EXCL)
mpi_amode |= MPI_MODE_EXCL;
#ifdef H5FDmpio_DEBUG
/* Check for debug commands in the info parameter */
if (MPI_INFO_NULL != info) {
char debug_str[128];
int flag;
MPI_Info_get(info, H5F_MPIO_DEBUG_KEY, sizeof(debug_str) - 1, debug_str, &flag);
if (flag)
H5FD__mpio_parse_debug_str(debug_str);
} /* end if */
#endif
if (MPI_SUCCESS != (mpi_code = MPI_File_open(comm, name, mpi_amode, info, &fh)))
HMPI_GOTO_ERROR(NULL, "MPI_File_open failed", mpi_code)
file_opened = TRUE;
/* Get the MPI-IO hints that actually used by MPI-IO underneath. */
if (MPI_SUCCESS != (mpi_code = MPI_File_get_info(fh, &info_used)))
HMPI_GOTO_ERROR(NULL, "MPI_File_get_info failed", mpi_code)
/* Copy hints in info_used into info. Note hints in info_used supersede
* info. There may be some hints set and used by HDF5 only, but not
* recognizable by MPI-IO. We need to keep them, as MPI_File_get_info()
* will remove any hints unrecognized by MPI-IO library underneath.
*/
if (info_used != MPI_INFO_NULL) {
int i, nkeys;
if (info == MPI_INFO_NULL) /* reuse info created from MPI_File_get_info() */
info = info_used;
else {
/* retrieve the number of hints */
if (MPI_SUCCESS != (mpi_code = MPI_Info_get_nkeys(info_used, &nkeys)))
HMPI_GOTO_ERROR(NULL, "MPI_Info_get_nkeys failed", mpi_code)
/* copy over each hint */
for (i = 0; i < nkeys; i++) {
char key[MPI_MAX_INFO_KEY], value[MPI_MAX_INFO_VAL];
int valuelen, flag;
/* retrieve the nth hint */
if (MPI_SUCCESS != (mpi_code = MPI_Info_get_nthkey(info_used, i, key)))
HMPI_GOTO_ERROR(NULL, "MPI_Info_get_nkeys failed", mpi_code)
/* retrieve the key of nth hint */
if (MPI_SUCCESS != (mpi_code = MPI_Info_get_valuelen(info_used, key, &valuelen, &flag)))
HMPI_GOTO_ERROR(NULL, "MPI_Info_get_valuelen failed", mpi_code)
/* retrieve the value of nth hint */
if (MPI_SUCCESS != (mpi_code = MPI_Info_get(info_used, key, valuelen + 1, value, &flag)))
HMPI_GOTO_ERROR(NULL, "MPI_Info_get failed", mpi_code)
/* copy the hint into info */
if (MPI_SUCCESS != (mpi_code = MPI_Info_set(info, key, value)))
HMPI_GOTO_ERROR(NULL, "MPI_Info_set failed", mpi_code)
}
/* Free info_used allocated in the call to MPI_File_get_info() */
if (MPI_SUCCESS != (mpi_code = MPI_Info_free(&info_used)))
HMPI_GOTO_ERROR(NULL, "MPI_Info_free failed", mpi_code)
}
/* Add info to the file access property list */
if (H5P_set(plist, H5F_ACS_MPI_PARAMS_INFO_NAME, &info) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTSET, NULL, "can't set MPI info object")
}
/* Build the return value and initialize it */
if (NULL == (file = (H5FD_mpio_t *)H5MM_calloc(sizeof(H5FD_mpio_t))))
HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, NULL, "memory allocation failed")
file->f = fh;
file->comm = comm;
file->info = info;
file->mpi_rank = mpi_rank;
file->mpi_size = mpi_size;
/* Only processor p0 will get the filesize and broadcast it. */
if (mpi_rank == 0) {
/* If MPI_File_get_size fails, broadcast file size as -1 to signal error */
if (MPI_SUCCESS != (mpi_code = MPI_File_get_size(fh, &file_size)))
file_size = (MPI_Offset)-1;
}
/* Broadcast file size */
if (MPI_SUCCESS != (mpi_code = MPI_Bcast(&file_size, (int)sizeof(MPI_Offset), MPI_BYTE, 0, comm)))
HMPI_GOTO_ERROR(NULL, "MPI_Bcast failed", mpi_code)
if (file_size < 0)
HMPI_GOTO_ERROR(NULL, "MPI_File_get_size failed", mpi_code)
/* Determine if the file should be truncated */
if (file_size && (flags & H5F_ACC_TRUNC)) {
/* Truncate the file */
if (MPI_SUCCESS != (mpi_code = MPI_File_set_size(fh, (MPI_Offset)0)))
HMPI_GOTO_ERROR(NULL, "MPI_File_set_size failed", mpi_code)
/* Don't let any proc return until all have truncated the file. */
if (MPI_SUCCESS != (mpi_code = MPI_Barrier(comm)))
HMPI_GOTO_ERROR(NULL, "MPI_Barrier failed", mpi_code)
/* File is zero size now */
file_size = 0;
} /* end if */
/* Set the size of the file (from library's perspective) */
file->eof = H5FD_mpi_MPIOff_to_haddr(file_size);
file->local_eof = file->eof;
/* Set return value */
ret_value = (H5FD_t *)file;
done:
if (ret_value == NULL) {
if (file_opened)
MPI_File_close(&fh);
if (H5_mpi_comm_free(&comm) < 0)
HDONE_ERROR(H5E_VFL, H5E_CANTFREE, NULL, "unable to free MPI communicator")
if (H5_mpi_info_free(&info) < 0)
HDONE_ERROR(H5E_VFL, H5E_CANTFREE, NULL, "unable to free MPI info object")
if (file)
H5MM_xfree(file);
} /* end if */
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Leaving\n", __func__, mpi_rank);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_open() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_close
*
* Purpose: Closes a file. This is collective.
*
* Return: SUCCEED/FAIL
*
* Programmer: Unknown
* January 30, 1998
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_close(H5FD_t *_file)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
int mpi_rank = file->mpi_rank;
#endif
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_PACKAGE
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
/* MPI_File_close sets argument to MPI_FILE_NULL */
if (MPI_SUCCESS != (mpi_code = MPI_File_close(&(file->f))))
HMPI_GOTO_ERROR(FAIL, "MPI_File_close failed", mpi_code)
/* Clean up other stuff */
H5_mpi_comm_free(&file->comm);
H5_mpi_info_free(&file->info);
H5MM_xfree(file);
done:
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Leaving\n", __func__, mpi_rank);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_close() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_query
*
* Purpose: Set the flags that this VFL driver is capable of supporting.
* (listed in H5FDpublic.h)
*
* Return: SUCCEED/FAIL
*
* Programmer: Quincey Koziol
* Friday, August 25, 2000
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_query(const H5FD_t H5_ATTR_UNUSED *_file, unsigned long *flags /* out */)
{
FUNC_ENTER_PACKAGE_NOERR
/* Set the VFL feature flags that this driver supports */
if (flags) {
*flags = 0;
*flags |= H5FD_FEAT_AGGREGATE_METADATA; /* OK to aggregate metadata allocations */
*flags |= H5FD_FEAT_AGGREGATE_SMALLDATA; /* OK to aggregate "small" raw data allocations */
*flags |= H5FD_FEAT_HAS_MPI; /* This driver uses MPI */
*flags |= H5FD_FEAT_DEFAULT_VFD_COMPATIBLE; /* VFD creates a file which can be opened with the default
VFD */
} /* end if */
FUNC_LEAVE_NOAPI(SUCCEED)
} /* end H5FD__mpio_query() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_get_eoa
*
* Purpose: Gets the end-of-address marker for the file. The EOA marker
* is the first address past the last byte allocated in the
* format address space.
*
* Return: Success: The end-of-address marker
* Failure: HADDR_UNDEF
*
* Programmer: Robb Matzke
* Friday, August 6, 1999
*
*-------------------------------------------------------------------------
*/
static haddr_t
H5FD__mpio_get_eoa(const H5FD_t *_file, H5FD_mem_t H5_ATTR_UNUSED type)
{
const H5FD_mpio_t *file = (const H5FD_mpio_t *)_file;
FUNC_ENTER_PACKAGE_NOERR
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
FUNC_LEAVE_NOAPI(file->eoa)
} /* end H5FD__mpio_get_eoa() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_set_eoa
*
* Purpose: Set the end-of-address marker for the file. This function is
* called shortly after an existing HDF5 file is opened in order
* to tell the driver where the end of the HDF5 data is located.
*
* Return: SUCCEED/FAIL
*
* Programmer: Robb Matzke
* Friday, August 6, 1999
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_set_eoa(H5FD_t *_file, H5FD_mem_t H5_ATTR_UNUSED type, haddr_t addr)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
FUNC_ENTER_PACKAGE_NOERR
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
file->eoa = addr;
FUNC_LEAVE_NOAPI(SUCCEED)
} /* end H5FD__mpio_set_eoa() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_get_eof
*
* Purpose: Gets the end-of-file marker for the file. The EOF marker
* is the real size of the file.
*
* The MPIO driver doesn't bother keeping this field updated
* since that's a relatively expensive operation. Fortunately
* the library only needs the EOF just after the file is opened
* in order to determine whether the file is empty, truncated,
* or okay. Therefore, any MPIO I/O function will set its value
* to HADDR_UNDEF which is the error return value of this
* function.
*
* Keeping the EOF updated (during write calls) is expensive
* because any process may extend the physical end of the
* file. -QAK
*
* Return: Success: The end-of-file marker
* Failure: HADDR_UNDEF
*
* Programmer: Robb Matzke
* Friday, August 6, 1999
*
*-------------------------------------------------------------------------
*/
static haddr_t
H5FD__mpio_get_eof(const H5FD_t *_file, H5FD_mem_t H5_ATTR_UNUSED type)
{
const H5FD_mpio_t *file = (const H5FD_mpio_t *)_file;
FUNC_ENTER_PACKAGE_NOERR
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
FUNC_LEAVE_NOAPI(file->eof)
} /* end H5FD__mpio_get_eof() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_get_handle
*
* Purpose: Returns the file handle of MPIO file driver.
*
* Returns: SUCCEED/FAIL
*
* Programmer: Raymond Lu
* Sept. 16, 2002
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_get_handle(H5FD_t *_file, hid_t H5_ATTR_UNUSED fapl, void **file_handle)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
herr_t ret_value = SUCCEED;
FUNC_ENTER_PACKAGE
if (!file_handle)
HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "file handle not valid")
*file_handle = &(file->f);
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_get_handle() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_read
*
* Purpose: Reads SIZE bytes of data from FILE beginning at address ADDR
* into buffer BUF according to data transfer properties in
* DXPL_ID using potentially complex file and buffer types to
* effect the transfer.
*
* Reading past the end of the MPI file returns zeros instead of
* failing. MPI is able to coalesce requests from different
* processes (collective or independent).
*
* Return: Success: SUCCEED. Result is stored in caller-supplied
* buffer BUF.
*
* Failure: FAIL. Contents of buffer BUF are undefined.
*
* Programmer: rky, 1998-01-30
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_read(H5FD_t *_file, H5FD_mem_t H5_ATTR_UNUSED type, hid_t H5_ATTR_UNUSED dxpl_id, haddr_t addr,
size_t size, void *buf /*out*/)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
MPI_Offset mpi_off;
MPI_Status mpi_stat; /* Status from I/O operation */
MPI_Datatype buf_type = MPI_BYTE; /* MPI description of the selection in memory */
int size_i; /* Integer copy of 'size' to read */
#if H5_CHECK_MPI_VERSION(3, 0)
MPI_Count bytes_read = 0; /* Number of bytes read in */
MPI_Count type_size; /* MPI datatype used for I/O's size */
MPI_Count io_size; /* Actual number of bytes requested */
MPI_Count n;
#else
int bytes_read = 0; /* Number of bytes read in */
int type_size; /* MPI datatype used for I/O's size */
int io_size; /* Actual number of bytes requested */
int n;
#endif
hbool_t use_view_this_time = FALSE;
hbool_t derived_type = FALSE;
hbool_t rank0_bcast = FALSE; /* If read-with-rank0-and-bcast flag was used */
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
hbool_t H5FD_mpio_debug_r_flag = (H5FD_mpio_debug_flags_s[(int)'r'] && H5FD_MPIO_TRACE_THIS_RANK(file));
#endif
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED;
FUNC_ENTER_PACKAGE
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
HDassert(buf);
/* Portably initialize MPI status variable */
HDmemset(&mpi_stat, 0, sizeof(MPI_Status));
/* some numeric conversions */
if (H5FD_mpi_haddr_to_MPIOff(addr, &mpi_off /*out*/) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off")
size_i = (int)size;
/* Only look for MPI views for raw data transfers */
if (type == H5FD_MEM_DRAW) {
H5FD_mpio_xfer_t xfer_mode; /* I/O transfer mode */
/* Get the transfer mode from the API context */
if (H5CX_get_io_xfer_mode(&xfer_mode) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O transfer mode")
/*
* Set up for a fancy xfer using complex types, or single byte block. We
* wouldn't need to rely on the use_view field if MPI semantics allowed
* us to test that btype=ftype=MPI_BYTE (or even MPI_TYPE_NULL, which
* could mean "use MPI_BYTE" by convention).
*/
if (xfer_mode == H5FD_MPIO_COLLECTIVE) {
MPI_Datatype file_type;
/* Remember that views are used */
use_view_this_time = TRUE;
/* Prepare for a full-blown xfer using btype, ftype, and disp */
if (H5CX_get_mpi_coll_datatypes(&buf_type, &file_type) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O datatypes")
/*
* Set the file view when we are using MPI derived types
*/
if (MPI_SUCCESS != (mpi_code = MPI_File_set_view(file->f, mpi_off, MPI_BYTE, file_type,
H5FD_mpi_native_g, file->info)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_view failed", mpi_code)
/* When using types, use the address as the displacement for
* MPI_File_set_view and reset the address for the read to zero
*/
mpi_off = 0;
} /* end if */
} /* end if */
/* Read the data. */
if (use_view_this_time) {
H5FD_mpio_collective_opt_t coll_opt_mode;
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stderr, "%s: (%d) using MPIO collective mode\n", __func__, file->mpi_rank);
#endif
/* Get the collective_opt property to check whether the application wants to do IO individually. */
if (H5CX_get_mpio_coll_opt(&coll_opt_mode) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O collective_op property")
if (coll_opt_mode == H5FD_MPIO_COLLECTIVE_IO) {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stderr, "%s: (%d) doing MPI collective IO\n", __func__, file->mpi_rank);
#endif
/* Check whether we should read from rank 0 and broadcast to other ranks */
if (H5CX_get_mpio_rank0_bcast()) {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stderr, "%s: (%d) doing read-rank0-and-MPI_Bcast\n", __func__, file->mpi_rank);
#endif
/* Indicate path we've taken */
rank0_bcast = TRUE;
/* Read on rank 0 Bcast to other ranks */
if (file->mpi_rank == 0) {
/* If MPI_File_read_at fails, push an error, but continue
* to participate in following MPI_Bcast */
if (MPI_SUCCESS !=
(mpi_code = MPI_File_read_at(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat)))
HMPI_DONE_ERROR(FAIL, "MPI_File_read_at failed", mpi_code)
}
if (MPI_SUCCESS != (mpi_code = MPI_Bcast(buf, size_i, buf_type, 0, file->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", mpi_code)
} /* end if */
else
/* Perform collective read operation */
if (MPI_SUCCESS !=
(mpi_code = MPI_File_read_at_all(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_read_at_all failed", mpi_code)
} /* end if */
else {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stderr, "%s: (%d) doing MPI independent IO\n", __func__, file->mpi_rank);
#endif
/* Perform independent read operation */
if (MPI_SUCCESS !=
(mpi_code = MPI_File_read_at(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_read_at failed", mpi_code)
} /* end else */
/*
* Reset the file view when we used MPI derived types
*/
if (MPI_SUCCESS != (mpi_code = MPI_File_set_view(file->f, (MPI_Offset)0, MPI_BYTE, MPI_BYTE,
H5FD_mpi_native_g, file->info)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_view failed", mpi_code)
} /* end if */
else {
if (size != (hsize_t)size_i) {
/* If HERE, then we need to work around the integer size limit
* of 2GB. The input size_t size variable cannot fit into an integer,
* but we can get around that limitation by creating a different datatype
* and then setting the integer size (or element count) to 1 when using
* the derived_type.
*/
if (H5_mpio_create_large_type(size, 0, MPI_BYTE, &buf_type) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_CANTGET, FAIL, "can't create MPI-I/O datatype")
derived_type = TRUE;
size_i = 1;
}
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stderr, "%s: (%d) doing MPI independent IO\n", __func__, file->mpi_rank);
#endif
/* Perform independent read operation */
if (MPI_SUCCESS != (mpi_code = MPI_File_read_at(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_read_at failed", mpi_code)
} /* end else */
/* Only retrieve bytes read if this rank _actually_ participated in I/O */
if (!rank0_bcast || (rank0_bcast && file->mpi_rank == 0)) {
/* How many bytes were actually read? */
#if H5_CHECK_MPI_VERSION(3, 0)
if (MPI_SUCCESS != (mpi_code = MPI_Get_elements_x(&mpi_stat, buf_type, &bytes_read))) {
#else
if (MPI_SUCCESS != (mpi_code = MPI_Get_elements(&mpi_stat, MPI_BYTE, &bytes_read))) {
#endif
if (rank0_bcast && file->mpi_rank == 0) {
/* If MPI_Get_elements(_x) fails for a rank 0 bcast strategy,
* push an error, but continue to participate in the following
* MPI_Bcast.
*/
bytes_read = -1;
HMPI_DONE_ERROR(FAIL, "MPI_Get_elements failed", mpi_code)
}
else
HMPI_GOTO_ERROR(FAIL, "MPI_Get_elements failed", mpi_code)
}
} /* end if */
/* If the rank0-bcast feature was used, broadcast the # of bytes read to
* other ranks, which didn't perform any I/O.
*/
/* NOTE: This could be optimized further to be combined with the broadcast
* of the data. (QAK - 2019/1/2)
*/
if (rank0_bcast)
#if H5_CHECK_MPI_VERSION(3, 0)
if (MPI_SUCCESS != MPI_Bcast(&bytes_read, 1, MPI_COUNT, 0, file->comm))
#else
if (MPI_SUCCESS != MPI_Bcast(&bytes_read, 1, MPI_INT, 0, file->comm))
#endif
HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", 0)
/* Get the type's size */
#if H5_CHECK_MPI_VERSION(3, 0)
if (MPI_SUCCESS != (mpi_code = MPI_Type_size_x(buf_type, &type_size)))
#else
if (MPI_SUCCESS != (mpi_code = MPI_Type_size(buf_type, &type_size)))
#endif
HMPI_GOTO_ERROR(FAIL, "MPI_Type_size failed", mpi_code)
/* Compute the actual number of bytes requested */
io_size = type_size * size_i;
/* Check for read failure */
if (bytes_read < 0 || bytes_read > io_size)
HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed")
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stderr, "%s: (%d) mpi_off = %ld bytes_read = %lld type = %s\n", __func__, file->mpi_rank,
(long)mpi_off, (long long)bytes_read, H5FD__mem_t_to_str(type));
#endif
/*
* This gives us zeroes beyond end of physical MPI file.
*/
if ((n = (io_size - bytes_read)) > 0)
HDmemset((char *)buf + bytes_read, 0, (size_t)n);
done:
if (derived_type)
MPI_Type_free(&buf_type);
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Leaving\n", __func__, file->mpi_rank);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_read() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_write
*
* Purpose: Writes SIZE bytes of data to FILE beginning at address ADDR
* from buffer BUF according to data transfer properties in
* DXPL_ID using potentially complex file and buffer types to
* effect the transfer.
*
* MPI is able to coalesce requests from different processes
* (collective and independent).
*
* Return: Success: SUCCEED. USE_TYPES and OLD_USE_TYPES in the
* access params are altered.
* Failure: FAIL. USE_TYPES and OLD_USE_TYPES in the
* access params may be altered.
*
* Programmer: Robert Kim Yates
* January 30, 1998
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_write(H5FD_t *_file, H5FD_mem_t type, hid_t H5_ATTR_UNUSED dxpl_id, haddr_t addr, size_t size,
const void *buf)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
MPI_Offset mpi_off;
MPI_Status mpi_stat; /* Status from I/O operation */
MPI_Datatype buf_type = MPI_BYTE; /* MPI description of the selection in memory */
#if H5_CHECK_MPI_VERSION(3, 0)
MPI_Count bytes_written;
MPI_Count type_size; /* MPI datatype used for I/O's size */
MPI_Count io_size; /* Actual number of bytes requested */
#else
int bytes_written;
int type_size; /* MPI datatype used for I/O's size */
int io_size; /* Actual number of bytes requested */
#endif
int size_i;
hbool_t use_view_this_time = FALSE;
hbool_t derived_type = FALSE;
H5FD_mpio_xfer_t xfer_mode; /* I/O transfer mode */
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
hbool_t H5FD_mpio_debug_w_flag = (H5FD_mpio_debug_flags_s[(int)'w'] && H5FD_MPIO_TRACE_THIS_RANK(file));
#endif
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED;
FUNC_ENTER_PACKAGE
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
HDassert(buf);
/* Verify that no data is written when between MPI_Barrier()s during file flush */
HDassert(!H5CX_get_mpi_file_flushing());
/* Portably initialize MPI status variable */
HDmemset(&mpi_stat, 0, sizeof(MPI_Status));
/* some numeric conversions */
if (H5FD_mpi_haddr_to_MPIOff(addr, &mpi_off) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off")
size_i = (int)size;
/* Get the transfer mode from the API context */
if (H5CX_get_io_xfer_mode(&xfer_mode) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O transfer mode")
/*
* Set up for a fancy xfer using complex types, or single byte block. We
* wouldn't need to rely on the use_view field if MPI semantics allowed
* us to test that btype=ftype=MPI_BYTE (or even MPI_TYPE_NULL, which
* could mean "use MPI_BYTE" by convention).
*/
if (xfer_mode == H5FD_MPIO_COLLECTIVE) {
MPI_Datatype file_type;
/* Remember that views are used */
use_view_this_time = TRUE;
/* Prepare for a full-blown xfer using btype, ftype, and disp */
if (H5CX_get_mpi_coll_datatypes(&buf_type, &file_type) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O datatypes")
/*
* Set the file view when we are using MPI derived types
*/
if (MPI_SUCCESS != (mpi_code = MPI_File_set_view(file->f, mpi_off, MPI_BYTE, file_type,
H5FD_mpi_native_g, file->info)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_view failed", mpi_code)
/* When using types, use the address as the displacement for
* MPI_File_set_view and reset the address for the read to zero
*/
mpi_off = 0;
} /* end if */
/* Write the data. */
if (use_view_this_time) {
H5FD_mpio_collective_opt_t coll_opt_mode;
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stderr, "%s: (%d) using MPIO collective mode\n", __func__, file->mpi_rank);
#endif
/* Get the collective_opt property to check whether the application wants to do IO individually. */
if (H5CX_get_mpio_coll_opt(&coll_opt_mode) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O collective_op property")
if (coll_opt_mode == H5FD_MPIO_COLLECTIVE_IO) {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stderr, "%s: (%d) doing MPI collective IO\n", __func__, file->mpi_rank);
#endif
/* Perform collective write operation */
if (MPI_SUCCESS !=
(mpi_code = MPI_File_write_at_all(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_write_at_all failed", mpi_code)
} /* end if */
else {
if (type != H5FD_MEM_DRAW)
HGOTO_ERROR(H5E_PLIST, H5E_BADTYPE, FAIL,
"Metadata Coll opt property should be collective at this point")
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stderr, "%s: (%d) doing MPI independent IO\n", __func__, file->mpi_rank);
#endif
/* Perform independent write operation */
if (MPI_SUCCESS !=
(mpi_code = MPI_File_write_at(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_write_at failed", mpi_code)
} /* end else */
/* Reset the file view when we used MPI derived types */
if (MPI_SUCCESS != (mpi_code = MPI_File_set_view(file->f, (MPI_Offset)0, MPI_BYTE, MPI_BYTE,
H5FD_mpi_native_g, file->info)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_view failed", mpi_code)
} /* end if */
else {
if (size != (hsize_t)size_i) {
/* If HERE, then we need to work around the integer size limit
* of 2GB. The input size_t size variable cannot fit into an integer,
* but we can get around that limitation by creating a different datatype
* and then setting the integer size (or element count) to 1 when using
* the derived_type.
*/
if (H5_mpio_create_large_type(size, 0, MPI_BYTE, &buf_type) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_CANTGET, FAIL, "can't create MPI-I/O datatype")
derived_type = TRUE;
size_i = 1;
}
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stderr, "%s: (%d) doing MPI independent IO\n", __func__, file->mpi_rank);
#endif
/* Perform independent write operation */
if (MPI_SUCCESS != (mpi_code = MPI_File_write_at(file->f, mpi_off, buf, size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_write_at failed", mpi_code)
} /* end else */
/* How many bytes were actually written? */
#if H5_CHECK_MPI_VERSION(3, 0)
if (MPI_SUCCESS != (mpi_code = MPI_Get_elements_x(&mpi_stat, buf_type, &bytes_written)))
#else
if (MPI_SUCCESS != (mpi_code = MPI_Get_elements(&mpi_stat, MPI_BYTE, &bytes_written)))
#endif
HMPI_GOTO_ERROR(FAIL, "MPI_Get_elements failed", mpi_code)
/* Get the type's size */
#if H5_CHECK_MPI_VERSION(3, 0)
if (MPI_SUCCESS != (mpi_code = MPI_Type_size_x(buf_type, &type_size)))
#else
if (MPI_SUCCESS != (mpi_code = MPI_Type_size(buf_type, &type_size)))
#endif
HMPI_GOTO_ERROR(FAIL, "MPI_Type_size failed", mpi_code)
/* Compute the actual number of bytes requested */
io_size = type_size * size_i;
/* Check for write failure */
if (bytes_written != io_size || bytes_written < 0)
HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "file write failed")
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stderr, "%s: (%d) mpi_off = %ld bytes_written = %lld type = %s\n", __func__,
file->mpi_rank, (long)mpi_off, (long long)bytes_written, H5FD__mem_t_to_str(type));
#endif
/* Each process will keep track of its perceived EOF value locally, and
* ultimately we will reduce this value to the maximum amongst all
* processes, but until then keep the actual eof at HADDR_UNDEF just in
* case something bad happens before that point. (rather have a value
* we know is wrong sitting around rather than one that could only
* potentially be wrong.) */
file->eof = HADDR_UNDEF;
if (bytes_written && (((haddr_t)bytes_written + addr) > file->local_eof))
file->local_eof = addr + (haddr_t)bytes_written;
done:
if (derived_type)
MPI_Type_free(&buf_type);
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Leaving: ret_value = %d\n", __func__, file->mpi_rank, ret_value);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_write() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_vector_build_types
*
* Purpose: Build MPI datatypes and calculate offset, base buffer, and
* size for MPIO vector I/O. Spun off from common code in
* H5FD__mpio_vector_read() and H5FD__mpio_vector_write().
*
* Return: Success: SUCCEED.
* Failure: FAIL.
*
* Programmer: Neil Fortner
* March 14, 2022
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_vector_build_types(uint32_t count, H5FD_mem_t types[], haddr_t addrs[], size_t sizes[],
H5_flexible_const_ptr_t bufs[], haddr_t *s_addrs[], size_t *s_sizes[],
H5_flexible_const_ptr_t *s_bufs[], hbool_t *vector_was_sorted,
MPI_Offset *mpi_off, H5_flexible_const_ptr_t *mpi_bufs_base, int *size_i,
MPI_Datatype *buf_type, hbool_t *buf_type_created, MPI_Datatype *file_type,
hbool_t *file_type_created, char *unused)
{
hsize_t bigio_count; /* Transition point to create derived type */
hbool_t fixed_size = FALSE;
size_t size;
H5FD_mem_t * s_types = NULL;
int * mpi_block_lengths = NULL;
MPI_Aint mpi_bufs_base_Aint;
MPI_Aint * mpi_bufs = NULL;
MPI_Aint * mpi_displacements = NULL;
MPI_Datatype *sub_types = NULL;
uint8_t * sub_types_created = NULL;
int i;
int j;
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED;
FUNC_ENTER_PACKAGE
/* Sanity checks */
HDassert(s_sizes);
HDassert(s_bufs);
HDassert(vector_was_sorted);
HDassert(*vector_was_sorted);
HDassert(mpi_off);
HDassert(mpi_bufs_base);
HDassert(size_i);
HDassert(buf_type);
HDassert(buf_type_created);
HDassert(!*buf_type_created);
HDassert(file_type);
HDassert(file_type_created);
HDassert(!*file_type_created);
HDassert(unused);
/* Get bio I/O transition point (may be lower than 2G for testing) */
bigio_count = H5_mpi_get_bigio_count();
if (count == 1) {
/* Single block. Just use a series of MPI_BYTEs for the file view.
*/
*size_i = (int)sizes[0];
*buf_type = MPI_BYTE;
*file_type = MPI_BYTE;
*mpi_bufs_base = bufs[0];
/* Setup s_addrs, s_sizes and s_bufs (needed for incomplete read filling code and eof
* calculation code) */
*s_addrs = addrs;
*s_sizes = sizes;
*s_bufs = bufs;
/* some numeric conversions */
if (H5FD_mpi_haddr_to_MPIOff(addrs[0], mpi_off) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't set MPI offset")
/* Check for size overflow */
if (sizes[0] > bigio_count) {
/* We need to work around the integer size limit of 2GB. The input size_t size
* variable cannot fit into an integer, but we can get around that limitation by
* creating a different datatype and then setting the integer size (or element
* count) to 1 when using the derived_type. */
if (H5_mpio_create_large_type(sizes[0], 0, MPI_BYTE, buf_type) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_CANTGET, FAIL, "can't create MPI-I/O datatype")
*buf_type_created = TRUE;
if (H5_mpio_create_large_type(sizes[0], 0, MPI_BYTE, file_type) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_CANTGET, FAIL, "can't create MPI-I/O datatype")
*file_type_created = TRUE;
*size_i = 1;
}
}
else if (count > 0) { /* create MPI derived types describing the vector write */
/* sort the vector I/O request into increasing address order if required
*
* If the vector is already sorted, the base addresses of types, addrs, sizes,
* and bufs will be returned in s_types, s_addrs, s_sizes, and s_bufs respectively.
*
* If the vector was not already sorted, new, sorted versions of types, addrs, sizes, and bufs
* are allocated, populated, and returned in s_types, s_addrs, s_sizes, and s_bufs respectively.
* In this case, this function must free the memory allocated for the sorted vectors.
*/
if (H5FD_sort_vector_io_req(vector_was_sorted, count, types, addrs, sizes, bufs, &s_types, s_addrs,
s_sizes, s_bufs) < 0)
HGOTO_ERROR(H5E_ARGS, H5E_BADVALUE, FAIL, "can't sort vector I/O request")
if ((NULL == (mpi_block_lengths = (int *)HDmalloc((size_t)count * sizeof(int)))) ||
(NULL == (mpi_displacements = (MPI_Aint *)HDmalloc((size_t)count * sizeof(MPI_Aint)))) ||
(NULL == (mpi_bufs = (MPI_Aint *)HDmalloc((size_t)count * sizeof(MPI_Aint))))) {
HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "can't alloc mpi block lengths / displacement")
}
/* when we setup mpi_bufs[] below, all addresses are offsets from
* mpi_bufs_base.
*
* Since these offsets must all be positive, we must scan through
* s_bufs[] to find the smallest value, and choose that for
* mpi_bufs_base.
*/
j = 0; /* guess at the index of the smallest value of s_bufs[] */
for (i = 1; i < (int)count; i++) {
if ((*s_bufs)[i].cvp < (*s_bufs)[j].cvp) {
j = i;
}
}
*mpi_bufs_base = (*s_bufs)[j];
if (MPI_SUCCESS != (mpi_code = MPI_Get_address(mpi_bufs_base->cvp, &mpi_bufs_base_Aint)))
HMPI_GOTO_ERROR(FAIL, "MPI_Get_address for s_bufs[] to mpi_bufs_base failed", mpi_code)
*size_i = 1;
fixed_size = FALSE;
/* load the mpi_block_lengths and mpi_displacements arrays */
for (i = 0; i < (int)count; i++) {
/* Determine size of this vector element */
if (!fixed_size) {
if ((*s_sizes)[i] == 0) {
HDassert(vector_was_sorted);
fixed_size = TRUE;
size = sizes[i - 1];
}
else {
size = (*s_sizes)[i];
}
}
/* Add to block lengths and displacements arrays */
mpi_block_lengths[i] = (int)size;
mpi_displacements[i] = (MPI_Aint)(*s_addrs)[i];
/* convert s_bufs[i] to MPI_Aint... */
if (MPI_SUCCESS != (mpi_code = MPI_Get_address((*s_bufs)[i].cvp, &(mpi_bufs[i]))))
HMPI_GOTO_ERROR(FAIL, "MPI_Get_address for s_bufs[] - mpi_bufs_base failed", mpi_code)
/*... and then subtract mpi_bufs_base_Aint from it. */
#if ((MPI_VERSION > 3) || ((MPI_VERSION == 3) && (MPI_SUBVERSION >= 1)))
mpi_bufs[i] = MPI_Aint_diff(mpi_bufs[i], mpi_bufs_base_Aint);
#else
mpi_bufs[i] = mpi_bufs[i] - mpi_bufs_base_Aint;
#endif
/* Check for size overflow */
if (size > bigio_count) {
/* We need to work around the integer size limit of 2GB. The input size_t size
* variable cannot fit into an integer, but we can get around that limitation by
* creating a different datatype and then setting the integer size (or element
* count) to 1 when using the derived_type. */
/* Allocate arrays to keep track of types and whether they were created, if
* necessary */
if (!sub_types) {
HDassert(!sub_types_created);
if (NULL == (sub_types = HDmalloc((size_t)count * sizeof(MPI_Datatype))))
HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "can't alloc sub types array")
if (NULL == (sub_types_created = (uint8_t *)HDcalloc((size_t)count, 1))) {
H5MM_free(sub_types);
sub_types = NULL;
HGOTO_ERROR(H5E_RESOURCE, H5E_CANTALLOC, FAIL, "can't alloc sub types created array")
}
/* Initialize sub_types to all MPI_BYTE */
for (j = 0; j < (int)count; j++)
sub_types[j] = MPI_BYTE;
}
HDassert(sub_types_created);
/* Create type for large block */
if (H5_mpio_create_large_type(size, 0, MPI_BYTE, &sub_types[i]) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_CANTGET, FAIL, "can't create MPI-I/O datatype")
sub_types_created[i] = TRUE;
/* Only one of these large types for this vector element */
mpi_block_lengths[i] = 1;
}
else
HDassert(size == (size_t)mpi_block_lengths[i]);
}
/* create the memory MPI derived types */
if (sub_types) {
if (MPI_SUCCESS != (mpi_code = MPI_Type_create_struct((int)count, mpi_block_lengths, mpi_bufs,
sub_types, buf_type)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_struct for buf_type failed", mpi_code)
}
else if (MPI_SUCCESS != (mpi_code = MPI_Type_create_hindexed((int)count, mpi_block_lengths, mpi_bufs,
MPI_BYTE, buf_type)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_hindexed for buf_type failed", mpi_code)
*buf_type_created = TRUE;
if (MPI_SUCCESS != (mpi_code = MPI_Type_commit(buf_type)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit for buf_type failed", mpi_code)
/* create the file MPI derived type */
if (sub_types) {
if (MPI_SUCCESS != (mpi_code = MPI_Type_create_struct((int)count, mpi_block_lengths,
mpi_displacements, sub_types, file_type)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_struct for file_type failed", mpi_code)
}
else if (MPI_SUCCESS != (mpi_code = MPI_Type_create_hindexed((int)count, mpi_block_lengths,
mpi_displacements, MPI_BYTE, file_type)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_create_hindexed for file_type failed", mpi_code)
*file_type_created = TRUE;
if (MPI_SUCCESS != (mpi_code = MPI_Type_commit(file_type)))
HMPI_GOTO_ERROR(FAIL, "MPI_Type_commit for file_type failed", mpi_code)
/* Free up memory used to build types */
HDassert(mpi_block_lengths);
HDfree(mpi_block_lengths);
mpi_block_lengths = NULL;
HDassert(mpi_displacements);
HDfree(mpi_displacements);
mpi_displacements = NULL;
HDassert(mpi_bufs);
HDfree(mpi_bufs);
mpi_bufs = NULL;
if (sub_types) {
HDassert(sub_types);
for (i = 0; i < (int)count; i++)
if (sub_types_created[i])
MPI_Type_free(&sub_types[i]);
HDfree(sub_types);
sub_types = NULL;
HDfree(sub_types_created);
sub_types_created = NULL;
}
/* some numeric conversions */
if (H5FD_mpi_haddr_to_MPIOff((haddr_t)0, mpi_off) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't set MPI off to 0")
}
else {
/* setup for null participation in the collective operation. */
*buf_type = MPI_BYTE;
*file_type = MPI_BYTE;
/* Set non-NULL pointer for I/O operation */
mpi_bufs_base->vp = unused;
/* MPI count to read */
*size_i = 0;
/* some numeric conversions */
if (H5FD_mpi_haddr_to_MPIOff((haddr_t)0, mpi_off) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't set MPI off to 0")
}
done:
/* free sorted vectors if they exist */
if (!vector_was_sorted)
if (s_types) {
HDfree(s_types);
s_types = NULL;
}
/* Clean up on error */
if (ret_value < 0) {
if (mpi_block_lengths) {
HDfree(mpi_block_lengths);
mpi_block_lengths = NULL;
}
if (mpi_displacements) {
HDfree(mpi_displacements);
mpi_displacements = NULL;
}
if (mpi_bufs) {
HDfree(mpi_bufs);
mpi_bufs = NULL;
}
if (sub_types) {
HDassert(sub_types_created);
for (i = 0; i < (int)count; i++)
if (sub_types_created[i])
MPI_Type_free(&sub_types[i]);
HDfree(sub_types);
sub_types = NULL;
HDfree(sub_types_created);
sub_types_created = NULL;
}
}
/* Make sure we cleaned up */
HDassert(!mpi_block_lengths);
HDassert(!mpi_displacements);
HDassert(!mpi_bufs);
HDassert(!sub_types);
HDassert(!sub_types_created);
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_vector_build_types() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_read_vector()
*
* Purpose: The behaviour of this function dependes on the value of
* the io_xfer_mode obtained from the context.
*
* If it is H5FD_MPIO_COLLECTIVE, this is a collective
* operation, which allows us to use MPI_File_set_view, and
* then perform the entire vector read in a single MPI call.
*
* Do this (if count is positive), by constructing memory
* and file derived types from the supplied vector, using
* file type to set the file view, and then reading the
* the memory type from file. Note that this read is
* either independent or collective depending on the
* value of mpio_coll_opt -- again obtained from the context.
*
* If count is zero, participate in the collective read
* (if so configured) with an empty read.
*
* Finally, set the file view back to its default state.
*
* In contrast, if io_xfer_mode is H5FD_MPIO_INDEPENDENT,
* this call is independent, and thus we cannot use
* MPI_File_set_view().
*
* In this case, simply walk the vector, and issue an
* independent read for each entry.
*
* Return: Success: SUCCEED.
* Failure: FAIL.
*
* Programmer: John Mainzer
* March 15, 2021
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_read_vector(H5FD_t *_file, hid_t H5_ATTR_UNUSED dxpl_id, uint32_t count, H5FD_mem_t types[],
haddr_t addrs[], size_t sizes[], void *bufs[])
{
H5FD_mpio_t * file = (H5FD_mpio_t *)_file;
hbool_t vector_was_sorted = TRUE;
haddr_t * s_addrs = NULL;
size_t * s_sizes = NULL;
void ** s_bufs = NULL;
char unused = 0; /* Unused, except for non-NULL pointer value */
void * mpi_bufs_base = NULL;
MPI_Datatype buf_type = MPI_BYTE; /* MPI description of the selection in memory */
hbool_t buf_type_created = FALSE;
MPI_Datatype file_type = MPI_BYTE; /* MPI description of the selection in file */
hbool_t file_type_created = FALSE;
int i;
int mpi_code; /* MPI return code */
MPI_Offset mpi_off = 0;
MPI_Status mpi_stat; /* Status from I/O operation */
H5FD_mpio_xfer_t xfer_mode; /* I/O transfer mode */
H5FD_mpio_collective_opt_t coll_opt_mode; /* whether we are doing collective or independent I/O */
int size_i;
#if MPI_VERSION >= 3
MPI_Count bytes_read = 0; /* Number of bytes read in */
MPI_Count type_size; /* MPI datatype used for I/O's size */
MPI_Count io_size; /* Actual number of bytes requested */
MPI_Count n;
#else
int bytes_read = 0; /* Number of bytes read in */
int type_size; /* MPI datatype used for I/O's size */
int io_size; /* Actual number of bytes requested */
int n;
#endif
hbool_t rank0_bcast = FALSE; /* If read-with-rank0-and-bcast flag was used */
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
hbool_t H5FD_mpio_debug_r_flag = (H5FD_mpio_debug_flags_s[(int)'r'] && H5FD_MPIO_TRACE_THIS_RANK(file));
#endif
herr_t ret_value = SUCCEED;
FUNC_ENTER_PACKAGE
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
HDassert((types) || (count == 0));
HDassert((addrs) || (count == 0));
HDassert((sizes) || (count == 0));
HDassert((bufs) || (count == 0));
/* verify that the first elements of the sizes and types arrays are
* valid.
*/
HDassert((count == 0) || (sizes[0] != 0));
HDassert((count == 0) || (types[0] != H5FD_MEM_NOLIST));
/* Get the transfer mode from the API context
*
* This flag is set to H5FD_MPIO_COLLECTIVE if the API call is
* collective, and to H5FD_MPIO_INDEPENDENT if it is not.
*
* While this doesn't mean that we are actually about to do a collective
* read, it does mean that all ranks are here, so we can use MPI_File_set_view().
*/
if (H5CX_get_io_xfer_mode(&xfer_mode) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O transfer mode")
if (xfer_mode == H5FD_MPIO_COLLECTIVE) {
/* Build MPI types, etc. */
if (H5FD__mpio_vector_build_types(count, types, addrs, sizes, (H5_flexible_const_ptr_t *)bufs,
&s_addrs, &s_sizes, (H5_flexible_const_ptr_t **)&s_bufs,
&vector_was_sorted, &mpi_off,
(H5_flexible_const_ptr_t *)&mpi_bufs_base, &size_i, &buf_type,
&buf_type_created, &file_type, &file_type_created, &unused) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't build MPI datatypes for I/O")
/* free sorted addrs vector if it exists */
if (!vector_was_sorted)
if (s_addrs) {
HDfree(s_addrs);
s_addrs = NULL;
}
/* Portably initialize MPI status variable */
HDmemset(&mpi_stat, 0, sizeof(mpi_stat));
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stdout, "%s: mpi_off = %ld size_i = %d\n", __func__, (long)mpi_off, size_i);
#endif
/* Setup the file view. */
if (MPI_SUCCESS != (mpi_code = MPI_File_set_view(file->f, mpi_off, MPI_BYTE, file_type,
H5FD_mpi_native_g, file->info)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_view failed", mpi_code)
/* Reset mpi_off to 0 since the view now starts at the data offset */
if (H5FD_mpi_haddr_to_MPIOff((haddr_t)0, &mpi_off) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't set MPI off to 0")
/* Get the collective_opt property to check whether the application wants to do IO individually.
*/
if (H5CX_get_mpio_coll_opt(&coll_opt_mode) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O collective_op property")
/* Read the data. */
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stdout, "%s: using MPIO collective mode\n", __func__);
#endif
if (coll_opt_mode == H5FD_MPIO_COLLECTIVE_IO) {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stdout, "%s: doing MPI collective IO\n", __func__);
#endif
/* Check whether we should read from rank 0 and broadcast to other ranks */
if (H5CX_get_mpio_rank0_bcast()) {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stdout, "%s: doing read-rank0-and-MPI_Bcast\n", __func__);
#endif
/* Indicate path we've taken */
rank0_bcast = TRUE;
/* Read on rank 0 Bcast to other ranks */
if (file->mpi_rank == 0)
if (MPI_SUCCESS != (mpi_code = MPI_File_read_at(file->f, mpi_off, mpi_bufs_base, size_i,
buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_read_at_all failed", mpi_code)
if (MPI_SUCCESS != (mpi_code = MPI_Bcast(mpi_bufs_base, size_i, buf_type, 0, file->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", mpi_code)
} /* end if */
else if (MPI_SUCCESS != (mpi_code = MPI_File_read_at_all(file->f, mpi_off, mpi_bufs_base, size_i,
buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_read_at_all failed", mpi_code)
} /* end if */
else if (size_i > 0) {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stdout, "%s: doing MPI independent IO\n", __func__);
#endif
if (MPI_SUCCESS !=
(mpi_code = MPI_File_read_at(file->f, mpi_off, mpi_bufs_base, size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_read_at failed", mpi_code)
} /* end else */
/* Reset the file view */
if (MPI_SUCCESS != (mpi_code = MPI_File_set_view(file->f, (MPI_Offset)0, MPI_BYTE, MPI_BYTE,
H5FD_mpi_native_g, file->info)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_view failed", mpi_code)
/* Only retrieve bytes read if this rank _actually_ participated in I/O */
if (!rank0_bcast || (rank0_bcast && file->mpi_rank == 0)) {
/* How many bytes were actually read? */
#if MPI_VERSION >= 3
if (MPI_SUCCESS != (mpi_code = MPI_Get_elements_x(&mpi_stat, buf_type, &bytes_read)))
#else
if (MPI_SUCCESS != (mpi_code = MPI_Get_elements(&mpi_stat, MPI_BYTE, &bytes_read)))
#endif
HMPI_GOTO_ERROR(FAIL, "MPI_Get_elements failed", mpi_code)
} /* end if */
/* If the rank0-bcast feature was used, broadcast the # of bytes read to
* other ranks, which didn't perform any I/O.
*/
/* NOTE: This could be optimized further to be combined with the broadcast
* of the data. (QAK - 2019/1/2)
* Or have rank 0 clear the unread parts of the buffer prior to
* the bcast. (NAF - 2021/9/15)
*/
if (rank0_bcast)
#if MPI_VERSION >= 3
if (MPI_SUCCESS != MPI_Bcast(&bytes_read, 1, MPI_COUNT, 0, file->comm))
#else
if (MPI_SUCCESS != MPI_Bcast(&bytes_read, 1, MPI_INT, 0, file->comm))
#endif
HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", 0)
/* Get the type's size */
#if MPI_VERSION >= 3
if (MPI_SUCCESS != (mpi_code = MPI_Type_size_x(buf_type, &type_size)))
#else
if (MPI_SUCCESS != (mpi_code = MPI_Type_size(buf_type, &type_size)))
#endif
HMPI_GOTO_ERROR(FAIL, "MPI_Type_size failed", mpi_code)
/* Compute the actual number of bytes requested */
io_size = type_size * size_i;
/* Check for read failure */
if (bytes_read < 0 || bytes_read > io_size)
HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed")
/* Check for incomplete read */
n = io_size - bytes_read;
if (n > 0) {
i = (int)count - 1;
/* Iterate over sorted array in reverse, filling in zeroes to
* sections of the buffers that were not read to */
do {
HDassert(i >= 0);
#if MPI_VERSION >= 3
io_size = MIN(n, (MPI_Count)s_sizes[i]);
bytes_read = (MPI_Count)s_sizes[i] - io_size;
#else
io_size = MIN(n, (int)s_sizes[i]);
bytes_read = (int)s_sizes[i] - io_size;
#endif
HDassert(bytes_read >= 0);
HDmemset((char *)s_bufs[i] + bytes_read, 0, (size_t)io_size);
n -= io_size;
i--;
} while (n > 0);
}
}
else if (count > 0) {
haddr_t max_addr = HADDR_MAX;
hbool_t fixed_size = FALSE;
size_t size;
/* The read is part of an independent operation. As a result,
* we can't use MPI_File_set_view() (since it it a collective operation),
* and thus we can't use the above code to construct the MPI datatypes.
* In the future, we could write code to detect when a contiguous slab
* in the file selection spans multiple vector elements and construct a
* memory datatype to match this larger block in the file, but for now
* just read in each element of the vector in a separate
* MPI_File_read_at() call.
*
* We could also just detect the case when the entire file selection is
* contiguous, which would allow us to use
* H5FD__mpio_vector_build_types() to construct the memory datatype.
*/
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_r_flag)
HDfprintf(stdout, "%s: doing MPI independent IO\n", __func__);
#endif
/* Loop over vector elements */
for (i = 0; i < (int)count; i++) {
/* Convert address to mpi offset */
if (H5FD_mpi_haddr_to_MPIOff(addrs[i], &mpi_off) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off")
/* Calculate I/O size */
if (!fixed_size) {
if (sizes[i] == 0) {
fixed_size = TRUE;
size = sizes[i - 1];
}
else {
size = sizes[i];
}
}
size_i = (int)size;
if (size != (size_t)size_i) {
/* If HERE, then we need to work around the integer size limit
* of 2GB. The input size_t size variable cannot fit into an integer,
* but we can get around that limitation by creating a different datatype
* and then setting the integer size (or element count) to 1 when using
* the derived_type.
*/
if (H5_mpio_create_large_type(size, 0, MPI_BYTE, &buf_type) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_CANTGET, FAIL, "can't create MPI-I/O datatype")
buf_type_created = TRUE;
size_i = 1;
}
/* Check if we actually need to do I/O */
if (addrs[i] < max_addr) {
/* Portably initialize MPI status variable */
HDmemset(&mpi_stat, 0, sizeof(mpi_stat));
/* Issue read */
if (MPI_SUCCESS !=
(mpi_code = MPI_File_read_at(file->f, mpi_off, bufs[i], size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_read_at failed", mpi_code)
/* How many bytes were actually read? */
#if MPI_VERSION >= 3
if (MPI_SUCCESS != (mpi_code = MPI_Get_elements_x(&mpi_stat, MPI_BYTE, &bytes_read)))
#else
if (MPI_SUCCESS != (mpi_code = MPI_Get_elements(&mpi_stat, MPI_BYTE, &bytes_read)))
#endif
HMPI_GOTO_ERROR(FAIL, "MPI_Get_elements failed", mpi_code)
/* Compute the actual number of bytes requested */
#if MPI_VERSION >= 3
io_size = (MPI_Count)size;
#else
io_size = (int)size;
#endif
/* Check for read failure */
if (bytes_read < 0 || bytes_read > io_size)
HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "file read failed")
/*
* If we didn't read the entire I/O, fill in zeroes beyond end of
* the physical MPI file and don't issue any more reads at higher
* addresses.
*/
if ((n = (io_size - bytes_read)) > 0) {
HDmemset((char *)bufs[i] + bytes_read, 0, (size_t)n);
max_addr = addrs[i] + (haddr_t)bytes_read;
}
}
else {
/* Read is past the max address, fill in zeroes */
HDmemset((char *)bufs[i], 0, size);
}
}
}
done:
if (buf_type_created) {
MPI_Type_free(&buf_type);
}
if (file_type_created) {
MPI_Type_free(&file_type);
}
/* free sorted vectors if they exist */
if (!vector_was_sorted) {
if (s_addrs) {
HDfree(s_addrs);
s_addrs = NULL;
}
if (s_sizes) {
HDfree(s_sizes);
s_sizes = NULL;
}
if (s_bufs) {
HDfree(s_bufs);
s_bufs = NULL;
}
}
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stdout, "%s: Leaving, proc %d: ret_value = %d\n", __func__, file->mpi_rank, ret_value);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_read_vector() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_write_vector
*
* Purpose: The behaviour of this function dependes on the value of
* the io_xfer_mode obtained from the context.
*
* If it is H5FD_MPIO_COLLECTIVE, this is a collective
* operation, which allows us to use MPI_File_set_view, and
* then perform the entire vector write in a single MPI call.
*
* Do this (if count is positive), by constructing memory
* and file derived types from the supplied vector, using
* file type to set the file view, and then writing the
* the memory type to file. Note that this write is
* either independent or collective depending on the
* value of mpio_coll_opt -- again obtained from the context.
*
* If count is zero, participate in the collective write
* (if so configured) with an empty write.
*
* Finally, set the file view back to its default state.
*
* In contrast, if io_xfer_mode is H5FD_MPIO_INDEPENDENT,
* this call is independent, and thus we cannot use
* MPI_File_set_view().
*
* In this case, simply walk the vector, and issue an
* independent write for each entry.
*
* Return: Success: SUCCEED.
* Failure: FAIL.
*
* Programmer: John Mainzer
* March 15, 2021
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_write_vector(H5FD_t *_file, hid_t H5_ATTR_UNUSED dxpl_id, uint32_t count, H5FD_mem_t types[],
haddr_t addrs[], size_t sizes[], const void *bufs[])
{
H5FD_mpio_t * file = (H5FD_mpio_t *)_file;
hbool_t vector_was_sorted = TRUE;
haddr_t * s_addrs = NULL;
size_t * s_sizes = NULL;
const void ** s_bufs = NULL;
char unused = 0; /* Unused, except for non-NULL pointer value */
const void * mpi_bufs_base = NULL;
MPI_Datatype buf_type = MPI_BYTE; /* MPI description of the selection in memory */
hbool_t buf_type_created = FALSE;
MPI_Datatype file_type = MPI_BYTE; /* MPI description of the selection in file */
hbool_t file_type_created = FALSE;
int i;
int mpi_code; /* MPI return code */
MPI_Offset mpi_off = 0;
MPI_Status mpi_stat; /* Status from I/O operation */
H5FD_mpio_xfer_t xfer_mode; /* I/O transfer mode */
H5FD_mpio_collective_opt_t coll_opt_mode; /* whether we are doing collective or independent I/O */
int size_i;
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
hbool_t H5FD_mpio_debug_w_flag = (H5FD_mpio_debug_flags_s[(int)'w'] && H5FD_MPIO_TRACE_THIS_RANK(file));
#endif
haddr_t max_addr = 0;
herr_t ret_value = SUCCEED;
FUNC_ENTER_PACKAGE
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
HDassert((types) || (count == 0));
HDassert((addrs) || (count == 0));
HDassert((sizes) || (count == 0));
HDassert((bufs) || (count == 0));
/* verify that the first elements of the sizes and types arrays are
* valid.
*/
HDassert((count == 0) || (sizes[0] != 0));
HDassert((count == 0) || (types[0] != H5FD_MEM_NOLIST));
/* Verify that no data is written when between MPI_Barrier()s during file flush */
HDassert(!H5CX_get_mpi_file_flushing());
/* Get the transfer mode from the API context
*
* This flag is set to H5FD_MPIO_COLLECTIVE if the API call is
* collective, and to H5FD_MPIO_INDEPENDENT if it is not.
*
* While this doesn't mean that we are actually about to do a collective
* write, it does mean that all ranks are here, so we can use MPI_File_set_view().
*/
if (H5CX_get_io_xfer_mode(&xfer_mode) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O transfer mode")
if (xfer_mode == H5FD_MPIO_COLLECTIVE) {
/* Build MPI types, etc. */
if (H5FD__mpio_vector_build_types(count, types, addrs, sizes, (H5_flexible_const_ptr_t *)bufs,
&s_addrs, &s_sizes, (H5_flexible_const_ptr_t **)&s_bufs,
&vector_was_sorted, &mpi_off,
(H5_flexible_const_ptr_t *)&mpi_bufs_base, &size_i, &buf_type,
&buf_type_created, &file_type, &file_type_created, &unused) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't build MPI datatypes for I/O")
/* Compute max addr writted to */
if (count > 0)
max_addr = s_addrs[count - 1] + (haddr_t)(s_sizes[count - 1]);
/* free sorted vectors if they exist */
if (!vector_was_sorted) {
if (s_addrs) {
HDfree(s_addrs);
s_addrs = NULL;
}
if (s_sizes) {
HDfree(s_sizes);
s_sizes = NULL;
}
if (s_bufs) {
HDfree(s_bufs);
s_bufs = NULL;
}
}
/* Portably initialize MPI status variable */
HDmemset(&mpi_stat, 0, sizeof(MPI_Status));
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stdout, "%s: mpi_off = %ld size_i = %d\n", __func__, (long)mpi_off, size_i);
#endif
/* Setup the file view. */
if (MPI_SUCCESS != (mpi_code = MPI_File_set_view(file->f, mpi_off, MPI_BYTE, file_type,
H5FD_mpi_native_g, file->info)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_view failed", mpi_code)
/* Reset mpi_off to 0 since the view now starts at the data offset */
if (H5FD_mpi_haddr_to_MPIOff((haddr_t)0, &mpi_off) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't set MPI off to 0")
/* Get the collective_opt property to check whether the application wants to do IO individually.
*/
if (H5CX_get_mpio_coll_opt(&coll_opt_mode) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI-I/O collective_op property")
/* Write the data. */
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stdout, "%s: using MPIO collective mode\n", __func__);
#endif
if (coll_opt_mode == H5FD_MPIO_COLLECTIVE_IO) {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stdout, "%s: doing MPI collective IO\n", __func__);
#endif
if (MPI_SUCCESS != (mpi_code = MPI_File_write_at_all(file->f, mpi_off, mpi_bufs_base, size_i,
buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_write_at_all failed", mpi_code)
} /* end if */
else if (size_i > 0) {
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stdout, "%s: doing MPI independent IO\n", __func__);
#endif
if (MPI_SUCCESS !=
(mpi_code = MPI_File_write_at(file->f, mpi_off, mpi_bufs_base, size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_write_at failed", mpi_code)
} /* end else */
/* Reset the file view */
if (MPI_SUCCESS != (mpi_code = MPI_File_set_view(file->f, (MPI_Offset)0, MPI_BYTE, MPI_BYTE,
H5FD_mpi_native_g, file->info)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_view failed", mpi_code)
}
else if (count > 0) {
hbool_t fixed_size = FALSE;
size_t size;
/* The read is part of an independent operation. As a result,
* we can't use MPI_File_set_view() (since it it a collective operation),
* and thus we can't use the above code to construct the MPI datatypes.
* In the future, we could write code to detect when a contiguous slab
* in the file selection spans multiple vector elements and construct a
* memory datatype to match this larger block in the file, but for now
* just read in each element of the vector in a separate
* MPI_File_read_at() call.
*
* We could also just detect the case when the entire file selection is
* contiguous, which would allow us to use
* H5FD__mpio_vector_build_types() to construct the memory datatype.
*/
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_w_flag)
HDfprintf(stdout, "%s: doing MPI independent IO\n", __func__);
#endif
/* Loop over vector elements */
for (i = 0; i < (int)count; i++) {
/* Convert address to mpi offset */
if (H5FD_mpi_haddr_to_MPIOff(addrs[i], &mpi_off) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "can't convert from haddr to MPI off")
/* Calculate I/O size */
if (!fixed_size) {
if (sizes[i] == 0) {
fixed_size = TRUE;
size = sizes[i - 1];
}
else {
size = sizes[i];
}
}
size_i = (int)size;
if (size != (size_t)size_i) {
/* If HERE, then we need to work around the integer size limit
* of 2GB. The input size_t size variable cannot fit into an integer,
* but we can get around that limitation by creating a different datatype
* and then setting the integer size (or element count) to 1 when using
* the derived_type.
*/
if (H5_mpio_create_large_type(size, 0, MPI_BYTE, &buf_type) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_CANTGET, FAIL, "can't create MPI-I/O datatype")
buf_type_created = TRUE;
size_i = 1;
}
/* Perform write */
if (MPI_SUCCESS !=
(mpi_code = MPI_File_write_at(file->f, mpi_off, bufs[i], size_i, buf_type, &mpi_stat)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_write_at failed", mpi_code)
/* Check if this is the highest address written to so far */
if (addrs[i] + size > max_addr)
max_addr = addrs[i] + size;
}
}
/* Each process will keep track of its perceived EOF value locally, and
* ultimately we will reduce this value to the maximum amongst all
* processes, but until then keep the actual eof at HADDR_UNDEF just in
* case something bad happens before that point. (rather have a value
* we know is wrong sitting around rather than one that could only
* potentially be wrong.)
*/
file->eof = HADDR_UNDEF;
/* check to see if the local eof has changed been extended, and update if so */
if (max_addr > file->local_eof)
file->local_eof = max_addr;
done:
if (buf_type_created)
MPI_Type_free(&buf_type);
if (file_type_created)
MPI_Type_free(&file_type);
/* Cleanup on error */
if (ret_value < 0 && !vector_was_sorted) {
if (s_addrs) {
HDfree(s_addrs);
s_addrs = NULL;
}
if (s_sizes) {
HDfree(s_sizes);
s_sizes = NULL;
}
if (s_bufs) {
HDfree(s_bufs);
s_bufs = NULL;
}
}
/* Make sure we cleaned up */
HDassert(vector_was_sorted || !s_addrs);
HDassert(vector_was_sorted || !s_sizes);
HDassert(vector_was_sorted || !s_bufs);
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stdout, "%s: Leaving, proc %d: ret_value = %d\n", __func__, file->mpi_rank, ret_value);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_write_vector() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_flush
*
* Purpose: Makes sure that all data is on disk. This is collective.
*
* Return: SUCCEED/FAIL
*
* Programmer: Robb Matzke
* January 30, 1998
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_flush(H5FD_t *_file, hid_t H5_ATTR_UNUSED dxpl_id, hbool_t closing)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
#endif
int mpi_code; /* mpi return code */
herr_t ret_value = SUCCEED;
FUNC_ENTER_PACKAGE
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
/* Only sync the file if we are not going to immediately close it */
if (!closing)
if (MPI_SUCCESS != (mpi_code = MPI_File_sync(file->f)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_sync failed", mpi_code)
done:
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Leaving\n", __func__, file->mpi_rank);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_flush() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_truncate
*
* Purpose: Make certain the file's size matches it's allocated size
*
* This is a little sticky in the mpio case, as it is not
* easy for us to track the current EOF by extracting it from
* write calls, since other ranks could have written to the
* file beyond the local EOF.
*
* Instead, we first check to see if the EOA has changed since
* the last call to this function. If it has, we call
* MPI_File_get_size() to determine the current EOF, and
* only call MPI_File_set_size() if this value disagrees
* with the current EOA.
*
* Return: SUCCEED/FAIL
*
* Programmer: Quincey Koziol
* January 31, 2008
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_truncate(H5FD_t *_file, hid_t H5_ATTR_UNUSED dxpl_id, hbool_t H5_ATTR_UNUSED closing)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
#ifdef H5FDmpio_DEBUG
hbool_t H5FD_mpio_debug_t_flag = (H5FD_mpio_debug_flags_s[(int)'t'] && H5FD_MPIO_TRACE_THIS_RANK(file));
#endif
herr_t ret_value = SUCCEED;
FUNC_ENTER_PACKAGE
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Entering\n", __func__, file->mpi_rank);
#endif
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
if (!H5F_addr_eq(file->eoa, file->last_eoa)) {
int mpi_code; /* mpi return code */
MPI_Offset size;
MPI_Offset needed_eof;
/* In principle, it is possible for the size returned by the
* call to MPI_File_get_size() to depend on whether writes from
* all proceeses have completed at the time process 0 makes the
* call.
*
* In practice, most (all?) truncate calls will come after a barrier
* and with no interviening writes to the file (with the possible
* exception of sueprblock / superblock extension message updates).
*
* Check the "MPI file closing" flag in the API context to determine
* if we can skip the barrier.
*/
if (!H5CX_get_mpi_file_flushing())
if (MPI_SUCCESS != (mpi_code = MPI_Barrier(file->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Barrier failed", mpi_code)
/* Only processor p0 will get the filesize and broadcast it. */
if (0 == file->mpi_rank) {
/* If MPI_File_get_size fails, broadcast file size as -1 to signal error */
if (MPI_SUCCESS != (mpi_code = MPI_File_get_size(file->f, &size)))
size = (MPI_Offset)-1;
}
/* Broadcast file size */
if (MPI_SUCCESS != (mpi_code = MPI_Bcast(&size, (int)sizeof(MPI_Offset), MPI_BYTE, 0, file->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Bcast failed", mpi_code)
if (size < 0)
HMPI_GOTO_ERROR(FAIL, "MPI_File_get_size failed", mpi_code)
if (H5FD_mpi_haddr_to_MPIOff(file->eoa, &needed_eof) < 0)
HGOTO_ERROR(H5E_INTERNAL, H5E_BADRANGE, FAIL, "cannot convert from haddr_t to MPI_Offset")
/* EOA != EOF. Set EOF to EOA */
if (size != needed_eof) {
/* Extend the file's size */
if (MPI_SUCCESS != (mpi_code = MPI_File_set_size(file->f, needed_eof)))
HMPI_GOTO_ERROR(FAIL, "MPI_File_set_size failed", mpi_code)
/* In general, we must wait until all processes have finished
* the truncate before any process can continue, since it is
* possible that a process would write at the end of the
* file, and this write would be discarded by the truncate.
*
* While this is an issue for a user initiated flush, it may
* not be an issue at file close. If so, we may be able to
* optimize out the following barrier in that case.
*/
if (MPI_SUCCESS != (mpi_code = MPI_Barrier(file->comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Barrier failed", mpi_code)
} /* end if */
/* Update the 'last' eoa value */
file->last_eoa = file->eoa;
} /* end if */
done:
#ifdef H5FDmpio_DEBUG
if (H5FD_mpio_debug_t_flag)
HDfprintf(stderr, "%s: (%d) Leaving\n", __func__, file->mpi_rank);
#endif
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_truncate() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_delete
*
* Purpose: Delete a file
*
* Return: SUCCEED/FAIL
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_delete(const char *filename, hid_t fapl_id)
{
H5P_genplist_t *plist; /* Property list pointer */
MPI_Comm comm = MPI_COMM_NULL;
MPI_Info info = MPI_INFO_NULL;
int mpi_rank = INT_MAX;
int mpi_code; /* MPI return code */
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_PACKAGE
HDassert(filename);
if (NULL == (plist = H5P_object_verify(fapl_id, H5P_FILE_ACCESS)))
HGOTO_ERROR(H5E_ARGS, H5E_BADTYPE, FAIL, "not a file access property list")
HDassert(H5FD_MPIO == H5P_peek_driver(plist));
if (H5FD_mpi_self_initialized) {
comm = MPI_COMM_WORLD;
}
else {
/* Get the MPI communicator and info from the fapl */
if (H5P_get(plist, H5F_ACS_MPI_PARAMS_INFO_NAME, &info) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI info object")
if (H5P_get(plist, H5F_ACS_MPI_PARAMS_COMM_NAME, &comm) < 0)
HGOTO_ERROR(H5E_VFL, H5E_CANTGET, FAIL, "can't get MPI communicator")
}
/* Get the MPI rank of this process */
if (MPI_SUCCESS != (mpi_code = MPI_Comm_rank(comm, &mpi_rank)))
HMPI_GOTO_ERROR(FAIL, "MPI_Comm_rank failed", mpi_code)
/* Set up a barrier */
if (MPI_SUCCESS != (mpi_code = MPI_Barrier(comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Barrier failed", mpi_code)
/* Delete the file */
if (mpi_rank == 0) {
/* If MPI_File_delete fails, push an error but
* still participate in the following MPI_Barrier
*/
if (MPI_SUCCESS != (mpi_code = MPI_File_delete(filename, info)))
HMPI_DONE_ERROR(FAIL, "MPI_File_delete failed", mpi_code)
}
/* Set up a barrier (don't want processes to run ahead of the delete) */
if (MPI_SUCCESS != (mpi_code = MPI_Barrier(comm)))
HMPI_GOTO_ERROR(FAIL, "MPI_Barrier failed", mpi_code)
done:
/* Free duplicated MPI Communicator and Info objects */
if (H5_mpi_comm_free(&comm) < 0)
HDONE_ERROR(H5E_VFL, H5E_CANTFREE, FAIL, "unable to free MPI communicator")
if (H5_mpi_info_free(&info) < 0)
HDONE_ERROR(H5E_VFL, H5E_CANTFREE, FAIL, "unable to free MPI info object")
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_delete() */
/*-------------------------------------------------------------------------
* Function: H5FD__mpio_ctl
*
* Purpose: MPIO version of the ctl callback.
*
* The desired operation is specified by the op_code
* parameter.
*
* The flags parameter controls management of op_codes that
* are unknown to the callback
*
* The input and output parameters allow op_code specific
* input and output
*
* At present, the supported op codes are:
*
* H5FD_CTL_GET_MPI_COMMUNICATOR_OPCODE
* H5FD_CTL_GET_MPI_RANK_OPCODE
* H5FD_CTL_GET_MPI_SIZE_OPCODE
*
* Note that these opcodes must be supported by all VFDs that
* support MPI.
*
* Return: Non-negative on success/Negative on failure
*
* Programmer: JRM -- 8/3/21
*
*-------------------------------------------------------------------------
*/
static herr_t
H5FD__mpio_ctl(H5FD_t *_file, uint64_t op_code, uint64_t flags, const void H5_ATTR_UNUSED *input,
void **output)
{
H5FD_mpio_t *file = (H5FD_mpio_t *)_file;
herr_t ret_value = SUCCEED; /* Return value */
FUNC_ENTER_NOAPI(FAIL)
/* Sanity checks */
HDassert(file);
HDassert(H5FD_MPIO == file->pub.driver_id);
switch (op_code) {
case H5FD_CTL_GET_MPI_COMMUNICATOR_OPCODE:
HDassert(output);
HDassert(*output);
**((MPI_Comm **)output) = file->comm;
break;
case H5FD_CTL_GET_MPI_RANK_OPCODE:
HDassert(output);
HDassert(*output);
**((int **)output) = file->mpi_rank;
break;
case H5FD_CTL_GET_MPI_SIZE_OPCODE:
HDassert(output);
HDassert(*output);
**((int **)output) = file->mpi_size;
break;
default: /* unknown op code */
if (flags & H5FD_CTL_FAIL_IF_UNKNOWN_FLAG) {
HGOTO_ERROR(H5E_VFL, H5E_FCNTL, FAIL, "unknown op_code and fail if unknown")
}
break;
}
done:
FUNC_LEAVE_NOAPI(ret_value)
} /* end H5FD__mpio_ctl() */
#endif /* H5_HAVE_PARALLEL */
|