mpid_nem_inline.h 29.8 KB
Newer Older
1
2
3
4
5
6
7
8
9
/* -*- Mode: C; c-basic-offset:4 ; -*- */
/*
 *  (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_POLLS_BEFORE_YIELD 1000
11
#define MPID_NEM_THREAD_POLLS_BEFORE_YIELD 10
12
13

#include "my_papi_defs.h"
14
#include "mpiiov.h"
15

16
extern int MPID_nem_lmt_shm_pending;
17
18
extern MPID_nem_cell_ptr_t MPID_nem_prefetched_cell;

19
static inline int MPID_nem_mpich2_send_header (void* buf, int size, MPIDI_VC_t *vc, int *again);
20
static inline int MPID_nem_mpich2_sendv (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *again);
21
22
static inline void MPID_nem_mpich2_dequeue_fastbox (int local_rank);
static inline void MPID_nem_mpich2_enqueue_fastbox (int local_rank);
23
static inline int MPID_nem_mpich2_sendv_header (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *again);
24
static inline int MPID_nem_recv_seqno_matches (MPID_nem_queue_ptr_t qhead);
25
static inline int MPID_nem_mpich2_test_recv (MPID_nem_cell_ptr_t *cell, int *in_fbox, int in_blocking_progress);
26
27
28
29
static inline int MPID_nem_mpich2_blocking_recv (MPID_nem_cell_ptr_t *cell, int *in_fbox);
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,
30
31
                                                           MPIDI_msg_sz_t segment_size, void *header, MPIDI_msg_sz_t header_sz,
                                                           MPIDI_VC_t *vc, int *again);
32
static inline void MPID_nem_mpich2_send_seg (MPID_Segment *segment, MPIDI_msg_sz_t *segment_first, MPIDI_msg_sz_t segment_size,
33
34
35
36
37
38
39
                                                    MPIDI_VC_t *vc, int *again);


#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_send_header
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
40
static inline int
41
42
43
44
45
46
47
48
49
50
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;
    MPIDI_CH3I_VC *vc_ch = (MPIDI_CH3I_VC *)vc->channel_private;

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

    MPIU_Assert (size == sizeof(MPIDI_CH3_Pkt_t));
51
    MPIU_Assert (vc_ch->is_local);
52
53
54
55
56
57
58
59
60
61

    my_rank = MPID_nem_mem_region.rank;

#ifdef USE_FASTBOX
    {
	MPID_nem_fbox_mpich2_t *pbox = vc_ch->fbox_out;
	int count = 10;

	if (MPID_nem_islocked ((MPID_nem_fbox_common_ptr_t)pbox, 0, count))
	    goto usequeue_l;
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
        
        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);
        
        OPA_write_barrier();
        pbox->flag.value = 1;
        
        MPIU_DBG_MSG (CH3_CHANNEL, VERBOSE, "--> Sent fbox ");
        MPIU_DBG_STMT (CH3_CHANNEL, VERBOSE, MPID_nem_dbg_dump_cell (&pbox->cell));
        
        goto return_success;
78
79
80
    }
 usequeue_l:
#endif /*USE_FASTBOX */
81

82
83
84
#ifdef PREFETCH_CELL
    DO_PAPI (PAPI_reset (PAPI_EventSet));
    el = MPID_nem_prefetched_cell;
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
    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);
    
114
    MPIU_Memcpy((void *)el->pkt.mpich2.payload, buf, size);
115
116
117
118
119
120
    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));
121
122
    MPID_nem_queue_enqueue (vc_ch->recv_queue, el);
    /*MPID_nem_rel_dump_queue( vc_ch->recv_queue ); */
123
    DO_PAPI (PAPI_accum_var (PAPI_EventSet, PAPI_vvalues12));
124
    DO_PAPI (PAPI_reset (PAPI_EventSet));
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

#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;
 fn_fail:
    goto fn_exit;
}


/*
  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)
163
static inline int
164
MPID_nem_mpich2_sendv (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *again)
165
166
167
168
169
170
171
{
    int mpi_errno = MPI_SUCCESS;
    MPID_nem_cell_ptr_t el;
    char *cell_buf;
    MPIDI_msg_sz_t payload_len;    
    int my_rank;
    MPIDI_CH3I_VC *vc_ch = (MPIDI_CH3I_VC *)vc->channel_private;
172
173
174
    MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_MPICH2_SENDV);

    MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_MPICH2_SENDV);
175

176
    MPIU_Assert (*n_iov > 0 && (*iov)->MPID_IOV_LEN > 0);
177
178
    MPIU_Assert(vc_ch->is_local);

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
    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 */
    
209
    while (*n_iov && payload_len >= (*iov)->MPID_IOV_LEN)
210
    {
211
	int _iov_len = (*iov)->MPID_IOV_LEN;
212
	MPIU_Memcpy (cell_buf, (*iov)->MPID_IOV_BUF, _iov_len);
213
214
215
216
217
218
219
220
	payload_len -= _iov_len;
	cell_buf += _iov_len;
	--(*n_iov);
	++(*iov);
    }
    
    if (*n_iov && payload_len > 0)
    {
221
	MPIU_Memcpy (cell_buf, (*iov)->MPID_IOV_BUF, payload_len);
222
223
	(*iov)->MPID_IOV_BUF = (char *)(*iov)->MPID_IOV_BUF + payload_len;
	(*iov)->MPID_IOV_LEN -= payload_len;
224
225
226
227
228
229
230
231
232
233
234
235
 	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));

236
237
    MPID_nem_queue_enqueue (vc_ch->recv_queue, el);
    /*MPID_nem_rel_dump_queue( vc_ch->recv_queue ); */
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252

#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:
253
    MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_MPICH2_SENDV);
254
255
256
257
258
259
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}

/* MPID_nem_mpich2_sendv_header (struct iovec **iov, int *n_iov, int dest)
260
   same as above but first iov element is an MPICH2 header */
261
262
263
264
#undef FUNCNAME
#define FUNCNAME MPID_nem_mpich2_sendv_header
#undef FCNAME
#define FCNAME MPIDI_QUOTE(FUNCNAME)
265
static inline int
266
MPID_nem_mpich2_sendv_header (MPID_IOV **iov, int *n_iov, MPIDI_VC_t *vc, int *again)
267
268
269
270
271
272
273
{
    int mpi_errno = MPI_SUCCESS;
    MPID_nem_cell_ptr_t el;
    char *cell_buf;
    MPIDI_msg_sz_t payload_len;    
    int my_rank;
    MPIDI_CH3I_VC *vc_ch = (MPIDI_CH3I_VC *)vc->channel_private;
274
275
276
    MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_MPICH2_SENDV_HEADER);
    
    MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_MPICH2_SENDV_HEADER);
277

278
279
    MPIU_Assert(vc_ch->is_local);

280
    DO_PAPI (PAPI_reset (PAPI_EventSet));
281
    MPIU_Assert (*n_iov > 0 && (*iov)->MPID_IOV_LEN == sizeof(MPIDI_CH3_Pkt_t));
282
283
284
285

    my_rank = MPID_nem_mem_region.rank;

#ifdef USE_FASTBOX
286
    if (*n_iov == 2 && (*iov)[1].MPID_IOV_LEN + sizeof(MPIDI_CH3_Pkt_t) <= MPID_NEM_FBOX_DATALEN)
287
288
289
290
291
292
    {
	MPID_nem_fbox_mpich2_t *pbox = vc_ch->fbox_out;
	int count = 10;

	if (MPID_nem_islocked ((MPID_nem_fbox_common_ptr_t)pbox, 0, count))
	    goto usequeue_l;
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
        
        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);
        
        OPA_write_barrier();
        pbox->flag.value = 1;
        *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;
310
311
    }
 usequeue_l:
312

313
314
315
316
#endif /*USE_FASTBOX */
	
#ifdef PREFETCH_CELL
    el = MPID_nem_prefetched_cell;
317

318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
    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 */

338
    MPIU_Memcpy((void *)el->pkt.mpich2.payload, (*iov)->MPID_IOV_BUF, sizeof(MPIDI_CH3_Pkt_t));
339
340
341
342
343
344

    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);
345
    while (*n_iov && payload_len >= (*iov)->MPID_IOV_LEN)
346
    {
347
	int _iov_len = (*iov)->MPID_IOV_LEN;
348
	MPIU_Memcpy (cell_buf, (*iov)->MPID_IOV_BUF, _iov_len);
349
350
351
352
353
354
355
356
	payload_len -= _iov_len;
	cell_buf += _iov_len;
	--(*n_iov);
	++(*iov);
    }
    
    if (*n_iov && payload_len > 0)
    {
357
	MPIU_Memcpy (cell_buf, (*iov)->MPID_IOV_BUF, payload_len);
358
359
	(*iov)->MPID_IOV_BUF = (char *)(*iov)->MPID_IOV_BUF + payload_len;
	(*iov)->MPID_IOV_LEN -= payload_len;
360
361
362
363
364
365
366
367
368
369
370
371
	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));

372
373
    MPID_nem_queue_enqueue (vc_ch->recv_queue, el);	
    /*MPID_nem_rel_dump_queue( vc_ch->recv_queue ); */
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389

#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:
390
    MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_MPICH2_SENDV_HEADER);
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
    return mpi_errno;
 fn_fail:
    goto fn_exit;
}

/* 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)
413
static inline void
414
415
416
417
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;
418
    MPIDI_msg_sz_t datalen;
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
    int my_rank;
    MPIDI_msg_sz_t last;
    MPIDI_CH3I_VC *vc_ch = (MPIDI_CH3I_VC *)vc->channel_private;

    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;
	int count = 10;

	if (MPID_nem_islocked ((MPID_nem_fbox_common_ptr_t)pbox, 0, count))
	    goto usequeue_l;
	{
	    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 */
447
            MPIU_Memcpy((void *)pbox->cell.pkt.mpich2.payload, header, header_sz);
448
            
449
450
451
452
            /* 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);
453
454

	    OPA_write_barrier();
455
456
457
458
459
460
461
462
463
464
465
	    pbox->flag.value = 1;

            *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:
466

467
468
469
470
#endif /*USE_FASTBOX */
	
#ifdef PREFETCH_CELL
    el = MPID_nem_prefetched_cell;
471

472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
    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 */
493
    MPIU_Memcpy((void *)el->pkt.mpich2.payload, header, header_sz);
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
    
    /* 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)
540
static inline void
541
542
543
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;
544
    MPIDI_msg_sz_t datalen;
545
546
547
548
    int my_rank;
    MPIDI_msg_sz_t last;
    MPIDI_CH3I_VC *vc_ch = (MPIDI_CH3I_VC *)vc->channel_private;

549
    MPIU_Assert(vc_ch->is_local); /* netmods will have their own implementation */    
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
    
    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)
627
static inline void MPID_nem_mpich2_dequeue_fastbox(int local_rank)
628
629
630
{
    MPID_nem_fboxq_elem_t *el;

631
632
    MPIU_Assert(local_rank < MPID_nem_mem_region.num_local);

633
    el = &MPID_nem_fboxq_elem_list[local_rank];    
634
    MPIU_Assert(el->fbox != NULL);
635

636
    MPIU_Assert(el->usage);
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657

    --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;
	}
658
    }    
659
660
661
662
663
664
665
666
667
668
669
670
}

/*
  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)
671
static inline void MPID_nem_mpich2_enqueue_fastbox(int local_rank)
672
673
674
{
    MPID_nem_fboxq_elem_t *el;

675
676
    MPIU_Assert(local_rank < MPID_nem_mem_region.num_local);

677
    el = &MPID_nem_fboxq_elem_list[local_rank];
678
679
    MPIU_Assert(el->fbox != NULL);

680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
    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;
700
    }
701
702
703
704
705
706
707
708
709
710
711
}
/*
  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)
712
static inline int
713
714
MPID_nem_recv_seqno_matches (MPID_nem_queue_ptr_t qhead)
{
715
716
717
    int source;
    MPID_nem_cell_ptr_t cell = MPID_nem_queue_head(qhead);
    source = cell->pkt.mpich2.source;
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
    
    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)
733
static inline int
734
MPID_nem_mpich2_test_recv(MPID_nem_cell_ptr_t *cell, int *in_fbox, int in_blocking_progress)
735
736
737
738
739
740
{
    int mpi_errno = MPI_SUCCESS;
    
    DO_PAPI (PAPI_reset (PAPI_EventSet));

#ifdef USE_FASTBOX
741
    if (poll_active_fboxes(cell)) goto fbox_l;
742
743
#endif/* USE_FASTBOX     */

744
    if (MPID_nem_num_netmods)
745
    {
746
	mpi_errno = MPID_nem_network_poll(in_blocking_progress);
747
748
749
750
751
752
        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
753
754
755
756
757
758
759
760
761
762
763
764
        /* 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);
        }
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
#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:
790
   *in_fbox = 1;
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
    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)
807
static inline int
808
809
810
811
812
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
813
    if (poll_active_fboxes(cell)) goto fbox_l;
814
815
#endif/* USE_FASTBOX     */

816
    if (MPID_nem_num_netmods)
817
    {
818
	mpi_errno = MPID_nem_network_poll(TRUE /* blocking */);
819
820
821
822
823
824
        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
825
	poll_next_fbox (cell, goto fbox_l);
826
827
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
#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)
865
static inline int
866
867
868
MPID_nem_mpich2_blocking_recv(MPID_nem_cell_ptr_t *cell, int *in_fbox)
{
    int mpi_errno = MPI_SUCCESS;
869
    unsigned completions = OPA_load_int(&MPIDI_CH3I_progress_completion_count);
870
#ifndef ENABLE_NO_YIELD
871
872
873
874
875
876
877
878
879
880
881
882
883
884
    int pollcount = 0;
#endif
    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
885
    if (poll_active_fboxes(cell)) goto fbox_l;
886
887
#endif /*USE_FASTBOX */

888
    if (MPID_nem_num_netmods)
889
    {
890
	mpi_errno = MPID_nem_network_poll(TRUE /* blocking */);
891
892
893
894
895
896
897
898
        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	
899
900
	poll_next_fbox (cell, goto fbox_l);
        if (poll_active_fboxes(cell)) goto fbox_l;
901
902
#endif /*USE_FASTBOX */

903
	if (MPID_nem_num_netmods)
904
	{            
905
	    mpi_errno = MPID_nem_network_poll(TRUE /* blocking */);
906
907
            if (mpi_errno) MPIU_ERR_POP (mpi_errno);

908
909
            if (completions != OPA_load_int(&MPIDI_CH3I_progress_completion_count) || MPID_nem_local_lmt_pending
                || MPIDI_CH3I_active_send[CH3_NORMAL_QUEUE] || MPIDI_CH3I_SendQ_head(CH3_NORMAL_QUEUE))
910
            {
911
                *cell = NULL;
912
913
914
915
                *in_fbox = 0;
                goto exit_l;
            }
	}
916
#if !defined ENABLE_NO_YIELD
917
918
919
	if (pollcount >= MPID_NEM_POLLS_BEFORE_YIELD)
	{
	    pollcount = 0;
920
	    MPIU_PW_Sched_yield();
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
956
957
958
959
	}
	++pollcount;
#endif
    }

    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)
960
static inline int
961
962
963
964
965
966
967
968
969
970
971
972
MPID_nem_mpich2_release_cell (MPID_nem_cell_ptr_t cell, MPIDI_VC_t *vc)
{
    int mpi_errno = MPI_SUCCESS;
    MPIDI_CH3I_VC *vc_ch = (MPIDI_CH3I_VC *)vc->channel_private;
    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*/