mpidrma.h 34.7 KB
Newer Older
1
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */
2
3
4
5
6
/*
 *  (C) 2001 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

7
8
#if !defined(MPID_RMA_H_INCLUDED)
#define MPID_RMA_H_INCLUDED
9

10
11
12
#include "mpid_rma_types.h"
#include "mpid_rma_oplist.h"
#include "mpid_rma_shm.h"
13
#include "mpid_rma_issue.h"
14
#include "mpid_rma_lockqueue.h"
15

16
#undef FUNCNAME
17
#define FUNCNAME send_lock_msg
18
19
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
20
static inline int send_lock_msg(int dest, int lock_type, MPID_Win * win_ptr)
21
{
22
23
24
25
26
27
28
    int mpi_errno = MPI_SUCCESS;
    MPIDI_CH3_Pkt_t upkt;
    MPIDI_CH3_Pkt_lock_t *lock_pkt = &upkt.lock;
    MPID_Request *req = NULL;
    MPIDI_VC_t *vc;
    MPIDI_STATE_DECL(MPID_STATE_SEND_LOCK_MSG);
    MPIDI_RMA_FUNC_ENTER(MPID_STATE_SEND_LOCK_MSG);
29

30
    MPIDI_Comm_get_vc_set_active(win_ptr->comm_ptr, dest, &vc);
31

32
33
34
35
    MPIDI_Pkt_init(lock_pkt, MPIDI_CH3_PKT_LOCK);
    lock_pkt->target_win_handle = win_ptr->all_win_handles[dest];
    lock_pkt->source_win_handle = win_ptr->handle;
    lock_pkt->lock_type = lock_type;
36
    lock_pkt->origin_rank = win_ptr->comm_ptr->rank;
37

38
39
40
41
    MPIU_THREAD_CS_ENTER(CH3COMM, vc);
    mpi_errno = MPIDI_CH3_iStartMsg(vc, lock_pkt, sizeof(*lock_pkt), &req);
    MPIU_THREAD_CS_EXIT(CH3COMM, vc);
    MPIU_ERR_CHKANDJUMP(mpi_errno != MPI_SUCCESS, mpi_errno, MPI_ERR_OTHER, "**ch3|rma_msg");
42

43
44
45
46
    /* release the request returned by iStartMsg */
    if (req != NULL) {
        MPID_Request_release(req);
    }
47

48
  fn_exit:
49
    MPIDI_RMA_FUNC_EXIT(MPID_STATE_SEND_LOCK_MSG);
50
    return mpi_errno;
51
    /* --BEGIN ERROR HANDLING-- */
52
  fn_fail:
53
    goto fn_exit;
54
    /* --END ERROR HANDLING-- */
55
56
}

57
#undef FUNCNAME
58
#define FUNCNAME send_unlock_msg
59
60
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
61
62
static inline int send_unlock_msg(int dest, MPID_Win * win_ptr,
                                  MPIDI_CH3_Pkt_flags_t flags)
63
64
{
    int mpi_errno = MPI_SUCCESS;
65
66
67
68
69
70
    MPIDI_CH3_Pkt_t upkt;
    MPIDI_CH3_Pkt_unlock_t *unlock_pkt = &upkt.unlock;
    MPID_Request *req = NULL;
    MPIDI_VC_t *vc;
    MPIDI_STATE_DECL(MPID_STATE_SEND_UNLOCK_MSG);
    MPIDI_RMA_FUNC_ENTER(MPID_STATE_SEND_UNLOCK_MSG);
71

72
    MPIDI_Comm_get_vc_set_active(win_ptr->comm_ptr, dest, &vc);
73

74
75
    /* Send a lock packet over to the target. wait for the lock_granted
     * reply. Then do all the RMA ops. */
76

77
78
    MPIDI_Pkt_init(unlock_pkt, MPIDI_CH3_PKT_UNLOCK);
    unlock_pkt->target_win_handle = win_ptr->all_win_handles[dest];
79
    unlock_pkt->source_win_handle = win_ptr->handle;
80
    unlock_pkt->flags = flags;
81

82
83
84
85
    MPIU_THREAD_CS_ENTER(CH3COMM, vc);
    mpi_errno = MPIDI_CH3_iStartMsg(vc, unlock_pkt, sizeof(*unlock_pkt), &req);
    MPIU_THREAD_CS_EXIT(CH3COMM, vc);
    MPIU_ERR_CHKANDJUMP(mpi_errno != MPI_SUCCESS, mpi_errno, MPI_ERR_OTHER, "**ch3|rma_msg");
86

87
88
89
    /* Release the request returned by iStartMsg */
    if (req != NULL) {
        MPID_Request_release(req);
90
    }
91

92
  fn_exit:
93
    MPIDI_RMA_FUNC_EXIT(MPID_STATE_SEND_UNLOCK_MSG);
94
95
    return mpi_errno;
    /* --BEGIN ERROR HANDLING-- */
96
  fn_fail:
97
98
99
100
    goto fn_exit;
    /* --END ERROR HANDLING-- */
}

101

102
#undef FUNCNAME
103
#define FUNCNAME MPIDI_CH3I_Send_lock_ack_pkt
104
105
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
106
107
108
static inline int MPIDI_CH3I_Send_lock_ack_pkt(MPIDI_VC_t * vc, MPID_Win * win_ptr,
                                               MPIDI_CH3_Pkt_flags_t flags,
                                               MPI_Win source_win_handle)
109
{
110
    MPIDI_CH3_Pkt_t upkt;
111
    MPIDI_CH3_Pkt_lock_ack_t *lock_ack_pkt = &upkt.lock_ack;
112
113
    MPID_Request *req = NULL;
    int mpi_errno;
114
    MPIDI_STATE_DECL(MPID_STATE_MPIDI_CH3I_SEND_LOCK_ACK_PKT);
115

116
    MPIDI_FUNC_ENTER(MPID_STATE_MPIDI_CH3I_SEND_LOCK_ACK_PKT);
117

118
119
120
121
    /* send lock ack packet */
    MPIDI_Pkt_init(lock_ack_pkt, MPIDI_CH3_PKT_LOCK_ACK);
    lock_ack_pkt->source_win_handle = source_win_handle;
    lock_ack_pkt->target_rank = win_ptr->comm_ptr->rank;
122
    lock_ack_pkt->flags = flags;
123
124

    MPIU_DBG_MSG_FMT(CH3_OTHER, VERBOSE,
125
126
                     (MPIU_DBG_FDEST, "sending lock ack pkt on vc=%p, source_win_handle=%#08x",
                      vc, lock_ack_pkt->source_win_handle));
127
128

    MPIU_THREAD_CS_ENTER(CH3COMM, vc);
129
    mpi_errno = MPIDI_CH3_iStartMsg(vc, lock_ack_pkt, sizeof(*lock_ack_pkt), &req);
130
131
132
    MPIU_THREAD_CS_EXIT(CH3COMM, vc);
    if (mpi_errno) {
        MPIU_ERR_SETANDJUMP(mpi_errno, MPI_ERR_OTHER, "**ch3|rmamsg");
133
134
    }

135
136
    if (req != NULL) {
        MPID_Request_release(req);
137
    }
138

139
  fn_fail:
140
    MPIDI_FUNC_EXIT(MPID_STATE_MPIDI_CH3I_SEND_LOCK_ACK_PKT);
141
142

    return mpi_errno;
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
#undef FUNCNAME
#define FUNCNAME MPIDI_CH3I_Send_lock_op_ack_pkt
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
static inline int MPIDI_CH3I_Send_lock_op_ack_pkt(MPIDI_VC_t * vc, MPID_Win * win_ptr,
                                                  MPIDI_CH3_Pkt_flags_t flags,
                                                  MPI_Win source_win_handle)
{
    MPIDI_CH3_Pkt_t upkt;
    MPIDI_CH3_Pkt_lock_op_ack_t *lock_op_ack_pkt = &upkt.lock_op_ack;
    MPID_Request *req = NULL;
    int mpi_errno;
    MPIDI_STATE_DECL(MPID_STATE_MPIDI_CH3I_SEND_LOCK_OP_ACK_PKT);

    MPIDI_FUNC_ENTER(MPID_STATE_MPIDI_CH3I_SEND_LOCK_OP_ACK_PKT);

    /* send lock ack packet */
    MPIDI_Pkt_init(lock_op_ack_pkt, MPIDI_CH3_PKT_LOCK_OP_ACK);
    lock_op_ack_pkt->source_win_handle = source_win_handle;
    lock_op_ack_pkt->target_rank = win_ptr->comm_ptr->rank;
    lock_op_ack_pkt->flags = flags;

    MPIU_DBG_MSG_FMT(CH3_OTHER, VERBOSE,
                     (MPIU_DBG_FDEST, "sending lock op ack pkt on vc=%p, source_win_handle=%#08x",
                      vc, lock_op_ack_pkt->source_win_handle));

    MPIU_THREAD_CS_ENTER(CH3COMM, vc);
    mpi_errno = MPIDI_CH3_iStartMsg(vc, lock_op_ack_pkt, sizeof(*lock_op_ack_pkt), &req);
    MPIU_THREAD_CS_EXIT(CH3COMM, vc);
    if (mpi_errno) {
        MPIU_ERR_SETANDJUMP(mpi_errno, MPI_ERR_OTHER, "**ch3|rmamsg");
    }

    if (req != NULL) {
        MPID_Request_release(req);
    }

  fn_fail:
    MPIDI_FUNC_EXIT(MPID_STATE_MPIDI_CH3I_SEND_LOCK_OP_ACK_PKT);
    return mpi_errno;
}

187
188

#undef FUNCNAME
Xin Zhao's avatar
Xin Zhao committed
189
#define FUNCNAME MPIDI_CH3I_Send_flush_ack_pkt
190
191
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
Xin Zhao's avatar
Xin Zhao committed
192
static inline int MPIDI_CH3I_Send_flush_ack_pkt(MPIDI_VC_t *vc, MPID_Win *win_ptr,
193
                                    MPI_Win source_win_handle)
194
{
195
    MPIDI_CH3_Pkt_t upkt;
Xin Zhao's avatar
Xin Zhao committed
196
    MPIDI_CH3_Pkt_flush_ack_t *flush_ack_pkt = &upkt.flush_ack;
197
198
    MPID_Request *req;
    int mpi_errno=MPI_SUCCESS;
Xin Zhao's avatar
Xin Zhao committed
199
    MPIDI_STATE_DECL(MPID_STATE_MPIDI_CH3I_SEND_FLUSH_ACK_PKT);
200

Xin Zhao's avatar
Xin Zhao committed
201
    MPIDI_FUNC_ENTER(MPID_STATE_MPIDI_CH3I_SEND_FLUSH_ACK_PKT);
202

Xin Zhao's avatar
Xin Zhao committed
203
204
205
    MPIDI_Pkt_init(flush_ack_pkt, MPIDI_CH3_PKT_FLUSH_ACK);
    flush_ack_pkt->source_win_handle = source_win_handle;
    flush_ack_pkt->target_rank = win_ptr->comm_ptr->rank;
206

207
208
    /* Because this is in a packet handler, it is already within a critical section */	
    /* MPIU_THREAD_CS_ENTER(CH3COMM,vc); */
Xin Zhao's avatar
Xin Zhao committed
209
    mpi_errno = MPIDI_CH3_iStartMsg(vc, flush_ack_pkt, sizeof(*flush_ack_pkt), &req);
210
211
212
    /* MPIU_THREAD_CS_EXIT(CH3COMM,vc); */
    if (mpi_errno != MPI_SUCCESS) {
	MPIU_ERR_SETANDJUMP(mpi_errno,MPI_ERR_OTHER,"**ch3|rmamsg");
213
214
    }

215
216
217
    if (req != NULL)
    {
        MPID_Request_release(req);
218
219
    }

220
 fn_fail:
Xin Zhao's avatar
Xin Zhao committed
221
    MPIDI_FUNC_EXIT(MPID_STATE_MPIDI_CH3I_SEND_FLUSH_ACK_PKT);
222
223
    return mpi_errno;
}
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
#undef FUNCNAME
#define FUNCNAME send_decr_at_cnt_msg
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
static inline int send_decr_at_cnt_msg(int dst, MPID_Win * win_ptr)
{
    MPIDI_CH3_Pkt_t upkt;
    MPIDI_CH3_Pkt_decr_at_counter_t *decr_at_cnt_pkt = &upkt.decr_at_cnt;
    MPIDI_VC_t * vc;
    MPID_Request *request = NULL;
    int mpi_errno = MPI_SUCCESS;
    MPIDI_STATE_DECL(MPID_STATE_SEND_DECR_AT_CNT_MSG);
    MPIDI_RMA_FUNC_ENTER(MPID_STATE_SEND_DECR_AT_CNT_MSG);

    MPIDI_Pkt_init(decr_at_cnt_pkt, MPIDI_CH3_PKT_DECR_AT_COUNTER);
    decr_at_cnt_pkt->target_win_handle = win_ptr->all_win_handles[dst];

    MPIDI_Comm_get_vc_set_active(win_ptr->comm_ptr, dst, &vc);

    MPIU_THREAD_CS_ENTER(CH3COMM,vc);
    mpi_errno = MPIDI_CH3_iStartMsg(vc, decr_at_cnt_pkt,
                                    sizeof(*decr_at_cnt_pkt), &request);
    MPIU_THREAD_CS_EXIT(CH3COMM,vc);
    if (mpi_errno != MPI_SUCCESS) {
        MPIU_ERR_SETANDJUMP(mpi_errno,MPI_ERR_OTHER,"**ch3|rmamsg" );
    }

    if (request != NULL) {
        MPID_Request_release(request);
    }

  fn_exit:
    MPIDI_RMA_FUNC_EXIT(MPID_STATE_SEND_DECR_AT_CNT_MSG);
    return mpi_errno;
    /* --BEGIN ERROR HANDLING-- */
  fn_fail:
    goto fn_exit;
    /* --END ERROR HANDLING-- */
}

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
#undef FUNCNAME
#define FUNCNAME send_flush_msg
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
static inline int send_flush_msg(int dest, MPID_Win * win_ptr)
{
    int mpi_errno = MPI_SUCCESS;
    MPIDI_CH3_Pkt_t upkt;
    MPIDI_CH3_Pkt_flush_t *flush_pkt = &upkt.flush;
    MPID_Request *req = NULL;
    MPIDI_VC_t *vc;
    MPIDI_STATE_DECL(MPID_STATE_SEND_FLUSH_MSG);
    MPIDI_RMA_FUNC_ENTER(MPID_STATE_SEND_FLUSH_MSG);

    MPIDI_Comm_get_vc_set_active(win_ptr->comm_ptr, dest, &vc);

    MPIDI_Pkt_init(flush_pkt, MPIDI_CH3_PKT_FLUSH);
    flush_pkt->target_win_handle = win_ptr->all_win_handles[dest];
    flush_pkt->source_win_handle = win_ptr->handle;

    MPIU_THREAD_CS_ENTER(CH3COMM, vc);
    mpi_errno = MPIDI_CH3_iStartMsg(vc, flush_pkt, sizeof(*flush_pkt), &req);
    MPIU_THREAD_CS_EXIT(CH3COMM, vc);
    MPIU_ERR_CHKANDJUMP(mpi_errno != MPI_SUCCESS, mpi_errno, MPI_ERR_OTHER, "**ch3|rma_msg");

    /* Release the request returned by iStartMsg */
    if (req != NULL) {
        MPID_Request_release(req);
    }

  fn_exit:
    MPIDI_RMA_FUNC_EXIT(MPID_STATE_SEND_FLUSH_MSG);
    return mpi_errno;
    /* --BEGIN ERROR HANDLING-- */
  fn_fail:
    goto fn_exit;
    /* --END ERROR HANDLING-- */
}

306
307

/* enqueue an unsatisfied origin in passive target at target side. */
308
309
310
311
static inline int enqueue_lock_origin(MPID_Win *win_ptr, MPIDI_VC_t *vc,
                                      MPIDI_CH3_Pkt_t *pkt,
                                      MPIDI_msg_sz_t *buflen,
                                      MPID_Request **reqp)
312
{
313
    MPIDI_RMA_Lock_entry_t *new_ptr = NULL;
314
315
316
    MPIDI_CH3_Pkt_flags_t flag;
    MPI_Win source_win_handle;
    int lock_discarded = 0, data_discarded = 0;
317
318
    int mpi_errno = MPI_SUCCESS;

319
    (*reqp) = NULL;
320

321
    new_ptr = MPIDI_CH3I_Win_lock_entry_alloc(win_ptr, pkt);
322
323
324
325
326
    if (new_ptr != NULL) {
        MPL_LL_APPEND(win_ptr->lock_queue, win_ptr->lock_queue_tail, new_ptr);
    }
    else {
        lock_discarded = 1;
327
    }
328

329
330
331
332
    if (pkt->type == MPIDI_CH3_PKT_LOCK ||
        pkt->type == MPIDI_CH3_PKT_GET ||
        pkt->type == MPIDI_CH3_PKT_FOP ||
        pkt->type == MPIDI_CH3_PKT_CAS) {
333

334
335
        /* return bytes of data processed in this pkt handler */
        (*buflen) = sizeof(MPIDI_CH3_Pkt_t);
336
337
338
339
340

        if (new_ptr != NULL)
            new_ptr->all_data_recved = 1;

        goto issue_ack;
341
342
343
344
345
346
347
    }
    else {
        MPI_Aint type_size = 0;
        MPIDI_msg_sz_t recv_data_sz = 0;
        MPID_Request *req = NULL;
        MPI_Datatype target_dtp;
        int target_count;
348
        int immed_len = 0;
349
350
351
352
353
354
355
356
357
358
359
360
361
362
        void *immed_data = NULL;
        int complete = 0;
        MPIDI_msg_sz_t data_len;
        char *data_buf = NULL;

        /* This is PUT, ACC, GACC */

        MPIDI_CH3_PKT_RMA_GET_TARGET_DATATYPE((*pkt), target_dtp, mpi_errno);
        MPIDI_CH3_PKT_RMA_GET_TARGET_COUNT((*pkt), target_count, mpi_errno);

        MPID_Datatype_get_size_macro(target_dtp, type_size);
        recv_data_sz = type_size * target_count;

        if (recv_data_sz <= MPIDI_RMA_IMMED_BYTES) {
363

364
365
            /* return bytes of data processed in this pkt handler */
            (*buflen) = sizeof(MPIDI_CH3_Pkt_t);
366
367
368
369
370

            if (new_ptr != NULL)
                new_ptr->all_data_recved = 1;

            goto issue_ack;
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
        if (new_ptr != NULL) {
            if (win_ptr->current_lock_data_bytes + recv_data_sz
                < MPIR_CVAR_CH3_RMA_LOCK_DATA_BYTES) {
                new_ptr->data = MPIU_Malloc(recv_data_sz);
            }

            if (new_ptr->data == NULL) {
                /* Note that there are two possible reasons to make new_ptr->data to be NULL:
                 * (1) win_ptr->current_lock_data_bytes + recv_data_sz >= MPIR_CVAR_CH3_RMA_LOCK_DATA_BYTES;
                 * (2) MPIU_Malloc(recv_data_sz) failed.
                 * In such cases, we cannot allocate memory for lock data, so we give up
                 * buffering lock data, however, we still buffer lock request.
                 */
                MPIDI_CH3_Pkt_t new_pkt;
                MPIDI_CH3_Pkt_lock_t *lock_pkt = &new_pkt.lock;
                MPI_Win target_win_handle;
                int lock_type, origin_rank;

                MPIDI_CH3_PKT_RMA_GET_TARGET_WIN_HANDLE((*pkt), target_win_handle, mpi_errno);
                MPIDI_CH3_PKT_RMA_GET_SOURCE_WIN_HANDLE((*pkt), source_win_handle, mpi_errno);
                MPIDI_CH3_PKT_RMA_GET_ORIGIN_RANK((*pkt), origin_rank, mpi_errno);
                MPIDI_CH3_PKT_RMA_GET_LOCK_TYPE((*pkt), lock_type, mpi_errno);

                MPIDI_Pkt_init(lock_pkt, MPIDI_CH3_PKT_LOCK);
                lock_pkt->target_win_handle = target_win_handle;
                lock_pkt->source_win_handle = source_win_handle;
                lock_pkt->lock_type = lock_type;
                lock_pkt->origin_rank = origin_rank;

                /* replace original pkt with lock pkt */
                new_ptr->pkt = new_pkt;
                new_ptr->all_data_recved = 1;

                data_discarded = 1;
            }
            else {
                win_ptr->current_lock_data_bytes += recv_data_sz;
                new_ptr->data_size = recv_data_sz;
            }
412
413
414
415
416
417
418
        }

        /* create request to receive upcoming requests */
        req = MPID_Request_create();
        MPIU_Object_set_ref(req, 1);

        /* fill in area in req that will be used in Receive_data_found() */
419
        if (lock_discarded || data_discarded) {
420
            req->dev.drop_data = TRUE;
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
            req->dev.user_buf = NULL;
            req->dev.user_count = target_count;
            req->dev.datatype = target_dtp;
            req->dev.recv_data_sz = recv_data_sz;
            req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_PiggybackLockOpRecvComplete;
            req->dev.OnFinal = MPIDI_CH3_ReqHandler_PiggybackLockOpRecvComplete;

            MPIDI_CH3_PKT_RMA_GET_IMMED_LEN((*pkt), immed_len, mpi_errno);
            if (immed_len > 0) {
                req->dev.recv_data_sz -= immed_len;
            }
            data_len = *buflen - sizeof(MPIDI_CH3_Pkt_t);
            data_buf = (char *) pkt + sizeof(MPIDI_CH3_Pkt_t);
            MPIU_Assert(req->dev.recv_data_sz > 0);
        }
        else {
            req->dev.user_buf = new_ptr->data;
            req->dev.user_count = target_count;
            req->dev.datatype = target_dtp;
            req->dev.recv_data_sz = recv_data_sz;
            req->dev.OnDataAvail = MPIDI_CH3_ReqHandler_PiggybackLockOpRecvComplete;
            req->dev.OnFinal = MPIDI_CH3_ReqHandler_PiggybackLockOpRecvComplete;
            req->dev.lock_queue_entry = new_ptr;

            MPIDI_CH3_PKT_RMA_GET_IMMED_LEN((*pkt), immed_len, mpi_errno);
            MPIDI_CH3_PKT_RMA_GET_IMMED_DATA_PTR((*pkt), immed_data, mpi_errno);

            if (immed_len > 0) {
                /* see if we can receive some data from packet header */
                MPIU_Memcpy(req->dev.user_buf, immed_data, (size_t)immed_len);
                req->dev.user_buf = (void*)((char*)req->dev.user_buf + immed_len);
                req->dev.recv_data_sz -= immed_len;
            }
            data_len = *buflen - sizeof(MPIDI_CH3_Pkt_t);
            data_buf = (char *) pkt + sizeof(MPIDI_CH3_Pkt_t);
            MPIU_Assert(req->dev.recv_data_sz > 0);
457
458
459
460
461
462
463
464
465
466
467
468
        }

        mpi_errno = MPIDI_CH3U_Receive_data_found(req, data_buf, &data_len, &complete);
        if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);

        /* return bytes of data processed in this pkt handler */
        (*buflen) = sizeof(MPIDI_CH3_Pkt_t) + data_len;

        if (complete) {
            mpi_errno = MPIDI_CH3_ReqHandler_PiggybackLockOpRecvComplete(vc, req, &complete);
            if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
            if (complete) {
469
                goto issue_ack;
470
471
472
473
474
475
            }
        }

        (*reqp) = req;
    }

476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
 issue_ack:
    MPIDI_CH3_PKT_RMA_GET_SOURCE_WIN_HANDLE((*pkt), source_win_handle, mpi_errno);
    if (pkt->type == MPIDI_CH3_PKT_LOCK) {
        if (lock_discarded) flag = MPIDI_CH3_PKT_FLAG_RMA_LOCK_DISCARDED;
        else flag = MPIDI_CH3_PKT_FLAG_RMA_LOCK_QUEUED_DATA_QUEUED;

        mpi_errno = MPIDI_CH3I_Send_lock_ack_pkt(vc, win_ptr, flag, source_win_handle);
        if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
    }
    else {
        if (lock_discarded) flag = MPIDI_CH3_PKT_FLAG_RMA_LOCK_DISCARDED;
        else if (data_discarded) flag = MPIDI_CH3_PKT_FLAG_RMA_LOCK_QUEUED_DATA_DISCARDED;
        else flag = MPIDI_CH3_PKT_FLAG_RMA_LOCK_QUEUED_DATA_QUEUED;

        mpi_errno = MPIDI_CH3I_Send_lock_op_ack_pkt(vc, win_ptr, flag, source_win_handle);
        if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
    }

494
495
496
497
498
499
500
 fn_exit:
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}


501
static inline int handle_lock_ack(MPID_Win *win_ptr, int target_rank,
502
                                        MPIDI_CH3_Pkt_flags_t flags)
503
{
504
    MPIDI_RMA_Target_t *t = NULL;
505
506
    int mpi_errno = MPI_SUCCESS;

507
508
509
510
511
512
513
514
515
516
    MPIU_Assert(win_ptr->states.access_state == MPIDI_RMA_PER_TARGET ||
                win_ptr->states.access_state == MPIDI_RMA_LOCK_ALL_CALLED ||
                win_ptr->states.access_state == MPIDI_RMA_LOCK_ALL_ISSUED);

    if (win_ptr->states.access_state == MPIDI_RMA_LOCK_ALL_CALLED) {
        MPIDI_VC_t *orig_vc = NULL, *target_vc = NULL;
        MPIDI_Comm_get_vc(win_ptr->comm_ptr, win_ptr->comm_ptr->rank, &orig_vc);
        MPIDI_Comm_get_vc(win_ptr->comm_ptr, target_rank, &target_vc);
        if (win_ptr->comm_ptr->rank == target_rank ||
            (win_ptr->shm_allocated == TRUE && orig_vc->node_id == target_vc->node_id)) {
517
518
519
520
521
522
523
524
525
            if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_GRANTED) {
                win_ptr->outstanding_locks--;
                MPIU_Assert(win_ptr->outstanding_locks >= 0);
            }
            else if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_DISCARDED) {
                /* re-send lock request message. */
                mpi_errno = send_lock_msg(target_rank, MPI_LOCK_SHARED, win_ptr);
                if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
            }
526
527
528
529
            goto fn_exit;
        }
    }
    else if (win_ptr->states.access_state == MPIDI_RMA_LOCK_ALL_ISSUED) {
530
531
532
533
534
535
536
537
538
        if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_GRANTED) {
            win_ptr->outstanding_locks--;
            MPIU_Assert(win_ptr->outstanding_locks >= 0);
        }
        else if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_DISCARDED) {
            /* re-send lock request message. */
            mpi_errno = send_lock_msg(target_rank, MPI_LOCK_SHARED, win_ptr);
            if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
        }
539
        goto fn_exit;
540
541
    }

542
543
544
    mpi_errno = MPIDI_CH3I_Win_find_target(win_ptr, target_rank, &t);
    if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
    MPIU_Assert(t != NULL);
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

    if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_GRANTED)
        t->access_state = MPIDI_RMA_LOCK_GRANTED;

    if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_DISCARDED)
        t->access_state = MPIDI_RMA_LOCK_CALLED;

 fn_exit:
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}

static inline int adjust_op_piggybacked_with_lock (MPID_Win *win_ptr,
                                                   int target_rank,
                                                   MPIDI_CH3_Pkt_flags_t flags) {
    MPIDI_RMA_Target_t *target = NULL;
    MPIDI_RMA_Op_t *op = NULL;
    MPIDI_CH3_Pkt_flags_t op_flags = MPIDI_CH3_PKT_FLAG_NONE;
    int mpi_errno = MPI_SUCCESS;

    mpi_errno = MPIDI_CH3I_Win_find_target(win_ptr, target_rank, &target);
    if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
    MPIU_Assert(target != NULL);

    op = target->pending_op_list;
    if (op != NULL) MPIDI_CH3_PKT_RMA_GET_FLAGS(op->pkt, op_flags, mpi_errno);

    if (op_flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK) {
        if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_GRANTED ||
            flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_QUEUED_DATA_QUEUED) {
            if (!op->request) {
                if (op->ureq) {
                    /* Complete user request and release the ch3 ref */
                    MPID_Request_set_completed(op->ureq);
                    MPID_Request_release(op->ureq);
                }

                MPIDI_CH3I_RMA_Ops_free_elem(win_ptr, &(target->pending_op_list),
                                             &(target->pending_op_list_tail), op);
            }
            else {
                MPIDI_CH3I_RMA_Ops_unlink(&(target->pending_op_list),
                                          &(target->pending_op_list_tail), op);
                if (op->is_dt) {
                    MPIDI_CH3I_RMA_Ops_append(&(target->dt_op_list),
                                              &(target->dt_op_list_tail), op);
                }
                else if (op->pkt.type == MPIDI_CH3_PKT_PUT ||
                         op->pkt.type == MPIDI_CH3_PKT_ACCUMULATE) {
                    MPIDI_CH3I_RMA_Ops_append(&(target->write_op_list),
                                              &(target->write_op_list_tail), op);
                }
                else {
                    MPIDI_CH3I_RMA_Ops_append(&(target->read_op_list),
                                              &(target->read_op_list_tail), op);
                }

                if (op->ureq) {
                    if (MPID_Request_is_complete(op->request)) {
                        /* Complete user request, let cleanup function to release
                           ch3 ref */
                        MPID_Request_set_completed(op->ureq);
                    }
                    else {
                        /* Increase ref for completion handler */
                        MPIU_Object_add_ref(op->ureq);
                        op->request->dev.request_handle = op->ureq->handle;
                        if (op->request->dev.OnDataAvail == NULL) {
                            op->request->dev.OnDataAvail = MPIDI_CH3_ReqHandler_ReqOpsComplete;
                        }
                        op->request->dev.OnFinal = MPIDI_CH3_ReqHandler_ReqOpsComplete;
                    }
                }
            }
        }
        else if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_QUEUED_DATA_DISCARDED ||
                 flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK_DISCARDED) {
            /* We need to re-transmit this operation, so we destroy
               the internal request and erase all flags in current
               operation. */
            if (op->request) {
                MPIDI_CH3_Request_destroy(op->request);
                op->request = NULL;
                win_ptr->active_req_cnt--;
            }
            MPIDI_CH3_PKT_RMA_ERASE_FLAGS(op->pkt, mpi_errno);

            target->next_op_to_issue = op;
        }
    }
636
637
638
639
640
641
642
643

 fn_exit:
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}


644
645
646
647
#undef FUNCNAME
#define FUNCNAME acquire_local_lock
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
648
static inline int acquire_local_lock(MPID_Win * win_ptr, int lock_type)
649
650
651
652
653
{
    int mpi_errno = MPI_SUCCESS;
    MPIDI_STATE_DECL(MPID_STATE_ACQUIRE_LOCAL_LOCK);
    MPIDI_RMA_FUNC_ENTER(MPID_STATE_ACQUIRE_LOCAL_LOCK);

Xin Zhao's avatar
Xin Zhao committed
654
655
    MPIR_T_PVAR_TIMER_START(RMA, rma_winlock_getlocallock);

656
    if (MPIDI_CH3I_Try_acquire_win_lock(win_ptr, lock_type) == 1) {
657
        mpi_errno = handle_lock_ack(win_ptr, win_ptr->comm_ptr->rank,
658
                                          MPIDI_CH3_PKT_FLAG_RMA_LOCK_GRANTED);
659
660
661
662
663
664
        if (mpi_errno) MPIU_ERR_POP(mpi_errno);
    }
    else {
        /* Queue the lock information. */
        MPIDI_CH3_Pkt_t pkt;
        MPIDI_CH3_Pkt_lock_t *lock_pkt = &pkt.lock;
665
        MPIDI_RMA_Lock_entry_t *new_ptr = NULL;
666
667
668
669
670

        MPIDI_Pkt_init(lock_pkt, MPIDI_CH3_PKT_LOCK);
        lock_pkt->lock_type = lock_type;
        lock_pkt->origin_rank = win_ptr->comm_ptr->rank;

671
672
        new_ptr = MPIDI_CH3I_Win_lock_entry_alloc(win_ptr, &pkt);
        if (new_ptr == NULL) {
673
            mpi_errno = handle_lock_ack(win_ptr, win_ptr->comm_ptr->rank,
674
675
676
                                              MPIDI_CH3_PKT_FLAG_RMA_LOCK_DISCARDED);
            if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
            goto fn_exit;
677
        }
678
679
680
        MPL_LL_APPEND(win_ptr->lock_queue, win_ptr->lock_queue_tail, new_ptr);

        new_ptr->all_data_recved = 1;
681
682
    }

683
  fn_exit:
Xin Zhao's avatar
Xin Zhao committed
684
    MPIR_T_PVAR_TIMER_END(RMA, rma_winlock_getlocallock);
685
    MPIDI_RMA_FUNC_EXIT(MPID_STATE_ACQUIRE_LOCAL_LOCK);
686
687
    return mpi_errno;
    /* --BEGIN ERROR HANDLING-- */
688
  fn_fail:
689
690
691
692
693
    goto fn_exit;
    /* --END ERROR HANDLING-- */
}


694
695
696
697
698
699
700
701
702
703
704
705
#undef FUNCNAME
#define FUNCNAME MPIDI_CH3I_RMA_Handle_flush_ack
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
static inline int MPIDI_CH3I_RMA_Handle_flush_ack(MPID_Win * win_ptr, int target_rank)
{
    int mpi_errno = MPI_SUCCESS;
    MPIDI_RMA_Target_t *t;

    mpi_errno = MPIDI_CH3I_Win_find_target(win_ptr, target_rank, &t);
    if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);

706
707
708
709
    t->sync.outstanding_acks--;
    MPIU_Assert(t->sync.outstanding_acks >= 0);

    t->put_acc_issued = 0; /* reset PUT_ACC_FLAG after FLUSH is completed */
710
711
712
713
714
715
716
717

 fn_exit:
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}


718
#undef FUNCNAME
719
#define FUNCNAME do_accumulate_op
720
721
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
722
723
static inline int do_accumulate_op(void *source_buf, void *target_buf,
                                   int acc_count, MPI_Datatype acc_dtp, MPI_Op acc_op)
724
725
{
    int mpi_errno = MPI_SUCCESS;
726
727
728
729
730
    MPI_User_function *uop;
    MPIDI_STATE_DECL(MPID_STATE_DO_ACCUMULATE_OP);

    MPIDI_FUNC_ENTER(MPID_STATE_DO_ACCUMULATE_OP);

731
    if (acc_op == MPI_REPLACE)
732
733
    {
        /* simply copy the data */
734
735
        mpi_errno = MPIR_Localcopy(source_buf, acc_count, acc_dtp,
                                   target_buf, acc_count, acc_dtp);
736
        if (mpi_errno) {
737
738
	    MPIU_ERR_POP(mpi_errno);
	}
739
740
741
        goto fn_exit;
    }

742
    if (HANDLE_GET_KIND(acc_op) == HANDLE_KIND_BUILTIN)
743
744
    {
        /* get the function by indexing into the op table */
745
        uop = MPIR_OP_HDL_TO_FN(acc_op);
746
747
748
749
    }
    else
    {
	/* --BEGIN ERROR HANDLING-- */
750
        mpi_errno = MPIR_Err_create_code( MPI_SUCCESS, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OP, "**opnotpredefined", "**opnotpredefined %d", acc_op );
751
752
        return mpi_errno;
	/* --END ERROR HANDLING-- */
753
754
    }

755
    if (MPIR_DATATYPE_IS_PREDEFINED(acc_dtp))
756
    {
757
        (*uop)(source_buf, target_buf, &acc_count, &acc_dtp);
758
759
760
761
    }
    else
    {
	/* derived datatype */
762
763
764
        MPID_Segment *segp;
        DLOOP_VECTOR *dloop_vec;
        MPI_Aint first, last;
765
766
        int vec_len, i, count;
        MPI_Aint type_size;
767
        MPI_Datatype type;
768
769
770
771
772
773
774
775
776
        MPID_Datatype *dtp;

        segp = MPID_Segment_alloc();
	/* --BEGIN ERROR HANDLING-- */
        if (!segp)
	{
            mpi_errno = MPIR_Err_create_code( MPI_SUCCESS, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**nomem", 0 );
	    MPIDI_FUNC_EXIT(MPID_STATE_DO_ACCUMULATE_OP);
            return mpi_errno;
777
        }
778
	/* --END ERROR HANDLING-- */
779
780
        MPID_Segment_init(NULL, acc_count,
			  acc_dtp, segp, 0);
781
782
783
        first = 0;
        last  = SEGMENT_IGNORE_LAST;

784
785
        MPID_Datatype_get_ptr(acc_dtp, dtp);
        vec_len = dtp->max_contig_blocks * acc_count + 1;
786
787
788
789
790
791
792
793
794
        /* +1 needed because Rob says so */
        dloop_vec = (DLOOP_VECTOR *)
            MPIU_Malloc(vec_len * sizeof(DLOOP_VECTOR));
	/* --BEGIN ERROR HANDLING-- */
        if (!dloop_vec)
	{
            mpi_errno = MPIR_Err_create_code( MPI_SUCCESS, MPIR_ERR_RECOVERABLE, FCNAME, __LINE__, MPI_ERR_OTHER, "**nomem", 0 );
	    MPIDI_FUNC_EXIT(MPID_STATE_DO_ACCUMULATE_OP);
            return mpi_errno;
795
        }
796
797
798
799
800
801
802
803
804
	/* --END ERROR HANDLING-- */

        MPID_Segment_pack_vector(segp, first, &last, dloop_vec, &vec_len);

        type = dtp->eltype;
        MPID_Datatype_get_size_macro(type, type_size);
        for (i=0; i<vec_len; i++)
	{
            MPIU_Assign_trunc(count, (dloop_vec[i].DLOOP_VECTOR_LEN)/type_size, int);
805
806
            (*uop)((char *)source_buf + MPIU_PtrToAint(dloop_vec[i].DLOOP_VECTOR_BUF),
                   (char *)target_buf + MPIU_PtrToAint(dloop_vec[i].DLOOP_VECTOR_BUF),
807
                   &count, &type);
808
809
        }

810
811
        MPID_Segment_free(segp);
        MPIU_Free(dloop_vec);
812
813
    }

814
815
 fn_exit:
    MPIDI_FUNC_EXIT(MPID_STATE_DO_ACCUMULATE_OP);
816
817

    return mpi_errno;
818
 fn_fail:
819
820
821
    goto fn_exit;
}

822

823
824
825
826
827
static inline int check_piggyback_lock(MPID_Win *win_ptr, MPIDI_VC_t *vc,
                                       MPIDI_CH3_Pkt_t *pkt,
                                       MPIDI_msg_sz_t *buflen,
                                       int *acquire_lock_fail,
                                       MPID_Request **reqp) {
828
829
830
831
832
    int lock_type;
    MPIDI_CH3_Pkt_flags_t flags;
    int mpi_errno = MPI_SUCCESS;

    (*acquire_lock_fail) = 0;
833
    (*reqp) = NULL;
834
835
836
837
838
839
840

    MPIDI_CH3_PKT_RMA_GET_FLAGS((*pkt), flags, mpi_errno);
    MPIDI_CH3_PKT_RMA_GET_LOCK_TYPE((*pkt), lock_type, mpi_errno);

    if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK) {
        if (MPIDI_CH3I_Try_acquire_win_lock(win_ptr, lock_type) == 0) {
            /* cannot acquire the lock, queue up this operation. */
841
            mpi_errno = enqueue_lock_origin(win_ptr, vc, pkt, buflen, reqp);
842
843
844
845
846
847
848
849
850
851
852
            if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
            (*acquire_lock_fail) = 1;
        }
    }

 fn_exit:
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}

853
854
855
856
857
858
859
860
static inline int finish_op_on_target(MPID_Win *win_ptr, MPIDI_VC_t *vc,
                                      MPIDI_CH3_Pkt_type_t type,
                                      MPIDI_CH3_Pkt_flags_t flags,
                                      MPI_Win source_win_handle) {
    int mpi_errno = MPI_SUCCESS;

    if (type == MPIDI_CH3_PKT_PUT || type == MPIDI_CH3_PKT_ACCUMULATE) {
        /* This is PUT or ACC */
861
        if (flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK) {
862
            MPIDI_CH3_Pkt_flags_t pkt_flags = MPIDI_CH3_PKT_FLAG_RMA_LOCK_GRANTED;
863
864
            if ((flags & MPIDI_CH3_PKT_FLAG_RMA_FLUSH) ||
                (flags & MPIDI_CH3_PKT_FLAG_RMA_UNLOCK))
865
                pkt_flags |= MPIDI_CH3_PKT_FLAG_RMA_FLUSH_ACK;
866
867
868
            mpi_errno = MPIDI_CH3I_Send_lock_op_ack_pkt(vc, win_ptr,
                                                        pkt_flags,
                                                        source_win_handle);
869
870
            if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);
            MPIDI_CH3_Progress_signal_completion();
871
872
        }
        if (flags & MPIDI_CH3_PKT_FLAG_RMA_FLUSH) {
873
874
875
876
877
878
879
            if (!(flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK)) {
                /* If op is piggybacked with both LOCK and FLUSH,
                   we only send LOCK ACK back, do not send FLUSH ACK. */
                mpi_errno = MPIDI_CH3I_Send_flush_ack_pkt(vc, win_ptr,
                                                          source_win_handle);
                if (mpi_errno) MPIU_ERR_POP(mpi_errno);
            }
880
881
882
883
884
885
886
887
888
889
            MPIDI_CH3_Progress_signal_completion();
        }
        if (flags & MPIDI_CH3_PKT_FLAG_RMA_DECR_AT_COUNTER) {
            win_ptr->at_completion_counter--;
            MPIU_Assert(win_ptr->at_completion_counter >= 0);
            /* Signal the local process when the op counter reaches 0. */
            if (win_ptr->at_completion_counter == 0)
                MPIDI_CH3_Progress_signal_completion();
        }
        if (flags & MPIDI_CH3_PKT_FLAG_RMA_UNLOCK) {
890
891
892
893
894
895
896
            if (!(flags & MPIDI_CH3_PKT_FLAG_RMA_LOCK)) {
                /* If op is piggybacked with both LOCK and UNLOCK,
                   we only send LOCK ACK back, do not send FLUSH (UNLOCK) ACK. */
                mpi_errno = MPIDI_CH3I_Send_flush_ack_pkt(vc, win_ptr,
                                                          source_win_handle);
                if (mpi_errno) MPIU_ERR_POP(mpi_errno);
            }
897
898
            mpi_errno = MPIDI_CH3I_Release_lock(win_ptr);
            if (mpi_errno) MPIU_ERR_POP(mpi_errno);
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
            MPIDI_CH3_Progress_signal_completion();
        }
    }
    else {
        /* This is GACC / GET / CAS / FOP */

        if (flags & MPIDI_CH3_PKT_FLAG_RMA_UNLOCK) {
            mpi_errno = MPIDI_CH3I_Release_lock(win_ptr);
            if (mpi_errno) MPIU_ERR_POP(mpi_errno);
            MPIDI_CH3_Progress_signal_completion();
        }

        if (flags & MPIDI_CH3_PKT_FLAG_RMA_DECR_AT_COUNTER) {
            win_ptr->at_completion_counter--;
            MPIU_Assert(win_ptr->at_completion_counter >= 0);
            /* Signal the local process when the op counter reaches 0. */
            if (win_ptr->at_completion_counter == 0)
                MPIDI_CH3_Progress_signal_completion();
        }
    }

 fn_exit:
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}
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

static inline int fill_ranks_in_win_grp(MPID_Win *win_ptr, MPID_Group *group_ptr,
                                        int *ranks_in_win_grp)
{
    int mpi_errno = MPI_SUCCESS;
    int i, *ranks_in_grp;
    MPID_Group *win_grp_ptr;
    MPIU_CHKLMEM_DECL(1);
    MPIDI_STATE_DECL(MPID_STATE_FILL_RANKS_IN_WIN_GRP);

    MPIDI_RMA_FUNC_ENTER(MPID_STATE_FILL_RANKS_IN_WIN_GRP);

    MPIU_CHKLMEM_MALLOC(ranks_in_grp, int *, group_ptr->size * sizeof(int),
                        mpi_errno, "ranks_in_grp");
    for (i = 0; i < group_ptr->size; i++) ranks_in_grp[i] = i;

    mpi_errno = MPIR_Comm_group_impl(win_ptr->comm_ptr, &win_grp_ptr);
    if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);

    mpi_errno = MPIR_Group_translate_ranks_impl(group_ptr, group_ptr->size,
                                                ranks_in_grp, win_grp_ptr, ranks_in_win_grp);
    if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);

    mpi_errno = MPIR_Group_free_impl(win_grp_ptr);
    if (mpi_errno != MPI_SUCCESS) MPIU_ERR_POP(mpi_errno);

  fn_exit:
    MPIU_CHKLMEM_FREEALL();
    MPIDI_RMA_FUNC_EXIT(MPID_STATE_FILL_RANKS_IN_WIN_GRP);
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}


961
static inline int wait_progress_engine(void)
962
963
{
    int mpi_errno = MPI_SUCCESS;
964
    MPID_Progress_state progress_state;
965

966
967
968
969
970
971
    MPID_Progress_start(&progress_state);
    mpi_errno = MPID_Progress_wait(&progress_state);
    /* --BEGIN ERROR HANDLING-- */
    if (mpi_errno != MPI_SUCCESS) {
        MPID_Progress_end(&progress_state);
        MPIU_ERR_SETANDJUMP(mpi_errno, MPI_ERR_OTHER, "**winnoprogress");
972
    }
973
974
    /* --END ERROR HANDLING-- */
    MPID_Progress_end(&progress_state);
975

976
  fn_exit:
977
    return mpi_errno;
978
  fn_fail:
979
980
981
    goto fn_exit;
}

982
static inline int poke_progress_engine(void)
983
984
{
    int mpi_errno = MPI_SUCCESS;
985
    MPID_Progress_state progress_state;
986

987
988
989
990
991
    MPID_Progress_start(&progress_state);
    mpi_errno = MPID_Progress_poke();
    if (mpi_errno != MPI_SUCCESS)
        MPIU_ERR_POP(mpi_errno);
    MPID_Progress_end(&progress_state);
992

993
  fn_exit:
994
    return mpi_errno;
995
  fn_fail:
996
997
998
    goto fn_exit;
}

999
#endif /* MPID_RMA_H_INCLUDED */