summaryrefslogtreecommitdiffstats
path: root/src/H5Fistore.c
blob: 1e4802e7cf7984dff4956e3b09ec368402e726e3 (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
/*
 * Copyright (C) 1997 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.
 */
#include <H5private.h>
#include <H5Dprivate.h>
#include <H5Eprivate.h>
#include <H5Fprivate.h>
#include <H5MFprivate.h>
#include <H5MMprivate.h>
#include <H5Oprivate.h>
#include <H5Vprivate.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 intn		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)	((intn)(((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		*/
    intn	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;

/* 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 *);
static intn H5F_istore_cmp2(H5F_t *f, void *_lt_key, void *_udata,
			    void *_rt_key);
static intn H5F_istore_cmp3(H5F_t *f, void *_lt_key, void *_udata,
			    void *_rt_key);
static herr_t H5F_istore_found(H5F_t *f, const haddr_t *addr,
			       const void *_lt_key, void *_udata,
			       const void *_rt_key);
static H5B_ins_t H5F_istore_insert(H5F_t *f, const 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_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, intn indent, intn fwidth,
				    const void *key, const void *udata);
#ifdef HAVE_PARALLEL
static herr_t H5F_istore_get_addr (H5F_t *f, const H5O_layout_t *layout,
				const hssize_t offset[], void *_udata/*out*/);
#endif

/*
 * 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*/
    uintn	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	*/
} 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?	*/
    NULL,					/*remove		*/
    NULL,					/*list			*/
    H5F_istore_decode_key,			/*decode		*/
    H5F_istore_encode_key,			/*encode		*/
    H5F_istore_debug_key,			/*debug			*/
}};

#define H5F_MIXUP(X)	{						      \
    (X) += (X)<<12;							      \
    (X) ^= (X)>>22;							      \
    (X) += (X)<<4;							      \
    (X) ^= (X)>>9;							      \
    (X) += (X)<<10;							      \
    (X) ^= (X)>>2;							      \
    (X) += (X)<<7;							      \
    (X) ^= (X)>>12;							      \
}



/*-------------------------------------------------------------------------
 * 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;
    intn		i;
    intn		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;
    intn		ndims = H5F_ISTORE_NDIMS(bt);
    intn		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, intn indent, intn 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;
    int				i;
    
    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 (i=0; i<udata->mesg.ndims; i++) {
	HDfprintf (stream, "%s%Hd", i?", ":"", key->offset[i]);
    }
    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 intn
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;
    intn		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 intn
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;
    intn		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/*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;
    intn		i;

    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);

    /* Allocate new storage */
    assert (udata->key.nbytes > 0);
#ifdef AKC
    printf("calling H5MF_alloc for new chunk\n");
#endif
    if (H5MF_alloc(f, H5MF_RAW, udata->key.nbytes, addr /*out */ ) < 0) {
	HRETURN_ERROR(H5E_IO, H5E_CANTINIT, FAIL,
		      "couldn't allocate new file storage");
    }
    udata->addr = *addr;

    /*
     * 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 (i=0; i<udata->mesg.ndims; i++) {
	lt_key->offset[i] = udata->key.offset[i];
    }

    /*
     * 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 (i=0; i<udata->mesg.ndims; i++) {
	    assert (udata->mesg.dim[i] < HSSIZET_MAX);
	    assert (udata->key.offset[i]+(hssize_t)(udata->mesg.dim[i]) >
		    udata->key.offset[i]);
	    rt_key->offset[i] = udata->key.offset[i] +
				(hssize_t)(udata->mesg.dim[i]);
	}
    }

    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:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_found(H5F_t __unused__ *f, const 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;
    int			   i;

    FUNC_ENTER(H5F_istore_found, FAIL);

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

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

    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:
 *
 *-------------------------------------------------------------------------
 */
static H5B_ins_t
H5F_istore_insert(H5F_t *f, const 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/*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;
    intn		i, cmp;
    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(addr && 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);

    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 (H5MF_realloc (f, H5MF_RAW, lt_key->nbytes, addr,
			      udata->key.nbytes, new_node/*out*/)<0) {
		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;
	    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 (i=0; i<udata->mesg.ndims; i++) {
	    assert(0 == udata->key.offset[i] % udata->mesg.dim[i]);
	    md_key->offset[i] = udata->key.offset[i];
	}

	/*
	 * Allocate storage for the new chunk
	 */
#ifdef AKC
	printf("calling H5MF_alloc for new chunk\n");
#endif
	if (H5MF_alloc(f, H5MF_RAW, udata->key.nbytes, new_node/*out*/)<0) {
	    HRETURN_ERROR(H5E_IO, H5E_CANTINIT, H5B_INS_ERROR,
			  "file allocation failed");
	}
	udata->addr = *new_node;
	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_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->access_parms->rdcc_nbytes>0 &&
	f->shared->access_parms->rdcc_nelmts>0) {
	rdcc->nslots = f->shared->access_parms->rdcc_nelmts;
	rdcc->slot = H5MM_calloc (rdcc->nslots*sizeof(H5F_rdcc_ent_t*));
	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		*/
    intn		i;		/*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;
	H5F_addr_undef(&(udata.addr));
	udata.key.nbytes = ent->chunk_size;
	for (i=0; i<ent->layout->ndims; i++) {
	    udata.key.offset[i] = ent->offset[i];
	}
	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 = H5MM_malloc(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, &(udata.addr), udata.key.nbytes, H5D_XFER_DFLT,
			     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;
	ent->chunk = H5MM_xfree(ent->chunk);
    }
    
    ret_value = SUCCEED;

 done:
    /* Free the temp buffer only if it's different than the entry chunk */
    if (buf!=ent->chunk) H5MM_xfree(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);
	ent->chunk = H5MM_xfree(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:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_preempt (H5F_t *f, H5F_rdcc_ent_t *ent)
{
    H5F_rdcc_t          *rdcc = &(f->shared->rdcc);
    
    FUNC_ENTER (H5F_istore_preempt, FAIL);

    assert(f);
    assert(ent);
    assert(!ent->locked);
    assert(ent->idx>=0 && ent->idx<rdcc->nslots);

    /* Flush */
    H5F_istore_flush_entry(f, ent, TRUE);

    /* 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 = -1;
    rdcc->nbytes -= ent->chunk_size;
    --rdcc->nused;

    /* Free */
    H5MM_xfree(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:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_flush (H5F_t *f, hbool_t preempt)
{
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);
    intn		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)<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:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_dest (H5F_t *f)
{
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);
    intn		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
	fputc('c', stderr);
	fflush(stderr);
#endif
	next = ent->next;
	if (H5F_istore_preempt(f, ent)<0) {
	    nerrors++;
	}
    }
    if (nerrors) {
	HRETURN_ERROR (H5E_IO, H5E_CANTFLUSH, FAIL,
		       "unable to flush one or more raw data chunks");
    }

    H5MM_xfree (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:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_prune (H5F_t *f, size_t size)
{
    intn		i, j, nerrors=0;
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);
    size_t		total = f->shared->access_parms->rdcc_nbytes;
    const int		nmeth=2;	/*number of methods		*/
    intn		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] = rdcc->nused * f->shared->access_parms->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
		putc('.', stderr);
		fflush(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
		putc(':', stderr);
		fflush(stderr);
#endif
		
	    } else {
		/* Nothing to preempt at this point */
		cur= NULL;
	    }

	    if (cur) {
		if (H5F_istore_preempt(f, cur)<0) nerrors++;
		for (j=0; j<nmeth; j++) {
		    if (p[j]==cur) p[j] = NULL;
		}
	    }
	}
	
	/* 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_rdcc_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:
 *
 *-------------------------------------------------------------------------
 */
static void *
H5F_istore_lock (H5F_t *f, const H5O_layout_t *layout,
		 const double split_ratios[], const H5O_pline_t *pline,
		 const H5O_fill_t *fill, const hssize_t offset[],
		 hbool_t relax, intn *idx_hint/*in,out*/)
{
    uintn		idx=0;			/*hash 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		*/
    intn		i;			/*counters		*/
    H5F_istore_ud1_t	udata;			/*B-tree pass-through	*/
    size_t		chunk_size=0;		/*size of a chunk	*/
    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		*/

    FUNC_ENTER (H5F_istore_lock, NULL);
    assert(split_ratios);

    if (rdcc->nslots>0) {
	/* We don't care about loss of precision in the following statement. */
	idx = (uintn)(layout->addr.offset);
	H5F_MIXUP(idx);
	for (i=0; i<layout->ndims; i++) {
	    idx += offset[i];
	    H5F_MIXUP(idx);
	}
	idx %= rdcc->nslots;
	ent = rdcc->slot[idx];
    
	if (ent &&
	    layout->ndims==ent->layout->ndims &&
	    H5F_addr_eq(&(layout->addr), &(ent->layout->addr))) {
	    for (i=0, found=TRUE; i<ent->layout->ndims; i++) {
		if (offset[i]!=ent->offset[i]) {
		    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
	putc('w', stderr);
	fflush(stderr);
#endif
	rdcc->nhits++;
	for (i=0, chunk_size=1; i<layout->ndims; i++) {
	    chunk_size *= layout->dim[i];
	}
	chunk_alloc = chunk_size;
	if (NULL==(chunk=H5MM_malloc (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 (i=0, chunk_size=1; i<layout->ndims; i++) {
	    udata.key.offset[i] = offset[i];
	    chunk_size *= layout->dim[i];
	}
	chunk_alloc = chunk_size;
	udata.mesg = *layout;
	H5F_addr_undef (&(udata.addr));
	status = H5B_find (f, H5B_ISTORE, &(layout->addr), &udata);
	H5E_clear ();
	if (NULL==(chunk = H5MM_malloc (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, &(udata.addr), udata.key.nbytes,
	        H5D_XFER_DFLT, 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->access_parms->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
	    putc('#', stderr);
	    fflush(stderr);
#endif
#if 0
	    HDfprintf(stderr, "\ncollision %3d %10a {",
		      idx, &(ent->layout->addr));
	    for (i=0; i<layout->ndims; i++) {
		HDfprintf(stderr, "%s%Zu", i?",":"", ent->offset[i]);
	    }
	    HDfprintf(stderr, "}\n              %10a {", &(layout->addr));
	    for (i=0; i<layout->ndims; i++) {
		HDfprintf(stderr, "%s%Zu", i?",":"", offset[i]);
	    }
	    fprintf(stderr, "}\n");
#endif
	    if (H5F_istore_preempt(f, ent)<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 = H5MM_malloc(sizeof(H5F_rdcc_ent_t));
	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 (i=0; i<layout->ndims; i++) {
	    ent->offset[i] = offset[i];
	}
	ent->rd_count = chunk_size;
	ent->wr_count = chunk_size;
	ent->chunk = chunk;
	ent->split_ratios[0] = split_ratios[0];
	ent->split_ratios[1] = split_ratios[1];
	ent->split_ratios[2] = split_ratios[2];

	/* 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 = -999;

    } 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) H5MM_xfree (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:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_unlock (H5F_t *f, const H5O_layout_t *layout,
		   const double split_ratios[],
		   const H5O_pline_t *pline, hbool_t dirty,
		   const hssize_t offset[], intn *idx_hint,
		   uint8_t *chunk, size_t naccessed)
{
    H5F_rdcc_t		*rdcc = &(f->shared->rdcc);
    H5F_rdcc_ent_t	*ent = NULL;
    intn		i, found = -1;
    
    FUNC_ENTER (H5F_istore_unlock, FAIL);

    if (-999==*idx_hint) {
	/*not in cache*/
    } else {
	assert(*idx_hint>=0 && *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;
	    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 (i=0, x.chunk_size=1; i<layout->ndims; i++) {
		x.offset[i] = offset[i];
		x.chunk_size *= layout->dim[i];
	    }
	    x.alloc_size = x.chunk_size;
	    x.chunk = chunk;
	    x.split_ratios[0] = split_ratios[0];
	    x.split_ratios[1] = split_ratios[1];
	    x.split_ratios[2] = split_ratios[2];
	    H5F_istore_flush_entry (f, &x, TRUE);
	} else {
	    H5MM_xfree (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:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_read(H5F_t *f, const H5D_xfer_t *xfer, const H5O_layout_t *layout,
		const H5O_pline_t *pline, const H5O_fill_t *fill,
		const hssize_t offset_f[], const hsize_t size[], void *buf)
{
    hssize_t		offset_m[H5O_LAYOUT_NDIMS];
    hsize_t		size_m[H5O_LAYOUT_NDIMS];
    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];
    intn		i, carry;
    size_t		naccessed;		/*bytes accessed in chnk*/
    uint8_t		*chunk=NULL;		/*ptr to a chunk buffer	*/
    intn		idx_hint=0;		/*cache index hint	*/

    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 (offset_f);
    assert (size);
    assert (buf);

    /*
     * For now, a hyperslab of the file must be read into an array in
     * memory.We do not yet support reading into a hyperslab of memory.
     */
    for (i=0; i<layout->ndims; i++) {
	offset_m[i] = 0;
	size_m[i] = size[i];
    }
    
#ifndef NDEBUG
    for (i=0; i<layout->ndims; i++) {
	assert (offset_f[i]>=0); /*negative offsets not supported*/
	assert (offset_m[i]>=0); /*negative offsets not supported*/
	assert (size[i]<SIZET_MAX);
	assert(offset_m[i]+(hssize_t)size[i]<=(hssize_t)size_m[i]);
	assert(layout->dim[i]>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 (i=0; i<layout->ndims; i++) {
	idx_min[i] = offset_f[i] / layout->dim[i];
	idx_max[i] = (offset_f[i]+size[i]-1) / layout->dim[i] + 1;
	idx_cur[i] = idx_min[i];
    }

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

	    /* The offset and size wrt the chunk */
	    offset_wrt_chunk[i] = MAX(offset_f[i], chunk_offset[i]) -
				  chunk_offset[i];
	    sub_size[i] = MIN((idx_cur[i]+1)*layout->dim[i],
			      offset_f[i]+size[i]) -
			  (chunk_offset[i] + offset_wrt_chunk[i]);
	    naccessed *= sub_size[i];
	    
	    /* Offset into mem buffer */
	    sub_offset_m[i] = chunk_offset[i] + offset_wrt_chunk[i] +
			      offset_m[i] - offset_f[i];
	}
#ifdef 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.
	 */
	 if (f->shared->access_parms->driver==H5F_LOW_MPIO){
	    H5F_istore_ud1_t	udata;
	    H5O_layout_t	l;	/* temporary layout */
	    H5D_xfer_t		tmp_xfer = *xfer;
	    if (H5F_istore_get_addr(f, layout, chunk_offset, &udata)<0){
		HRETURN_ERROR (H5E_IO, H5E_WRITEERROR, FAIL,
				"unable to locate raw data chunk");
	    };
	    /*
	     * 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 (i=l.ndims; i-- > 0;)
		l.dim[i] = layout->dim[i];
	    l.addr = udata.addr;
	    tmp_xfer.xfer_mode = H5D_XFER_DFLT;
	    if (H5F_arr_read(f, &tmp_xfer, &l, pline, fill, NULL/*no efl*/,
			     sub_size, size_m, sub_offset_m, offset_wrt_chunk,
			     buf)<0){
		HRETURN_ERROR (H5E_IO, H5E_READERROR, FAIL,
			     "unable to read raw data from file");
	    }
	} else {
#endif

#ifdef AKC
	    printf("Locking chunk( ");
	    for (i=0; i<layout->ndims; i++){
		printf("%ld ", chunk_offset[i]);
	    }
	    printf(")\n");
#endif
	    /*
	     * Lock the chunk, transfer data to the application, then unlock
	     * the chunk.
	     */
	    if (NULL==(chunk=H5F_istore_lock (f, layout, xfer->split_ratios, 
					      pline, fill, chunk_offset,
					      FALSE, &idx_hint))) {
		HRETURN_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);
	    if (H5F_istore_unlock (f, layout, xfer->split_ratios, pline,
				   FALSE, chunk_offset, &idx_hint, chunk,
				   naccessed)<0) {
		HRETURN_ERROR (H5E_IO, H5E_READERROR, FAIL,
			       "unable to unlock raw data chunk");
	    }
#ifdef HAVE_PARALLEL
	}
#endif

	/* 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;
	}
	if (carry) break;
    }
    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * 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:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_write(H5F_t *f, const H5D_xfer_t *xfer, const H5O_layout_t *layout,
		 const H5O_pline_t *pline, const H5O_fill_t *fill,
		 const hssize_t offset_f[], const hsize_t size[],
		 const void *buf)
{
    hssize_t		offset_m[H5O_LAYOUT_NDIMS];
    hsize_t		size_m[H5O_LAYOUT_NDIMS];
    intn		i, carry;
    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;
    intn		idx_hint=0;
    size_t		chunk_size, naccessed;
    
    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(offset_f);
    assert(size);
    assert(buf);

    /*
     * For now the source must not be a hyperslab.  It must be an entire
     * memory buffer.
     */
    for (i=0, chunk_size=1; i<layout->ndims; i++) {
	offset_m[i] = 0;
	size_m[i] = size[i];
	chunk_size *= layout->dim[i];
    }

#ifndef NDEBUG
    for (i=0; i<layout->ndims; i++) {
	assert (offset_f[i]>=0); /*negative offsets not supported*/
	assert (offset_m[i]>=0); /*negative offsets not supported*/
	assert(size[i]<SIZET_MAX);
	assert(offset_m[i]+(hssize_t)size[i]<=(hssize_t)size_m[i]);
	assert(layout->dim[i]>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 (i=0; i<layout->ndims; i++) {
	idx_min[i] = offset_f[i] / layout->dim[i];
	idx_max[i] = (offset_f[i]+size[i]-1) / layout->dim[i] + 1;
	idx_cur[i] = idx_min[i];
    }


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

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

#ifdef 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.
	 */
	 if (f->shared->access_parms->driver==H5F_LOW_MPIO){
	    H5F_istore_ud1_t	udata;
	    H5O_layout_t	l;	/* temporary layout */
	    H5D_xfer_t		tmp_xfer = *xfer;
	    if (H5F_istore_get_addr(f, layout, chunk_offset, &udata)<0){
		HRETURN_ERROR (H5E_IO, H5E_WRITEERROR, FAIL,
				"unable to locate raw data chunk");
	    };
	    /*
	     * 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 (i=l.ndims; i-- > 0;)
		l.dim[i] = layout->dim[i];
	    l.addr = udata.addr;
	    tmp_xfer.xfer_mode = H5D_XFER_DFLT;
	    if (H5F_arr_write(f, &tmp_xfer, &l, pline, fill, NULL/*no efl*/,
			      sub_size, size_m, sub_offset_m, offset_wrt_chunk,
			      buf)<0){
		HRETURN_ERROR (H5E_IO, H5E_WRITEERROR, FAIL,
			       "unable to write raw data to file");
	    }
	} else {
#endif

#ifdef AKC
	    printf("Locking chunk( ");
	    for (i=0; i<layout->ndims; i++){
		printf("%ld ", chunk_offset[i]);
	    }
	    printf(")\n");
#endif
	    /*
	     * Lock the chunk, copy from application to chunk, then unlock the
	     * chunk.
	     */
	    if (NULL==(chunk=H5F_istore_lock (f, layout, xfer->split_ratios,
					      pline, fill, chunk_offset,
					      naccessed==chunk_size,
					      &idx_hint))) {
		HRETURN_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);
	    if (H5F_istore_unlock (f, layout, xfer->split_ratios, pline, TRUE,
				   chunk_offset, &idx_hint, chunk,
				   naccessed)<0) {
		HRETURN_ERROR (H5E_IO, H5E_WRITEERROR, FAIL,
			       "uanble to unlock raw data chunk");
	    }
#ifdef HAVE_PARALLEL
	}
#endif
	
	/* 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;
	}
	if (carry) break;
    }

    FUNC_LEAVE(SUCCEED);
}


/*-------------------------------------------------------------------------
 * 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
    int			i;
#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 (i = 0; i < layout->ndims; i++) {
	assert(layout->dim[i] > 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_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:
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_debug(H5F_t *f, const haddr_t *addr, FILE * stream, intn indent,
		 intn 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:
 *
 *-------------------------------------------------------------------------
 */
static herr_t
H5F_istore_get_addr(H5F_t *f, const H5O_layout_t *layout,
		    const hssize_t offset[], void *_udata/*out*/)
{
    H5F_istore_ud1_t	*udata = _udata;
    intn		i;
    herr_t		status;			/*func return status	*/
    
    FUNC_ENTER (H5F_istore_get_addr, FAIL);

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


    for (i=0; i<layout->ndims; i++) {
	udata->key.offset[i] = offset[i];
    }
    udata->mesg = *layout;
    H5F_addr_undef (&(udata->addr));
    status = H5B_find (f, H5B_ISTORE, &(layout->addr), udata);
    H5E_clear ();
    if (status>=0 && H5F_addr_defined (&(udata->addr)))
	HRETURN(SUCCEED);

    FUNC_LEAVE (FAIL);
}


/*-------------------------------------------------------------------------
 * 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 980923
 * Added barrier to preclude racing with data writes.
 *
 * rky 19981207
 * Added Wait-Signal wrapper around unlock-lock critical region
 * to prevent race condition (unlock reads, lock writes the chunk).
 *
 *-------------------------------------------------------------------------
 */
herr_t
H5F_istore_allocate (H5F_t *f, const H5O_layout_t *layout,
		     const hsize_t *space_dim, const double split_ratios[],
		     const H5O_pline_t *pline, const H5O_fill_t *fill)
{

    intn		i, carry;
    hssize_t		chunk_offset[H5O_LAYOUT_NDIMS];
    uint8_t		*chunk=NULL;
    intn		idx_hint=0;
    size_t		chunk_size;
#ifdef AKC
    H5F_istore_ud1_t	udata;
#endif
    
    FUNC_ENTER(H5F_istore_allocate, FAIL);
#ifdef AKC
    printf("Enter %s:\n", FUNC);
#endif

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

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

    /* Loop over all chunks */
    while (1) {
	
#ifdef AKC
	printf("Checking allocation for chunk( ");
	for (i=0; i<layout->ndims; i++){
	    printf("%ld ", chunk_offset[i]);
	}
	printf(")\n");
#endif
#ifdef NO
	if (H5F_istore_get_addr(f, layout, chunk_offset, &udata)<0){
#endif
	    /* 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. */

#ifdef AKC
	    printf("need allocation\n");
#endif
	    /*
	     * Lock the chunk, copy from application to chunk, then unlock the
	     * chunk.
	     */

#ifdef HAVE_PARALLEL
	    /* rky 981207 Serialize access to this critical region. */
	    if (SUCCEED!=
		H5PC_Wait_for_left_neighbor(f->shared->access_parms->u.mpio.comm))
	    {
		HRETURN_ERROR (H5E_IO, H5E_WRITEERROR, FAIL,
			       "unable to lock the data chunk");
	    }
#endif
	    if (NULL==(chunk=H5F_istore_lock (f, layout, split_ratios, pline,
					      fill, chunk_offset, FALSE,
					      &idx_hint))) {
		HRETURN_ERROR (H5E_IO, H5E_WRITEERROR, FAIL,
			       "unable to read raw data chunk");
	    }
	    if (H5F_istore_unlock (f, layout, split_ratios, pline, TRUE,
				   chunk_offset, &idx_hint, chunk,
				   chunk_size)<0) {
		HRETURN_ERROR (H5E_IO, H5E_WRITEERROR, FAIL,
			       "uanble to unlock raw data chunk");
	    }
#ifdef HAVE_PARALLEL
	    if (SUCCEED!=
		H5PC_Signal_right_neighbor(f->shared->access_parms->u.mpio.comm))
	    {
		HRETURN_ERROR (H5E_IO, H5E_WRITEERROR, FAIL,
			       "unable to unlock the data chunk");
	    }
#endif
#ifdef NO
	} else {
#ifdef AKC
	    printf("NO need for allocation\n");
	    printf("udata.addr.offset=%d\n", udata.addr.offset);
#endif
	}
#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;
	    }
	}
	if (carry) break;
    }

#ifdef 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( f->shared->access_parms->u.mpio.comm )) {
	HRETURN_ERROR(H5E_INTERNAL, H5E_MPI, FAIL, "MPI_Barrier failed");
    }
#endif

    FUNC_LEAVE(SUCCEED);
}