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
|
/*
* Copyright (C) 1997 NCSA
* All rights reserved.
*
* Programmer: Robb Matzke <matzke@llnl.gov>
* Friday, October 10, 1997
*/
#include <H5private.h>
#include <H5Eprivate.h>
#include <H5Oprivate.h>
#include <H5Vprivate.h>
#define H5V_HYPER_NDIMS H5O_LAYOUT_NDIMS
#define PABLO_MASK H5V_mask
static hbool_t interface_initialize_g = TRUE;
#define INTERFACE_INIT NULL
/*-------------------------------------------------------------------------
* Function: H5V_stride_optimize1
*
* Purpose: Given a stride vector which references elements of the
* specified size, optimize the dimensionality, the stride
* vector, and the element size to minimize the dimensionality
* and the number of memory accesses.
*
* All arguments are passed by reference and their values may be
* modified by this function.
*
* Return: Success: SUCCEED
*
* Failure: FAIL
*
* Programmer: Robb Matzke
* Saturday, October 11, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5V_stride_optimize1(intn *np/*in,out*/, hsize_t *elmt_size/*in,out*/,
hsize_t *size, hssize_t *stride1)
{
FUNC_ENTER(H5V_stride_optimize1, FAIL);
/*
* This has to be true because if we optimize the dimensionality down to
* zero we still must make one reference.
*/
assert(1 == H5V_vector_reduce_product(0, NULL));
/*
* Combine adjacent memory accesses
*/
while (*np &&
stride1[*np-1]>0 &&
(hsize_t)(stride1[*np-1])==*elmt_size) {
*elmt_size *= size[*np-1];
if (--*np) {
stride1[*np-1] += size[*np] * stride1[*np];
}
}
FUNC_LEAVE(SUCCEED);
}
/*-------------------------------------------------------------------------
* Function: H5V_stride_optimize2
*
* Purpose: Given two stride vectors which reference elements of the
* specified size, optimize the dimensionality, the stride
* vectors, and the element size to minimize the dimensionality
* and the number of memory accesses.
*
* All arguments are passed by reference and their values may be
* modified by this function.
*
* Return: Success: SUCCEED
*
* Failure: FAIL
*
* Programmer: Robb Matzke
* Saturday, October 11, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5V_stride_optimize2(intn *np/*in,out*/, hsize_t *elmt_size/*in,out*/,
hsize_t *size, hssize_t *stride1, hssize_t *stride2)
{
FUNC_ENTER(H5V_stride_optimize2, FAIL);
/*
* This has to be true because if we optimize the dimensionality down to
* zero we still must make one reference.
*/
assert(1 == H5V_vector_reduce_product(0, NULL));
assert (*elmt_size>0);
/*
* Combine adjacent memory accesses
*/
while (*np &&
stride1[*np-1] > 0 &&
(hsize_t)(stride1[*np-1]) == *elmt_size &&
stride2[*np-1] > 0 &&
(hsize_t)(stride2[*np-1]) == *elmt_size) {
*elmt_size *= size[*np-1];
if (--*np) {
stride1[*np-1] += size[*np] * stride1[*np];
stride2[*np-1] += size[*np] * stride2[*np];
}
}
FUNC_LEAVE(SUCCEED);
}
/*-------------------------------------------------------------------------
* Function: H5V_hyper_stride
*
* Purpose: Given a description of a hyperslab, this function returns
* (through STRIDE[]) the byte strides appropriate for accessing
* all bytes of the hyperslab and the byte offset where the
* striding will begin. The SIZE can be passed to the various
* stride functions.
*
* The dimensionality of the whole array, the hyperslab, and the
* returned stride array is N. The whole array dimensions are
* TOTAL_SIZE and the hyperslab is at offset OFFSET and has
* dimensions SIZE.
*
* The stride and starting point returned will cause the
* hyperslab elements to be referenced in C order.
*
* Return: Success: Byte offset from beginning of array to start
* of striding.
*
* Failure: abort() -- should never fail
*
* Programmer: Robb Matzke
* Saturday, October 11, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
hsize_t
H5V_hyper_stride(intn n, const hsize_t *size,
const hsize_t *total_size, const hssize_t *offset,
hssize_t *stride/*out*/)
{
hsize_t skip; /*starting point byte offset */
hsize_t acc; /*accumulator */
int i; /*counter */
FUNC_ENTER(H5V_hyper_stride, (abort(), 0));
assert(n >= 0 && n < H5V_HYPER_NDIMS);
assert(size);
assert(total_size);
assert(stride);
/* init */
stride[n-1] = 1;
skip = offset ? offset[n-1] : 0;
/* others */
for (i=n-2, acc=1; i>=0; --i) {
hsize_t tmp = acc * (total_size[i+1] - size[i+1]);
assert (tmp<((hsize_t)1<<(8*sizeof(hssize_t)-1)));
stride[i] = (hssize_t)tmp; /*overflow checked*/
acc *= total_size[i+1];
skip += acc * (offset ? offset[i] : 0);
}
FUNC_LEAVE(skip);
}
/*-------------------------------------------------------------------------
* Function: H5V_hyper_eq
*
* Purpose: Determines whether two hyperslabs are equal. This function
* assumes that both hyperslabs are relative to the same array,
* for if not, they could not possibly be equal.
*
* Return: Success: TRUE if the hyperslabs are equal (that is,
* both refer to exactly the same elements of an
* array)
*
* FALSE otherwise.
*
* Failure: TRUE the rank is zero or if both hyperslabs
* are of zero size.
*
* Programmer: Robb Matzke
* Friday, October 17, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
hbool_t
H5V_hyper_eq(intn n,
const hssize_t *offset1, const hsize_t *size1,
const hssize_t *offset2, const hsize_t *size2)
{
hsize_t nelmts1 = 1, nelmts2 = 1;
intn i;
if (n <= 0) return TRUE;
for (i=0; i<n; i++) {
if ((offset1 ? offset1[i] : 0) != (offset2 ? offset2[i] : 0)) {
return FALSE;
}
if ((size1 ? size1[i] : 0) != (size2 ? size2[i] : 0)) {
return FALSE;
}
if (0 == (nelmts1 *= (size1 ? size1[i] : 0))) return FALSE;
if (0 == (nelmts2 *= (size2 ? size2[i] : 0))) return FALSE;
}
return TRUE;
}
/*-------------------------------------------------------------------------
* Function: H5V_hyper_disjointp
*
* Purpose: Determines if two hyperslabs are disjoint.
*
* Return: Success: FALSE if they are not disjoint.
* TRUE if they are disjoint.
*
* Failure: A hyperslab of zero size is disjoint from all
* other hyperslabs.
*
* Programmer: Robb Matzke
* Thursday, October 16, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
hbool_t
H5V_hyper_disjointp(intn n,
const hssize_t *offset1, const hsize_t *size1,
const hssize_t *offset2, const hsize_t *size2)
{
intn i;
if (!n || !size1 || !size2) return TRUE;
for (i=0; i<n; i++) {
assert (size1[i]<MAX_HSSIZET);
assert (size2[i]<MAX_HSSIZET);
if (0==size1[i] || 0==size2[i]) return TRUE;
if (((offset1?offset1[i]:0) < (offset2?offset2[i]:0) &&
((offset1?offset1[i]:0) + (hssize_t)size1[i] <=
(offset2?offset2[i]:0))) ||
((offset2?offset2[i]:0) < (offset1?offset1[i]:0) &&
((offset2?offset2[i]:0) + (hssize_t)size2[i] <=
(offset1?offset1[i]:0)))) {
return TRUE;
}
}
return FALSE;
}
/*-------------------------------------------------------------------------
* Function: H5V_hyper_fill
*
* Purpose: Similar to memset() except it operates on hyperslabs...
*
* Fills a hyperslab of array BUF with some value VAL. BUF
* is treated like a C-order array with N dimensions where the
* size of each dimension is TOTAL_SIZE[]. The hyperslab which
* will be filled with VAL begins at byte offset OFFSET[] from
* the minimum corner of BUF and continues for SIZE[] bytes in
* each dimension.
*
* Return: Success: SUCCEED
*
* Failure: FAIL
*
* Programmer: Robb Matzke
* Friday, October 10, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5V_hyper_fill(intn n, const hsize_t *_size,
const hsize_t *total_size, const hssize_t *offset, void *_dst,
uintn fill_value)
{
uint8 *dst = (uint8 *) _dst; /*cast for ptr arithmetic */
hsize_t size[H5V_HYPER_NDIMS]; /*a modifiable copy of _size */
hssize_t dst_stride[H5V_HYPER_NDIMS]; /*destination stride info */
hsize_t dst_start; /*byte offset to start of stride*/
hsize_t elmt_size = 1; /*bytes per element */
herr_t status; /*function return status */
#ifndef NDEBUG
int i;
#endif
FUNC_ENTER(H5V_hyper_fill, FAIL);
/* check args */
assert(n > 0 && n <= H5V_HYPER_NDIMS);
assert(_size);
assert(total_size);
assert(dst);
#ifndef NDEBUG
for (i = 0; i < n; i++) {
assert(_size[i] > 0);
assert(total_size[i] > 0);
}
#endif
/* Copy the size vector so we can modify it */
H5V_vector_cpy(n, size, _size);
/* Compute an optimal destination stride vector */
dst_start = H5V_hyper_stride(n, size, total_size, offset, dst_stride);
H5V_stride_optimize1(&n, &elmt_size, size, dst_stride);
/* Copy */
status = H5V_stride_fill(n, elmt_size, size, dst_stride, dst+dst_start,
fill_value);
FUNC_LEAVE(status);
}
/*-------------------------------------------------------------------------
* Function: H5V_hyper_copy
*
* Purpose: Copies a hyperslab from the source to the destination.
*
* A hyperslab is a logically contiguous region of
* multi-dimensional size SIZE of an array whose dimensionality
* is N and whose total size is DST_TOTAL_SIZE or SRC_TOTAL_SIZE.
* The minimum corner of the hyperslab begins at a
* multi-dimensional offset from the minimum corner of the DST
* (destination) or SRC (source) array. The sizes and offsets
* are assumed to be in C order, that is, the first size/offset
* varies the slowest while the last varies the fastest in the
* mapping from N-dimensional space to linear space. This
* function assumes that the array elements are single bytes (if
* your array has multi-byte elements then add an additional
* dimension whose size is that of your element).
*
* The SRC and DST array may be the same array, but the results
* are undefined if the source hyperslab overlaps the
* destination hyperslab.
*
* Return: Success: SUCCEED
*
* Failure: FAIL
*
* Programmer: Robb Matzke
* Friday, October 10, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5V_hyper_copy(intn n, const hsize_t *_size,
/*destination*/
const hsize_t *dst_size, const hssize_t *dst_offset,
void *_dst,
/*source*/
const hsize_t *src_size, const hssize_t *src_offset,
const void *_src)
{
const uint8 *src = (const uint8 *)_src; /*cast for ptr arithmtc */
uint8 *dst = (uint8 *) _dst; /*cast for ptr arithmtc */
hsize_t size[H5V_HYPER_NDIMS]; /*a modifiable _size */
hssize_t src_stride[H5V_HYPER_NDIMS]; /*source stride info */
hssize_t dst_stride[H5V_HYPER_NDIMS]; /*dest stride info */
hsize_t dst_start, src_start; /*offset to start at */
hsize_t elmt_size = 1; /*element size in bytes */
herr_t status; /*return status */
#ifndef NDEBUG
intn i;
#endif
FUNC_ENTER(H5V_hyper_copy, FAIL);
/* check args */
assert(n > 0 && n <= H5V_HYPER_NDIMS);
assert(_size);
assert(dst_size);
assert(src_size);
assert(dst);
assert(src);
#ifndef NDEBUG
for (i = 0; i < n; i++) {
assert(_size[i] > 0);
assert(dst_size[i] > 0);
assert(src_size[i] > 0);
}
#endif
#ifdef QAK
{
intn i;
printf("%s: n=%d, _dst=%p, _src=%p\n",FUNC,(int)n,_dst,_src);
for(i=0; i<n; i++) {
printf("%d: size=%d, dst_size=%d, dst_offset=%d, src_size=%d, src_offset=%d\n",i,(int)size[i],(int)dst_size[i],(int)dst_offset[i],(int)src_size[i],(int)src_offset[i]);
} /* end for */
}
#endif /* QAK */
/* Copy the size vector so we can modify it */
H5V_vector_cpy(n, size, _size);
/* Compute stride vectors for source and destination */
dst_start = H5V_hyper_stride(n, size, dst_size, dst_offset, dst_stride);
src_start = H5V_hyper_stride(n, size, src_size, src_offset, src_stride);
/* Optimize the strides as a pair */
H5V_stride_optimize2(&n, &elmt_size, size, dst_stride, src_stride);
/* Perform the copy in terms of stride */
status = H5V_stride_copy(n, elmt_size, size,
dst_stride, dst+dst_start,
src_stride, src+src_start);
FUNC_LEAVE(status);
}
/*-------------------------------------------------------------------------
* Function: H5V_stride_fill
*
* Purpose: Fills all bytes of a hyperslab with the same value using
* memset().
*
* Return: Success: SUCCEED
*
* Failure: FAIL
*
* Programmer: Robb Matzke
* Saturday, October 11, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5V_stride_fill(intn n, hsize_t elmt_size, const hsize_t *size,
const hssize_t *stride, void *_dst, uintn fill_value)
{
uint8 *dst = (uint8 *) _dst; /*cast for ptr arithmetic */
hsize_t idx[H5V_HYPER_NDIMS]; /*1-origin indices */
hsize_t nelmts; /*number of elements to fill */
hsize_t i; /*counter */
intn j; /*counter */
hbool_t carry; /*subtraction carray value */
FUNC_ENTER(H5V_stride_fill, FAIL);
assert (elmt_size < MAX_SIZET);
H5V_vector_cpy(n, idx, size);
nelmts = H5V_vector_reduce_product(n, size);
for (i=0; i<nelmts; i++) {
/* Copy an element */
HDmemset(dst, (signed)fill_value, (size_t)elmt_size);
/* Decrement indices and advance pointer */
for (j=n-1, carry=TRUE; j>=0 && carry; --j) {
dst += stride[j];
if (--idx[j]) carry = FALSE;
else idx[j] = size[j];
}
}
FUNC_LEAVE(SUCCEED);
}
/*-------------------------------------------------------------------------
* Function: H5V_stride_copy
*
* Purpose: Uses DST_STRIDE and SRC_STRIDE to advance through the arrays
* DST and SRC while copying bytes from SRC to DST. This
* function minimizes the number of calls to memcpy() by
* combining various strides, but it will never touch memory
* outside the hyperslab defined by the strides.
*
* Note: If the src_stride is all zero and elmt_size is one, then it's
* probably more efficient to use H5V_stride_fill() instead.
*
* Return: Success: SUCCEED
*
* Failure: FAIL
*
* Programmer: Robb Matzke
* Saturday, October 11, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5V_stride_copy(int n, hsize_t elmt_size, const hsize_t *size,
const hssize_t *dst_stride, void *_dst,
const hssize_t *src_stride, const void *_src)
{
uint8 *dst = (uint8 *) _dst; /*cast for ptr arithmetic*/
const uint8 *src = (const uint8 *) _src; /*cast for ptr arithmetic*/
hsize_t idx[H5V_HYPER_NDIMS]; /*1-origin indices */
hsize_t nelmts; /*num elements to copy */
hsize_t i; /*counter */
intn j; /*counters */
hbool_t carry; /*carray for subtraction*/
FUNC_ENTER(H5V_stride_copy, FAIL);
assert (elmt_size<MAX_SIZET);
if (n) {
H5V_vector_cpy(n, idx, size);
nelmts = H5V_vector_reduce_product(n, size);
for (i=0; i<nelmts; i++) {
/* Copy an element */
HDmemcpy(dst, src, (size_t)elmt_size);
/* Decrement indices and advance pointers */
for (j=n-1, carry=TRUE; j>=0 && carry; --j) {
src += src_stride[j];
dst += dst_stride[j];
if (--idx[j]) carry = FALSE;
else idx[j] = size[j];
}
}
} else {
HDmemcpy (dst, src, (size_t)elmt_size);
HRETURN (SUCCEED);
}
FUNC_LEAVE(SUCCEED);
}
/*-------------------------------------------------------------------------
* Function: H5V_stride_copy2
*
* Purpose: Similar to H5V_stride_copy() except the source and
* destination each have their own dimensionality and size and
* we copy exactly NELMTS elements each of size ELMT_SIZE. The
* size counters wrap if NELMTS is more than a size counter.
*
* Return: Success: SUCCEED
*
* Failure: FAIL
*
* Programmer: Robb Matzke
* Saturday, October 11, 1997
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5V_stride_copy2(hsize_t nelmts, hsize_t elmt_size,
/* destination */
intn dst_n, const hsize_t *dst_size,
const hssize_t *dst_stride,
void *_dst,
/* source */
intn src_n, const hsize_t *src_size,
const hssize_t *src_stride,
const void *_src)
{
uint8 *dst = (uint8 *) _dst;
const uint8 *src = (const uint8 *) _src;
hsize_t dst_idx[H5V_HYPER_NDIMS];
hsize_t src_idx[H5V_HYPER_NDIMS];
hsize_t i;
intn j;
hbool_t carry;
FUNC_ENTER(H5V_stride_copy2, FAIL);
assert (elmt_size < MAX_SIZET);
H5V_vector_cpy(dst_n, dst_idx, dst_size);
H5V_vector_cpy(src_n, src_idx, src_size);
for (i=0; i<nelmts; i++) {
/* Copy an element */
HDmemcpy(dst, src, (size_t)elmt_size);
/* Decrement indices and advance pointers */
for (j=dst_n-1, carry=TRUE; j>=0 && carry; --j) {
dst += dst_stride[j];
if (--dst_idx[j]) carry = FALSE;
else dst_idx[j] = dst_size[j];
}
for (j=src_n-1, carry=TRUE; j>=0 && carry; --j) {
src += src_stride[j];
if (--src_idx[j]) carry = FALSE;
else src_idx[j] = src_size[j];
}
}
FUNC_LEAVE(SUCCEED);
}
/*-------------------------------------------------------------------------
* Function: H5V_array_fill
*
* Purpose: Fills all bytes of an array with the same value using memset().
* Increases amount copied by power of two until the halfway point is
* crossed, then copies the rest in one swoop.
*
* Return: Success: SUCCEED
*
* Failure: FAIL
*
* Programmer: Quincey Koziol
* Thursday, June 18, 1998
*
* Modifications:
*
*-------------------------------------------------------------------------
*/
herr_t
H5V_array_fill(void *_dst, const void *src, size_t size, size_t count)
{
size_t copy_size; /* size of the buffer to copy */
size_t copy_items; /* number of items currently copying */
size_t items_left; /* number of items left to copy */
uint8 *dst=(uint8 *)_dst; /* alias for pointer arithmetic */
FUNC_ENTER(H5V_array_fill, FAIL);
assert (dst);
assert (src);
assert (size < MAX_SIZET && size > 0);
assert (count < MAX_SIZET && count > 0);
HDmemcpy(dst, src, size); /* copy first item */
/* Initialize counters, etc. while compensating for first element copied */
copy_size = size;
copy_items = 1;
items_left = count - 1;
dst += size;
/* copy until we've copied at least half of the items */
while (items_left >= copy_items)
{
HDmemcpy(dst, _dst, copy_size); /* copy the current chunk */
dst += copy_size; /* move the offset for the next chunk */
items_left -= copy_items; /* decrement the number of items left */
copy_size *= 2; /* increase the size of the chunk to copy */
copy_items *= 2; /* increase the count of items we are copying */
} /* end while */
if (items_left > 0) /* if there are any items left to copy */
HDmemcpy(dst, _dst, items_left * size);
FUNC_LEAVE(SUCCEED);
} /* H5V_array_fill() */
|