summaryrefslogtreecommitdiffstats
path: root/src/H5Fistore.c
blob: 1c2b13716714a00a45ce938dacfb0222dd54d30f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
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
/*
 * Copyright (C) 1997-2001 NCSA
 *		           All rights reserved.
 *
 * Programmer: 	Robb Matzke <matzke@llnl.gov>
 *	       	Wednesday, October  8, 1997
 *
 * Purpose:	Indexed (chunked) I/O functions.  The logical
 *		multi-dimensional data space is regularly partitioned into
 *		same-sized "chunks", the first of which is aligned with the
 *		logical origin.  The chunks are given a multi-dimensional
 *		index which is used as a lookup key in a B-tree that maps
 *		chunk index to disk address.  Each chunk can be compressed
 *		independently and the chunks may move around in the file as
 *		their storage requirements change.
 *
 * Cache:	Disk I/O is performed in units of chunks and H5MF_alloc()
 *		contains code to optionally align chunks on disk block
 *		boundaries for performance.
 *
 *		The chunk cache is an extendible hash indexed by a function
 *		of storage B-tree address and chunk N-dimensional offset
 *		within the dataset.  Collisions are not resolved -- one of
 *		the two chunks competing for the hash slot must be preempted
 *		from the cache.  All entries in the hash also participate in
 *		a doubly-linked list and entries are penalized by moving them
 *		toward the front of the list.  When a new chunk is about to
 *		be added to the cache the heap is pruned by preempting
 *		entries near the front of the list to make room for the new
 *		entry which is added to the end of the list.
 */

#define H5F_PACKAGE		/*suppress error about including H5Fpkg	  */

#include "H5private.h"
#include "H5Dprivate.h"
#include "H5Eprivate.h"
#include "H5Fpkg.h"
#include "H5FLprivate.h"	/*Free Lists	  */
#include "H5Iprivate.h"
#include "H5MFprivate.h"
#include "H5MMprivate.h"
#include "H5Oprivate.h"
#include "H5Pprivate.h"         /* Property lists */
#include "H5Sprivate.h"         /* Dataspaces */
#include "H5Vprivate.h"

/* MPIO driver needed for special checks */
#include "H5FDmpio.h"

/*
 * Feature: If this constant is defined then every cache preemption and load
 *	    causes a character to be printed on the standard error stream:
 *
 *     `.': Entry was preempted because it has been completely read or
 *	    completely written but not partially read and not partially
 *	    written. This is often a good reason for preemption because such
 *	    a chunk will be unlikely to be referenced in the near future.
 *
 *     `:': Entry was preempted because it hasn't been used recently.
 *
 *     `#': Entry was preempted because another chunk collided with it. This
 *	    is usually a relatively bad thing.  If there are too many of
 *	    these then the number of entries in the cache can be increased.
 *
 *       c: Entry was preempted because the file is closing.
 *
 *	 w: A chunk read operation was eliminated because the library is
 *	    about to write new values to the entire chunk.  This is a good
 *	    thing, especially on files where the chunk size is the same as
 *	    the disk block size, chunks are aligned on disk block boundaries,
 *	    and the operating system can also eliminate a read operation.
 */

/*#define H5F_ISTORE_DEBUG */

/* Interface initialization */
#define PABLO_MASK	H5Fistore_mask
static int		interface_initialize_g = 0;
#define INTERFACE_INIT NULL

/*
 * Given a B-tree node return the dimensionality of the chunks pointed to by
 * that node.
 */
#define H5F_ISTORE_NDIMS(X)	((int)(((X)->sizeof_rkey-8)/8))

/* Raw data chunks are cached.  Each entry in the cache is: */
typedef struct H5F_rdcc_ent_t {
    hbool_t	locked;		/*entry is locked in cache		*/
    hbool_t	dirty;		/*needs to be written to disk?		*/
    H5O_layout_t *layout;	/*the layout message			*/
    double	split_ratios[3];/*B-tree node splitting ratios		*/
    H5O_pline_t	*pline;		/*filter pipeline message		*/
    hssize_t	offset[H5O_LAYOUT_NDIMS]; /*chunk name			*/
    size_t	rd_count;	/*bytes remaining to be read		*/
    size_t	wr_count;	/*bytes remaining to be written		*/
    size_t	chunk_size;	/*size of a chunk			*/
    size_t	alloc_size;	/*amount allocated for the chunk	*/
    uint8_t	*chunk;		/*the unfiltered chunk data		*/
    unsigned	idx;		/*index in hash table			*/
    struct H5F_rdcc_ent_t *next;/*next item in doubly-linked list	*/
    struct H5F_rdcc_ent_t *prev;/*previous item in doubly-linked list	*/
} H5F_rdcc_ent_t;
typedef H5F_rdcc_ent_t *H5F_rdcc_ent_ptr_t; /* For free lists */

/* Private prototypes */
static size_t H5F_istore_sizeof_rkey(H5F_t *f, const void *_udata);
static herr_t H5F_istore_new_node(H5F_t *f, H5B_ins_t, void *_lt_key,
				  void *_udata, void *_rt_key,
				  haddr_t *addr_p /*out*/);
static int H5F_istore_cmp2(H5F_t *f, void *_lt_key, void *_udata,
			    void *_rt_key);
static int H5F_istore_cmp3(H5F_t *f, void *_lt_key, void *_udata,
			    void *_rt_key);
static herr_t H5F_istore_found(H5F_t *f, haddr_t addr, const void *_lt_key,
			       void *_udata, const void *_rt_key);
static H5B_ins_t H5F_istore_insert(H5F_t *f, haddr_t addr, void *_lt_key,
				   hbool_t *lt_key_changed, void *_md_key,
				   void *_udata, void *_rt_key,
				   hbool_t *rt_key_changed,
				   haddr_t *new_node/*out*/);
static herr_t H5F_istore_iter_allocated(H5F_t *f, void *left_key, haddr_t addr,
				 void *right_key, void *_udata);
static herr_t H5F_istore_iter_dump(H5F_t *f, void *left_key, haddr_t addr,
				 void *right_key, void *_udata);
static herr_t H5F_istore_decode_key(H5F_t *f, H5B_t *bt, uint8_t *raw,
				    void *_key);
static herr_t H5F_istore_encode_key(H5F_t *f, H5B_t *bt, uint8_t *raw,
				    void *_key);
static herr_t H5F_istore_debug_key(FILE *stream, int indent, int fwidth,
				   const void *key, const void *udata);
static haddr_t H5F_istore_get_addr(H5F_t *f, const H5O_layout_t *layout,
				  const hssize_t offset[]);
static herr_t H5F_istore_prune_extent(H5F_t *f, void *_lt_key, haddr_t addr,
        void *_rt_key, void *_udata);
static H5B_ins_t H5F_istore_remove( H5F_t *f, haddr_t addr, void *_lt_key,
                  hbool_t *lt_key_changed, void *_udata, void *_rt_key,
                  hbool_t *rt_key_changed);

/*
 * B-tree key.	A key contains the minimum logical N-dimensional address and
 * the logical size of the chunk to which this key refers.  The
 * fastest-varying dimension is assumed to reference individual bytes of the
 * array, so a 100-element 1-d array of 4-byte integers would really be a 2-d
 * array with the slow varying dimension of size 100 and the fast varying
 * dimension of size 4 (the storage dimensionality has very little to do with
 * the real dimensionality).
 *
 * Only the first few values of the OFFSET and SIZE fields are actually
 * stored on disk, depending on the dimensionality.
 *
 * The chunk's file address is part of the B-tree and not part of the key.
 */
typedef struct H5F_istore_key_t {
    size_t	nbytes;				/*size of stored data	*/
    hssize_t	offset[H5O_LAYOUT_NDIMS];	/*logical offset to start*/
    unsigned	filter_mask;			/*excluded filters	*/
} H5F_istore_key_t;

typedef struct H5F_istore_ud1_t {
    H5F_istore_key_t	key;	                /*key values		*/
    haddr_t		addr;			/*file address of chunk */
    H5O_layout_t	mesg;		        /*layout message	*/
    hsize_t		total_storage;	        /*output from iterator	*/
    FILE		*stream;		/*debug output stream	*/
    hsize_t		*dims;		        /*dataset dimensions	*/
} H5F_istore_ud1_t;

/* inherits B-tree like properties from H5B */
H5B_class_t H5B_ISTORE[1] = {{
    H5B_ISTORE_ID,		/*id			*/
    sizeof(H5F_istore_key_t),	/*sizeof_nkey		*/
    H5F_istore_sizeof_rkey, 	/*get_sizeof_rkey	*/
    H5F_istore_new_node,	/*new			*/
    H5F_istore_cmp2,		/*cmp2			*/
    H5F_istore_cmp3,		/*cmp3			*/
    H5F_istore_found,		/*found			*/
    H5F_istore_insert,		/*insert		*/
    FALSE,			/*follow min branch?	*/
    FALSE,			/*follow max branch?	*/
    H5F_istore_remove,          /*remove		*/
    H5F_istore_decode_key,	/*decode		*/
    H5F_istore_encode_key,	/*encode		*/
    H5F_istore_debug_key,	/*debug			*/
}};

#define H5F_HASH_DIVISOR 8     /* Attempt to spread out the hashing */
                                /* This should be the same size as the alignment of */
                                /* of the smallest file format object written to the file.  */
#define H5F_HASH(F,ADDR) H5F_addr_hash((ADDR/H5F_HASH_DIVISOR),(F)->shared->rdcc.nslots)


/* Declare a free list to manage the chunk information */
H5FL_BLK_DEFINE_STATIC(istore_chunk);

/* Declare a free list to manage H5F_rdcc_ent_t objects */
H5FL_DEFINE_STATIC(H5F_rdcc_ent_t);

/* Declare a PQ free list to manage the H5F_rdcc_ent_ptr_t array information */
H5FL_ARR_DEFINE_STATIC(H5F_rdcc_ent_ptr_t,-1);


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_chunk_alloc
 *
 * Purpose:	Allocates memory for a chunk of a dataset.  This routine is used
 *      instead of malloc because the chunks can be kept on a free list so
 *      they don't thrash malloc/free as much.
 *
 * Return:	Success:	valid pointer to the chunk
 *
 *		Failure:	NULL
 *
 * Programmer:	Quincey Koziol
 *		Tuesday, March  21, 2000
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
void *
H5F_istore_chunk_alloc(size_t chunk_size)
{
    void *ret_value;                    /* Pointer to the chunk to return to the user */

    FUNC_ENTER(H5F_istore_chunk_alloc, NULL);

    ret_value=H5FL_BLK_ALLOC(istore_chunk,chunk_size,0);

    FUNC_LEAVE(ret_value);
} /* end H5F_istore_chunk_alloc() */


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_chunk_free
 *
 * Purpose:	Releases memory for a chunk of a dataset.  This routine is used
 *      instead of free because the chunks can be kept on a free list so
 *      they don't thrash malloc/free as much.
 *
 * Return:	Success:	NULL
 *
 *		Failure:	never fails
 *
 * Programmer:	Quincey Koziol
 *		Tuesday, March  21, 2000
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
void *
H5F_istore_chunk_free(void *chunk)
{
    FUNC_ENTER(H5F_istore_chunk_free, NULL);

    H5FL_BLK_FREE(istore_chunk,chunk);

    FUNC_LEAVE(NULL);
} /* end H5F_istore_chunk_free() */


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_chunk_realloc
 *
 * Purpose:	Resizes a chunk in chunking memory allocation system.  This
 *      does things the straightforward, simple way, not actually using
 *      realloc.
 *
 * Return:	Success:	NULL
 *
 *		Failure:	never fails
 *
 * Programmer:	Quincey Koziol
 *		Tuesday, March  21, 2000
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
void *
H5F_istore_chunk_realloc(void *chunk, size_t new_size)
{
    void *ret_value=NULL;               /* Return value */

    FUNC_ENTER(H5F_istore_chunk_realloc, NULL);

    ret_value=H5FL_BLK_REALLOC(istore_chunk,chunk,new_size);

    FUNC_LEAVE(ret_value);
} /* end H5F_istore_chunk_realloc() */


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_sizeof_rkey
 *
 * Purpose:	Returns the size of a raw key for the specified UDATA.	The
 *		size of the key is dependent on the number of dimensions for
 *		the object to which this B-tree points.	 The dimensionality
 *		of the UDATA is the only portion that's referenced here.
 *
 * Return:	Success:	Size of raw key in bytes.
 *
 *		Failure:	abort()
 *
 * Programmer:	Robb Matzke
 *		Wednesday, October  8, 1997
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static size_t
H5F_istore_sizeof_rkey(H5F_t UNUSED *f, const void *_udata)
{
    const H5F_istore_ud1_t *udata = (const H5F_istore_ud1_t *) _udata;
    size_t		    nbytes;

    assert(udata);
    assert(udata->mesg.ndims > 0 && udata->mesg.ndims <= H5O_LAYOUT_NDIMS);

    nbytes = 4 +			/*storage size		*/
	     4 +			/*filter mask		*/
	     udata->mesg.ndims*8;	/*dimension indices	*/

    return nbytes;
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_decode_key
 *
 * Purpose:	Decodes a raw key into a native key for the B-tree
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *		Friday, October 10, 1997
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_decode_key(H5F_t UNUSED *f, H5B_t *bt, uint8_t *raw, void *_key)
{
    H5F_istore_key_t	*key = (H5F_istore_key_t *) _key;
    int		i;
    int		ndims = H5F_ISTORE_NDIMS(bt);

    FUNC_ENTER(H5F_istore_decode_key, FAIL);

    /* check args */
    assert(f);
    assert(bt);
    assert(raw);
    assert(key);
    assert(ndims>0 && ndims<=H5O_LAYOUT_NDIMS);

    /* decode */
    UINT32DECODE(raw, key->nbytes);
    UINT32DECODE(raw, key->filter_mask);
    for (i=0; i<ndims; i++) {
	UINT64DECODE(raw, key->offset[i]);
    }

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_encode_key
 *
 * Purpose:	Encode a key from native format to raw format.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *		Friday, October 10, 1997
 *
 * Modifications:
 *	
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_encode_key(H5F_t UNUSED *f, H5B_t *bt, uint8_t *raw, void *_key)
{
    H5F_istore_key_t	*key = (H5F_istore_key_t *) _key;
    int		ndims = H5F_ISTORE_NDIMS(bt);
    int		i;

    FUNC_ENTER(H5F_istore_encode_key, FAIL);

    /* check args */
    assert(f);
    assert(bt);
    assert(raw);
    assert(key);
    assert(ndims>0 && ndims<=H5O_LAYOUT_NDIMS);

    /* encode */
    UINT32ENCODE(raw, key->nbytes);
    UINT32ENCODE(raw, key->filter_mask);
    for (i=0; i<ndims; i++) {
	UINT64ENCODE(raw, key->offset[i]);
    }

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_debug_key
 *
 * Purpose:	Prints a key.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Thursday, April 16, 1998
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_debug_key (FILE *stream, int indent, int fwidth,
		      const void *_key, const void *_udata)
{
    const H5F_istore_key_t	*key = (const H5F_istore_key_t *)_key;
    const H5F_istore_ud1_t	*udata = (const H5F_istore_ud1_t *)_udata;
    unsigned		u;
    
    FUNC_ENTER (H5F_istore_debug_key, FAIL);
    assert (key);

    HDfprintf(stream, "%*s%-*s %Zd bytes\n", indent, "", fwidth,
	      "Chunk size:", key->nbytes);
    HDfprintf(stream, "%*s%-*s 0x%08x\n", indent, "", fwidth,
	      "Filter mask:", key->filter_mask);
    HDfprintf(stream, "%*s%-*s {", indent, "", fwidth,
	      "Logical offset:");
    for (u=0; u<udata->mesg.ndims; u++) {
        HDfprintf (stream, "%s%Hd", u?", ":"", key->offset[u]);
    }
    HDfputs ("}\n", stream);

    FUNC_LEAVE (SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_cmp2
 *
 * Purpose:	Compares two keys sort of like strcmp().  The UDATA pointer
 *		is only to supply extra information not carried in the keys
 *		(in this case, the dimensionality) and is not compared
 *		against the keys.
 *
 * Return:	Success:	-1 if LT_KEY is less than RT_KEY;
 *				1 if LT_KEY is greater than RT_KEY;
 *				0 if LT_KEY and RT_KEY are equal.
 *
 *		Failure:	FAIL (same as LT_KEY<RT_KEY)
 *
 * Programmer:	Robb Matzke
 *		Thursday, November  6, 1997
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static int
H5F_istore_cmp2(H5F_t UNUSED *f, void *_lt_key, void *_udata,
		void *_rt_key)
{
    H5F_istore_key_t	*lt_key = (H5F_istore_key_t *) _lt_key;
    H5F_istore_key_t	*rt_key = (H5F_istore_key_t *) _rt_key;
    H5F_istore_ud1_t	*udata = (H5F_istore_ud1_t *) _udata;
    int		cmp;

    FUNC_ENTER(H5F_istore_cmp2, FAIL);

    assert(lt_key);
    assert(rt_key);
    assert(udata);
    assert(udata->mesg.ndims > 0 && udata->mesg.ndims <= H5O_LAYOUT_NDIMS);

    /* Compare the offsets but ignore the other fields */
    cmp = H5V_vector_cmp_s(udata->mesg.ndims, lt_key->offset, rt_key->offset);

    FUNC_LEAVE(cmp);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_cmp3
 *
 * Purpose:	Compare the requested datum UDATA with the left and right
 *		keys of the B-tree.
 *
 * Return:	Success:	negative if the min_corner of UDATA is less
 *				than the min_corner of LT_KEY.
 *
 *				positive if the min_corner of UDATA is
 *				greater than or equal the min_corner of
 *				RT_KEY.
 *
 *				zero otherwise.	 The min_corner of UDATA is
 *				not necessarily contained within the address
 *				space represented by LT_KEY, but a key that
 *				would describe the UDATA min_corner address
 *				would fall lexicographically between LT_KEY
 *				and RT_KEY.
 *				
 *		Failure:	FAIL (same as UDATA < LT_KEY)
 *
 * Programmer:	Robb Matzke
 *		Wednesday, October  8, 1997
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static int
H5F_istore_cmp3(H5F_t UNUSED *f, void *_lt_key, void *_udata,
		void *_rt_key)
{
    H5F_istore_key_t	*lt_key = (H5F_istore_key_t *) _lt_key;
    H5F_istore_key_t	*rt_key = (H5F_istore_key_t *) _rt_key;
    H5F_istore_ud1_t	*udata = (H5F_istore_ud1_t *) _udata;
    int		cmp = 0;

    FUNC_ENTER(H5F_istore_cmp3, FAIL);

    assert(lt_key);
    assert(rt_key);
    assert(udata);
    assert(udata->mesg.ndims > 0 && udata->mesg.ndims <= H5O_LAYOUT_NDIMS);

    if (H5V_vector_lt_s(udata->mesg.ndims, udata->key.offset,
			lt_key->offset)) {
	cmp = -1;
    } else if (H5V_vector_ge_s(udata->mesg.ndims, udata->key.offset,
			     rt_key->offset)) {
	cmp = 1;
    }
    FUNC_LEAVE(cmp);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_new_node
 *
 * Purpose:	Adds a new entry to an i-storage B-tree.  We can assume that
 *		the domain represented by UDATA doesn't intersect the domain
 *		already represented by the B-tree.
 *
 * Return:	Success:	Non-negative. The address of leaf is returned
 *				through the ADDR argument.  It is also added
 *				to the UDATA.
 *
 * 		Failure:	Negative
 *
 * Programmer:	Robb Matzke
 *		Tuesday, October 14, 1997
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_new_node(H5F_t *f, H5B_ins_t op,
		    void *_lt_key, void *_udata, void *_rt_key,
		    haddr_t *addr_p/*out*/)
{
    H5F_istore_key_t	*lt_key = (H5F_istore_key_t *) _lt_key;
    H5F_istore_key_t	*rt_key = (H5F_istore_key_t *) _rt_key;
    H5F_istore_ud1_t	*udata = (H5F_istore_ud1_t *) _udata;
    unsigned		u;

    FUNC_ENTER(H5F_istore_new_node, FAIL);
#ifdef AKC
    printf("%s: Called\n", FUNC);
#endif
    /* check args */
    assert(f);
    assert(lt_key);
    assert(rt_key);
    assert(udata);
    assert(udata->mesg.ndims > 0 && udata->mesg.ndims < H5O_LAYOUT_NDIMS);
    assert(addr_p);

    /* Allocate new storage */
    assert (udata->key.nbytes > 0);
#ifdef AKC
    printf("calling H5MF_alloc for new chunk\n");
#endif
    if (HADDR_UNDEF==(*addr_p=H5MF_alloc(f, H5FD_MEM_DRAW, (hsize_t)udata->key.nbytes))) {
        HRETURN_ERROR(H5E_IO, H5E_CANTINIT, FAIL,
		      "couldn't allocate new file storage");
    }
    udata->addr = *addr_p;

    /*
     * The left key describes the storage of the UDATA chunk being
     * inserted into the tree.
     */
    lt_key->nbytes = udata->key.nbytes;
    lt_key->filter_mask = udata->key.filter_mask;
    for (u=0; u<udata->mesg.ndims; u++) {
        lt_key->offset[u] = udata->key.offset[u];
    }

    /*
     * The right key might already be present.  If not, then add a zero-width
     * chunk.
     */
    if (H5B_INS_LEFT != op) {
        rt_key->nbytes = 0;
        rt_key->filter_mask = 0;
        for (u=0; u<udata->mesg.ndims; u++) {
            assert (udata->mesg.dim[u] < HSSIZET_MAX);
            assert (udata->key.offset[u]+(hssize_t)(udata->mesg.dim[u]) >
                udata->key.offset[u]);
            rt_key->offset[u] = udata->key.offset[u] +
                    (hssize_t)(udata->mesg.dim[u]);
        }
    }

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_found
 *
 * Purpose:	This function is called when the B-tree search engine has
 *		found the leaf entry that points to a chunk of storage that
 *		contains the beginning of the logical address space
 *		represented by UDATA.  The LT_KEY is the left key (the one
 *		that describes the chunk) and RT_KEY is the right key (the
 *		one that describes the next or last chunk).
 *
 * Note:	It's possible that the chunk isn't really found.  For
 *		instance, in a sparse dataset the requested chunk might fall
 *		between two stored chunks in which case this function is
 *		called with the maximum stored chunk indices less than the
 *		requested chunk indices.
 *
 * Return:	Non-negative on success with information about the chunk
 *		returned through the UDATA argument. Negative on failure.
 *
 * Programmer:	Robb Matzke
 *		Thursday, October  9, 1997
 *
 * Modifications:
 *		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_found(H5F_t UNUSED *f, haddr_t addr, const void *_lt_key,
		 void *_udata, const void UNUSED *_rt_key)
{
    H5F_istore_ud1_t	   *udata = (H5F_istore_ud1_t *) _udata;
    const H5F_istore_key_t *lt_key = (const H5F_istore_key_t *) _lt_key;
    unsigned		u;

    FUNC_ENTER(H5F_istore_found, FAIL);

    /* Check arguments */
    assert(f);
    assert(H5F_addr_defined(addr));
    assert(udata);
    assert(lt_key);

    /* Is this *really* the requested chunk? */
    for (u=0; u<udata->mesg.ndims; u++) {
        if (udata->key.offset[u] >= lt_key->offset[u]+(hssize_t)(udata->mesg.dim[u])) {
            HRETURN(FAIL);
        }
    }

    /* Initialize return values */
    udata->addr = addr;
    udata->key.nbytes = lt_key->nbytes;
    udata->key.filter_mask = lt_key->filter_mask;
    assert (lt_key->nbytes>0);
    for (u = 0; u < udata->mesg.ndims; u++) {
        udata->key.offset[u] = lt_key->offset[u];
    }

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_insert
 *
 * Purpose:	This function is called when the B-tree insert engine finds
 *		the node to use to insert new data.  The UDATA argument
 *		points to a struct that describes the logical addresses being
 *		added to the file.  This function allocates space for the
 *		data and returns information through UDATA describing a
 *		file chunk to receive (part of) the data.
 *
 *		The LT_KEY is always the key describing the chunk of file
 *		memory at address ADDR. On entry, UDATA describes the logical
 *		addresses for which storage is being requested (through the
 *		`offset' and `size' fields). On return, UDATA describes the
 *		logical addresses contained in a chunk on disk.
 *
 * Return:	Success:	An insertion command for the caller, one of
 *				the H5B_INS_* constants.  The address of the
 *				new chunk is returned through the NEW_NODE
 *				argument.
 *
 *		Failure:	H5B_INS_ERROR
 *
 * Programmer:	Robb Matzke
 *		Thursday, October  9, 1997
 *
 * Modifications:
 *		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value. The NEW_NODE argument
 *		is renamed NEW_NODE_P.
 *-------------------------------------------------------------------------
 */
static H5B_ins_t
H5F_istore_insert(H5F_t *f, haddr_t addr, void *_lt_key,
		  hbool_t UNUSED *lt_key_changed,
		  void *_md_key, void *_udata, void *_rt_key,
		  hbool_t UNUSED *rt_key_changed,
		  haddr_t *new_node_p/*out*/)
{
    H5F_istore_key_t	*lt_key = (H5F_istore_key_t *) _lt_key;
    H5F_istore_key_t	*md_key = (H5F_istore_key_t *) _md_key;
    H5F_istore_key_t	*rt_key = (H5F_istore_key_t *) _rt_key;
    H5F_istore_ud1_t	*udata = (H5F_istore_ud1_t *) _udata;
    int		cmp;
    unsigned		u;
    H5B_ins_t		ret_value = H5B_INS_ERROR;

    FUNC_ENTER(H5F_istore_insert, H5B_INS_ERROR);
#ifdef AKC
    printf("%s: Called\n", FUNC);
#endif

    /* check args */
    assert(f);
    assert(H5F_addr_defined(addr));
    assert(lt_key);
    assert(lt_key_changed);
    assert(md_key);
    assert(udata);
    assert(rt_key);
    assert(rt_key_changed);
    assert(new_node_p);

    cmp = H5F_istore_cmp3(f, lt_key, udata, rt_key);
    assert(cmp <= 0);

    if (cmp < 0) {
        /* Negative indices not supported yet */
        assert("HDF5 INTERNAL ERROR -- see rpm" && 0);
        HRETURN_ERROR(H5E_STORAGE, H5E_UNSUPPORTED, H5B_INS_ERROR,
		      "internal error");
	
    } else if (H5V_vector_eq_s (udata->mesg.ndims,
				udata->key.offset, lt_key->offset) &&
	       lt_key->nbytes>0) {
        /*
         * Already exists.  If the new size is not the same as the old size
         * then we should reallocate storage.
         */
        if (lt_key->nbytes != udata->key.nbytes) {
#ifdef AKC
            printf("calling H5MF_realloc for new chunk\n");
#endif
            if (HADDR_UNDEF==(*new_node_p=H5MF_realloc(f, H5FD_MEM_DRAW, addr,
                                  (hsize_t)lt_key->nbytes,
                                  (hsize_t)udata->key.nbytes))) {
                HRETURN_ERROR (H5E_STORAGE, H5E_WRITEERROR, H5B_INS_ERROR,
                       "unable to reallocate chunk storage");
            }
            lt_key->nbytes = udata->key.nbytes;
            lt_key->filter_mask = udata->key.filter_mask;
            *lt_key_changed = TRUE;
            udata->addr = *new_node_p;
            ret_value = H5B_INS_CHANGE;
        } else {
            udata->addr = addr;
            ret_value = H5B_INS_NOOP;
        }

    } else if (H5V_hyper_disjointp(udata->mesg.ndims,
				   lt_key->offset, udata->mesg.dim,
				   udata->key.offset, udata->mesg.dim)) {
        assert(H5V_hyper_disjointp(udata->mesg.ndims,
				   rt_key->offset, udata->mesg.dim,
				   udata->key.offset, udata->mesg.dim));
        /*
         * Split this node, inserting the new new node to the right of the
         * current node.  The MD_KEY is where the split occurs.
         */
        md_key->nbytes = udata->key.nbytes;
        md_key->filter_mask = udata->key.filter_mask;
        for (u=0; u<udata->mesg.ndims; u++) {
            assert(0 == udata->key.offset[u] % udata->mesg.dim[u]);
            md_key->offset[u] = udata->key.offset[u];
        }

        /*
         * Allocate storage for the new chunk
         */
#ifdef AKC
        printf("calling H5MF_alloc for new chunk\n");
#endif
        if (HADDR_UNDEF==(*new_node_p=H5MF_alloc(f, H5FD_MEM_DRAW,
                             (hsize_t)udata->key.nbytes))) {
            HRETURN_ERROR(H5E_IO, H5E_CANTINIT, H5B_INS_ERROR,
                  "file allocation failed");
        }
        udata->addr = *new_node_p;
        ret_value = H5B_INS_RIGHT;

    } else {
        assert("HDF5 INTERNAL ERROR -- see rpm" && 0);
        HRETURN_ERROR(H5E_IO, H5E_UNSUPPORTED, H5B_INS_ERROR,
		      "internal error");
    }

    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_iter_allocated
 *
 * Purpose:	Simply counts the number of chunks for a dataset.
 *
 * Return:	Success:	Non-negative
 *
 *		Failure:	Negative
 *
 * Programmer:	Robb Matzke
 *              Wednesday, April 21, 1999
 *
 * Modifications:
 *		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *
 *		Quincey Koziol, 2002-04-22
 *		Changed to callback from H5B_iterate
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_iter_allocated (H5F_t UNUSED *f, void *_lt_key, haddr_t UNUSED addr,
		    void UNUSED *_rt_key, void *_udata)
{
    H5F_istore_ud1_t	*bt_udata = (H5F_istore_ud1_t *)_udata;
    H5F_istore_key_t	*lt_key = (H5F_istore_key_t *)_lt_key;

    FUNC_ENTER(H5F_istore_iterate, FAIL);

    bt_udata->total_storage += lt_key->nbytes;
    FUNC_LEAVE(SUCCEED);
} /* H5F_istore_iter_allocated() */


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_iter_dump
 *
 * Purpose:	If the UDATA.STREAM member is non-null then debugging
 *              information is written to that stream.
 *
 * Return:	Success:	Non-negative
 *
 *		Failure:	Negative
 *
 * Programmer:	Robb Matzke
 *              Wednesday, April 21, 1999
 *
 * Modifications:
 *		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *
 *		Quincey Koziol, 2002-04-22
 *		Changed to callback from H5B_iterate
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_iter_dump (H5F_t UNUSED *f, void *_lt_key, haddr_t UNUSED addr,
		    void UNUSED *_rt_key, void *_udata)
{
    H5F_istore_ud1_t	*bt_udata = (H5F_istore_ud1_t *)_udata;
    H5F_istore_key_t	*lt_key = (H5F_istore_key_t *)_lt_key;
    unsigned		u;

    FUNC_ENTER(H5F_istore_iterate, FAIL);

    if (bt_udata->stream) {
        if (0==bt_udata->total_storage) {
            fprintf(bt_udata->stream, "    Address:\n");
            fprintf(bt_udata->stream,
                "             Flags    Bytes    Address Logical Offset\n");
            fprintf(bt_udata->stream,
                "        ========== ======== ========== "
                "==============================\n");
        }
        HDfprintf(bt_udata->stream, "        0x%08x %8Zu %10a [",
              lt_key->filter_mask, lt_key->nbytes, addr);
        for (u=0; u<bt_udata->mesg.ndims; u++) {
            HDfprintf(bt_udata->stream, "%s%Hd", u?", ":"", lt_key->offset[u]);
        }
        HDfputs("]\n", bt_udata->stream);
    }

    FUNC_LEAVE(SUCCEED);
} /* H5F_istore_iter_dump() */


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_init
 *
 * Purpose:	Initialize the raw data chunk cache for a file.  This is
 *		called when the file handle is initialized.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Monday, May 18, 1998
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_init (H5F_t *f)
{
    H5F_rdcc_t	*rdcc = &(f->shared->rdcc);
    
    FUNC_ENTER (H5F_istore_init, FAIL);

    HDmemset (rdcc, 0, sizeof(H5F_rdcc_t));
    if (f->shared->rdcc_nbytes>0 && f->shared->rdcc_nelmts>0) {
	rdcc->nslots = f->shared->rdcc_nelmts;
	rdcc->slot = H5FL_ARR_ALLOC (H5F_rdcc_ent_ptr_t,rdcc->nslots,1);
	if (NULL==rdcc->slot) {
	    HRETURN_ERROR (H5E_RESOURCE, H5E_NOSPACE, FAIL,
			   "memory allocation failed");
	}
    }

    FUNC_LEAVE (SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_flush_entry
 *
 * Purpose:	Writes a chunk to disk.  If RESET is non-zero then the
 *		entry is cleared -- it's slightly faster to flush a chunk if
 *		the RESET flag is turned on because it results in one fewer
 *		memory copy.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Thursday, May 21, 1998
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_flush_entry(H5F_t *f, H5F_rdcc_ent_t *ent, hbool_t reset)
{
    herr_t		ret_value=FAIL;	/*return value			*/
    H5F_istore_ud1_t 	udata;		/*pass through B-tree		*/
    unsigned		u;		/*counters			*/
    void		*buf=NULL;	/*temporary buffer		*/
    size_t		alloc;		/*bytes allocated for BUF	*/
    hbool_t		point_of_no_return = FALSE;
    
    FUNC_ENTER(H5F_istore_flush_entry, FAIL);
    assert(f);
    assert(ent);
    assert(!ent->locked);

    buf = ent->chunk;
    if (ent->dirty) {
        udata.mesg = *(ent->layout);
        udata.key.filter_mask = 0;
        udata.addr = HADDR_UNDEF;
        udata.key.nbytes = ent->chunk_size;
        for (u=0; u<ent->layout->ndims; u++) {
            udata.key.offset[u] = ent->offset[u];
        }
        alloc = ent->alloc_size;

        /* Should the chunk be filtered before writing it to disk? */
        if (ent->pline && ent->pline->nfilters) {
            if (!reset) {
                /*
                 * Copy the chunk to a new buffer before running it through
                 * the pipeline because we'll want to save the original buffer
                 * for later.
                 */
                alloc = ent->chunk_size;
                if (NULL==(buf = H5F_istore_chunk_alloc(alloc))) {
                    HGOTO_ERROR(H5E_RESOURCE, H5E_NOSPACE, FAIL,
                        "memory allocation failed for pipeline");
                }
                HDmemcpy(buf, ent->chunk, ent->chunk_size);
            } else {
                /*
                 * If we are reseting and something goes wrong after this
                 * point then it's too late to recover because we may have
                 * destroyed the original data by calling H5Z_pipeline().
                 * The only safe option is to continue with the reset
                 * even if we can't write the data to disk.
                 */
                point_of_no_return = TRUE;
                ent->chunk = NULL;
            }
            if (H5Z_pipeline(f, ent->pline, 0, &(udata.key.filter_mask),
                     &(udata.key.nbytes), &alloc, &buf)<0) {
                HGOTO_ERROR(H5E_PLINE, H5E_WRITEERROR, FAIL,
                    "output pipeline failed");
            }
        }

        /*
         * Create the chunk it if it doesn't exist, or reallocate the chunk if
         * its size changed.  Then write the data into the file.
         */
        if (H5B_insert(f, H5B_ISTORE, ent->layout->addr, ent->split_ratios,
                   &udata)<0) {
            HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL,
                "unable to allocate chunk");
        }
        if (H5F_block_write(f, H5FD_MEM_DRAW, udata.addr, udata.key.nbytes, H5P_DATASET_XFER_DEFAULT,
                    buf)<0) {
            HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL,
                "unable to write raw data to file");
        }

        /* Mark cache entry as clean */
        ent->dirty = FALSE;
        f->shared->rdcc.nflushes++;
    }
    
    /* Reset, but do not free or removed from list */
    if (reset) {
        point_of_no_return = FALSE;
        ent->layout = H5O_free(H5O_LAYOUT, ent->layout);
        ent->pline = H5O_free(H5O_PLINE, ent->pline);
        if (buf==ent->chunk) buf = NULL;
        if(ent->chunk!=NULL)
            ent->chunk = H5F_istore_chunk_free(ent->chunk);
    }
    
    ret_value = SUCCEED;

done:
    /* Free the temp buffer only if it's different than the entry chunk */
    if (buf!=ent->chunk)
        H5F_istore_chunk_free(buf);
    
    /*
     * If we reached the point of no return then we have no choice but to
     * reset the entry.  This can only happen if RESET is true but the
     * output pipeline failed.  Do not free the entry or remove it from the
     * list.
     */
    if (ret_value<0 && point_of_no_return) {
        ent->layout = H5O_free(H5O_LAYOUT, ent->layout);
        ent->pline = H5O_free(H5O_PLINE, ent->pline);
        if(ent->chunk)
            ent->chunk = H5F_istore_chunk_free(ent->chunk);
    }
    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:    H5F_istore_preempt
 *
 * Purpose:     Preempts the specified entry from the cache, flushing it to
 *              disk if necessary.
 *
 * Return:      Non-negative on success/Negative on failure
 *
 * Programmer:  Robb Matzke
 *              Thursday, May 21, 1998
 *
 * Modifications:
 *      Pedro Vicente, March 28, 2002
 *      Added flush parameter that switches the call to H5F_istore_flush_entry 
 *      The call with FALSE is used by the H5F_istore_prune_by_extent function
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_preempt(H5F_t *f, H5F_rdcc_ent_t * ent, hbool_t flush)
{
    H5F_rdcc_t             *rdcc = &(f->shared->rdcc);

    FUNC_ENTER(H5F_istore_preempt, FAIL);

    assert(f);
    assert(ent);
    assert(!ent->locked);
    assert(ent->idx < rdcc->nslots);

    if(flush) {

	/* Flush */
	if(H5F_istore_flush_entry(f, ent, TRUE) < 0) {
	    HRETURN_ERROR(H5E_IO, H5E_WRITEERROR, FAIL,
		    "cannot flush indexed storage buffer");
	}
    }

    else {

	/* Reset, but do not free or remove from list */
	ent->layout = H5O_free(H5O_LAYOUT, ent->layout);
	ent->pline = H5O_free(H5O_PLINE, ent->pline);
	if(ent->chunk != NULL)
	    ent->chunk = H5F_istore_chunk_free(ent->chunk);

    }

    /* Unlink from list */
    if(ent->prev) {
	ent->prev->next = ent->next;
    }
    else {
	rdcc->head = ent->next;
    }
    if(ent->next) {
	ent->next->prev = ent->prev;
    }
    else {
	rdcc->tail = ent->prev;
    }
    ent->prev = ent->next = NULL;

    /* Remove from cache */
    rdcc->slot[ent->idx] = NULL;
    ent->idx = UINT_MAX;
    rdcc->nbytes -= ent->chunk_size;
    --rdcc->nused;

    /* Free */
    H5FL_FREE(H5F_rdcc_ent_t, ent);

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_flush
 *
 * Purpose:	Writes all dirty chunks to disk and optionally preempts them
 *		from the cache.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Thursday, May 21, 1998
 *
 * Modifications:
 *      Pedro Vicente, March 28, 2002
 *      Added TRUE parameter to the call to H5F_istore_preempt
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_flush (H5F_t *f, hbool_t preempt)
{
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);
    int		nerrors=0;
    H5F_rdcc_ent_t	*ent=NULL, *next=NULL;
    
    FUNC_ENTER (H5F_istore_flush, FAIL);

    for (ent=rdcc->head; ent; ent=next) {
	next = ent->next;
	if (preempt) {
	    if (H5F_istore_preempt(f, ent, TRUE )<0) {
		nerrors++;
	    }
	} else {
	    if (H5F_istore_flush_entry(f, ent, FALSE)<0) {
		nerrors++;
	    }
	}
    }
    
    if (nerrors) {
	HRETURN_ERROR (H5E_IO, H5E_CANTFLUSH, FAIL,
		       "unable to flush one or more raw data chunks");
    }
    FUNC_LEAVE (SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_dest
 *
 * Purpose:	Destroy the entire chunk cache by flushing dirty entries,
 *		preempting all entries, and freeing the cache itself.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Thursday, May 21, 1998
 *
 * Modifications:
 *      Pedro Vicente, March 28, 2002
 *      Added TRUE parameter to the call to H5F_istore_preempt
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_dest (H5F_t *f)
{
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);
    int		nerrors=0;
    H5F_rdcc_ent_t	*ent=NULL, *next=NULL;
    
    FUNC_ENTER (H5F_istore_dest, FAIL);

    for (ent=rdcc->head; ent; ent=next) {
#ifdef H5F_ISTORE_DEBUG
	HDfputc('c', stderr);
	HDfflush(stderr);
#endif
	next = ent->next;
	if (H5F_istore_preempt(f, ent, TRUE )<0) {
	    nerrors++;
	}
    }
    if (nerrors) {
	HRETURN_ERROR (H5E_IO, H5E_CANTFLUSH, FAIL,
		       "unable to flush one or more raw data chunks");
    }

    H5FL_ARR_FREE (H5F_rdcc_ent_ptr_t,rdcc->slot);
    HDmemset (rdcc, 0, sizeof(H5F_rdcc_t));
    FUNC_LEAVE (SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_prune
 *
 * Purpose:	Prune the cache by preempting some things until the cache has
 *		room for something which is SIZE bytes.  Only unlocked
 *		entries are considered for preemption.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Thursday, May 21, 1998
 *
 * Modifications:
 *      Pedro Vicente, March 28, 2002
 *      TRUE parameter to the call to H5F_istore_preempt
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_prune (H5F_t *f, size_t size)
{
    int		i, j, nerrors=0;
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);
    size_t		total = f->shared->rdcc_nbytes;
    const int		nmeth=2;	/*number of methods		*/
    int		        w[1];		/*weighting as an interval	*/
    H5F_rdcc_ent_t	*p[2], *cur;	/*list pointers			*/
    H5F_rdcc_ent_t	*n[2];		/*list next pointers		*/

    FUNC_ENTER (H5F_istore_prune, FAIL);

    /*
     * Preemption is accomplished by having multiple pointers (currently two)
     * slide down the list beginning at the head. Pointer p(N+1) will start
     * traversing the list when pointer pN reaches wN percent of the original
     * list.  In other words, preemption method N gets to consider entries in
     * approximate least recently used order w0 percent before method N+1
     * where 100% means tha method N will run to completion before method N+1
     * begins.  The pointers participating in the list traversal are each
     * given a chance at preemption before any of the pointers are advanced.
     */
    w[0] = (int)(rdcc->nused * f->shared->rdcc_w0);
    p[0] = rdcc->head;
    p[1] = NULL;

    while ((p[0] || p[1]) && rdcc->nbytes+size>total) {

	/* Introduce new pointers */
	for (i=0; i<nmeth-1; i++) if (0==w[i]) p[i+1] = rdcc->head;
	
	/* Compute next value for each pointer */
	for (i=0; i<nmeth; i++) n[i] = p[i] ? p[i]->next : NULL;

	/* Give each method a chance */
	for (i=0; i<nmeth && rdcc->nbytes+size>total; i++) {
	    if (0==i && p[0] && !p[0]->locked &&
		((0==p[0]->rd_count && 0==p[0]->wr_count) ||
		 (0==p[0]->rd_count && p[0]->chunk_size==p[0]->wr_count) ||
		 (p[0]->chunk_size==p[0]->rd_count && 0==p[0]->wr_count))) {
		/*
		 * Method 0: Preempt entries that have been completely written
		 * and/or completely read but not entries that are partially
		 * written or partially read.
		 */
		cur = p[0];
#ifdef H5F_ISTORE_DEBUG
		HDputc('.', stderr);
		HDfflush(stderr);
#endif
		
	    } else if (1==i && p[1] && !p[1]->locked) {
		/*
		 * Method 1: Preempt the entry without regard to
		 * considerations other than being locked.  This is the last
		 * resort preemption.
		 */
		cur = p[1];
#ifdef H5F_ISTORE_DEBUG
		HDputc(':', stderr);
		HDfflush(stderr);
#endif
		
	    } else {
		/* Nothing to preempt at this point */
		cur= NULL;
	    }

	    if (cur) {
		for (j=0; j<nmeth; j++) {
		    if (p[j]==cur) p[j] = NULL;
		    if (n[j]==cur) n[j] = cur->next;
		}
		if (H5F_istore_preempt(f, cur, TRUE)<0) nerrors++;
	    }
	}
	
	/* Advance pointers */
	for (i=0; i<nmeth; i++) p[i] = n[i];
	for (i=0; i<nmeth-1; i++) w[i] -= 1;
    }

    if (nerrors) {
	HRETURN_ERROR (H5E_IO, H5E_CANTFLUSH, FAIL,
		       "unable to preempt one or more raw data cache entry");
    }
    FUNC_LEAVE (SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_lock
 *
 * Purpose:	Return a pointer to a dataset chunk.  The pointer points
 *		directly into the chunk cache and should not be freed
 *		by the caller but will be valid until it is unlocked.  The
 *		input value IDX_HINT is used to speed up cache lookups and
 *		it's output value should be given to H5F_istore_unlock().
 *		IDX_HINT is ignored if it is out of range, and if it points
 *		to the wrong entry then we fall back to the normal search
 *		method.
 *
 *		If RELAX is non-zero and the chunk isn't in the cache then
 *		don't try to read it from the file, but just allocate an
 *		uninitialized buffer to hold the result.  This is intended
 *		for output functions that are about to overwrite the entire
 *		chunk.
 *
 * Return:	Success:	Ptr to a file chunk.
 *
 *		Failure:	NULL
 *
 * Programmer:	Robb Matzke
 *              Thursday, May 21, 1998
 *
 * Modifications:
 *		Robb Matzke, 1999-08-02
 *		The split ratios are passed in as part of the data transfer
 *		property list.
 *
 *              Pedro Vicente, March 28, 2002
 *              TRUE parameter to the call to H5F_istore_preempt
 *-------------------------------------------------------------------------
 */
static void *
H5F_istore_lock(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
		const H5O_pline_t *pline, const H5O_fill_t *fill,
		const hssize_t offset[], hbool_t relax,
		unsigned *idx_hint/*in,out*/)
{
    int		idx=0;			/*hash index number	*/
    hsize_t	temp_idx=0;			/* temporary index number	*/
    hbool_t		found = FALSE;		/*already in cache?	*/
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);/*raw data chunk cache*/
    H5F_rdcc_ent_t	*ent = NULL;		/*cache entry		*/
    unsigned		u;			/*counters		*/
    H5F_istore_ud1_t	udata;			/*B-tree pass-through	*/
    size_t		chunk_size=0;		/*size of a chunk	*/
    hsize_t             tempchunk_size;
    size_t		chunk_alloc=0;		/*allocated chunk size	*/
    herr_t		status;			/*func return status	*/
    void		*chunk=NULL;		/*the file chunk	*/
    void		*ret_value=NULL;	/*return value		*/
    H5P_genplist_t *plist=NULL;                 /* Property list */

    FUNC_ENTER (H5F_istore_lock, NULL);

    if (rdcc->nslots>0) {
        /* We don't care about loss of precision in the following statement. */
        for (u=0, temp_idx=0; u<layout->ndims; u++) {
            temp_idx *= layout->dim[u];
            temp_idx += offset[u];
        }
        temp_idx += (hsize_t)(layout->addr);
        idx=H5F_HASH(f,temp_idx);
        ent = rdcc->slot[idx];
        
        if (ent && layout->ndims==ent->layout->ndims &&
                H5F_addr_eq(layout->addr, ent->layout->addr)) {
            for (u=0, found=TRUE; u<ent->layout->ndims; u++) {
                if (offset[u]!=ent->offset[u]) {
                    found = FALSE;
                    break;
                }
            }
        }
    }

    if (found) {
        /*
         * Already in the cache.  Count a hit.
         */
        rdcc->nhits++;

    } else if (!found && relax) {
        /*
         * Not in the cache, but we're about to overwrite the whole thing
         * anyway, so just allocate a buffer for it but don't initialize that
         * buffer with the file contents. Count this as a hit instead of a
         * miss because we saved ourselves lots of work.
         */
#ifdef H5F_ISTORE_DEBUG
        HDputc('w', stderr);
        HDfflush(stderr);
#endif
        rdcc->nhits++;
        for (u=0, tempchunk_size=1; u<layout->ndims; u++) {
            tempchunk_size *= layout->dim[u];
        }
        H5_ASSIGN_OVERFLOW(chunk_size,tempchunk_size,hsize_t,size_t);
        chunk_alloc = chunk_size;
        if (NULL==(chunk=H5F_istore_chunk_alloc (chunk_alloc))) {
            HGOTO_ERROR (H5E_RESOURCE, H5E_NOSPACE, NULL,
                 "memory allocation failed for raw data chunk");
        }
        
    } else {
        /*
         * Not in the cache.  Read it from the file and count this as a miss
         * if it's in the file or an init if it isn't.
         */
        for (u=0, tempchunk_size=1; u<layout->ndims; u++) {
            udata.key.offset[u] = offset[u];
            tempchunk_size *= layout->dim[u];
        }
        H5_ASSIGN_OVERFLOW(chunk_size,tempchunk_size,hsize_t,size_t);
        chunk_alloc = chunk_size;
        udata.mesg = *layout;
        udata.addr = HADDR_UNDEF;
        status = H5B_find (f, H5B_ISTORE, layout->addr, &udata);
        H5E_clear ();
        if (NULL==(chunk = H5F_istore_chunk_alloc (chunk_alloc))) {
            HGOTO_ERROR (H5E_RESOURCE, H5E_NOSPACE, NULL,
                 "memory allocation failed for raw data chunk");
        }
        if (status>=0 && H5F_addr_defined(udata.addr)) {
            /*
             * The chunk exists on disk.
             */
            if (H5F_block_read(f, H5FD_MEM_DRAW, udata.addr, udata.key.nbytes, H5P_DATASET_XFER_DEFAULT,
                       chunk)<0) {
                HGOTO_ERROR (H5E_IO, H5E_READERROR, NULL,
                     "unable to read raw data chunk");
            }
            if (H5Z_pipeline(f, pline, H5Z_FLAG_REVERSE,
                     &(udata.key.filter_mask), &(udata.key.nbytes),
                     &chunk_alloc, &chunk)<0 || udata.key.nbytes!=chunk_size) {
                HGOTO_ERROR(H5E_PLINE, H5E_READERROR, NULL,
                    "data pipeline read failed");
            }
            rdcc->nmisses++;
        } else if (fill && fill->buf) {
            /*
             * The chunk doesn't exist in the file.  Replicate the fill
             * value throughout the chunk.
             */
            assert(0==chunk_size % fill->size);
            H5V_array_fill(chunk, fill->buf, fill->size, chunk_size/fill->size);
            rdcc->ninits++;
        } else {
            /*
             * The chunk doesn't exist in the file and no fill value was
             * specified.  Assume all zeros.
             */
            HDmemset (chunk, 0, chunk_size);
            rdcc->ninits++;
        }
    }
    assert (found || chunk_size>0);
    
    if (!found && rdcc->nslots>0 && chunk_size<=f->shared->rdcc_nbytes &&
            (!ent || !ent->locked)) {
        /*
         * Add the chunk to the cache only if the slot is not already locked.
         * Preempt enough things from the cache to make room.
         */
        if (ent) {
#ifdef H5F_ISTORE_DEBUG
            HDputc('#', stderr);
            HDfflush(stderr);
#endif
#if 0
            HDfprintf(stderr, "\ncollision %3d %10a {",
                  idx, ent->layout->addr);
            for (u=0; u<layout->ndims; u++) {
                HDfprintf(stderr, "%s%Zu", u?",":"", ent->offset[u]);
            }
            HDfprintf(stderr, "}\n              %10a {", layout->addr);
            for (u=0; u<layout->ndims; u++) {
                HDfprintf(stderr, "%s%Zu", u?",":"", offset[u]);
            }
            fprintf(stderr, "}\n");
#endif
            if (H5F_istore_preempt(f, ent, TRUE)<0) {
                HGOTO_ERROR(H5E_IO, H5E_CANTINIT, NULL,
                    "unable to preempt chunk from cache");
            }
        }
        if (H5F_istore_prune(f, chunk_size)<0) {
            HGOTO_ERROR(H5E_IO, H5E_CANTINIT, NULL,
                "unable to preempt chunk(s) from cache");
        }

        /* Create a new entry */
        ent = H5FL_ALLOC(H5F_rdcc_ent_t,0);
        ent->locked = 0;
        ent->dirty = FALSE;
        ent->chunk_size = chunk_size;
        ent->alloc_size = chunk_size;
        ent->layout = H5O_copy(H5O_LAYOUT, layout, NULL);
        ent->pline = H5O_copy(H5O_PLINE, pline, NULL);
        for (u=0; u<layout->ndims; u++) {
            ent->offset[u] = offset[u];
        }
        ent->rd_count = chunk_size;
        ent->wr_count = chunk_size;
        ent->chunk = chunk;
        
        assert(H5I_GENPROP_LST==H5I_get_type(dxpl_id));
        assert(TRUE==H5P_isa_class(dxpl_id,H5P_DATASET_XFER));
        plist=H5I_object(dxpl_id);
        assert(plist!=NULL);
        H5P_get(plist,H5D_XFER_BTREE_SPLIT_RATIO_NAME,&(ent->split_ratios));
        
        /* Add it to the cache */
        assert(NULL==rdcc->slot[idx]);
        rdcc->slot[idx] = ent;
        ent->idx = idx;
        rdcc->nbytes += chunk_size;
        rdcc->nused++;

        /* Add it to the linked list */
        ent->next = NULL;
        if (rdcc->tail) {
            rdcc->tail->next = ent;
            ent->prev = rdcc->tail;
            rdcc->tail = ent;
        } else {
            rdcc->head = rdcc->tail = ent;
            ent->prev = NULL;
        }
        found = TRUE;
        
    } else if (!found) {
        /*
         * The chunk is larger than the entire cache so we don't cache it.
         * This is the reason all those arguments have to be repeated for the
         * unlock function.
         */
        ent = NULL;
        idx = UINT_MAX;

    } else if (found) {
        /*
         * The chunk is not at the beginning of the cache; move it backward
         * by one slot.  This is how we implement the LRU preemption
         * algorithm.
         */
        if (ent->next) {
            if (ent->next->next) {
                ent->next->next->prev = ent;
            } else {
                rdcc->tail = ent;
            }
            ent->next->prev = ent->prev;
            if (ent->prev) {
                ent->prev->next = ent->next;
            } else {
                rdcc->head = ent->next;
            }
            ent->prev = ent->next;
            ent->next = ent->next->next;
            ent->prev->next = ent;
        }
    }

    /* Lock the chunk into the cache */
    if (ent) {
        assert (!ent->locked);
        ent->locked = TRUE;
        chunk = ent->chunk;
    }

    if (idx_hint)
        *idx_hint = idx;
    ret_value = chunk;
    
 done:
    if (!ret_value)
        H5F_istore_chunk_free (chunk);
    FUNC_LEAVE (ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_unlock
 *
 * Purpose:	Unlocks a previously locked chunk. The LAYOUT, COMP, and
 *		OFFSET arguments should be the same as for H5F_rdcc_lock().
 *		The DIRTY argument should be set to non-zero if the chunk has
 *		been modified since it was locked. The IDX_HINT argument is
 *		the returned index hint from the lock operation and BUF is
 *		the return value from the lock.
 *
 *		The NACCESSED argument should be the number of bytes accessed
 *		for reading or writing (depending on the value of DIRTY).
 *		It's only purpose is to provide additional information to the
 *		preemption policy.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Thursday, May 21, 1998
 *
 * Modifications:
 *		Robb Matzke, 1999-08-02
 *		The split_ratios are passed as part of the data transfer
 *		property list.
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_unlock(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
		  const H5O_pline_t *pline, hbool_t dirty,
		  const hssize_t offset[], unsigned *idx_hint,
		  uint8_t *chunk, size_t naccessed)
{
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);
    H5F_rdcc_ent_t	*ent = NULL;
    int		found = -1;
    unsigned		u;
    H5P_genplist_t *plist;      /* Property list */
    
    FUNC_ENTER (H5F_istore_unlock, FAIL);

    if (UINT_MAX==*idx_hint) {
	/*not in cache*/
    } else {
	assert(*idx_hint<rdcc->nslots);
	assert(rdcc->slot[*idx_hint]);
	assert(rdcc->slot[*idx_hint]->chunk==chunk);
	found = *idx_hint;
    }
    
    if (found<0) {
        /*
         * It's not in the cache, probably because it's too big.  If it's
         * dirty then flush it to disk.  In any case, free the chunk.
         * Note: we have to copy the layout and filter messages so we
         *	 don't discard the `const' qualifier.
         */
        if (dirty) {
            H5F_rdcc_ent_t x;
            hsize_t tempchunk_size;

            HDmemset (&x, 0, sizeof x);
            x.dirty = TRUE;
            x.layout = H5O_copy (H5O_LAYOUT, layout, NULL);
            x.pline = H5O_copy (H5O_PLINE, pline, NULL);
            for (u=0, tempchunk_size=1; u<layout->ndims; u++) {
                x.offset[u] = offset[u];
                tempchunk_size *= layout->dim[u];
            }
            H5_ASSIGN_OVERFLOW(x.chunk_size,tempchunk_size,hsize_t,size_t);
            x.alloc_size = x.chunk_size;
            x.chunk = chunk;

            assert(H5I_GENPROP_LST==H5I_get_type(dxpl_id));
            assert(TRUE==H5P_isa_class(dxpl_id,H5P_DATASET_XFER));
            plist=H5I_object(dxpl_id);
            assert(plist!=NULL);
            H5P_get(plist,H5D_XFER_BTREE_SPLIT_RATIO_NAME,&(x.split_ratios));
            
            H5F_istore_flush_entry (f, &x, TRUE);
        } else {
            if(chunk)
                H5F_istore_chunk_free (chunk);
        }
    } else {
        /*
         * It's in the cache so unlock it.
         */
        ent = rdcc->slot[found];
        assert (ent->locked);
        if (dirty) {
            ent->dirty = TRUE;
            ent->wr_count -= MIN (ent->wr_count, naccessed);
        } else {
            ent->rd_count -= MIN (ent->rd_count, naccessed);
        }
        ent->locked = FALSE;
    }
    
    FUNC_LEAVE (SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_read
 *
 * Purpose:	Reads a multi-dimensional buffer from (part of) an indexed raw
 *		storage array.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *		Wednesday, October 15, 1997
 *
 * Modifications:
 *		Robb Matzke, 1999-08-02
 *		The data transfer property list is passed as an object ID
 *		since that's how the virtual file layer wants it.
 *
 *              Quincey Koziol, 2002-04-02
 *              Enable hyperslab I/O into memory buffer
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_read(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
                H5P_genplist_t *dc_plist,
                const hsize_t size_m[], const hssize_t offset_m[],
		const hssize_t offset_f[], const hsize_t size[], void *buf)
{
    hsize_t		idx_cur[H5O_LAYOUT_NDIMS];
    hsize_t		idx_min[H5O_LAYOUT_NDIMS];
    hsize_t		idx_max[H5O_LAYOUT_NDIMS];
    hsize_t		sub_size[H5O_LAYOUT_NDIMS];
    hssize_t		offset_wrt_chunk[H5O_LAYOUT_NDIMS];
    hssize_t		sub_offset_m[H5O_LAYOUT_NDIMS];
    hssize_t		chunk_offset[H5O_LAYOUT_NDIMS];
    int		i, carry;
    unsigned		u;
    hsize_t		naccessed;		/*bytes accessed in chnk*/
    uint8_t		*chunk=NULL;		/*ptr to a chunk buffer	*/
    unsigned		idx_hint=0;		/*cache index hint	*/
    hsize_t		chunk_size;     /* Bytes in chunk */
    haddr_t	        chunk_addr;     /* Chunk address on disk */
    H5O_pline_t         pline;          /* I/O pipeline information */
    H5O_fill_t          fill;           /* Fill value information */
    herr_t              ret_value = SUCCEED;            /* Return value */

    FUNC_ENTER(H5F_istore_read, FAIL);

    /* Check args */
    assert(f);
    assert(layout && H5D_CHUNKED==layout->type);
    assert(layout->ndims>0 && layout->ndims<=H5O_LAYOUT_NDIMS);
    assert(H5F_addr_defined(layout->addr));
    assert(size_m);
    assert(offset_m);
    assert(offset_f);
    assert(size);
    assert(buf);

    /* Get necessary properties from property list */
    if(H5P_get(dc_plist, H5D_CRT_FILL_VALUE_NAME, &fill) < 0)
        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get fill value");
    if(H5P_get(dc_plist, H5D_CRT_DATA_PIPELINE_NAME, &pline) < 0)
        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get data pipeline");

    /* Compute chunk size */
    for (u=0, chunk_size=1; u<layout->ndims; u++)
        chunk_size *= layout->dim[u];
    
#ifndef NDEBUG
    for (u=0; u<layout->ndims; u++) {
        assert(offset_f[u]>=0); /*negative offsets not supported*/
        assert(offset_m[u]>=0); /*negative offsets not supported*/
        assert(size[u]<SIZET_MAX);
        assert(offset_m[u]+(hssize_t)size[u]<=(hssize_t)size_m[u]);
        assert(layout->dim[u]>0);
    }
#endif

    /*
     * Set up multi-dimensional counters (idx_min, idx_max, and idx_cur) and
     * loop through the chunks copying each to its final destination in the
     * application buffer.
     */
    for (u=0; u<layout->ndims; u++) {
        idx_min[u] = offset_f[u] / layout->dim[u];
        idx_max[u] = (offset_f[u]+size[u]-1) / layout->dim[u] + 1;
        idx_cur[u] = idx_min[u];
    }

    /* Loop over all chunks */
    carry=0;
    while (carry==0) {
        for (u=0, naccessed=1; u<layout->ndims; u++) {
            /* The location and size of the chunk being accessed */
            assert(layout->dim[u] < HSSIZET_MAX);
            chunk_offset[u] = idx_cur[u] * (hssize_t)(layout->dim[u]);

            /* The offset and size wrt the chunk */
            offset_wrt_chunk[u] = MAX(offset_f[u], chunk_offset[u]) -
                      chunk_offset[u];
            sub_size[u] = MIN((idx_cur[u]+1)*layout->dim[u],
                      offset_f[u]+size[u]) -
                  (chunk_offset[u] + offset_wrt_chunk[u]);
            naccessed *= sub_size[u];
            
            /* Offset into mem buffer */
            sub_offset_m[u] = chunk_offset[u] + offset_wrt_chunk[u] +
                      offset_m[u] - offset_f[u];
        }

        /* Get the address of this chunk on disk */
        chunk_addr=H5F_istore_get_addr(f, layout, chunk_offset);

        /*
         * If the chunk is too large to load into the cache and it has no
         * filters in the pipeline (i.e. not compressed) and if the address
         * for the chunk has been defined, then don't load the chunk into the
         * cache, just read the data from it directly.
         */
        if ((chunk_size>f->shared->rdcc_nbytes && pline.nfilters==0 &&
                chunk_addr!=HADDR_UNDEF)

#ifdef H5_HAVE_PARALLEL
        /*
         * If MPIO is used, must bypass the chunk-cache scheme because other
         * MPI processes could be writing to other elements in the same chunk.
         * Do a direct write-through of only the elements requested.
         */
            || IS_H5FD_MPIO(f)
#endif /* H5_HAVE_PARALLEL */
            ) {
            H5O_layout_t	l;	/* temporary layout */

#ifdef H5_HAVE_PARALLEL
            /* Additional sanity checks when operating in parallel */
            if (chunk_addr==HADDR_UNDEF || pline.nfilters>0)
                HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "unable to locate raw data chunk");
#endif /* H5_HAVE_PARALLEL */
            
            /*
             * use default transfer mode as we do not support collective
             * transfer mode since each data write could decompose into
             * multiple chunk writes and we are not doing the calculation yet.
             */
            l.type = H5D_CONTIGUOUS;
            l.ndims = layout->ndims;
            for (u=l.ndims; u-- > 0; /*void*/)
                l.dim[u] = layout->dim[u];
            l.addr = chunk_addr;
            if (H5F_arr_read(f, H5P_DATASET_XFER_DEFAULT, &l, dc_plist,
                     sub_size, size_m, sub_offset_m, offset_wrt_chunk, buf)<0)
                HGOTO_ERROR (H5E_IO, H5E_READERROR, FAIL, "unable to read raw data from file");
        } /* end if */
        else {
            /*
             * Lock the chunk, transfer data to the application, then unlock
             * the chunk.
             */
            if (NULL==(chunk=H5F_istore_lock(f, dxpl_id, layout, &pline, &fill,
                         chunk_offset, FALSE, &idx_hint)))
                HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "unable to read raw data chunk");
            H5V_hyper_copy(layout->ndims, sub_size, size_m, sub_offset_m,
                   (void*)buf, layout->dim, offset_wrt_chunk, chunk);
            H5_CHECK_OVERFLOW(naccessed,hsize_t,size_t);
            if (H5F_istore_unlock(f, dxpl_id, layout, &pline, FALSE,
                      chunk_offset, &idx_hint, chunk, (size_t)naccessed)<0)
                HGOTO_ERROR(H5E_IO, H5E_READERROR, FAIL, "unable to unlock raw data chunk");
        } /* end else */

        /* Increment indices */
        for (i=(int)(layout->ndims-1), carry=1; i>=0 && carry; --i) {
            if (++idx_cur[i]>=idx_max[i])
                idx_cur[i] = idx_min[i];
            else
                carry = 0;
        }
    }

done:
    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_write
 *
 * Purpose:	Writes a multi-dimensional buffer to (part of) an indexed raw
 *		storage array.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *		Wednesday, October 15, 1997
 *
 * Modifications:
 *		Robb Matzke, 1999-08-02
 *		The data transfer property list is passed as an object ID
 *		since that's how the virtual file layer wants it.
 *
 *              Quincey Koziol, 2002-04-02
 *              Enable hyperslab I/O into memory buffer
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_write(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
                 H5P_genplist_t *dc_plist,
                 const hsize_t size_m[], const hssize_t offset_m[],
		 const hssize_t offset_f[], const hsize_t size[],
		 const void *buf)
{
    int		i, carry;
    unsigned		u;
    hsize_t		idx_cur[H5O_LAYOUT_NDIMS];
    hsize_t		idx_min[H5O_LAYOUT_NDIMS];
    hsize_t		idx_max[H5O_LAYOUT_NDIMS];
    hsize_t		sub_size[H5O_LAYOUT_NDIMS];
    hssize_t	chunk_offset[H5O_LAYOUT_NDIMS];
    hssize_t	offset_wrt_chunk[H5O_LAYOUT_NDIMS];
    hssize_t	sub_offset_m[H5O_LAYOUT_NDIMS];
    uint8_t		*chunk=NULL;
    unsigned		idx_hint=0;
    hsize_t		chunk_size, naccessed;
    haddr_t	        chunk_addr;     /* Chunk address on disk */
    H5O_pline_t         pline;          /* I/O pipeline information */
    H5O_fill_t          fill;           /* Fill value information */
    herr_t              ret_value = SUCCEED;            /* Return value */
    
    FUNC_ENTER(H5F_istore_write, FAIL);

    /* Check args */
    assert(f);
    assert(layout && H5D_CHUNKED==layout->type);
    assert(layout->ndims>0 && layout->ndims<=H5O_LAYOUT_NDIMS);
    assert(H5F_addr_defined(layout->addr));
    assert(size_m);
    assert(offset_m);
    assert(offset_f);
    assert(size);
    assert(buf);

    /* Get necessary properties from property list */
    if(H5P_get(dc_plist, H5D_CRT_FILL_VALUE_NAME, &fill) < 0)
        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get fill value");
    if(H5P_get(dc_plist, H5D_CRT_DATA_PIPELINE_NAME, &pline) < 0)
        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get data pipeline");

    /* Compute chunk size */
    for (u=0, chunk_size=1; u<layout->ndims; u++)
        chunk_size *= layout->dim[u];

#ifndef NDEBUG
    for (u=0; u<layout->ndims; u++) {
        assert(offset_f[u]>=0); /*negative offsets not supported*/
        assert(offset_m[u]>=0); /*negative offsets not supported*/
        assert(size[u]<SIZET_MAX);
        assert(offset_m[u]+(hssize_t)size[u]<=(hssize_t)size_m[u]);
        assert(layout->dim[u]>0);
    }
#endif

    /*
     * Set up multi-dimensional counters (idx_min, idx_max, and idx_cur) and
     * loop through the chunks copying each chunk from the application to the
     * chunk cache.
     */
    for (u=0; u<layout->ndims; u++) {
        idx_min[u] = offset_f[u] / layout->dim[u];
        idx_max[u] = (offset_f[u]+size[u]-1) / layout->dim[u] + 1;
        idx_cur[u] = idx_min[u];
    }


    /* Loop over all chunks */
    carry=0;
    while (carry==0) {
        for (u=0, naccessed=1; u<layout->ndims; u++) {
            /* The location and size of the chunk being accessed */
            assert(layout->dim[u] < HSSIZET_MAX);
            chunk_offset[u] = idx_cur[u] * (hssize_t)(layout->dim[u]);

            /* The offset and size wrt the chunk */
            offset_wrt_chunk[u] = MAX(offset_f[u], chunk_offset[u]) -
                      chunk_offset[u];
            sub_size[u] = MIN((idx_cur[u]+1)*layout->dim[u],
                      offset_f[u]+size[u]) -
                  (chunk_offset[u] + offset_wrt_chunk[u]);
            naccessed *= sub_size[u];

            /* Offset into mem buffer */
            sub_offset_m[u] = chunk_offset[u] + offset_wrt_chunk[u] +
                      offset_m[u] - offset_f[u];
        }

        /* Get the address of this chunk on disk */
        chunk_addr=H5F_istore_get_addr(f, layout, chunk_offset);

        /*
         * If the chunk is too large to load into the cache and it has no
         * filters in the pipeline (i.e. not compressed) and if the address
         * for the chunk has been defined, then don't load the chunk into the
         * cache, just write the data to it directly.
         */
        if ((chunk_size>f->shared->rdcc_nbytes && pline.nfilters==0 &&
                chunk_addr!=HADDR_UNDEF)

#ifdef H5_HAVE_PARALLEL
        /*
         * If MPIO is used, must bypass the chunk-cache scheme because other
         * MPI processes could be writing to other elements in the same chunk.
         * Do a direct write-through of only the elements requested.
         */
            || IS_H5FD_MPIO(f)
#endif /* H5_HAVE_PARALLEL */
            ) {
            H5O_layout_t	l;	/* temporary layout */

#ifdef H5_HAVE_PARALLEL
            /* Additional sanity check when operating in parallel */
            if (chunk_addr==HADDR_UNDEF || pline.nfilters>0)
                HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "unable to locate raw data chunk");
#endif /* H5_HAVE_PARALLEL */
            
            /*
             * use default transfer mode as we do not support collective
             * transfer mode since each data write could decompose into
             * multiple chunk writes and we are not doing the calculation yet.
             */
            l.type = H5D_CONTIGUOUS;
            l.ndims = layout->ndims;
            for (u=l.ndims; u-- > 0; /*void*/)
                l.dim[u] = layout->dim[u];
            l.addr = chunk_addr;
            if (H5F_arr_write(f, H5P_DATASET_XFER_DEFAULT, &l, dc_plist,
                     sub_size, size_m, sub_offset_m, offset_wrt_chunk, buf)<0)
                HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "unable to write raw data to file");
        } /* end if */
        else {
            /*
             * Lock the chunk, copy from application to chunk, then unlock the
             * chunk.
             */
            if (NULL==(chunk=H5F_istore_lock(f, dxpl_id, layout, &pline, &fill,
                     chunk_offset, (hbool_t)(naccessed==chunk_size), &idx_hint)))
                HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "unable to read raw data chunk");
            H5V_hyper_copy(layout->ndims, sub_size,
               layout->dim, offset_wrt_chunk, chunk, size_m, sub_offset_m, buf);
            H5_CHECK_OVERFLOW(naccessed,hsize_t,size_t);
            if (H5F_istore_unlock(f, dxpl_id, layout, &pline, TRUE,
                      chunk_offset, &idx_hint, chunk, (size_t)naccessed)<0)
                HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "uanble to unlock raw data chunk");
        } /* end else */
        
        /* Increment indices */
        for (i=layout->ndims-1, carry=1; i>=0 && carry; --i) {
            if (++idx_cur[i]>=idx_max[i])
                idx_cur[i] = idx_min[i];
            else
                carry = 0;
        }
    } /* end while */

done:
    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_create
 *
 * Purpose:	Creates a new indexed-storage B-tree and initializes the
 *		istore struct with information about the storage.  The
 *		struct should be immediately written to the object header.
 *
 *		This function must be called before passing ISTORE to any of
 *		the other indexed storage functions!
 *
 * Return:	Non-negative on success (with the ISTORE argument initialized
 *		and ready to write to an object header). Negative on failure.
 *
 * Programmer:	Robb Matzke
 *		Tuesday, October 21, 1997
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_create(H5F_t *f, H5O_layout_t *layout /*out */ )
{
    H5F_istore_ud1_t	udata;
#ifndef NDEBUG
    unsigned			u;
#endif

    FUNC_ENTER(H5F_istore_create, FAIL);

    /* Check args */
    assert(f);
    assert(layout && H5D_CHUNKED == layout->type);
    assert(layout->ndims > 0 && layout->ndims <= H5O_LAYOUT_NDIMS);
#ifndef NDEBUG
    for (u = 0; u < layout->ndims; u++) {
	assert(layout->dim[u] > 0);
    }
#endif

    udata.mesg.ndims = layout->ndims;
    if (H5B_create(f, H5B_ISTORE, &udata, &(layout->addr)/*out*/) < 0) {
	HRETURN_ERROR(H5E_IO, H5E_CANTINIT, FAIL, "can't create B-tree");
    }
    
    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_allocated
 *
 * Purpose:	Return the number of bytes allocated in the file for storage
 *		of raw data under the specified B-tree (ADDR is the address
 *		of the B-tree).
 *
 * Return:	Success:	Number of bytes stored in all chunks.
 *
 *		Failure:	0
 *
 * Programmer:	Robb Matzke
 *              Wednesday, April 21, 1999
 *
 * Modifications:
 *		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *-------------------------------------------------------------------------
 */
hsize_t
H5F_istore_allocated(H5F_t *f, unsigned ndims, haddr_t addr)
{
    H5F_istore_ud1_t	udata;

    FUNC_ENTER(H5F_istore_nchunks, 0);

    HDmemset(&udata, 0, sizeof udata);
    udata.mesg.ndims = ndims;
    if (H5B_iterate(f, H5B_ISTORE, H5F_istore_iter_allocated, addr, &udata)<0) {
        HRETURN_ERROR(H5E_IO, H5E_CANTINIT, 0,
		      "unable to iterate over chunk B-tree");
    }
    FUNC_LEAVE(udata.total_storage);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_dump_btree
 *
 * Purpose:	Prints information about the storage B-tree to the specified
 *		stream.
 *
 * Return:	Success:	Non-negative
 *
 *		Failure:	negative
 *
 * Programmer:	Robb Matzke
 *              Wednesday, April 28, 1999
 *
 * Modifications:
 *		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_dump_btree(H5F_t *f, FILE *stream, unsigned ndims, haddr_t addr)
{
    H5F_istore_ud1_t	udata;

    FUNC_ENTER(H5F_istore_dump_btree, FAIL);

    HDmemset(&udata, 0, sizeof udata);
    udata.mesg.ndims = ndims;
    udata.stream = stream;
    if (H5B_iterate(f, H5B_ISTORE, H5F_istore_iter_dump, addr, &udata)<0) {
        HRETURN_ERROR(H5E_IO, H5E_CANTINIT, 0,
		      "unable to iterate over chunk B-tree");
    }
    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_stats
 *
 * Purpose:	Print raw data cache statistics to the debug stream.  If
 *		HEADERS is non-zero then print table column headers,
 *		otherwise assume that the H5AC layer has already printed them.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Thursday, May 21, 1998
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_stats (H5F_t *f, hbool_t headers)
{
    H5F_rdcc_t	*rdcc = &(f->shared->rdcc);
    double	miss_rate;
    char	ascii[32];
    
    FUNC_ENTER (H5F_istore_stats, FAIL);
    if (!H5DEBUG(AC)) HRETURN(SUCCEED);

    if (headers) {
        fprintf(H5DEBUG(AC), "H5F: raw data cache statistics for file %s\n",
            f->name);
        fprintf(H5DEBUG(AC), "   %-18s %8s %8s %8s %8s+%-8s\n",
            "Layer", "Hits", "Misses", "MissRate", "Inits", "Flushes");
        fprintf(H5DEBUG(AC), "   %-18s %8s %8s %8s %8s-%-8s\n",
            "-----", "----", "------", "--------", "-----", "-------");
    }

#ifdef H5AC_DEBUG
    if (H5DEBUG(AC)) headers = TRUE;
#endif

    if (headers) {
        if (rdcc->nhits>0 || rdcc->nmisses>0) {
            miss_rate = 100.0 * rdcc->nmisses /
                    (rdcc->nhits + rdcc->nmisses);
        } else {
            miss_rate = 0.0;
        }
        if (miss_rate > 100) {
            sprintf(ascii, "%7d%%", (int) (miss_rate + 0.5));
        } else {
            sprintf(ascii, "%7.2f%%", miss_rate);
        }

        fprintf(H5DEBUG(AC), "   %-18s %8u %8u %7s %8d+%-9ld\n",
            "raw data chunks", rdcc->nhits, rdcc->nmisses, ascii,
            rdcc->ninits, (long)(rdcc->nflushes)-(long)(rdcc->ninits));
    }

    FUNC_LEAVE (SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_debug
 *
 * Purpose:	Debugs a B-tree node for indexed raw data storage.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Robb Matzke
 *              Thursday, April 16, 1998
 *
 * Modifications:
 *		Robb Matzke, 1999-07-28
 *		The ADDR argument is passed by value.
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_debug(H5F_t *f, haddr_t addr, FILE * stream, int indent,
		 int fwidth, int ndims)
{
    H5F_istore_ud1_t	udata;
    
    FUNC_ENTER (H5F_istore_debug, FAIL);

    HDmemset (&udata, 0, sizeof udata);
    udata.mesg.ndims = ndims;

    H5B_debug (f, addr, stream, indent, fwidth, H5B_ISTORE, &udata);

    FUNC_LEAVE (SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_get_addr
 *
 * Purpose:	Get the file address of a chunk if file space has been
 *		assigned.  Save the retrieved information in the udata
 *		supplied.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Programmer:	Albert Cheng
 *              June 27, 1998
 *
 * Modifications:
 *              Modified to return the address instead of returning it through
 *              a parameter - QAK, 1/30/02
 *
 *-------------------------------------------------------------------------
 */
static haddr_t
H5F_istore_get_addr(H5F_t *f, const H5O_layout_t *layout,
		    const hssize_t offset[])
{
    H5F_istore_ud1_t	udata;                  /* Information about a chunk */
    unsigned	u;
    haddr_t	ret_value=HADDR_UNDEF;		/* Return value */
    
    FUNC_ENTER (H5F_istore_get_addr, HADDR_UNDEF);

    assert(f);
    assert(layout && (layout->ndims > 0));
    assert(offset);

    /* Initialize the information about the chunk we are looking for */
    for (u=0; u<layout->ndims; u++)
	udata.key.offset[u] = offset[u];
    udata.mesg = *layout;
    udata.addr = HADDR_UNDEF;

    /* Go get the chunk information */
    if (H5B_find (f, H5B_ISTORE, layout->addr, &udata)<0) {
        H5E_clear();
	HGOTO_ERROR(H5E_BTREE,H5E_NOTFOUND,HADDR_UNDEF,"Can't locate chunk info");
    } /* end if */

    /* Success!  Set the return value */
    ret_value=udata.addr;

done:
    FUNC_LEAVE (ret_value);
} /* H5F_istore_get_addr() */


/*-------------------------------------------------------------------------
 * Function:	H5F_istore_allocate
 *
 * Purpose:	Allocate file space for all chunks that are not allocated yet.
 *		Return SUCCEED if all needed allocation succeed, otherwise
 *		FAIL.
 *
 * Return:	Non-negative on success/Negative on failure
 *
 * Note:	Current implementation relies on cache_size being 0,
 *		thus no chunk is cashed and written to disk immediately
 *		when a chunk is unlocked (via H5F_istore_unlock)
 *		This should be changed to do a direct flush independent
 *		of the cache value.
 *
 * Programmer:	Albert Cheng
 *		June 26, 1998
 *
 * Modifications:
 *		rky, 1998-09-23
 *		Added barrier to preclude racing with data writes.
 *
 *		rky, 1998-12-07
 *		Added Wait-Signal wrapper around unlock-lock critical region
 *		to prevent race condition (unlock reads, lock writes the
 *		chunk).
 *
 * 		Robb Matzke, 1999-08-02
 *		The split_ratios are passed in as part of the data transfer
 *		property list.
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_allocate(H5F_t *f, hid_t dxpl_id, const H5O_layout_t *layout,
		    const hsize_t *space_dim, H5P_genplist_t *dc_plist)
{

    int		i, carry;
    unsigned	u;
    hssize_t	chunk_offset[H5O_LAYOUT_NDIMS];
    uint8_t	*chunk=NULL;
    unsigned	idx_hint=0;
    hsize_t	chunk_size;
    H5O_pline_t pline;         /* I/O pipeline information */
    H5O_fill_t fill;           /* Fill value information */
    herr_t	ret_value=SUCCEED;	/* Return value */
    
    FUNC_ENTER(H5F_istore_allocate, FAIL);

    /* Check args */
    assert(f);
    assert(space_dim);
    assert(layout && H5D_CHUNKED==layout->type);
    assert(layout->ndims>0 && layout->ndims<=H5O_LAYOUT_NDIMS);
    assert(H5F_addr_defined(layout->addr));

    /* Get necessary properties from property list */
    if(H5P_get(dc_plist, H5D_CRT_FILL_VALUE_NAME, &fill) < 0)
        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get fill value");
    if(H5P_get(dc_plist, H5D_CRT_DATA_PIPELINE_NAME, &pline) < 0)
        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get data pipeline");

    /*
     * Setup indice to go through all chunks. (Future improvement
     * should allocate only chunks that have no file space assigned yet.
     */
    for (u=0, chunk_size=1; u<layout->ndims; u++) {
        chunk_offset[u]=0;
        chunk_size *= layout->dim[u];
    } /* end for */

    /* Loop over all chunks */
    carry=0;
    while (carry==0) {
        /* No file space assigned yet.  Allocate it. */
        /* The following needs improvement like calling the */
        /* allocation directly rather than indirectly using the */
        /* allocation effect in the unlock process. */

        /*
         * Lock the chunk, copy from application to chunk, then unlock the
         * chunk.
         */

#ifdef H5_HAVE_PARALLEL
        /* rky 981207 Serialize access to this critical region. */
        if (SUCCEED!= H5FD_mpio_wait_for_left_neighbor(f->shared->lf))
            HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "unable to lock the data chunk");
#endif
        if (NULL==(chunk=H5F_istore_lock(f, dxpl_id, layout, &pline,
                      &fill, chunk_offset, FALSE, &idx_hint)))
            HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "unable to read raw data chunk");

        H5_CHECK_OVERFLOW(chunk_size,hsize_t,size_t);
        if (H5F_istore_unlock(f, dxpl_id, layout, &pline, TRUE,
                  chunk_offset, &idx_hint, chunk, (size_t)chunk_size)<0)
            HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "uanble to unlock raw data chunk");
#ifdef H5_HAVE_PARALLEL
        if (SUCCEED!= H5FD_mpio_signal_right_neighbor(f->shared->lf))
            HGOTO_ERROR (H5E_IO, H5E_WRITEERROR, FAIL, "unable to unlock the data chunk");
#endif
	
        /* Increment indices */
        for (i=layout->ndims-1, carry=1; i>=0 && carry; --i) {
            chunk_offset[i] += layout->dim[i];
            if (chunk_offset[i] >= (hssize_t)(space_dim[i])) {
                chunk_offset[i] = 0;
            } else {
                carry = 0;
            }
        } /* end for */
    } /* end while */

#ifdef H5_HAVE_PARALLEL
    /*
     * rky 980923
     * 
     * The following barrier is a temporary fix to prevent overwriting real
     * data caused by a race between one proc's call of H5F_istore_allocate
     * (from H5D_init_storage, ultimately from H5Dcreate and H5Dextend) and
     * another proc's call of H5Dwrite.  Eventually, this barrier should be
     * removed, when H5D_init_storage is changed to call H5MF_alloc directly
     * to allocate space, instead of calling H5F_istore_unlock.
     */
    if (MPI_Barrier(H5FD_mpio_communicator(f->shared->lf)))
        HGOTO_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_Barrier failed");
#endif

done:
    FUNC_LEAVE(ret_value);
}


/*-------------------------------------------------------------------------
 * Function: H5F_istore_prune_by_extent
 *
 * Purpose: This function searches for chunks that are no longer necessary both in the
 *  raw data cache and in the B-tree. 
 *
 * Return: Success: 0, Failure: -1
 *
 * Programmer: Pedro Vicente, pvn@ncsa.uiuc.edu
 * Algorithm: Robb Matzke
 *
 * Date: March 27, 2002
 *
 * The algorithm is:
 *
 *  For chunks that are no longer necessary:
 *
 *  1. Search in the raw data cache for each chunk 
 *  2. If found then preempt it from the cache
 *  3. Search in the B-tree for each chunk 
 *  4. If found then remove it from the B-tree and deallocate file storage for the chunk
 *
 * This example shows a 2d dataset of 90x90 with a chunk size of 20x20. 
 *
 *
 *     0         20        40        60        80    90   100
 *    0 +---------+---------+---------+---------+-----+...+
 *      |:::::X::::::::::::::         :         :     |   :
 *      |:::::::X::::::::::::         :         :     |   :   Key
 *      |::::::::::X:::::::::         :         :     |   :   --------
 *      |::::::::::::X:::::::         :         :     |   :  +-+ Dataset
 *    20+::::::::::::::::::::.........:.........:.....+...:  | | Extent
 *      |         :::::X:::::         :         :     |   :  +-+
 *      |         :::::::::::         :         :     |   :
 *      |         :::::::::::         :         :     |   :  ... Chunk
 *      |         :::::::X:::         :         :     |   :  : : Boundary
 *    40+.........:::::::::::.........:.........:.....+...:  :.:
 *      |         :         :         :         :     |   :
 *      |         :         :         :         :     |   :  ... Allocated
 *      |         :         :         :         :     |   :  ::: & Filled
 *      |         :         :         :         :     |   :  ::: Chunk
 *    60+.........:.........:.........:.........:.....+...:
 *      |         :         :::::::X:::         :     |   :   X  Element
 *      |         :         :::::::::::         :     |   :      Written
 *      |         :         :::::::::::         :     |   :
 *      |         :         :::::::::::         :     |   :
 *    80+.........:.........:::::::::::.........:.....+...:   O  Fill Val
 *      |         :         :         :::::::::::     |   :      Explicitly
 *      |         :         :         ::::::X::::     |   :      Written    
 *    90+---------+---------+---------+---------+-----+   :
 *      :         :         :         :::::::::::         :
 *   100:.........:.........:.........:::::::::::.........:
 *
 *
 * We have 25 total chunks for this dataset, 5 of which have space
 * allocated in the file because they were written to one or more
 * elements. These five chunks (and only these five) also have entries in
 * the storage B-tree for this dataset.
 *
 * Now lets say we want to shrink the dataset down to 70x70:
 *
 *
 *      0         20        40        60   70   80    90   100
 *    0 +---------+---------+---------+----+----+-----+...+
 *      |:::::X::::::::::::::         :    |    :     |   :
 *      |:::::::X::::::::::::         :    |    :     |   :    Key           
 *      |::::::::::X:::::::::         :    |    :     |   :    --------      
 *      |::::::::::::X:::::::         :    |    :     |   :   +-+ Dataset    
 *    20+::::::::::::::::::::.........:....+....:.....|...:   | | Extent     
 *      |         :::::X:::::         :    |    :     |   :   +-+            
 *      |         :::::::::::         :    |    :     |   :                  
 *      |         :::::::::::         :    |    :     |   :   ... Chunk      
 *      |         :::::::X:::         :    |    :     |   :   : : Boundary   
 *    40+.........:::::::::::.........:....+....:.....|...:   :.:            
 *      |         :         :         :    |    :     |   :                  
 *      |         :         :         :    |    :     |   :   ... Allocated  
 *      |         :         :         :    |    :     |   :   ::: & Filled   
 *      |         :         :         :    |    :     |   :   ::: Chunk      
 *    60+.........:.........:.........:....+....:.....|...:                  
 *      |         :         :::::::X:::    |    :     |   :    X  Element    
 *      |         :         :::::::::::    |    :     |   :       Written    
 *      +---------+---------+---------+----+    :     |   :                  
 *      |         :         :::::::::::         :     |   :                  
 *    80+.........:.........:::::::::X:.........:.....|...:    O  Fill Val   
 *      |         :         :         :::::::::::     |   :       Explicitly 
 *      |         :         :         ::::::X::::     |   :       Written    
 *    90+---------+---------+---------+---------+-----+   :
 *      :         :         :         :::::::::::         :
 *   100:.........:.........:.........:::::::::::.........:
 *
 *
 * That means that the nine chunks along the bottom and right side should
 * no longer exist. Of those nine chunks, (0,80), (20,80), (40,80),
 * (60,80), (80,80), (80,60), (80,40), (80,20), and (80,0), one is actually allocated 
 * that needs to be released.
 * To release the chunks, we traverse the B-tree to obtain a list of unused
 * allocated chunks, and then call H5B_remove() for each chunk.
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_prune_by_extent(H5F_t *f, const H5O_layout_t *layout, const H5S_t * space)
{
    H5F_rdcc_t             *rdcc = &(f->shared->rdcc);	/*raw data chunk cache */
    H5F_rdcc_ent_t         *ent = NULL, *next = NULL;	/*cache entry  */
    unsigned                u;	/*counters  */
    int                     found = 0;	/*remove this entry  */
    H5F_istore_ud1_t        udata;	/*B-tree pass-through */
    hsize_t                 curr_dims[H5O_LAYOUT_NDIMS];	/*current dataspace dimensions */

    FUNC_ENTER(H5F_istore_prune_by_extent, FAIL);

    /* Check args */
    assert(f);
    assert(layout && H5D_CHUNKED == layout->type);
    assert(layout->ndims > 0 && layout->ndims <= H5O_LAYOUT_NDIMS);
    assert(H5F_addr_defined(layout->addr));
    assert(space);

    /* Go get the rank & dimensions */
    if(H5S_get_simple_extent_dims(space, curr_dims, NULL) < 0)
	HRETURN_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get dataset dimensions");

 /*-------------------------------------------------------------------------
  * Figure out what chunks are no longer in use for the specified extent 
  * and release them from the linked list raw data cache
  *-------------------------------------------------------------------------
  */
    for(ent = rdcc->head; ent; ent = next) {
	next = ent->next;

	found = 0;
	for(u = 0; u < ent->layout->ndims - 1; u++) {
	    if((hsize_t)ent->offset[u] > curr_dims[u]) {
		found = 1;
		break;
	    }
	}

	if(found) {
#if defined (H5F_ISTORE_DEBUG)
	    HDfputs("cache:remove:[", stdout);
	    for(u = 0; u < ent->layout->ndims - 1; u++) {
		HDfprintf(stdout, "%s%Hd", u ? ", " : "", ent->offset[u]);
	    }
	    HDfputs("]\n", stdout);
#endif

	    /* Preempt the entry from the cache, but do not flush it to disk */
	    if(H5F_istore_preempt(f, ent, FALSE) < 0) {
		HRETURN_ERROR(H5E_IO, H5E_CANTINIT, 0,
			"unable to preempt chunk");
	    }

	}

    }

/*-------------------------------------------------------------------------
 * Check if there are any chunks on the B-tree
 *-------------------------------------------------------------------------
 */

    HDmemset(&udata, 0, sizeof udata);
    udata.stream = stdout;
    udata.mesg.addr = layout->addr;
    udata.mesg.ndims = layout->ndims;
    for(u = 0; u < udata.mesg.ndims; u++)
	udata.mesg.dim[u] = layout->dim[u];
    udata.dims = curr_dims;

    if(H5B_iterate(f, H5B_ISTORE, H5F_istore_prune_extent, layout->addr, &udata) < 0)
	HRETURN_ERROR(H5E_IO, H5E_CANTINIT, 0, "unable to iterate over B-tree");

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function: H5F_istore_prune_extent
 *
 * Purpose: Search for chunks that are no longer necessary in the B-tree. 
 *
 * Return: Success: 0, Failure: -1
 *
 * Programmer: Pedro Vicente, pvn@ncsa.uiuc.edu
 *
 * Date: March 26, 2002
 *
 * Comments: Called by H5B_prune_by_extent, part of H5B_ISTORE
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_prune_extent(H5F_t *f, void *_lt_key, haddr_t UNUSED addr,
        void UNUSED *_rt_key, void *_udata)
{
    H5F_istore_ud1_t       *bt_udata = (H5F_istore_ud1_t *)_udata;
    H5F_istore_key_t       *lt_key = (H5F_istore_key_t *)_lt_key;
    unsigned                u;
    H5F_istore_ud1_t        udata;

    /* The LT_KEY is the left key (the one that describes the chunk). It points to a chunk of 
     * storage that contains the beginning of the logical address space represented by UDATA.
     */

    FUNC_ENTER(H5F_istore_prune_extent, FAIL);

    /* Figure out what chunks are no longer in use for the specified extent and release them */
    for(u = 0; u < bt_udata->mesg.ndims - 1; u++)
	if((hsize_t)lt_key->offset[u] > bt_udata->dims[u]) {

#if defined (H5F_ISTORE_DEBUG)
            HDfputs("b-tree:remove:[", bt_udata->stream);
            for(u = 0; u < bt_udata->mesg.ndims - 1; u++) {
                HDfprintf(bt_udata->stream, "%s%Hd", u ? ", " : "",
                        lt_key->offset[u]);
            }
            HDfputs("]\n", bt_udata->stream);
#endif

            HDmemset(&udata, 0, sizeof udata);
            udata.key = *lt_key;
            udata.mesg = bt_udata->mesg;

            /* Remove */
            if(H5B_remove(f, H5B_ISTORE, bt_udata->mesg.addr, &udata) < 0)
                HRETURN_ERROR(H5E_SYM, H5E_CANTINIT, FAIL, "unable to remove entry");
	    break;
	} /* end if */

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * Function: H5F_istore_remove
 *
 * Purpose: Removes chunks that are no longer necessary in the B-tree. 
 *
 * Return: Success: 0, Failure: -1
 *
 * Programmer: Robb Matzke
 *             Pedro Vicente, pvn@ncsa.uiuc.edu
 *
 * Date: March 28, 2002
 *
 * Comments: Part of H5B_ISTORE
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
static H5B_ins_t
H5F_istore_remove(H5F_t *f, haddr_t addr, void *_lt_key /*in,out */ ,
	hbool_t *lt_key_changed /*out */ ,
	void UNUSED * _udata /*in,out */ ,
	void UNUSED * _rt_key /*in,out */ ,
	hbool_t *rt_key_changed /*out */ )
{
    H5F_istore_key_t       *lt_key = (H5F_istore_key_t *)_lt_key;

    H5FD_free(f->shared->lf, H5FD_MEM_DRAW, addr, (hsize_t)lt_key->nbytes);
    *lt_key_changed = FALSE;
    *rt_key_changed = FALSE;
    return H5B_INS_REMOVE;
}


/*-------------------------------------------------------------------------
 * Function: H5F_istore_initialize_by_extent
 *
 * Purpose:  This function searches for chunks that have to be initialized with the fill
 *   value both in the raw data cache and in the B-tree. 
 *
 * Return: Success: 0, Failure: -1
 *
 * Programmer: Pedro Vicente, pvn@ncsa.uiuc.edu
 *
 * Date: April 4, 2002
 *
 * Comments: 
 *
 * (See the example of H5F_istore_prune_by_extent)
 * Next, there are seven chunks where the database extent boundary is
 * within the chunk. We find those seven just like we did with the previous nine.
 * Fot the ones that are allocated we initialize the part that lies outside the boundary 
 * with the fill value.
 *
 * Modifications:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_initialize_by_extent(H5F_t *f, const H5O_layout_t *layout,
        H5P_genplist_t *dc_plist, const H5S_t * space)
{
    hid_t                   dxpl_id;	/*dataset transfer property list */
    uint8_t                *chunk = NULL;	/*the file chunk  */
    unsigned                idx_hint = 0;	/*input value for H5F_istore_lock */
    hssize_t                chunk_offset[H5O_LAYOUT_NDIMS];	/*logical location of the chunks */
    hsize_t                 idx_cur[H5O_LAYOUT_NDIMS];	/*multi-dimensional counters */
    hsize_t                 idx_min[H5O_LAYOUT_NDIMS];
    hsize_t                 idx_max[H5O_LAYOUT_NDIMS];
    hsize_t                 sub_size[H5O_LAYOUT_NDIMS];
    hsize_t                 naccessed;	/*bytes accessed in chunk */
    hsize_t                 elm_size;	/*size of an element in bytes */
    hsize_t                 end_chunk;	/*chunk position counter */
    hssize_t                start[H5O_LAYOUT_NDIMS];	/*starting location of hyperslab */
    hsize_t                 count[H5O_LAYOUT_NDIMS];	/*element count of hyperslab */
    hsize_t                 size[H5O_LAYOUT_NDIMS];	/*current size of dimensions */
    H5S_t                  *space_chunk = NULL;	/*dataspace for a chunk */
    hsize_t                 curr_dims[H5O_LAYOUT_NDIMS];	/*current dataspace dimensions */
    int                     rank;	/*current # of dimensions */
    int                     i, carry;	/*counters  */
    unsigned                u;
    int                     found = 0;	/*initialize this entry  */
    H5O_pline_t             pline;      /* I/O pipeline information */
    H5O_fill_t              fill;       /* Fill value information */
    herr_t	            ret_value=SUCCEED;	/* Return value */

    FUNC_ENTER(H5F_istore_initialize_by_extent, FAIL);

    /* Check args */
    assert(f);
    assert(layout && H5D_CHUNKED == layout->type);
    assert(layout->ndims > 0 && layout->ndims <= H5O_LAYOUT_NDIMS);
    assert(H5F_addr_defined(layout->addr));
    assert(space);

    /* Get necessary properties from property list */
    if(H5P_get(dc_plist, H5D_CRT_FILL_VALUE_NAME, &fill) < 0)
        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get fill value");
    if(H5P_get(dc_plist, H5D_CRT_DATA_PIPELINE_NAME, &pline) < 0)
        HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get data pipeline");

    /* Reset start & count arrays */
    HDmemset(start, 0, sizeof(start));
    HDmemset(count, 0, sizeof(count));

    /* Go get the rank & dimensions */
    if((rank = H5S_get_simple_extent_dims(space, curr_dims, NULL)) < 0)
	HGOTO_ERROR(H5E_DATASET, H5E_CANTGET, FAIL, "can't get dataset dimensions");

    /* Copy current dimensions */
    for(i = 0; i < rank; i++)
	size[i] = curr_dims[i];
    size[i] = layout->dim[i];
    elm_size = size[i];

    /* Default dataset transfer property list */
    dxpl_id = H5P_DATASET_XFER_DEFAULT;

    /* Create a data space for a chunk & set the extent */
    if(NULL == (space_chunk = H5S_create(H5S_SIMPLE)))
	HGOTO_ERROR(H5E_DATASPACE, H5E_CANTCREATE, FAIL, "can't create simple dataspace");
    if(H5S_set_extent_simple(space_chunk, (unsigned)rank, layout->dim, NULL) < 0)
	HGOTO_ERROR(H5E_DATASPACE, H5E_CANTINIT, FAIL, "can't set dimensions");

/*
 * Set up multi-dimensional counters (idx_min, idx_max, and idx_cur) and
 * loop through the chunks copying each chunk from the application to the
 * chunk cache.
 */
    for(u = 0; u < layout->ndims; u++) {
	idx_min[u] = 0;
	idx_max[u] = (size[u] - 1) / layout->dim[u] + 1;
	idx_cur[u] = idx_min[u];
    } /* end for */

    /* Loop over all chunks */
    carry=0;
    while(carry==0) {

	for(u = 0, naccessed = 1; u < layout->ndims; u++) {
	    /* The location and size of the chunk being accessed */
	    chunk_offset[u] = idx_cur[u] * (hssize_t)(layout->dim[u]);
	    sub_size[u] = MIN((idx_cur[u] + 1) * layout->dim[u],
		    size[u]) - chunk_offset[u];
	    naccessed *= sub_size[u];
	} /* end for */

	/* 
	 * Figure out what chunks have to be initialized. These are the chunks where the database 
	 * extent boundary is within the chunk
	 */
	for(u = 0, found = 0; u < layout->ndims - 1; u++) {
	    end_chunk = chunk_offset[u] + layout->dim[u];
	    if(end_chunk > size[u]) {
		found = 1;
		break;
	    }
	} /* end for */

	if(found) {

	    if(NULL == (chunk = H5F_istore_lock(f, dxpl_id, layout, &pline, &fill,
			    chunk_offset, FALSE, &idx_hint)))
		HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "unable to read raw data chunk");

	    if(H5S_select_all(space_chunk) < 0)
		HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "unable to select space");

	    for(i = 0; i < rank; i++)
		count[i] = MIN((idx_cur[i] + 1) * layout->dim[i], size[i] - chunk_offset[i]);

#if defined (H5F_ISTORE_DEBUG)
	    HDfputs("cache:initialize:offset:[", stdout);
	    for(u = 0; u < layout->ndims - 1; u++)
		HDfprintf(stdout, "%s%Hd", u ? ", " : "", chunk_offset[u]);
	    HDfputs("]", stdout);
	    HDfputs(":count:[", stdout);
	    for(u = 0; u < layout->ndims - 1; u++)
		HDfprintf(stdout, "%s%Hd", u ? ", " : "", count[u]);
	    HDfputs("]\n", stdout);
#endif

	    if(H5S_select_hyperslab(space_chunk, H5S_SELECT_NOTB, start, NULL,
			count, NULL) < 0)
		HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "unable to select hyperslab");

	    /* Fill the selection in the memory buffer */
	    if(H5S_select_fill(fill.buf, fill.size, space_chunk, chunk) < 0)
		HGOTO_ERROR(H5E_DATASET, H5E_CANTENCODE, FAIL, "filling selection failed");

	    if(H5F_istore_unlock(f, dxpl_id, layout, &pline, TRUE,
                    chunk_offset, &idx_hint, chunk, (size_t)naccessed) < 0)
		HGOTO_ERROR(H5E_IO, H5E_WRITEERROR, FAIL, "unable to unlock raw data chunk");

	} /*found */

	/* Increment indices */
	for(i = layout->ndims - 1, carry = 1; i >= 0 && carry; --i) {
	    if(++idx_cur[i] >= idx_max[i])
		idx_cur[i] = idx_min[i];
	    else
		carry = 0;
	} /* end for */
    } /* end while */

done:
    if(space_chunk)
	H5S_close(space_chunk);

    FUNC_LEAVE(ret_value);
}