Source

mpi3-fortran / opal / datatype / opal_convertor.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
/* -*- Mode: C; c-basic-offset:4 ; -*- */
/*
 * Copyright (c) 2004-2006 The Trustees of Indiana University and Indiana
 *                         University Research and Technology
 *                         Corporation.  All rights reserved.
 * Copyright (c) 2004-2010 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) 2011      NVIDIA Corporation.  All rights reserved.
 * $COPYRIGHT$
 *
 * Additional copyrights may follow
 *
 * $HEADER$
 */

#include "opal_config.h"

#include <stddef.h>
#include <stdio.h>

#ifdef HAVE_STDINT_H
#include <stdint.h>
#endif

#include "opal/prefetch.h"
#include "opal/util/arch.h"
#include "opal/util/output.h"

#include "opal/datatype/opal_datatype_internal.h"
#include "opal/datatype/opal_datatype.h"
#include "opal/datatype/opal_convertor.h"
#include "opal/datatype/opal_datatype_checksum.h"
#include "opal/datatype/opal_datatype_prototypes.h"
#include "opal/datatype/opal_convertor_internal.h"
#if OPAL_CUDA_SUPPORT
#include "opal/datatype/opal_datatype_cuda.h"
#define MEMCPY_CUDA( DST, SRC, BLENGTH, CONVERTOR ) \
    CONVERTOR->cbmemcpy( (DST), (SRC), (BLENGTH), (CONVERTOR) )
#endif

extern int opal_convertor_create_stack_with_pos_general( opal_convertor_t* convertor,
                                                         int starting_point, const int* sizes );

static void opal_convertor_construct( opal_convertor_t* convertor )
{
    convertor->pStack         = convertor->static_stack;
    convertor->stack_size     = DT_STATIC_STACK_SIZE;
    convertor->partial_length = 0;
    convertor->remoteArch     = opal_local_arch;
    convertor->flags          = OPAL_DATATYPE_FLAG_NO_GAPS | CONVERTOR_COMPLETED;
#if OPAL_CUDA_SUPPORT
    convertor->cbmemcpy       = &opal_cuda_memcpy;
#endif
}


static void opal_convertor_destruct( opal_convertor_t* convertor )
{
    opal_convertor_cleanup( convertor );
}

OBJ_CLASS_INSTANCE(opal_convertor_t, opal_object_t, opal_convertor_construct, opal_convertor_destruct );

static opal_convertor_master_t* opal_convertor_master_list = NULL;

extern conversion_fct_t opal_datatype_heterogeneous_copy_functions[OPAL_DATATYPE_MAX_PREDEFINED];
extern conversion_fct_t opal_datatype_copy_functions[OPAL_DATATYPE_MAX_PREDEFINED];

void opal_convertor_destroy_masters( void )
{
    opal_convertor_master_t* master = opal_convertor_master_list;

    while( NULL != master ) {
        opal_convertor_master_list = master->next;
        master->next = NULL;
        /* Cleanup the conversion function if not one of the defaults */
        if( (master->pFunctions != opal_datatype_heterogeneous_copy_functions) &&
            (master->pFunctions != opal_datatype_copy_functions) )
            free( master->pFunctions );

        free( master );
        master = opal_convertor_master_list;
    }
}

/**
 * Find or create a convertor suitable for the remote architecture. If there
 * is already a master convertor for this architecture then return it.
 * Otherwise, create and initialize a full featured master convertor.
 */
opal_convertor_master_t* opal_convertor_find_or_create_master( uint32_t remote_arch )
{
    opal_convertor_master_t* master = opal_convertor_master_list;
    int i;
    size_t* remote_sizes;

    while( NULL != master ) {
        if( master->remote_arch == remote_arch )
            return master;
        master = master->next;
    }
    /**
     * Create a new convertor matching the specified architecture and add it to the
     * master convertor list.
     */
    master = (opal_convertor_master_t*)malloc( sizeof(opal_convertor_master_t) );
    master->next = opal_convertor_master_list;
    opal_convertor_master_list = master;
    master->remote_arch = remote_arch;
    master->flags       = 0;
    master->hetero_mask = 0;
    /**
     * Most of the sizes will be identical, so for now just make a copy of
     * the local ones. As master->remote_sizes is defined as being an array of
     * consts we have to manually cast it before using it for writing purposes.
     */
    remote_sizes = (size_t*)master->remote_sizes;
    memcpy(remote_sizes, opal_datatype_local_sizes, sizeof(size_t) * OPAL_DATATYPE_MAX_PREDEFINED);
    /**
     * If the local and remote architecture are the same there is no need
     * to check for the remote data sizes. They will always be the same as
     * the local ones.
     */
    if( master->remote_arch == opal_local_arch ) {
        master->pFunctions = opal_datatype_copy_functions;
        master->flags |= CONVERTOR_HOMOGENEOUS;
        return master;
    }

    /* Find out the remote bool size */
    if( opal_arch_checkmask( &master->remote_arch, OPAL_ARCH_BOOLIS8 ) ) {
        remote_sizes[OPAL_DATATYPE_BOOL] = 1;
    } else if( opal_arch_checkmask( &master->remote_arch, OPAL_ARCH_BOOLIS16 ) ) {
        remote_sizes[OPAL_DATATYPE_BOOL] = 2;
    } else if( opal_arch_checkmask( &master->remote_arch, OPAL_ARCH_BOOLIS32 ) ) {
        remote_sizes[OPAL_DATATYPE_BOOL] = 4;
    } else {
        opal_output( 0, "Unknown sizeof(bool) for the remote architecture\n" );
    }

    /**
     * Now we can compute the conversion mask. For all sizes where the remote
     * and local architecture differ a conversion is needed. Moreover, if the
     * 2 architectures don't have the same endianess all data with a length
     * over 2 bytes (with the exception of logicals) have to be byte-swapped.
     */
    for( i = OPAL_DATATYPE_FIRST_TYPE; i < OPAL_DATATYPE_MAX_PREDEFINED; i++ ) {
        if( remote_sizes[i] != opal_datatype_local_sizes[i] )
            master->hetero_mask |= (((uint32_t)1) << i);
    }
    if( opal_arch_checkmask( &master->remote_arch, OPAL_ARCH_ISBIGENDIAN ) !=
        opal_arch_checkmask( &opal_local_arch, OPAL_ARCH_ISBIGENDIAN ) ) {
        uint32_t hetero_mask = 0;

        for( i = OPAL_DATATYPE_FIRST_TYPE; i < OPAL_DATATYPE_MAX_PREDEFINED; i++ ) {
            if( remote_sizes[i] > 1 )
                hetero_mask |= (((uint32_t)1) << i);
        }
        hetero_mask &= ~(((uint32_t)1) << OPAL_DATATYPE_BOOL);
        master->hetero_mask |= hetero_mask;
    }
    master->pFunctions = (conversion_fct_t*)malloc( sizeof(opal_datatype_heterogeneous_copy_functions) );
    /**
     * Usually the heterogeneous functions are slower than the copy ones. Let's
     * try to minimize the usage of the heterogeneous versions.
     */
    for( i = OPAL_DATATYPE_FIRST_TYPE; i < OPAL_DATATYPE_MAX_PREDEFINED; i++ ) {
        if( master->hetero_mask & (((uint32_t)1) << i) )
            master->pFunctions[i] = opal_datatype_heterogeneous_copy_functions[i];
        else
            master->pFunctions[i] = opal_datatype_copy_functions[i];
    }

    /* We're done so far, return the mater convertor */
    return master;
}


opal_convertor_t* opal_convertor_create( int32_t remote_arch, int32_t mode )
{
    opal_convertor_t* convertor = OBJ_NEW(opal_convertor_t);
    opal_convertor_master_t* master;

    master = opal_convertor_find_or_create_master( remote_arch );

    convertor->remoteArch = remote_arch;
    convertor->stack_pos  = 0;
    convertor->flags      = master->flags;
    convertor->master     = master;

    return convertor;
}

#define OPAL_CONVERTOR_SET_STATUS_BEFORE_PACK_UNPACK( CONVERTOR, IOV, OUT, MAX_DATA ) \
    do {                                                                \
        /* protect against over packing data */                         \
        if( OPAL_UNLIKELY((CONVERTOR)->flags & CONVERTOR_COMPLETED) ) { \
            (IOV)[0].iov_len = 0;                                       \
            *(OUT) = 0;                                                 \
            *(MAX_DATA) = 0;                                            \
            return 1;  /* nothing to do */                              \
        }                                                               \
        (CONVERTOR)->checksum = OPAL_CSUM_ZERO;                         \
        (CONVERTOR)->csum_ui1 = 0;                                      \
        (CONVERTOR)->csum_ui2 = 0;                                      \
        assert( (CONVERTOR)->bConverted < (CONVERTOR)->local_size );    \
    } while(0)

/** 
 * Return 0 if everything went OK and if there is still room before the complete
 *          conversion of the data (need additional call with others input buffers )
 *        1 if everything went fine and the data was completly converted
 *       -1 something wrong occurs.
 */
int32_t opal_convertor_pack( opal_convertor_t* pConv,
                             struct iovec* iov, uint32_t* out_size,
                             size_t* max_data )
{
    OPAL_CONVERTOR_SET_STATUS_BEFORE_PACK_UNPACK( pConv, iov, out_size, max_data );

    if( OPAL_LIKELY(pConv->flags & CONVERTOR_NO_OP) ) {
        /**
         * We are doing conversion on a contiguous datatype on a homogeneous
         * environment. The convertor contain minimal informations, we only
         * use the bConverted to manage the conversion.
         */
        uint32_t i;
        unsigned char* base_pointer;
        size_t pending_length = pConv->local_size - pConv->bConverted;

        *max_data = pending_length;
        opal_convertor_get_current_pointer( pConv, (void**)&base_pointer );

        for( i = 0; i < *out_size; i++ ) {
            if( iov[i].iov_len >= pending_length ) {
                goto complete_contiguous_data_pack;
            }
            if( OPAL_LIKELY(NULL == iov[i].iov_base) )
                iov[i].iov_base = (IOVBASE_TYPE *) base_pointer;
            else
#if OPAL_CUDA_SUPPORT
                MEMCPY_CUDA( iov[i].iov_base, base_pointer, iov[i].iov_len, pConv );
#else
                MEMCPY( iov[i].iov_base, base_pointer, iov[i].iov_len );
#endif
            pending_length -= iov[i].iov_len;
            base_pointer += iov[i].iov_len;
        }
        *max_data -= pending_length;
        pConv->bConverted += (*max_data);
        return 0;

complete_contiguous_data_pack:
        iov[i].iov_len = pending_length;
        if( OPAL_LIKELY(NULL == iov[i].iov_base) )
            iov[i].iov_base = (IOVBASE_TYPE *) base_pointer;
        else
#if OPAL_CUDA_SUPPORT
            MEMCPY_CUDA( iov[i].iov_base, base_pointer, iov[i].iov_len, pConv );
#else
            MEMCPY( iov[i].iov_base, base_pointer, iov[i].iov_len );
#endif
        pConv->bConverted = pConv->local_size;
        *out_size = i + 1;
        pConv->flags |= CONVERTOR_COMPLETED;
        return 1;
    }

    return pConv->fAdvance( pConv, iov, out_size, max_data );
}


int32_t opal_convertor_unpack( opal_convertor_t* pConv,
                               struct iovec* iov, uint32_t* out_size,
                               size_t* max_data )
{
    OPAL_CONVERTOR_SET_STATUS_BEFORE_PACK_UNPACK( pConv, iov, out_size, max_data );

    if( OPAL_LIKELY(pConv->flags & CONVERTOR_NO_OP) ) {
        /**
         * We are doing conversion on a contiguous datatype on a homogeneous
         * environment. The convertor contain minimal informations, we only
         * use the bConverted to manage the conversion.
         */
        uint32_t i;
        unsigned char* base_pointer;
        size_t pending_length = pConv->local_size - pConv->bConverted;

        *max_data = pending_length;
        opal_convertor_get_current_pointer( pConv, (void**)&base_pointer );

        for( i = 0; i < *out_size; i++ ) {
            if( iov[i].iov_len >= pending_length ) {
                goto complete_contiguous_data_unpack;
            }
#if OPAL_CUDA_SUPPORT
            MEMCPY_CUDA( base_pointer, iov[i].iov_base, iov[i].iov_len, pConv );
#else
            MEMCPY( base_pointer, iov[i].iov_base, iov[i].iov_len );
#endif
            pending_length -= iov[i].iov_len;
            base_pointer += iov[i].iov_len;
        }
        *max_data -= pending_length;
        pConv->bConverted += (*max_data);
        return 0;

complete_contiguous_data_unpack:
        iov[i].iov_len = pending_length;
#if OPAL_CUDA_SUPPORT
        MEMCPY_CUDA( base_pointer, iov[i].iov_base, iov[i].iov_len, pConv );
#else
        MEMCPY( base_pointer, iov[i].iov_base, iov[i].iov_len );
#endif
        pConv->bConverted = pConv->local_size;
        *out_size = i + 1;
        pConv->flags |= CONVERTOR_COMPLETED;
        return 1;
    }

    return pConv->fAdvance( pConv, iov, out_size, max_data );
}

static inline int opal_convertor_create_stack_with_pos_contig( opal_convertor_t* pConvertor,
                                                               size_t starting_point, const size_t* sizes )
{
    dt_stack_t* pStack;   /* pointer to the position on the stack */
    const opal_datatype_t* pData = pConvertor->pDesc;
    dt_elem_desc_t* pElems;
    uint32_t count;
    OPAL_PTRDIFF_TYPE extent;

    pStack = pConvertor->pStack;
    /**
     * The prepare function already make the selection on which data representation
     * we have to use: normal one or the optimized version ?
     */
    pElems = pConvertor->use_desc->desc;

    count = (uint32_t)(starting_point / pData->size);
    extent = pData->ub - pData->lb;

    pStack[0].type     = OPAL_DATATYPE_LOOP;  /* the first one is always the loop */
    pStack[0].count    = pConvertor->count - count;
    pStack[0].index    = -1;
    pStack[0].disp     = count * extent;

    /* now compute the number of pending bytes */
    count = (uint32_t)(starting_point - count * pData->size);
    /**
     * We save the current displacement starting from the begining
     * of this data.
     */
    if( OPAL_LIKELY(0 == count) ) {
        pStack[1].type     = pElems->elem.common.type;
        pStack[1].count    = pElems->elem.count;
        pStack[1].disp     = pElems->elem.disp;
    } else {
        pStack[1].type  = OPAL_DATATYPE_UINT1;
        pStack[1].count = pData->size - count;
        pStack[1].disp  = pData->true_lb + count;
    }
    pStack[1].index    = 0;  /* useless */

    pConvertor->bConverted = starting_point;
    pConvertor->stack_pos = 1;
    assert( 0 == pConvertor->partial_length );
    return OPAL_SUCCESS;
}

static inline
int opal_convertor_create_stack_at_begining( opal_convertor_t* convertor,
                                             const size_t* sizes )
{
    dt_stack_t* pStack = convertor->pStack;
    dt_elem_desc_t* pElems;

    /**
     * The prepare function already make the selection on which data representation
     * we have to use: normal one or the optimized version ?
     */
    pElems = convertor->use_desc->desc;

    convertor->stack_pos      = 1;
    convertor->partial_length = 0;
    convertor->bConverted     = 0;
    /**
     * Fill the first position on the stack. This one correspond to the
     * last fake OPAL_DATATYPE_END_LOOP that we add to the data representation and
     * allow us to move quickly inside the datatype when we have a count.
     */
    pStack[0].index = -1;
    pStack[0].count = convertor->count;
    pStack[0].disp  = 0;

    pStack[1].index = 0;
    pStack[1].disp = 0;
    if( pElems[0].elem.common.type == OPAL_DATATYPE_LOOP ) {
        pStack[1].count = pElems[0].loop.loops;
    } else {
        pStack[1].count = pElems[0].elem.count;
    }
    return OPAL_SUCCESS;
}


int32_t opal_convertor_set_position_nocheck( opal_convertor_t* convertor,
                                             size_t* position )
{
    int32_t rc;

    /**
     * If we plan to rollback the convertor then first we have to set it
     * at the beginning.
     */
    if( (0 == (*position)) || ((*position) < convertor->bConverted) ) {
        rc = opal_convertor_create_stack_at_begining( convertor, opal_datatype_local_sizes );
        if( 0 == (*position) ) return rc;
    }
    if( OPAL_LIKELY(convertor->flags & OPAL_DATATYPE_FLAG_CONTIGUOUS) ) {
        rc = opal_convertor_create_stack_with_pos_contig( convertor, (*position),
                                                          opal_datatype_local_sizes );
    } else {
        rc = opal_convertor_generic_simple_position( convertor, position );
    }
    *position = convertor->bConverted;
    return rc;
}


/**
 * Compute the remote size.
 */
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
#define OPAL_CONVERTOR_COMPUTE_REMOTE_SIZE(convertor, datatype, bdt_mask) \
{                                                                         \
    if( OPAL_UNLIKELY(0 != (bdt_mask)) ) {                                \
        opal_convertor_master_t* master;                                  \
        int i;                                                            \
        uint32_t mask = datatype->bdt_used;                               \
        convertor->flags ^= CONVERTOR_HOMOGENEOUS;                        \
        master = convertor->master;                                       \
        convertor->remote_size = 0;                                       \
        for( i = OPAL_DATATYPE_FIRST_TYPE; mask && (i < OPAL_DATATYPE_MAX_PREDEFINED); i++ ) { \
            if( mask & ((uint32_t)1 << i) ) {                             \
                convertor->remote_size += (datatype->btypes[i] *          \
                                           master->remote_sizes[i]);      \
                mask ^= ((uint32_t)1 << i);                               \
            }                                                             \
        }                                                                 \
        convertor->remote_size *= convertor->count;                       \
        convertor->use_desc = &(datatype->desc);                          \
    }                                                                     \
}
#else
#define OPAL_CONVERTOR_COMPUTE_REMOTE_SIZE(convertor, datatype, bdt_mask) \
    assert(0 == (bdt_mask))
#endif  /* OPAL_ENABLE_HETEROGENEOUS_SUPPORT */

/**
 * This macro will initialize a convertor based on a previously created
 * convertor. The idea is the move outside these function the heavy
 * selection of architecture features for the convertors. I consider
 * here that the convertor is clean, either never initialized or already
 * cleaned.
 */
#define OPAL_CONVERTOR_PREPARE( convertor, datatype, count, pUserBuf )  \
    {                                                                   \
        uint32_t bdt_mask;                                              \
                                                                        \
        /* If the data is empty we just mark the convertor as           \
         * completed. With this flag set the pack and unpack functions  \
         * will not do anything.                                        \
         */                                                             \
        if( OPAL_UNLIKELY((0 == count) || (0 == datatype->size)) ) {    \
            convertor->flags |= OPAL_DATATYPE_FLAG_NO_GAPS | CONVERTOR_COMPLETED;  \
            convertor->local_size = convertor->remote_size = 0;         \
            return OPAL_SUCCESS;                                        \
        }                                                               \
        /* Compute the local in advance */                              \
        convertor->local_size = count * datatype->size;                 \
        convertor->pBaseBuf   = (unsigned char*)pUserBuf;               \
        convertor->count      = count;                                  \
                                                                        \
        /* Grab the datatype part of the flags */                       \
        convertor->flags     &= CONVERTOR_TYPE_MASK;                    \
        convertor->flags     |= (CONVERTOR_DATATYPE_MASK & datatype->flags); \
        convertor->flags     |= (CONVERTOR_NO_OP | CONVERTOR_HOMOGENEOUS); \
        convertor->pDesc      = (opal_datatype_t*)datatype;             \
        convertor->bConverted = 0;                                      \
        /* By default consider the optimized description */             \
        convertor->use_desc = &(datatype->opt_desc);                    \
                                                                        \
        convertor->remote_size = convertor->local_size;                 \
        if( OPAL_LIKELY(convertor->remoteArch == opal_local_arch) ) {   \
            if( (convertor->flags & (CONVERTOR_WITH_CHECKSUM | OPAL_DATATYPE_FLAG_NO_GAPS)) == OPAL_DATATYPE_FLAG_NO_GAPS ) { \
                return OPAL_SUCCESS;                                    \
            }                                                           \
            if( ((convertor->flags & (CONVERTOR_WITH_CHECKSUM | OPAL_DATATYPE_FLAG_CONTIGUOUS)) \
                 == OPAL_DATATYPE_FLAG_CONTIGUOUS) && (1 == count) ) {             \
                return OPAL_SUCCESS;                                    \
            }                                                           \
        }                                                               \
                                                                        \
        bdt_mask = datatype->bdt_used & convertor->master->hetero_mask; \
        OPAL_CONVERTOR_COMPUTE_REMOTE_SIZE( convertor, datatype,        \
                                            bdt_mask );                 \
        assert( NULL != convertor->use_desc->desc );                    \
        /* For predefined datatypes (contiguous) do nothing more */     \
        /* if checksum is enabled then always continue */               \
        if( ((convertor->flags & (CONVERTOR_WITH_CHECKSUM | OPAL_DATATYPE_FLAG_NO_GAPS)) \
             == OPAL_DATATYPE_FLAG_NO_GAPS) &&                                     \
            (convertor->flags & (CONVERTOR_SEND | CONVERTOR_HOMOGENEOUS)) ) { \
            return OPAL_SUCCESS;                                        \
        }                                                               \
        convertor->flags &= ~CONVERTOR_NO_OP;                           \
        {                                                               \
            uint32_t required_stack_length = datatype->btypes[OPAL_DATATYPE_LOOP] + 1; \
                                                                        \
            if( required_stack_length > convertor->stack_size ) {       \
                convertor->stack_size = required_stack_length;          \
                convertor->pStack     = (dt_stack_t*)malloc(sizeof(dt_stack_t) * \
                                                            convertor->stack_size ); \
            } else {                                                    \
                convertor->pStack = convertor->static_stack;            \
                convertor->stack_size = DT_STATIC_STACK_SIZE;           \
            }                                                           \
        }                                                               \
        opal_convertor_create_stack_at_begining( convertor, opal_datatype_local_sizes ); \
    }


int32_t opal_convertor_prepare_for_recv( opal_convertor_t* convertor,
                                         const struct opal_datatype_t* datatype,
                                         int32_t count,
                                         const void* pUserBuf )
{
    /* Here I should check that the data is not overlapping */

    convertor->flags |= CONVERTOR_RECV;
#if OPAL_CUDA_SUPPORT
    mca_cuda_convertor_init(convertor, pUserBuf);
#endif

    OPAL_CONVERTOR_PREPARE( convertor, datatype, count, pUserBuf );

    if( convertor->flags & CONVERTOR_WITH_CHECKSUM ) {
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
        if( !(convertor->flags & CONVERTOR_HOMOGENEOUS) ) {
            convertor->fAdvance = opal_unpack_general_checksum;
        } else
#endif
        if( convertor->pDesc->flags & OPAL_DATATYPE_FLAG_CONTIGUOUS ) {
            convertor->fAdvance = opal_unpack_homogeneous_contig_checksum;
        } else {
            convertor->fAdvance = opal_generic_simple_unpack_checksum;
        }
    } else {
#if OPAL_ENABLE_HETEROGENEOUS_SUPPORT
        if( !(convertor->flags & CONVERTOR_HOMOGENEOUS) ) {
            convertor->fAdvance = opal_unpack_general;
        } else
#endif
        if( convertor->pDesc->flags & OPAL_DATATYPE_FLAG_CONTIGUOUS ) {
            convertor->fAdvance = opal_unpack_homogeneous_contig;
        } else {
            convertor->fAdvance = opal_generic_simple_unpack;
        }
    }
    return OPAL_SUCCESS;
}


int32_t opal_convertor_prepare_for_send( opal_convertor_t* convertor,
                                         const struct opal_datatype_t* datatype,
                                         int32_t count,
                                         const void* pUserBuf )
{
    convertor->flags |= CONVERTOR_SEND;
#if OPAL_CUDA_SUPPORT
    mca_cuda_convertor_init(convertor, pUserBuf);
#endif

    OPAL_CONVERTOR_PREPARE( convertor, datatype, count, pUserBuf );

    if( convertor->flags & CONVERTOR_WITH_CHECKSUM ) {
        if( datatype->flags & OPAL_DATATYPE_FLAG_CONTIGUOUS ) {
            if( ((datatype->ub - datatype->lb) == (OPAL_PTRDIFF_TYPE)datatype->size)
                || (1 >= convertor->count) )
                convertor->fAdvance = opal_pack_homogeneous_contig_checksum;
            else
                convertor->fAdvance = opal_pack_homogeneous_contig_with_gaps_checksum;
        } else {
            convertor->fAdvance = opal_generic_simple_pack_checksum;
        }
    } else {
        if( datatype->flags & OPAL_DATATYPE_FLAG_CONTIGUOUS ) {
            if( ((datatype->ub - datatype->lb) == (OPAL_PTRDIFF_TYPE)datatype->size)
                || (1 >= convertor->count) )
                convertor->fAdvance = opal_pack_homogeneous_contig;
            else
                convertor->fAdvance = opal_pack_homogeneous_contig_with_gaps;
        } else {
            convertor->fAdvance = opal_generic_simple_pack;
        }
    }
    return OPAL_SUCCESS;
}

/*
 * These functions can be used in order to create an IDENTICAL copy of one convertor. In this
 * context IDENTICAL means that the datatype and count and all other properties of the basic
 * convertor get replicated on this new convertor. However, the references to the datatype
 * are not increased. This function take special care about the stack. If all the cases the
 * stack is created with the correct number of entries but if the copy_stack is true (!= 0)
 * then the content of the old stack is copied on the new one. The result will be a convertor
 * ready to use starting from the old position. If copy_stack is false then the convertor
 * is created with a empty stack (you have to use opal_convertor_set_position before using it).
 */
int opal_convertor_clone( const opal_convertor_t* source,
                          opal_convertor_t* destination,
                          int32_t copy_stack )
{
    destination->remoteArch        = source->remoteArch;
    destination->flags             = source->flags;
    destination->pDesc             = source->pDesc;
    destination->use_desc          = source->use_desc;
    destination->count             = source->count;
    destination->pBaseBuf          = source->pBaseBuf;
    destination->fAdvance          = source->fAdvance;
    destination->master            = source->master;
    destination->local_size        = source->local_size;
    destination->remote_size       = source->remote_size;
    /* create the stack */
    if( OPAL_UNLIKELY(source->stack_size > DT_STATIC_STACK_SIZE) ) {
        destination->pStack = (dt_stack_t*)malloc(sizeof(dt_stack_t) * source->stack_size );
    } else {
        destination->pStack = destination->static_stack;
    }
    destination->stack_size = source->stack_size;

    /* initialize the stack */
    if( OPAL_LIKELY(0 == copy_stack) ) {
        destination->bConverted = -1;
        destination->stack_pos  = -1;
    } else {
        memcpy( destination->pStack, source->pStack, sizeof(dt_stack_t) * (source->stack_pos+1) );
        destination->bConverted = source->bConverted;
        destination->stack_pos  = source->stack_pos;
    }
#if OPAL_CUDA_SUPPORT
    destination->cbmemcpy   = source->cbmemcpy;
#endif
    return OPAL_SUCCESS;
}


void opal_convertor_dump( opal_convertor_t* convertor )
{
    printf( "Convertor %p count %d stack position %d bConverted %ld\n", (void*)convertor,
            convertor->count, convertor->stack_pos, (unsigned long)convertor->bConverted );
    printf( "\tlocal_size %ld remote_size %ld flags %X stack_size %d pending_length %d\n",
            (unsigned long)convertor->local_size, (unsigned long)convertor->remote_size,
            convertor->flags, convertor->stack_size, convertor->partial_length );
    opal_datatype_dump( convertor->pDesc );
    printf( "Actual stack representation\n" );
    opal_datatype_dump_stack( convertor->pStack, convertor->stack_pos,
                              convertor->pDesc->desc.desc, convertor->pDesc->name );
}


void opal_datatype_dump_stack( const dt_stack_t* pStack, int stack_pos,
                               const union dt_elem_desc* pDesc, const char* name )
{
    opal_output( 0, "\nStack %p stack_pos %d name %s\n", (void*)pStack, stack_pos, name );
    for( ; stack_pos >= 0; stack_pos-- ) {
        opal_output( 0, "%d: pos %d count %d disp %ld ", stack_pos, pStack[stack_pos].index,
                     (int)pStack[stack_pos].count, (long)pStack[stack_pos].disp );
        if( pStack->index != -1 )
            opal_output( 0, "\t[desc count %d disp %ld extent %ld]\n",
                         pDesc[pStack[stack_pos].index].elem.count,
                         (long)pDesc[pStack[stack_pos].index].elem.disp,
                         (long)pDesc[pStack[stack_pos].index].elem.extent );
        else
            opal_output( 0, "\n" );
    }
    opal_output( 0, "\n" );
}
Tip: Filter by directory path e.g. /media app.js to search for public/media/app.js.
Tip: Use camelCasing e.g. ProjME to search for ProjectModifiedEvent.java.
Tip: Filter by extension type e.g. /repo .js to search for all .js files in the /repo directory.
Tip: Separate your search with spaces e.g. /ssh pom.xml to search for src/ssh/pom.xml.
Tip: Use ↑ and ↓ arrow keys to navigate and return to view the file.
Tip: You can also navigate files with Ctrl+j (next) and Ctrl+k (previous) and view the file with Ctrl+o.
Tip: You can also navigate files with Alt+j (next) and Alt+k (previous) and view the file with Alt+o.