mpid_nem_inline.h 29.7 KB
Newer Older
1
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */
2
3
4
5
6
7
8
9
/*
 *  (C) 2006 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

#ifndef _MPID_NEM_INLINE_H
#define _MPID_NEM_INLINE_H

10
#define MPID_NEM_THREAD_POLLS_BEFORE_YIELD 10
11
12

#include "my_papi_defs.h"
13
#include "mpiiov.h"
14

15
extern int MPID_nem_lmt_shm_pending;
16
17
extern MPID_nem_cell_ptr_t MPID_nem_prefetched_cell;

18
static inline int MPID_nem_mpich2_send_header (void* buf, int size, MPIDI_VC_t *vc, int *again);
19
static inline int MPID_nem_mpich2_sendv (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *again);
20
21
static inline void MPID_nem_mpich2_dequeue_fastbox (int local_rank);
static inline void MPID_nem_mpich2_enqueue_fastbox (int local_rank);
22
static inline int MPID_nem_mpich2_sendv_header (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *again);
23
static inline int MPID_nem_recv_seqno_matches (MPID_nem_queue_ptr_t qhead);
24
static inline int MPID_nem_mpich2_test_recv (MPID_nem_cell_ptr_t *cell, int *in_fbox, int in_blocking_progress);
25
static inline int MPID_nem_mpich2_blocking_recv (MPID_nem_cell_ptr_t *cell, int *in_fbox, int completions);
26
27
28
static inline int MPID_nem_mpich2_test_recv_wait (MPID_nem_cell_ptr_t *cell, int *in_fbox, int timeout);
static inline int MPID_nem_mpich2_release_cell (MPID_nem_cell_ptr_t cell, MPIDI_VC_t *vc);
static inline void MPID_nem_mpich2_send_seg_header (MPID_Segment *segment, MPIDI_msg_sz_t *segment_first,
29
30
                                                           MPIDI_msg_sz_t segment_size, void *header, MPIDI_msg_sz_t header_sz,
                                                           MPIDI_VC_t *vc, int *again);
31
static inline void MPID_nem_mpich2_send_seg (MPID_Segment *segment, MPIDI_msg_sz_t *segment_first, MPIDI_msg_sz_t segment_size,
32
33
                                                    MPIDI_VC_t *vc, int *again);

34
35
/* evaluates to TRUE if it is safe to block on recv operations in the progress
 * loop, FALSE otherwise */
36
37
38
39
#define MPID_nem_safe_to_block_recv()           \
    (!MPID_nem_local_lmt_pending &&             \
     !MPIDI_CH3I_shm_active_send &&             \
     !MPIDI_CH3I_Sendq_head(MPIDI_CH3I_shm_sendq) &&       \
40
     !MPIDU_Sched_are_pending())
41
42
43
44
45

#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_send_header
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
46
static inline int
47
48
49
50
51
MPID_nem_mpich2_send_header (void* buf, int size, MPIDI_VC_t *vc, int *again)
{
    int mpi_errno = MPI_SUCCESS;
    MPID_nem_cell_ptr_t el;
    int my_rank;
52
    MPIDI_CH3I_VC *vc_ch = VC_CH(vc);
53
54
55
56

    /*DO_PAPI (PAPI_reset (PAPI_EventSet)); */

    MPIU_Assert (size == sizeof(MPIDI_CH3_Pkt_t));
57
    MPIU_Assert (vc_ch->is_local);
58
59
60
61
62
63
64

    my_rank = MPID_nem_mem_region.rank;

#ifdef USE_FASTBOX
    {
	MPID_nem_fbox_mpich2_t *pbox = vc_ch->fbox_out;

65
66
67
68
        /* _is_full contains acquire barrier */
        if (MPID_nem_fbox_is_full((MPID_nem_fbox_common_ptr_t)pbox))
            goto usequeue_l;

69
70
71
72
73
74
75
        pbox->cell.pkt.mpich2.source  = MPID_nem_mem_region.local_rank;
        pbox->cell.pkt.mpich2.datalen = size;
        pbox->cell.pkt.mpich2.seqno   = vc_ch->send_seqno++;
        
        MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, pbox->cell.pkt.mpich2.type = MPID_NEM_PKT_MPICH2_HEAD);
        
        MPIU_Memcpy((void *)pbox->cell.pkt.mpich2.payload, buf, size);
76
77
78

        OPA_store_release_int(&pbox->flag.value, 1);

79
80
81
82
        MPIU_DBG_MSG (CH3_CHANNEL, VERBOSE, "--> Sent fbox ");
        MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (&pbox->cell));
        
        goto return_success;
83
84
85
    }
 usequeue_l:
#endif /*USE_FASTBOX */
86

87
88
89
#ifdef PREFETCH_CELL
    DO_PAPI (PAPI_reset (PAPI_EventSet));
    el = MPID_nem_prefetched_cell;
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
    if (!el)
    {
	if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	    goto return_again;
	
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &el);
    }
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues14));
#else /* PREFETCH_CELL */
    DO_PAPI (PAPI_reset (PAPI_EventSet));
    if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
    {
	goto return_again;
    }
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues14));

    DO_PAPI (PAPI_reset (PAPI_EventSet));
    MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ , &el);
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues10));
#endif /* PREFETCH_CELL */

    DO_PAPI (PAPI_reset (PAPI_EventSet));
    el->pkt.mpich2.source  = my_rank;
    el->pkt.mpich2.dest    = vc->lpid;
    el->pkt.mpich2.datalen = size;
    el->pkt.mpich2.seqno   = vc_ch->send_seqno++;
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, el->pkt.mpich2.type = MPID_NEM_PKT_MPICH2_HEAD);
    
119
    MPIU_Memcpy((void *)el->pkt.mpich2.payload, buf, size);
120
121
122
123
124
125
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues11));

    MPIU_DBG_MSG (CH3_CHANNEL, VERBOSE, "--> Sent queue");
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (el));

    DO_PAPI (PAPI_reset (PAPI_EventSet));
126
127
    MPID_nem_queue_enqueue (vc_ch->recv_queue, el);
    /*MPID_nem_rel_dump_queue( vc_ch->recv_queue ); */
128
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues12));
129
    DO_PAPI (PAPI_reset (PAPI_EventSet));
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149

#ifdef PREFETCH_CELL
    DO_PAPI (PAPI_reset (PAPI_EventSet));
    if (!MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &MPID_nem_prefetched_cell);
    else
	MPID_nem_prefetched_cell = 0;
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues10));
#endif /*PREFETCH_CELL */

    /*DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues14)); */

 return_success:
    *again = 0;
    goto fn_exit;
 return_again:
    *again = 1;
    goto fn_exit;
 fn_exit:
    return mpi_errno;
150
151
 fn_fail:
    goto fn_exit;
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
}


/*
  int MPID_nem_mpich2_sendv (struct iovec **iov, int *n_iov, MPIDI_VC_t *vc);

  sends iov to vc
  Non-blocking
  if iov specifies more than MPID_NEM_MPICH2_DATA_LEN of data, the iov will be truncated, so that after MPID_nem_mpich2_sendv returns,
  iov will describe unsent data
  sets again to 1 if it can't get a free cell, 0 otherwise
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_sendv
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
168
static inline int
169
MPID_nem_mpich2_sendv (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *again)
170
171
172
173
174
175
{
    int mpi_errno = MPI_SUCCESS;
    MPID_nem_cell_ptr_t el;
    char *cell_buf;
    MPIDI_msg_sz_t payload_len;    
    int my_rank;
176
    MPIDI_CH3I_VC *vc_ch = VC_CH(vc);
177
178
179
    MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_MPICH2_SENDV);

    MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_MPICH2_SENDV);
180

181
    MPIU_Assert (*n_iov > 0 && (*iov)->MPID_IOV_LEN > 0);
182
183
    MPIU_Assert(vc_ch->is_local);

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
    DO_PAPI (PAPI_reset (PAPI_EventSet));

    my_rank = MPID_nem_mem_region.rank;
	
#ifdef PREFETCH_CELL
    el = MPID_nem_prefetched_cell;
    
    if (!el)
    {
	if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	{
	    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));
            goto return_again;
	}
	
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &el);
    }
#else /*PREFETCH_CELL     */
    if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
    {
	DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));
        goto return_again;
    }

    MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ , &el);
#endif /*PREFETCH_CELL     */

    payload_len = MPID_NEM_MPICH2_DATA_LEN;
    cell_buf    = (char *) el->pkt.mpich2.payload; /* cast away volatile */
    
214
    while (*n_iov && payload_len >= (*iov)->MPID_IOV_LEN)
215
    {
216
	int _iov_len = (*iov)->MPID_IOV_LEN;
217
	MPIU_Memcpy (cell_buf, (*iov)->MPID_IOV_BUF, _iov_len);
218
219
220
221
222
223
224
225
	payload_len -= _iov_len;
	cell_buf += _iov_len;
	--(*n_iov);
	++(*iov);
    }
    
    if (*n_iov && payload_len > 0)
    {
226
	MPIU_Memcpy (cell_buf, (*iov)->MPID_IOV_BUF, payload_len);
227
228
	(*iov)->MPID_IOV_BUF = (char *)(*iov)->MPID_IOV_BUF + payload_len;
	(*iov)->MPID_IOV_LEN -= payload_len;
229
230
231
232
233
234
235
236
237
238
239
240
 	payload_len = 0;
    }

    el->pkt.mpich2.source  = my_rank;
    el->pkt.mpich2.dest    = vc->lpid;
    el->pkt.mpich2.datalen = MPID_NEM_MPICH2_DATA_LEN - payload_len;
    el->pkt.mpich2.seqno   = vc_ch->send_seqno++;
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, el->pkt.mpich2.type = MPID_NEM_PKT_MPICH2);

    MPIU_DBG_MSG (CH3_CHANNEL, VERBOSE, "--> Sent queue");
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (el));

241
242
    MPID_nem_queue_enqueue (vc_ch->recv_queue, el);
    /*MPID_nem_rel_dump_queue( vc_ch->recv_queue ); */
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257

#ifdef PREFETCH_CELL
    if (!MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &MPID_nem_prefetched_cell);
    else
	MPID_nem_prefetched_cell = 0;
#endif /*PREFETCH_CELL */
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));

    *again = 0;
    goto fn_exit;
 return_again:
    *again = 1;
    goto fn_exit;
 fn_exit:
258
    MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_MPICH2_SENDV);
259
    return mpi_errno;
260
261
 fn_fail:
    goto fn_exit;
262
263
264
}

/* MPID_nem_mpich2_sendv_header (struct iovec **iov, int *n_iov, int dest)
265
   same as above but first iov element is an MPICH2 header */
266
267
268
269
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_sendv_header
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
270
static inline int
271
MPID_nem_mpich2_sendv_header (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *again)
272
273
274
275
276
277
{
    int mpi_errno = MPI_SUCCESS;
    MPID_nem_cell_ptr_t el;
    char *cell_buf;
    MPIDI_msg_sz_t payload_len;    
    int my_rank;
278
    MPIDI_CH3I_VC *vc_ch = VC_CH(vc);
279
280
281
    MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_MPICH2_SENDV_HEADER);
    
    MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_MPICH2_SENDV_HEADER);
282

283
284
    MPIU_Assert(vc_ch->is_local);

285
    DO_PAPI (PAPI_reset (PAPI_EventSet));
286
    MPIU_Assert (*n_iov > 0 && (*iov)->MPID_IOV_LEN == sizeof(MPIDI_CH3_Pkt_t));
287
288
289
290

    my_rank = MPID_nem_mem_region.rank;

#ifdef USE_FASTBOX
291
    if (*n_iov == 2 && (*iov)[1].MPID_IOV_LEN + sizeof(MPIDI_CH3_Pkt_t) <= MPID_NEM_FBOX_DATALEN)
292
293
294
    {
	MPID_nem_fbox_mpich2_t *pbox = vc_ch->fbox_out;

295
296
297
        if (MPID_nem_fbox_is_full((MPID_nem_fbox_common_ptr_t)pbox))
            goto usequeue_l;

298
299
300
301
302
303
304
305
        pbox->cell.pkt.mpich2.source  = MPID_nem_mem_region.local_rank;
        pbox->cell.pkt.mpich2.datalen = (*iov)[1].MPID_IOV_LEN + sizeof(MPIDI_CH3_Pkt_t);
        pbox->cell.pkt.mpich2.seqno   = vc_ch->send_seqno++;
        MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, pbox->cell.pkt.mpich2.type = MPID_NEM_PKT_MPICH2_HEAD);
        
        MPIU_Memcpy((void *)pbox->cell.pkt.mpich2.payload, (*iov)[0].MPID_IOV_BUF, (*iov)[0].MPID_IOV_LEN);
        MPIU_Memcpy ((char *)pbox->cell.pkt.mpich2.payload + (*iov)[0].MPID_IOV_LEN, (*iov)[1].MPID_IOV_BUF, (*iov)[1].MPID_IOV_LEN);
        
306
        OPA_store_release_int(&pbox->flag.value, 1);
307
308
309
310
311
312
        *n_iov = 0;

        MPIU_DBG_MSG (CH3_CHANNEL, VERBOSE, "--> Sent fbox ");
        MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (&pbox->cell));
        
        goto return_success;
313
314
    }
 usequeue_l:
315

316
317
318
319
#endif /*USE_FASTBOX */
	
#ifdef PREFETCH_CELL
    el = MPID_nem_prefetched_cell;
320

321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
    if (!el)
    {
	if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	{
	    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));
            goto return_again;
	}
	
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &el);
    }
#else /*PREFETCH_CELL    */
    if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
    {
	DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));
        goto return_again;
    }

    MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &el);
#endif /*PREFETCH_CELL */

341
    MPIU_Memcpy((void *)el->pkt.mpich2.payload, (*iov)->MPID_IOV_BUF, sizeof(MPIDI_CH3_Pkt_t));
342
343
344
345
346
347

    cell_buf = (char *)(el->pkt.mpich2.payload) + sizeof(MPIDI_CH3_Pkt_t);
    ++(*iov);
    --(*n_iov);

    payload_len = MPID_NEM_MPICH2_DATA_LEN - sizeof(MPIDI_CH3_Pkt_t);
348
    while (*n_iov && payload_len >= (*iov)->MPID_IOV_LEN)
349
    {
350
	int _iov_len = (*iov)->MPID_IOV_LEN;
351
	MPIU_Memcpy (cell_buf, (*iov)->MPID_IOV_BUF, _iov_len);
352
353
354
355
356
357
358
359
	payload_len -= _iov_len;
	cell_buf += _iov_len;
	--(*n_iov);
	++(*iov);
    }
    
    if (*n_iov && payload_len > 0)
    {
360
	MPIU_Memcpy (cell_buf, (*iov)->MPID_IOV_BUF, payload_len);
361
362
	(*iov)->MPID_IOV_BUF = (char *)(*iov)->MPID_IOV_BUF + payload_len;
	(*iov)->MPID_IOV_LEN -= payload_len;
363
364
365
366
367
368
369
370
371
372
373
374
	payload_len = 0;
    }

    el->pkt.mpich2.source  = my_rank;
    el->pkt.mpich2.dest    = vc->lpid;
    el->pkt.mpich2.datalen = MPID_NEM_MPICH2_DATA_LEN - payload_len;
    el->pkt.mpich2.seqno   = vc_ch->send_seqno++;
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, el->pkt.mpich2.type = MPID_NEM_PKT_MPICH2_HEAD);

    MPIU_DBG_MSG (CH3_CHANNEL, VERBOSE, "--> Sent queue");
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (el));

375
376
    MPID_nem_queue_enqueue (vc_ch->recv_queue, el);	
    /*MPID_nem_rel_dump_queue( vc_ch->recv_queue ); */
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392

#ifdef PREFETCH_CELL
    if (!MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &MPID_nem_prefetched_cell);
    else
	MPID_nem_prefetched_cell = 0;
#endif /*PREFETCH_CELL */
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));

 return_success:
    *again = 0;
    goto fn_exit;
 return_again:
    *again = 1;
    goto fn_exit;
 fn_exit:
393
    MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_MPICH2_SENDV_HEADER);
394
    return mpi_errno;
395
396
 fn_fail:
    goto fn_exit;
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
}

/* send the header and data described by the segment in one cell.  If
   there is no cell available, *again is set to 1.  If all of the data
   cannot be sent, *segment_first is set to the index of the first
   unsent byte.
   Pre condition:  This must be the first packet of a message (i.e.,
                       *segment first == 0)
                   The destination process is local
   Post conditions:  the header has been sent iff *again == 0
                     if there is data to send (segment_size > 0) then
                         (the header has been sent iff any data has
                         been sent (i.e., *segment_first > 0) )
                     i.e.: we will never send only the header
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_send_seg_header
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
416
static inline void
417
418
419
420
MPID_nem_mpich2_send_seg_header (MPID_Segment *segment, MPIDI_msg_sz_t *segment_first, MPIDI_msg_sz_t segment_size,
                                 void *header, MPIDI_msg_sz_t header_sz, MPIDI_VC_t *vc, int *again)
{
    MPID_nem_cell_ptr_t el;
421
    MPIDI_msg_sz_t datalen;
422
423
    int my_rank;
    MPIDI_msg_sz_t last;
424
    MPIDI_CH3I_VC *vc_ch = VC_CH(vc);
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439

    MPIU_Assert(vc_ch->is_local); /* netmods will have their own implementation */
    MPIU_Assert(header_sz <= sizeof(MPIDI_CH3_Pkt_t));
    MPIU_Assert(*segment_first == 0); /* this routine is only called for new messages */
    
    
    DO_PAPI (PAPI_reset (PAPI_EventSet));

    my_rank = MPID_nem_mem_region.rank;

#ifdef USE_FASTBOX
    if (sizeof(MPIDI_CH3_Pkt_t) + segment_size <= MPID_NEM_FBOX_DATALEN)
    {
	MPID_nem_fbox_mpich2_t *pbox = vc_ch->fbox_out;

440
441
442
        if (MPID_nem_fbox_is_full((MPID_nem_fbox_common_ptr_t)pbox))
            goto usequeue_l;

443
444
445
446
447
448
449
	{
	    pbox->cell.pkt.mpich2.source  = MPID_nem_mem_region.local_rank;
	    pbox->cell.pkt.mpich2.datalen = sizeof(MPIDI_CH3_Pkt_t) + segment_size;
	    pbox->cell.pkt.mpich2.seqno   = vc_ch->send_seqno++;
            MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, pbox->cell.pkt.mpich2.type = MPID_NEM_PKT_MPICH2_HEAD);

            /* copy header */
450
            MPIU_Memcpy((void *)pbox->cell.pkt.mpich2.payload, header, header_sz);
451
            
452
453
454
455
            /* copy data */
            last = segment_size;
            MPID_Segment_pack(segment, *segment_first, &last, (char *)pbox->cell.pkt.mpich2.payload + sizeof(MPIDI_CH3_Pkt_t));
            MPIU_Assert(last == segment_size);
456

457
            OPA_store_release_int(&pbox->flag.value, 1);
458
459
460
461
462
463
464
465
466
467

            *segment_first = last;

	    MPIU_DBG_MSG(CH3_CHANNEL, VERBOSE, "--> Sent fbox ");
	    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (&pbox->cell));

            goto return_success;
	}
    }
 usequeue_l:
468

469
470
471
472
#endif /*USE_FASTBOX */
	
#ifdef PREFETCH_CELL
    el = MPID_nem_prefetched_cell;
473

474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
    if (!el)
    {
	if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	{
	    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));
            goto return_again;
	}
	
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &el);
    }
#else /*PREFETCH_CELL    */
    if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
    {
	DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));
        goto return_again;
    }

    MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &el);
#endif /*PREFETCH_CELL */

    /* copy header */
495
    MPIU_Memcpy((void *)el->pkt.mpich2.payload, header, header_sz);
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
    
    /* copy data */
    if (segment_size - *segment_first <= MPID_NEM_MPICH2_DATA_LEN - sizeof(MPIDI_CH3_Pkt_t))
        last = segment_size;
    else
        last = *segment_first + MPID_NEM_MPICH2_DATA_LEN - sizeof(MPIDI_CH3_Pkt_t);
    
    MPID_Segment_pack(segment, *segment_first, &last, (char *)el->pkt.mpich2.payload + sizeof(MPIDI_CH3_Pkt_t));
    datalen = sizeof(MPIDI_CH3_Pkt_t) + last - *segment_first;
    *segment_first = last;
    
    el->pkt.mpich2.source  = my_rank;
    el->pkt.mpich2.dest    = vc->lpid;
    el->pkt.mpich2.datalen = datalen;
    el->pkt.mpich2.seqno   = vc_ch->send_seqno++;
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, el->pkt.mpich2.type = MPID_NEM_PKT_MPICH2_HEAD);

    MPIU_DBG_MSG (CH3_CHANNEL, VERBOSE, "--> Sent queue");
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (el));

    MPID_nem_queue_enqueue (vc_ch->recv_queue, el);	

#ifdef PREFETCH_CELL
    if (!MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &MPID_nem_prefetched_cell);
    else
	MPID_nem_prefetched_cell = 0;
#endif /*PREFETCH_CELL */
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));

 return_success:
    *again = 0;
    goto fn_exit;
 return_again:
    *again = 1;
    goto fn_exit;
 fn_exit:
    return;
}

/* similar to MPID_nem_mpich2_send_seg_header, except there is no
   header to send.  This need not be the first packet of a message. */
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_send_seg
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
542
static inline void
543
544
545
MPID_nem_mpich2_send_seg (MPID_Segment *segment, MPIDI_msg_sz_t *segment_first, MPIDI_msg_sz_t segment_size, MPIDI_VC_t *vc, int *again)
{
    MPID_nem_cell_ptr_t el;
546
    MPIDI_msg_sz_t datalen;
547
548
    int my_rank;
    MPIDI_msg_sz_t last;
549
    MPIDI_CH3I_VC *vc_ch = VC_CH(vc);
550

551
    MPIU_Assert(vc_ch->is_local); /* netmods will have their own implementation */    
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
    
    DO_PAPI (PAPI_reset (PAPI_EventSet));

    my_rank = MPID_nem_mem_region.rank;
	
#ifdef PREFETCH_CELL
    el = MPID_nem_prefetched_cell;
    
    if (!el)
    {
	if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	{
	    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));
            goto return_again;
	}
	
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &el);
    }
#else /*PREFETCH_CELL    */
    if (MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
    {
	DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));
        goto return_again;
    }

    MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &el);
#endif /*PREFETCH_CELL */

    /* copy data */
    if (segment_size - *segment_first <= MPID_NEM_MPICH2_DATA_LEN)
        last = segment_size;
    else
        last = *segment_first + MPID_NEM_MPICH2_DATA_LEN;
    
    MPID_Segment_pack(segment, *segment_first, &last, (char *)el->pkt.mpich2.payload);
    datalen = last - *segment_first;
    *segment_first = last;
    
    el->pkt.mpich2.source  = my_rank;
    el->pkt.mpich2.dest    = vc->lpid;
    el->pkt.mpich2.datalen = datalen;
    el->pkt.mpich2.seqno   = vc_ch->send_seqno++;
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, el->pkt.mpich2.type = MPID_NEM_PKT_MPICH2_HEAD);

    MPIU_DBG_MSG (CH3_CHANNEL, VERBOSE, "--> Sent queue");
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (el));

    MPID_nem_queue_enqueue (vc_ch->recv_queue, el);	

#ifdef PREFETCH_CELL
    if (!MPID_nem_queue_empty (MPID_nem_mem_region.my_freeQ))
	MPID_nem_queue_dequeue (MPID_nem_mem_region.my_freeQ, &MPID_nem_prefetched_cell);
    else
	MPID_nem_prefetched_cell = 0;
#endif /*PREFETCH_CELL */
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues5));

    *again = 0;
    goto fn_exit;
 return_again:
    *again = 1;
    goto fn_exit;
 fn_exit:
    return;
}

/*
  MPID_nem_mpich2_dequeue_fastbox (int local_rank)
  decrements usage count on fastbox for process with local rank local_rank and
  dequeues it from fbox queue if usage is 0.
  This function is called whenever a receive for a process on this node is matched.
  Fastboxes on fbox queue are polled regularly for incoming messages.
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_dequeue_fastbox
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
629
static inline void MPID_nem_mpich2_dequeue_fastbox(int local_rank)
630
631
632
{
    MPID_nem_fboxq_elem_t *el;

633
634
    MPIU_Assert(local_rank < MPID_nem_mem_region.num_local);

635
    el = &MPID_nem_fboxq_elem_list[local_rank];    
636
    MPIU_Assert(el->fbox != NULL);
637

638
    MPIU_Assert(el->usage);
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659

    --el->usage;
    if (el->usage == 0)
    {
	if (el->prev == NULL)
	    MPID_nem_fboxq_head = el->next;
	else
	    el->prev->next = el->next;

	if (el->next == NULL)
	    MPID_nem_fboxq_tail = el->prev;
	else
	    el->next->prev = el->prev;

	if (el == MPID_nem_curr_fboxq_elem)
	{
	    if (el->next == NULL)
		MPID_nem_curr_fboxq_elem = MPID_nem_fboxq_head;
	    else
		MPID_nem_curr_fboxq_elem = el->next;
	}
660
    }    
661
662
663
664
665
666
667
668
669
670
671
672
}

/*
  MPID_nem_mpich2_enqueue_fastbox (int local_rank)
  enqueues fastbox for process with local rank local_rank on fbox queue
  This function is called whenever a receive is posted for a process on this node.
  Fastboxes on fbox queue are polled regularly for incoming messages.
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_dequeue_fastbox
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
673
static inline void MPID_nem_mpich2_enqueue_fastbox(int local_rank)
674
675
676
{
    MPID_nem_fboxq_elem_t *el;

677
678
    MPIU_Assert(local_rank < MPID_nem_mem_region.num_local);

679
    el = &MPID_nem_fboxq_elem_list[local_rank];
680
681
    MPIU_Assert(el->fbox != NULL);

682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
    if (el->usage)
    {
	++el->usage;
    }
    else
    {
	el->usage = 1;
	if (MPID_nem_fboxq_tail == NULL)
	{
	    el->prev = NULL;
	    MPID_nem_curr_fboxq_elem = MPID_nem_fboxq_head = el;
	}
	else
	{
	    el->prev = MPID_nem_fboxq_tail;
	    MPID_nem_fboxq_tail->next = el;
	}
	    
	el->next = NULL;
	MPID_nem_fboxq_tail = el;
702
    }
703
704
705
706
707
708
709
710
711
712
713
}
/*
  MPID_nem_recv_seqno_matches (MPID_nem_queue_ptr_t qhead)
  check whether the sequence number for the cell at the head of qhead is the one
  expected from the sender of that cell
  We only check these for processes in COMM_WORLD (i.e. the ones initially allocated)
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_recv_seqno_matches
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
714
static inline int
715
716
MPID_nem_recv_seqno_matches (MPID_nem_queue_ptr_t qhead)
{
717
718
719
    int source;
    MPID_nem_cell_ptr_t cell = MPID_nem_queue_head(qhead);
    source = cell->pkt.mpich2.source;
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
    
    return (cell->pkt.mpich2.seqno == MPID_nem_recv_seqno[source]);
}

/*
  int MPID_nem_mpich2_test_recv (MPID_nem_cell_ptr_t *cell, int *in_fbox);

  non-blocking receive
  sets cell to the received cell, or NULL if there is nothing to receive. in_fbox is true iff the cell was found in a fbox
  the cell must be released back to the subsystem with MPID_nem_mpich2_release_cell() once the packet has been copied out
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_test_recv
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
735
static inline int
736
MPID_nem_mpich2_test_recv(MPID_nem_cell_ptr_t *cell, int *in_fbox, int in_blocking_progress)
737
738
739
740
741
742
{
    int mpi_errno = MPI_SUCCESS;
    
    DO_PAPI (PAPI_reset (PAPI_EventSet));

#ifdef USE_FASTBOX
743
    if (poll_active_fboxes(cell)) goto fbox_l;
744
745
#endif/* USE_FASTBOX     */

746
    if (MPID_nem_num_netmods)
747
    {
748
	mpi_errno = MPID_nem_network_poll(in_blocking_progress);
749
750
751
752
753
754
        if (mpi_errno) MPIU_ERR_POP (mpi_errno);
    }

    if (MPID_nem_queue_empty (MPID_nem_mem_region.my_recvQ) || !MPID_nem_recv_seqno_matches (MPID_nem_mem_region.my_recvQ))
    {
#ifdef USE_FASTBOX
755
756
757
758
759
760
761
762
763
764
765
766
        /* check for messages from any process (even those from which
           we don't expect messages).  If we're nonblocking, check all
           fboxes at once, if we're in a blocking loop we'll keep
           iterating, so just check them one at a time. */
        if (!in_blocking_progress) {
            int found;
            found = poll_every_fbox(cell);
            if (found)
                goto fbox_l;
        } else {
            poll_next_fbox (cell, goto fbox_l);
        }
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
#endif/* USE_FASTBOX     */
	*cell = NULL;
	goto fn_exit;
    }
    
    MPID_nem_queue_dequeue (MPID_nem_mem_region.my_recvQ, cell);

    ++MPID_nem_recv_seqno[(*cell)->pkt.mpich2.source];
    *in_fbox = 0;

 fn_exit:
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues6));
    
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, {
	if (*cell)
	{
	    MPIU_DBG_MSG_S (CH3_CHANNEL, VERBOSE, "<-- Recv %s", (*in_fbox) ? "fbox " : "queue");
	    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (*cell));
	}
    });

 fn_fail:
    return mpi_errno;

 fbox_l:
792
   *in_fbox = 1;
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
    goto fn_exit;

}

/*
  int MPID_nem_mpich2_test_recv_wait (MPID_nem_cell_ptr_t *cell, int *in_fbox, int timeout);

  blocking receive with timeout
  waits up to timeout iterations to receive a cell
  sets cell to the received cell, or NULL if there is nothing to receive. in_fbox is true iff the cell was found in a fbox
  the cell must be released back to the subsystem with MPID_nem_mpich2_release_cell() once the packet has been copied out
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_test_recv_wait
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
809
static inline int
810
811
812
813
814
MPID_nem_mpich2_test_recv_wait (MPID_nem_cell_ptr_t *cell, int *in_fbox, int timeout)
{
    int mpi_errno = MPI_SUCCESS;
    
#ifdef USE_FASTBOX
815
    if (poll_active_fboxes(cell)) goto fbox_l;
816
817
#endif/* USE_FASTBOX     */

818
    if (MPID_nem_num_netmods)
819
    {
820
	mpi_errno = MPID_nem_network_poll(TRUE /* blocking */);
821
822
823
824
825
826
        if (mpi_errno) MPIU_ERR_POP (mpi_errno);
    }

    while ((--timeout > 0) && (MPID_nem_queue_empty (MPID_nem_mem_region.my_recvQ) || !MPID_nem_recv_seqno_matches (MPID_nem_mem_region.my_recvQ)))
    {
#ifdef USE_FASTBOX
827
	poll_next_fbox (cell, goto fbox_l);
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
#endif/* USE_FASTBOX     */
	*cell = NULL;
	goto exit_l;
    }
    
    MPID_nem_queue_dequeue (MPID_nem_mem_region.my_recvQ, cell);

    ++MPID_nem_recv_seqno[(*cell)->pkt.mpich2.source];
    *in_fbox = 0;
 exit_l:
    
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, {
            if (*cell)
            {
                MPIU_DBG_MSG_S (CH3_CHANNEL, VERBOSE, "<-- Recv %s", (*in_fbox) ? "fbox " : "queue");
                MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (*cell));
            }
        });

 fn_fail:
    return mpi_errno;

 fbox_l:
    *in_fbox = 1;
    goto exit_l;
}

/*
  int MPID_nem_mpich2_blocking_recv (MPID_nem_cell_ptr_t *cell, int *in_fbox);

  blocking receive waits until there is something to receive, or then
  sets cell to the received cell. in_fbox is true iff the cell was
  found in a fbox the cell must be released back to the subsystem with
  MPID_nem_mpich2_release_cell() once the packet has been copied out
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_blocking_recv
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
867
static inline int
868
MPID_nem_mpich2_blocking_recv(MPID_nem_cell_ptr_t *cell, int *in_fbox, int completions)
869
870
871
872
873
874
875
876
877
878
879
880
881
882
{
    int mpi_errno = MPI_SUCCESS;
    DO_PAPI (PAPI_reset (PAPI_EventSet));

#ifdef MPICH_IS_THREADED
    /* We should never enter this function in a multithreaded app */
#ifdef HAVE_RUNTIME_THREADCHECK
    MPIU_Assert(!MPIR_ThreadInfo.isThreaded);
#else
    MPIU_Assert(0);
#endif
#endif

#ifdef USE_FASTBOX
883
    if (poll_active_fboxes(cell)) goto fbox_l;
884
885
#endif /*USE_FASTBOX */

886
    if (MPID_nem_num_netmods)
887
    {
888
	mpi_errno = MPID_nem_network_poll(TRUE /* blocking */);
889
890
891
892
893
894
895
896
        if (mpi_errno) MPIU_ERR_POP (mpi_errno);
    }

    while (MPID_nem_queue_empty (MPID_nem_mem_region.my_recvQ) || !MPID_nem_recv_seqno_matches (MPID_nem_mem_region.my_recvQ))
    {
	DO_PAPI (PAPI_reset (PAPI_EventSet));

#ifdef USE_FASTBOX	
897
898
	poll_next_fbox (cell, goto fbox_l);
        if (poll_active_fboxes(cell)) goto fbox_l;
899
900
#endif /*USE_FASTBOX */

901
	if (MPID_nem_num_netmods)
902
	{            
903
	    mpi_errno = MPID_nem_network_poll(TRUE /* blocking */);
904
905
            if (mpi_errno) MPIU_ERR_POP (mpi_errno);

906
            if (!MPID_nem_safe_to_block_recv())
907
            {
908
                *cell = NULL;
909
910
911
912
                *in_fbox = 0;
                goto exit_l;
            }
	}
913
914
915
916
917
918

        if (completions != OPA_load_int(&MPIDI_CH3I_progress_completion_count)) {
            *cell = NULL;
            *in_fbox = 0;
            goto exit_l;
        }
919
        MPIU_Busy_wait();
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
    }

    MPID_nem_queue_dequeue (MPID_nem_mem_region.my_recvQ, cell);

    ++MPID_nem_recv_seqno[(*cell)->pkt.mpich2.source];
    *in_fbox = 0;

 exit_l:    

    DO_PAPI (PAPI_accum_var (PAPI_EventSet,PAPI_vvalues8));
    
    MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, {
            if (*cell)
            {
                MPIU_DBG_MSG_S (CH3_CHANNEL, VERBOSE, "<-- Recv %s", (*in_fbox) ? "fbox " : "queue");
                MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell(*cell));
            }
        });

 fn_fail:
    return mpi_errno;

 fbox_l:
    *in_fbox = 1;
    goto exit_l;
}

/*
  int MPID_nem_mpich2_release_cell (MPID_nem_cell_ptr_t cell, MPIDI_VC_t *vc);

  releases the cell back to the subsystem to be used for subsequent receives
*/
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_release_cell
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
956
static inline int
957
958
959
MPID_nem_mpich2_release_cell (MPID_nem_cell_ptr_t cell, MPIDI_VC_t *vc)
{
    int mpi_errno = MPI_SUCCESS;
960
    MPIDI_CH3I_VC *vc_ch = VC_CH(vc);
961
962
963
964
965
966
967
968
    DO_PAPI (PAPI_reset (PAPI_EventSet));
    MPID_nem_queue_enqueue (vc_ch->free_queue, cell);
    DO_PAPI (PAPI_accum_var (PAPI_EventSet,PAPI_vvalues9));
    return mpi_errno;
}

#endif /*_MPID_NEM_INLINE_H*/