-
Notifications
You must be signed in to change notification settings - Fork 841
/
ompi_datatype_args.c
838 lines (761 loc) · 31 KB
/
ompi_datatype_args.c
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
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
* University Research and Technology
* Corporation. All rights reserved.
* Copyright (c) 2004-2013 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2006 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2006 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2009 Oak Ridge National Labs. All rights reserved.
* Copyright (c) 2013 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2015 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include <stddef.h>
#include "opal/align.h"
#include "opal/types.h"
#include "opal/util/arch.h"
#include "opal/datatype/opal_datatype.h"
#include "opal/datatype/opal_datatype_internal.h"
#include "ompi/constants.h"
#include "ompi/datatype/ompi_datatype.h"
#include "ompi/datatype/ompi_datatype_internal.h"
#include "ompi/proc/proc.h"
static inline int
__ompi_datatype_pack_description( ompi_datatype_t* datatype,
void** packed_buffer, int* next_index );
static ompi_datatype_t*
__ompi_datatype_create_from_args( int32_t* i, OPAL_PTRDIFF_TYPE * a,
ompi_datatype_t** d, int32_t type );
typedef struct __dt_args {
int ref_count;
int create_type;
size_t total_pack_size;
int ci;
int ca;
int cd;
int* i;
OPAL_PTRDIFF_TYPE* a;
ompi_datatype_t** d;
} ompi_datatype_args_t;
/**
* Some architectures really don't like having unaligned
* accesses. We'll be int aligned, because any sane system will
* require that. But we might not be long aligned, and some
* architectures will complain if a long is accessed on int
* alignment (but not long alignment). On those architectures,
* copy the buffer into an aligned buffer first.
*/
#if OPAL_ALIGN_WORD_SIZE_INTEGERS
#define OMPI_DATATYPE_ALIGN_INT(VALUE, TYPE) \
(VALUE) = OPAL_ALIGN((VALUE), sizeof(OPAL_PTRDIFF_TYPE), TYPE)
#define OMPI_DATATYPE_ALIGN_PTR(PTR, TYPE) \
(PTR) = OPAL_ALIGN_PTR((PTR), sizeof(OPAL_PTRDIFF_TYPE), TYPE)
#else
#define OMPI_DATATYPE_ALIGN_INT(VALUE, TYPE)
#define OMPI_DATATYPE_ALIGN_PTR(PTR, TYPE)
#endif /* OPAL_ALIGN_WORD_SIZE_INTEGERS */
/**
* Some architecture require that 64 bits pointers (to pointers) has to
* be 64 bits aligned. As in the ompi_datatype_args_t structure we have 2 such
* pointers and one to an array of ints, if we start by setting the 64
* bits aligned one we will not have any trouble. Problem arise on
* SPARC 64.
*/
#define ALLOC_ARGS(PDATA, IC, AC, DC) \
do { \
int length = sizeof(ompi_datatype_args_t) + (IC) * sizeof(int) + \
(AC) * sizeof(OPAL_PTRDIFF_TYPE) + (DC) * sizeof(MPI_Datatype); \
char* buf = (char*)malloc( length ); \
ompi_datatype_args_t* pArgs = (ompi_datatype_args_t*)buf; \
pArgs->ci = (IC); \
pArgs->ca = (AC); \
pArgs->cd = (DC); \
buf += sizeof(ompi_datatype_args_t); \
if( pArgs->ca == 0 ) pArgs->a = NULL; \
else { \
pArgs->a = (OPAL_PTRDIFF_TYPE*)buf; \
buf += pArgs->ca * sizeof(OPAL_PTRDIFF_TYPE); \
} \
if( pArgs->cd == 0 ) pArgs->d = NULL; \
else { \
pArgs->d = (ompi_datatype_t**)buf; \
buf += pArgs->cd * sizeof(MPI_Datatype); \
} \
if( pArgs->ci == 0 ) pArgs->i = NULL; \
else pArgs->i = (int*)buf; \
pArgs->ref_count = 1; \
pArgs->total_pack_size = (4 + (IC)) * sizeof(int) + \
(AC) * sizeof(OPAL_PTRDIFF_TYPE) + (DC) * sizeof(int); \
OMPI_DATATYPE_ALIGN_INT( pArgs->total_pack_size, int ); \
(PDATA)->args = (void*)pArgs; \
(PDATA)->packed_description = NULL; \
} while(0)
int32_t ompi_datatype_set_args( ompi_datatype_t* pData,
int32_t ci, const int32_t** i,
int32_t ca, const OPAL_PTRDIFF_TYPE* a,
int32_t cd, ompi_datatype_t* const * d, int32_t type)
{
int pos;
ompi_datatype_args_t* pArgs;
assert( NULL == pData->args );
ALLOC_ARGS( pData, ci, ca, cd );
pArgs = (ompi_datatype_args_t*)pData->args;
pArgs->create_type = type;
switch(type) {
case MPI_COMBINER_DUP:
pArgs->total_pack_size = 0; /* store no extra data */
break;
case MPI_COMBINER_CONTIGUOUS:
pArgs->i[0] = i[0][0];
break;
case MPI_COMBINER_VECTOR:
pArgs->i[0] = i[0][0];
pArgs->i[1] = i[1][0];
pArgs->i[2] = i[2][0];
break;
case MPI_COMBINER_HVECTOR_INTEGER:
case MPI_COMBINER_HVECTOR:
pArgs->i[0] = i[0][0];
pArgs->i[1] = i[1][0];
break;
case MPI_COMBINER_INDEXED:
pos = 1;
pArgs->i[0] = i[0][0];
memcpy( pArgs->i + pos, i[1], i[0][0] * sizeof(int) );
pos += i[0][0];
memcpy( pArgs->i + pos, i[2], i[0][0] * sizeof(int) );
break;
case MPI_COMBINER_HINDEXED_INTEGER:
case MPI_COMBINER_HINDEXED:
pArgs->i[0] = i[0][0];
memcpy( pArgs->i + 1, i[1], i[0][0] * sizeof(int) );
break;
case MPI_COMBINER_INDEXED_BLOCK:
pArgs->i[0] = i[0][0];
pArgs->i[1] = i[1][0];
memcpy( pArgs->i + 2, i[2], i[0][0] * sizeof(int) );
break;
case MPI_COMBINER_STRUCT_INTEGER:
case MPI_COMBINER_STRUCT:
pArgs->i[0] = i[0][0];
memcpy( pArgs->i + 1, i[1], i[0][0] * sizeof(int) );
break;
case MPI_COMBINER_SUBARRAY:
pos = 1;
pArgs->i[0] = i[0][0];
memcpy( pArgs->i + pos, i[1], pArgs->i[0] * sizeof(int) );
pos += pArgs->i[0];
memcpy( pArgs->i + pos, i[2], pArgs->i[0] * sizeof(int) );
pos += pArgs->i[0];
memcpy( pArgs->i + pos, i[3], pArgs->i[0] * sizeof(int) );
pos += pArgs->i[0];
pArgs->i[pos] = i[4][0];
break;
case MPI_COMBINER_DARRAY:
pos = 3;
pArgs->i[0] = i[0][0];
pArgs->i[1] = i[1][0];
pArgs->i[2] = i[2][0];
memcpy( pArgs->i + pos, i[3], i[2][0] * sizeof(int) );
pos += i[2][0];
memcpy( pArgs->i + pos, i[4], i[2][0] * sizeof(int) );
pos += i[2][0];
memcpy( pArgs->i + pos, i[5], i[2][0] * sizeof(int) );
pos += i[2][0];
memcpy( pArgs->i + pos, i[6], i[2][0] * sizeof(int) );
pos += i[2][0];
pArgs->i[pos] = i[7][0];
break;
case MPI_COMBINER_F90_REAL:
case MPI_COMBINER_F90_COMPLEX:
pArgs->i[0] = i[0][0];
pArgs->i[1] = i[1][0];
break;
case MPI_COMBINER_F90_INTEGER:
pArgs->i[0] = i[0][0];
break;
case MPI_COMBINER_RESIZED:
break;
case MPI_COMBINER_HINDEXED_BLOCK:
pArgs->i[0] = i[0][0];
pArgs->i[1] = i[1][0];
break;
default:
break;
}
/* copy the array of MPI_Aint, aka OPAL_PTRDIFF_TYPE */
if( pArgs->a != NULL )
memcpy( pArgs->a, a, ca * sizeof(OPAL_PTRDIFF_TYPE) );
for( pos = 0; pos < cd; pos++ ) {
pArgs->d[pos] = d[pos];
if( !(ompi_datatype_is_predefined(d[pos])) ) {
/* We handle a user defined datatype. We should make sure that the
* user will not have the oportunity to destroy it before all derived
* datatypes are destroyed. As we keep pointers to every datatype
* (for MPI_Type_get_content and MPI_Type_get_envelope) we have to make
* sure that those datatype will be available if the user ask for them.
* However, there is no easy way to free them in this case ...
*/
OBJ_RETAIN( d[pos] );
pArgs->total_pack_size += ((ompi_datatype_args_t*)d[pos]->args)->total_pack_size;
#if OPAL_ALIGN_WORD_SIZE_INTEGERS
/*
* as total_pack_size is always aligned to
* MPI_Aint (aka OPAL_PTRDIFF_TYPE) size their sum
* will be aligned to ...
*/
assert( pArgs->total_pack_size ==
OPAL_ALIGN(pArgs->total_pack_size, sizeof(OPAL_PTRDIFF_TYPE), int) );
#endif /* OPAL_ALIGN_WORD_SIZE_INTEGERS */
} else {
pArgs->total_pack_size += 2 * sizeof(int); /* _NAMED + predefined id */
}
}
return OMPI_SUCCESS;
}
int32_t ompi_datatype_print_args( const ompi_datatype_t* pData )
{
int32_t i;
ompi_datatype_args_t* pArgs = (ompi_datatype_args_t*)pData->args;
if( ompi_datatype_is_predefined(pData) ) {
/* nothing to do for predefined data-types */
return OMPI_SUCCESS;
}
if( pArgs == NULL ) return MPI_ERR_INTERN;
printf( "type %d count ints %d count disp %d count datatype %d\n",
pArgs->create_type, pArgs->ci, pArgs->ca, pArgs->cd );
if( pArgs->i != NULL ) {
printf( "ints: " );
for( i = 0; i < pArgs->ci; i++ ) {
printf( "%d ", pArgs->i[i] );
}
printf( "\n" );
}
if( pArgs->a != NULL ) {
printf( "MPI_Aint: " );
for( i = 0; i < pArgs->ca; i++ ) {
printf( "%ld ", (long)pArgs->a[i] );
}
printf( "\n" );
}
if( pArgs->d != NULL ) {
int count = 1;
ompi_datatype_t *temp, *old;
printf( "types: " );
old = pArgs->d[0];
for( i = 1; i < pArgs->cd; i++ ) {
temp = pArgs->d[i];
if( old == temp ) {
count++;
continue;
}
if( count <= 1 ) {
if( ompi_datatype_is_predefined(old) )
printf( "%s ", old->name );
else
printf( "%p ", (void*)old );
} else {
if( ompi_datatype_is_predefined(old) )
printf( "(%d * %s) ", count, old->name );
else
printf( "(%d * %p) ", count, (void*)old );
}
count = 1;
old = temp;
}
if( count <= 1 ) {
if( ompi_datatype_is_predefined(old) )
printf( "%s ", old->name );
else
printf( "%p ", (void*)old );
} else {
if( ompi_datatype_is_predefined(old) )
printf( "(%d * %s) ", count, old->name );
else
printf( "(%d * %p) ", count, (void*)old );
}
printf( "\n" );
}
return OMPI_SUCCESS;
}
int32_t ompi_datatype_get_args( const ompi_datatype_t* pData, int32_t which,
int32_t* ci, int32_t* i,
int32_t* ca, OPAL_PTRDIFF_TYPE* a,
int32_t* cd, ompi_datatype_t** d, int32_t* type)
{
ompi_datatype_args_t* pArgs = (ompi_datatype_args_t*)pData->args;
if( NULL == pArgs ) { /* only for predefined datatypes */
if( ompi_datatype_is_predefined(pData) ) {
switch(which){
case 0:
*ci = 0;
*ca = 0;
*cd = 0;
*type = MPI_COMBINER_NAMED;
break;
default:
return MPI_ERR_INTERN;
}
return OMPI_SUCCESS;
}
return MPI_ERR_INTERN;
}
switch(which){
case 0: /* GET THE LENGTHS */
*ci = pArgs->ci;
*ca = pArgs->ca;
*cd = pArgs->cd;
*type = pArgs->create_type;
break;
case 1: /* GET THE ARGUMENTS */
if(*ci < pArgs->ci || *ca < pArgs->ca || *cd < pArgs->cd) {
return MPI_ERR_ARG;
}
if( (NULL != i) && (NULL != pArgs->i) ) {
memcpy( i, pArgs->i, pArgs->ci * sizeof(int) );
}
if( (NULL != a) && (NULL != pArgs->a) ) {
memcpy( a, pArgs->a, pArgs->ca * sizeof(OPAL_PTRDIFF_TYPE) );
}
if( (NULL != d) && (NULL != pArgs->d) ) {
memcpy( d, pArgs->d, pArgs->cd * sizeof(MPI_Datatype) );
}
break;
default:
return MPI_ERR_INTERN;
}
return OMPI_SUCCESS;
}
int32_t ompi_datatype_copy_args( const ompi_datatype_t* source_data,
ompi_datatype_t* dest_data )
{
ompi_datatype_args_t* pArgs = (ompi_datatype_args_t*)source_data->args;
/* If required then increase the reference count of the arguments. This avoid us
* to make one more copy for a read only piece of memory.
*/
assert( NULL != source_data->args );
pArgs->ref_count++;
dest_data->args = pArgs;
return OMPI_SUCCESS;
}
/* In the dt_add function we increase the reference count for all datatypes
* (except for the predefined ones) that get added to another datatype. This
* insure that they cannot get released until all the references to them
* get removed.
*/
int32_t ompi_datatype_release_args( ompi_datatype_t* pData )
{
int i;
ompi_datatype_args_t* pArgs = (ompi_datatype_args_t*)pData->args;
assert( 0 < pArgs->ref_count );
pArgs->ref_count--;
if( 0 == pArgs->ref_count ) {
/* There are some duplicated datatypes around that have a pointer to this
* args. We will release them only when the last datatype will dissapear.
*/
for( i = 0; i < pArgs->cd; i++ ) {
if( !(ompi_datatype_is_predefined(pArgs->d[i])) ) {
OBJ_RELEASE( pArgs->d[i] );
}
}
free( pData->args );
}
pData->args = NULL;
return OMPI_SUCCESS;
}
static inline int __ompi_datatype_pack_description( ompi_datatype_t* datatype,
void** packed_buffer, int* next_index )
{
int i, *position = (int*)*packed_buffer;
ompi_datatype_args_t* args = (ompi_datatype_args_t*)datatype->args;
char* next_packed = (char*)*packed_buffer;
if( ompi_datatype_is_predefined(datatype) ) {
position[0] = MPI_COMBINER_NAMED;
position[1] = datatype->id; /* On the OMPI - layer, copy the ompi_datatype.id */
next_packed += (2 * sizeof(int));
*packed_buffer = next_packed;
return OMPI_SUCCESS;
}
/* For duplicated datatype we don't have to store all the information */
if( MPI_COMBINER_DUP == args->create_type ) {
ompi_datatype_t* temp_data = args->d[0];
return __ompi_datatype_pack_description(temp_data,
packed_buffer,
next_index );
}
position[0] = args->create_type;
position[1] = args->ci;
position[2] = args->ca;
position[3] = args->cd;
next_packed += (4 * sizeof(int));
/* So far there are 4 integers in the array, so we're still 64 bits aligned
* if we suppose that the original buffer was 64 bits aligned.
*
* In order to solve issues with the Sparc 64 which require 64 bits pointers
* to be correctly aligned, we have to start adding the data in a smart way,
* just to keep everything as aligned as possible. Therefore, the first
* array we have to copy is the array of displacements, followed by the
* array of datatypes (both of them might be arrays of pointers) and then
* finally the array of counts.
*/
if( 0 < args->ca ) {
memcpy( next_packed, args->a, sizeof(OPAL_PTRDIFF_TYPE) * args->ca );
next_packed += sizeof(OPAL_PTRDIFF_TYPE) * args->ca;
}
position = (int*)next_packed;
next_packed += sizeof(int) * args->cd;
/* copy the aray of counts (32 bits aligned) */
memcpy( next_packed, args->i, sizeof(int) * args->ci );
next_packed += args->ci * sizeof(int);
/* description of next datatype should be 64 bits aligned */
OMPI_DATATYPE_ALIGN_PTR(next_packed, char*);
/* copy the rest of the data */
for( i = 0; i < args->cd; i++ ) {
ompi_datatype_t* temp_data = args->d[i];
if( ompi_datatype_is_predefined(temp_data) ) {
position[i] = temp_data->id; /* On the OMPI - layer, copy the ompi_datatype.id */
} else {
position[i] = *next_index;
(*next_index)++;
__ompi_datatype_pack_description( temp_data,
(void**)&next_packed,
next_index );
}
}
*packed_buffer = next_packed;
return OMPI_SUCCESS;
}
int ompi_datatype_get_pack_description( ompi_datatype_t* datatype,
const void** packed_buffer )
{
ompi_datatype_args_t* args = (ompi_datatype_args_t*)datatype->args;
int next_index = OMPI_DATATYPE_MAX_PREDEFINED;
void* recursive_buffer;
if( NULL == datatype->packed_description ) {
if( ompi_datatype_is_predefined(datatype) ) {
datatype->packed_description = malloc(2 * sizeof(int));
} else if( NULL == args ) {
return OMPI_ERROR;
} else {
datatype->packed_description = malloc(args->total_pack_size);
}
recursive_buffer = datatype->packed_description;
__ompi_datatype_pack_description( datatype, &recursive_buffer, &next_index );
if( !ompi_datatype_is_predefined(datatype) ) {
args->total_pack_size = (uintptr_t)((char*)recursive_buffer - (char*)datatype->packed_description);
OMPI_DATATYPE_ALIGN_PTR(args->total_pack_size, char*);
}
}
*packed_buffer = (const void*)datatype->packed_description;
return OMPI_SUCCESS;
}
size_t ompi_datatype_pack_description_length( ompi_datatype_t* datatype )
{
if( ompi_datatype_is_predefined(datatype) ) {
return 2 * sizeof(int);
}
if( NULL == datatype->packed_description ) {
const void* buf;
int rc;
rc = ompi_datatype_get_pack_description(datatype, &buf);
if( OMPI_SUCCESS != rc )
return 0;
}
assert( NULL != (ompi_datatype_args_t*)datatype->args );
assert( NULL != (ompi_datatype_args_t*)datatype->packed_description );
return ((ompi_datatype_args_t*)datatype->args)->total_pack_size;
}
static ompi_datatype_t* __ompi_datatype_create_from_packed_description( void** packed_buffer,
const struct ompi_proc_t* remote_processor )
{
int* position;
ompi_datatype_t* datatype = NULL;
ompi_datatype_t** array_of_datatype;
OPAL_PTRDIFF_TYPE* array_of_disp;
int* array_of_length;
int number_of_length, number_of_disp, number_of_datatype, data_id;
int create_type, i;
char* next_buffer;
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
bool need_swap = false;
if( (remote_processor->super.proc_arch ^ ompi_proc_local()->super.proc_arch) &
OPAL_ARCH_ISBIGENDIAN ) {
need_swap = true;
}
#endif
next_buffer = (char*)*packed_buffer;
/* The pointer should always be aligned on MPI_Aint, aka OPAL_PTRDIFF_TYPE */
OMPI_DATATYPE_ALIGN_PTR(next_buffer, char*);
position = (int*)next_buffer;
create_type = position[0];
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
if (need_swap) {
create_type = opal_swap_bytes4(create_type);
}
#endif
if( MPI_COMBINER_NAMED == create_type ) {
/* there we have a simple predefined datatype */
data_id = position[1];
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
if (need_swap) {
data_id = opal_swap_bytes4(data_id);
}
#endif
assert( data_id < OMPI_DATATYPE_MAX_PREDEFINED );
*packed_buffer = position + 2;
return (ompi_datatype_t*)ompi_datatype_basicDatatypes[data_id];
}
number_of_length = position[1];
number_of_disp = position[2];
number_of_datatype = position[3];
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
if (need_swap) {
number_of_length = opal_swap_bytes4(number_of_length);
number_of_disp = opal_swap_bytes4(number_of_disp);
number_of_datatype = opal_swap_bytes4(number_of_datatype);
}
#endif
array_of_datatype = (ompi_datatype_t**)malloc( sizeof(ompi_datatype_t*) *
number_of_datatype );
next_buffer += (4 * sizeof(int)); /* move after the header */
array_of_disp = (OPAL_PTRDIFF_TYPE*)next_buffer;
next_buffer += number_of_disp * sizeof(OPAL_PTRDIFF_TYPE);
/* the other datatypes */
position = (int*)next_buffer;
next_buffer += number_of_datatype * sizeof(int);
/* the array of lengths (32 bits aligned) */
array_of_length = (int*)next_buffer;
next_buffer += (number_of_length * sizeof(int));
for( i = 0; i < number_of_datatype; i++ ) {
data_id = position[i];
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
if (need_swap) {
data_id = opal_swap_bytes4(data_id);
}
#endif
if( data_id < OMPI_DATATYPE_MAX_PREDEFINED ) {
array_of_datatype[i] = (ompi_datatype_t*)ompi_datatype_basicDatatypes[data_id];
continue;
}
array_of_datatype[i] =
__ompi_datatype_create_from_packed_description( (void**)&next_buffer,
remote_processor );
if( NULL == array_of_datatype[i] ) {
/* don't cleanup more than required. We can now modify these
* values as we already know we have failed to rebuild the
* datatype.
*/
array_of_datatype[i] = (ompi_datatype_t*)ompi_datatype_basicDatatypes[OPAL_DATATYPE_INT1]; /*XXX TODO */
number_of_datatype = i;
goto cleanup_and_exit;
}
}
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
if (need_swap) {
for (i = 0 ; i < number_of_length ; ++i) {
array_of_length[i] = opal_swap_bytes4(array_of_length[i]);
}
for (i = 0 ; i < number_of_disp ; ++i) {
#if SIZEOF_PTRDIFF_T == 4
array_of_disp[i] = opal_swap_bytes4(array_of_disp[i]);
#elif SIZEOF_PTRDIFF_T == 8
array_of_disp[i] = (MPI_Aint)opal_swap_bytes8(array_of_disp[i]);
#else
#error "Unknown size of ptrdiff_t"
#endif
}
}
#endif
datatype = __ompi_datatype_create_from_args( array_of_length, array_of_disp,
array_of_datatype, create_type );
*packed_buffer = next_buffer;
cleanup_and_exit:
for( i = 0; i < number_of_datatype; i++ ) {
if( !(ompi_datatype_is_predefined(array_of_datatype[i])) ) {
OBJ_RELEASE(array_of_datatype[i]);
}
}
free( array_of_datatype );
return datatype;
}
static ompi_datatype_t* __ompi_datatype_create_from_args( int32_t* i, MPI_Aint* a,
ompi_datatype_t** d, int32_t type )
{
ompi_datatype_t* datatype = NULL;
switch(type){
/******************************************************************/
case MPI_COMBINER_DUP:
/* should we duplicate d[0]? */
/* ompi_datatype_set_args( datatype, 0, NULL, 0, NULL, 1, d[0], MPI_COMBINER_DUP ); */
assert(0); /* shouldn't happen */
break;
/******************************************************************/
case MPI_COMBINER_CONTIGUOUS:
ompi_datatype_create_contiguous( i[0], d[0], &datatype );
ompi_datatype_set_args( datatype, 1, (const int **) &i, 0, NULL, 1, d, MPI_COMBINER_CONTIGUOUS );
break;
/******************************************************************/
case MPI_COMBINER_VECTOR:
ompi_datatype_create_vector( i[0], i[1], i[2], d[0], &datatype );
{
const int* a_i[3] = {&i[0], &i[1], &i[2]};
ompi_datatype_set_args( datatype, 3, a_i, 0, NULL, 1, d, MPI_COMBINER_VECTOR );
}
break;
/******************************************************************/
case MPI_COMBINER_HVECTOR_INTEGER:
case MPI_COMBINER_HVECTOR:
ompi_datatype_create_hvector( i[0], i[1], a[0], d[0], &datatype );
{
const int* a_i[2] = {&i[0], &i[1]};
ompi_datatype_set_args( datatype, 2, a_i, 1, a, 1, d, MPI_COMBINER_HVECTOR );
}
break;
/******************************************************************/
case MPI_COMBINER_INDEXED: /* TO CHECK */
ompi_datatype_create_indexed( i[0], &(i[1]), &(i[1+i[0]]), d[0], &datatype );
{
const int* a_i[3] = {&i[0], &i[1], &(i[1+i[0]])};
ompi_datatype_set_args( datatype, 2 * i[0] + 1, a_i, 0, NULL, 1, d, MPI_COMBINER_INDEXED );
}
break;
/******************************************************************/
case MPI_COMBINER_HINDEXED_INTEGER:
case MPI_COMBINER_HINDEXED:
ompi_datatype_create_hindexed( i[0], &(i[1]), a, d[0], &datatype );
{
const int* a_i[2] = {&i[0], &i[1]};
ompi_datatype_set_args( datatype, i[0] + 1, a_i, i[0], a, 1, d, MPI_COMBINER_HINDEXED );
}
break;
/******************************************************************/
case MPI_COMBINER_INDEXED_BLOCK:
ompi_datatype_create_indexed_block( i[0], i[1], &(i[2]), d[0], &datatype );
{
const int* a_i[3] = {&i[0], &i[1], &i[2]};
ompi_datatype_set_args( datatype, i[0] + 2, a_i, 0, NULL, 1, d, MPI_COMBINER_INDEXED_BLOCK );
}
break;
/******************************************************************/
case MPI_COMBINER_STRUCT_INTEGER:
case MPI_COMBINER_STRUCT:
ompi_datatype_create_struct( i[0], &(i[1]), a, d, &datatype );
{
const int* a_i[2] = {&i[0], &i[1]};
ompi_datatype_set_args( datatype, i[0] + 1, a_i, i[0], a, i[0], d, MPI_COMBINER_STRUCT );
}
break;
/******************************************************************/
case MPI_COMBINER_SUBARRAY:
ompi_datatype_create_subarray( i[0], &i[1 + 0 * i[0]], &i[1 + 1 * i[0]],
&i[1 + 2 * i[0]], i[1 + 3 * i[0]],
d[0], &datatype );
{
const int* a_i[5] = {&i[0], &i[1 + 0 * i[0]], &i[1 + 1 * i[0]], &i[1 + 2 * i[0]], &i[1 + 3 * i[0]]};
ompi_datatype_set_args( datatype, 3 * i[0] + 2, a_i, 0, NULL, 1, d, MPI_COMBINER_SUBARRAY);
}
break;
/******************************************************************/
case MPI_COMBINER_DARRAY:
ompi_datatype_create_darray( i[0] /* size */, i[1] /* rank */, i[2] /* ndims */,
&i[3 + 0 * i[0]], &i[3 + 1 * i[0]],
&i[3 + 2 * i[0]], &i[3 + 3 * i[0]],
i[3 + 4 * i[0]], d[0], &datatype );
{
const int* a_i[8] = {&i[0], &i[1], &i[2], &i[3 + 0 * i[0]], &i[3 + 1 * i[0]], &i[3 + 2 * i[0]],
&i[3 + 3 * i[0]], &i[3 + 4 * i[0]]};
ompi_datatype_set_args( datatype, 4 * i[0] + 4,a_i, 0, NULL, 1, d, MPI_COMBINER_DARRAY);
}
break;
/******************************************************************/
case MPI_COMBINER_F90_REAL:
case MPI_COMBINER_F90_COMPLEX:
/*pArgs->i[0] = i[0][0];
pArgs->i[1] = i[1][0];
*/
break;
/******************************************************************/
case MPI_COMBINER_F90_INTEGER:
/*pArgs->i[0] = i[0][0];*/
break;
/******************************************************************/
case MPI_COMBINER_RESIZED:
/*ompi_datatype_set_args( datatype, 0, NULL, 2, a, 1, d, MPI_COMBINER_RESIZED );*/
break;
/******************************************************************/
case MPI_COMBINER_HINDEXED_BLOCK:
ompi_datatype_create_hindexed_block( i[0], i[1], a, d[0], &datatype );
{
const int* a_i[2] = {&i[0], &i[1]};
ompi_datatype_set_args( datatype, 2 + i[0], a_i, i[0], a, 1, d, MPI_COMBINER_HINDEXED_BLOCK );
}
break;
/******************************************************************/
default:
break;
}
return datatype;
}
ompi_datatype_t* ompi_datatype_create_from_packed_description( void** packed_buffer,
struct ompi_proc_t* remote_processor )
{
ompi_datatype_t* datatype;
datatype = __ompi_datatype_create_from_packed_description( packed_buffer,
remote_processor );
/* Keep the pointer aligned to MPI_Aint */
OMPI_DATATYPE_ALIGN_PTR(*packed_buffer, void*);
if( NULL == datatype ) {
return NULL;
}
ompi_datatype_commit( &datatype );
return datatype;
}
/**
* Parse the datatype description from the args and find if the
* datatype is created from a single predefined type. If yes,
* return the type, otherwise return NULL.
*/
ompi_datatype_t* ompi_datatype_get_single_predefined_type_from_args( ompi_datatype_t* type )
{
ompi_datatype_t *predef = NULL, *current_type, *current_predef;
ompi_datatype_args_t* args = (ompi_datatype_args_t*)type->args;
int i;
if( ompi_datatype_is_predefined(type) )
return type;
for( i = 0; i < args->cd; i++ ) {
current_type = args->d[i];
if( ompi_datatype_is_predefined(current_type) ) {
current_predef = current_type;
} else {
current_predef = ompi_datatype_get_single_predefined_type_from_args(current_type);
if( NULL == current_predef ) { /* No single predefined datatype */
return NULL;
}
}
if( NULL == predef ) { /* This is the first iteration */
predef = current_predef;
} else {
/**
* What exactly should we consider as identical types? If they are
* the same MPI level type, or if they map to the same OPAL datatype?
* In other words, MPI_FLOAT and MPI_REAL4 are they identical?
*/
if( predef != current_predef ) {
return NULL;
}
}
}
return predef;
}