mpidimpl.h 78.4 KB
Newer Older
1
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil ; -*- */
2
3
4
5
6
7
8
9
10
11
12
13
14
15
/*
 *  (C) 2001 by Argonne National Laboratory.
 *      See COPYRIGHT in top-level directory.
 */

/*
 * WARNING: Functions and macros in this file are for internal use only.  
 * As such, they are only visible to the device and
 * channel.  Do not include them in the MPID macros.
 */

#if !defined(MPICH_MPIDIMPL_H_INCLUDED)
#define MPICH_MPIDIMPL_H_INCLUDED

16
17
#include "mpichconf.h"

18
19
20
21
22
23
24
25
26
27
#if defined(HAVE_ASSERT_H)
#include <assert.h>
#endif

#include "mpiimpl.h"

#if !defined(MPICH_MPIDPRE_H_INCLUDED)
#include "mpidpre.h"
#endif

28
29
#include "mpidftb.h"

30
31
32
/* Add the ch3 packet definitions */
#include "mpidpkt.h"

33
34
#include "mpid_sched.h"

35
/* We need to match the size of MPIR_Pint to the relevant Format control
36
 */
37
#define MPIDI_MSG_SZ_FMT MPIR_PINT_FMT_DEC_SPEC
38
39
40
41
42
43
44
45
46
47

#if !defined(MPIDI_IOV_DENSITY_MIN)
#   define MPIDI_IOV_DENSITY_MIN (16 * 1024)
#endif

#if defined(HAVE_GETHOSTNAME) && defined(NEEDS_GETHOSTNAME_DECL) && \
   !defined(gethostname)
int gethostname(char *name, size_t len);
# endif

48
49
50
51
/* group of processes detected to have failed.  This is a subset of
   comm_world group. */
extern MPID_Group *MPIDI_Failed_procs_group;

52
53
extern int MPIDI_Use_pmi2_api;

54
55
56
57
58
#define MPIDI_CHANGE_VC_STATE(vc, new_state) do {               \
        MPIU_DBG_VCSTATECHANGE(vc, VC_STATE_##new_state);       \
        (vc)->state = MPIDI_VC_STATE_##new_state;               \
    } while (0)

59
60
61
62
63
64
65
66
67
68
69
70
71
/*S
  MPIDI_PG_t - Process group description

  Notes:
  Every 'MPI_COMM_WORLD' known to this process has an associated process 
  group.  
  S*/
typedef struct MPIDI_PG
{
    /* MPIU_Object field.  MPIDI_PG_t objects are not allocated using the 
       MPIU_Object system, but we do use the associated reference counting 
       routines.  Therefore, handle must be present, but is not used 
       except by debugging routines */
72
    MPIU_OBJECT_HEADER; /* adds handle and ref_count fields */
73
74
75
76
77
78
79
80
81
82

    /* Next pointer used to maintain a list of all process groups known to 
       this process */
    struct MPIDI_PG * next;

    /* Number of processes in the process group */
    int size;

    /* VC table.  At present this is a pointer to an array of VC structures. 
       Someday we may want make this a pointer to an array
83
       of VC references.  Thus, it is important to use MPIDI_PG_Get_vc() 
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
       instead of directly referencing this field. */
    struct MPIDI_VC * vct;

    /* Pointer to the process group ID.  The actual ID is defined and 
       allocated by the process group.  The pointer is kept in the
       device space because it is necessary for the device to be able to 
       find a particular process group. */
    void * id;

    /* Replacement abstraction for connection information */
    /* Connection information needed to access processes in this process 
       group and to share the data with other processes.  The items are
       connData - pointer for data used to implement these functions 
                  (e.g., a pointer to an array of process group info)
       getConnInfo( rank, buf, bufsize, self ) - function to store into
                  buf the connection information for rank in this process 
                  group
       connInfoToString( buf_p, size, self ) - return in buf_p a string
                  that can be sent to another process to recreate the
                  connection information (the info needed to support
                  getConnInfo)
       connInfoFromString( buf, self ) - setup the information needed
                  to implement getConnInfo
       freeConnInfo( self ) - free any storage or resources associated
                  with the connection information.

       See ch3/src/mpidi_pg.c 
    */
    void *connData;
    int  (*getConnInfo)( int, char *, int, struct MPIDI_PG * );
    int  (*connInfoToString)( char **, int *, struct MPIDI_PG * );
    int  (*connInfoFromString)( const char *,  struct MPIDI_PG * );
    int  (*freeConnInfo)( struct MPIDI_PG * );

    /* Rather than have each channel define its own fields for the 
       channel-specific data, we provide a fixed-sized scratchpad.  Currently,
       this has a very generous size, though this may shrink later (a channel
       can always allocate storage and hang it off of the end).  This 
       is necessary to allow dynamic loading of channels at MPI_Init time. */
#define MPIDI_CH3_PG_SIZE 48
    int32_t channel_private[MPIDI_CH3_PG_SIZE];
#if defined(MPIDI_CH3_PG_DECL)
    MPIDI_CH3_PG_DECL
#endif    
}
MPIDI_PG_t;


/*S
  MPIDI_Process_t - The information required about this process by the CH3 
  device.

  S*/
typedef struct MPIDI_Process
{
    MPIDI_PG_t * my_pg;
    int my_pg_rank;
}
MPIDI_Process_t;

extern MPIDI_Process_t MPIDI_Process;

/*----------------------
  BEGIN DATATYPE SECTION
  ----------------------*/
/* FIXME: We want to avoid even storing information about the builtins
   if we can */
#define MPIDI_Datatype_get_info(count_, datatype_, dt_contig_out_, data_sz_out_, dt_ptr_, dt_true_lb_)\
{									\
    if (HANDLE_GET_KIND(datatype_) == HANDLE_KIND_BUILTIN)		\
    {									\
	(dt_ptr_) = NULL;						\
	(dt_contig_out_) = TRUE;					\
        (dt_true_lb_)    = 0;                                           \
158
	(data_sz_out_) = (MPIDI_msg_sz_t) (count_) * MPID_Datatype_get_basic_size(datatype_); \
159
160
161
162
163
164
165
	MPIDI_DBG_PRINTF((15, FCNAME, "basic datatype: dt_contig=%d, dt_sz=%d, data_sz=" MPIDI_MSG_SZ_FMT,\
			  (dt_contig_out_), MPID_Datatype_get_basic_size(datatype_), (data_sz_out_)));\
    }									\
    else								\
    {									\
	MPID_Datatype_get_ptr((datatype_), (dt_ptr_));			\
	(dt_contig_out_) = (dt_ptr_)->is_contig;			\
166
	(data_sz_out_) = (MPIDI_msg_sz_t) (count_) * (dt_ptr_)->size;	\
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
        (dt_true_lb_)    = (dt_ptr_)->true_lb;                          \
	MPIDI_DBG_PRINTF((15, FCNAME, "user defined datatype: dt_contig=%d, dt_sz=%d, data_sz=" MPIDI_MSG_SZ_FMT,\
			  (dt_contig_out_), (dt_ptr_)->size, (data_sz_out_)));\
    }									\
}
/*--------------------
  END DATATYPE SECTION
  --------------------*/


/*---------------------
  BEGIN REQUEST SECTION
  ---------------------*/

/*
 * MPID_Requests
 *
 * MPI Requests are handles to MPID_Request structures.  These are used
 * for most communication operations to provide a uniform way in which to
 * define pending operations.  As such, they contain many fields that are 
 * only used by some operations (logically, an MPID_Request is a union type).
 *
 * There are several kinds of requests.  They are
 *    Send, Receive, RMA, User, Persistent
 * In addition, send and RMA requests may be "incomplete"; this means that
 * they have not sent their initial packet, and they may store additional
 * data about the operation that will be used when the initial packet 
 * can be sent.
 *
 * Also, requests that are used internally within blocking MPI routines
 * (only Send and Receive requests) do not require references to
 * (or increments of the reference counts) communicators or datatypes.
 * Thus, freeing these requests also does not require testing or 
 * decrementing these fields.
 *
 * Finally, we want to avoid multiple tests for a failure to allocate
 * a request.  Thus, the request allocation macros will jump to fn_fail
 * if there is an error.  This is akin to using a "throw" in C++.
 * 
 * For example, a posted (unmatched) receive queue entry needs only:
 *     match info
 *     buffer info (address, count, datatype)
 *     if nonblocking, communicator (used for finding error handler)
 *     if nonblocking, cancelled state
 * Once matched, a receive queue entry also needs
 *     actual match info
 *     message type (eager, rndv, eager-sync)
 *     completion state (is all data available)
 *        If destination datatype is non-contiguous, it also needs
 *        current unpack state.
 * An unexpected message (in the unexpected receive queue) needs only:
 *     match info
 *     message type (eager, rndv, eager-sync)
 *     if (eager, eager-sync), data
 *     completion state (is all data available?)
 * A send request requires only
 *     message type (eager, rndv, eager-sync)
 *     completion state (has all data been sent?)
 *     canceled state
 *     if nonblocking, communicator (used for finding error handler)
 *     if the initial envelope is still pending (e.g., could not write yet)
 *         match info
 *     if the data is still pending (rndv or would not send eager)
 *         buffer info (address, count, datatype)
 * RMA requests require (what)?
 * User (generalized) requests require
 *     function pointers for operations
 *     completion state
 *     cancelled state
 */
237

238
/* FIXME XXX DJG for TLS hack */
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
#define MPID_REQUEST_TLS_MAX 128

#if MPIU_HANDLE_ALLOCATION_METHOD == MPIU_HANDLE_ALLOCATION_THREAD_LOCAL
#  define MPIDI_Request_tls_alloc(req) \
    do { \
        int i;                                                         \
        MPIU_THREADPRIV_DECL;                                          \
        MPIU_THREADPRIV_GET;                                           \
        if (!MPIU_THREADPRIV_FIELD(request_handles)) {                 \
            MPID_Request *prev, *cur;                                  \
            /* batch allocate a linked list of requests */             \
            MPIU_THREAD_CS_ENTER(HANDLEALLOC,);                        \
            prev = MPIU_Handle_obj_alloc_unsafe(&MPID_Request_mem);    \
            prev->next = NULL;                                         \
            assert(prev);                                              \
            for (i = 1; i < MPID_REQUEST_TLS_MAX; ++i) {               \
                cur = MPIU_Handle_obj_alloc_unsafe(&MPID_Request_mem); \
                assert(cur);                                           \
                cur->next = prev;                                      \
                prev = cur;                                            \
            }                                                          \
            MPIU_THREAD_CS_EXIT(HANDLEALLOC,);                         \
            MPIU_THREADPRIV_FIELD(request_handles) = cur;              \
            MPIU_THREADPRIV_FIELD(request_handle_count) += MPID_REQUEST_TLS_MAX;    \
        }                                                              \
        (req) = MPIU_THREADPRIV_FIELD(request_handles);                \
        MPIU_THREADPRIV_FIELD(request_handles) = req->next;            \
        MPIU_THREADPRIV_FIELD(request_handle_count) -= 1;              \
    } while (0)
#elif MPIU_HANDLE_ALLOCATION_METHOD == MPIU_HANDLE_ALLOCATION_MUTEX
269
270
271
272
273
274
#  define MPIDI_Request_tls_alloc(req_) \
    do { \
	(req_) = MPIU_Handle_obj_alloc(&MPID_Request_mem); \
        MPIU_DBG_MSG_P(CH3_CHANNEL,VERBOSE,		\
	       "allocated request, handle=0x%08x", req_);\
    } while (0)
275
276
277
278
#else
#  error MPIU_HANDLE_ALLOCATION_METHOD not defined
#endif

279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
#define MPIDI_CH3U_Request_complete(req_)			\
{								\
    int incomplete__;						\
								\
    MPIDI_CH3U_Request_decrement_cc((req_), &incomplete__);	\
    if (!incomplete__)						\
    {								\
	MPID_Request_release(req_);				\
	MPIDI_CH3_Progress_signal_completion();			\
    }								\
}


/* If the channel doesn't initialize anything in the request, 
   provide a dummy */
#ifndef MPIDI_CH3_REQUEST_INIT
#define MPIDI_CH3_REQUEST_INIT(a_)
#endif

/* FIXME: Why does a send request need the match information?
   Is that for debugging information?  In case the initial envelope
   cannot be sent? Ditto for the dev.user_buf, count, and datatype 
   fields when the data is sent eagerly.  

   The following fields needed to be set:
   datatype_ptr
   status.MPI_ERROR

   Note that this macro requires that rank, tag, context_offset,
   comm, buf, datatype, and count all be available with those names
   (they are not arguments to the routine)
*/
#define MPIDI_Request_create_sreq(sreq_, mpi_errno_, FAIL_)	\
{								\
    (sreq_) = MPIU_Handle_obj_alloc(&MPID_Request_mem);         \
    if ((sreq_) == NULL)					\
    {								\
	MPIU_DBG_MSG(CH3_CHANNEL,TYPICAL,"unable to allocate a request");\
	(mpi_errno_) = MPIR_ERR_MEMALLOCFAILED;			\
	FAIL_;							\
    }								\
    MPIU_DBG_MSG_P(CH3_CHANNEL,VERBOSE,                         \
	       "allocated request, handle=0x%08x", (sreq_)->handle);\
    								\
    MPIU_Object_set_ref((sreq_), 2);				\
    (sreq_)->kind = MPID_REQUEST_SEND;				\
    (sreq_)->comm = comm;					\
326
    (sreq_)->greq_fns = NULL;                                   \
327
    MPID_cc_set(&(sreq_)->cc, 1);                               \
328
    (sreq_)->cc_ptr		   = &(sreq_)->cc;              \
329
    (sreq_)->partner_request   = NULL;                          \
330
331
332
333
334
    MPIR_Comm_add_ref(comm);					\
    (sreq_)->status.MPI_ERROR	   = MPI_SUCCESS;               \
    (sreq_)->status.cancelled	   = FALSE;		        \
    (sreq_)->dev.state = 0;                                     \
    (sreq_)->dev.cancel_pending = FALSE;                        \
335
336
337
    (sreq_)->dev.match.parts.rank = rank;			\
    (sreq_)->dev.match.parts.tag = tag;				\
    (sreq_)->dev.match.parts.context_id = comm->context_id + context_offset;	\
338
339
340
341
342
    (sreq_)->dev.user_buf = (void *) buf;			\
    (sreq_)->dev.user_count = count;				\
    (sreq_)->dev.datatype = datatype;				\
    (sreq_)->dev.datatype_ptr	   = NULL;                      \
    (sreq_)->dev.segment_ptr	   = NULL;                      \
343
344
    (sreq_)->dev.OnDataAvail	   = NULL;                      \
    (sreq_)->dev.OnFinal	   = NULL;                      \
345
346
    (sreq_)->dev.iov_count	   = 0;                         \
    (sreq_)->dev.iov_offset	   = 0;                         \
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
}

/* This is the receive request version of MPIDI_Request_create_sreq */
#define MPIDI_Request_create_rreq(rreq_, mpi_errno_, FAIL_)	\
{								\
    (rreq_) = MPIU_Handle_obj_alloc(&MPID_Request_mem);         \
    if ((rreq_) == NULL)					\
    {								\
	MPIU_DBG_MSG(CH3_CHANNEL,TYPICAL,"unable to allocate a request");\
	(mpi_errno_) = MPIR_ERR_MEMALLOCFAILED;			\
	FAIL_;							\
    }								\
    MPIU_DBG_MSG_P(CH3_CHANNEL,VERBOSE,                         \
	       "allocated request, handle=0x%08x", (rreq_)->handle);\
    								\
    MPIU_Object_set_ref((rreq_), 2);				\
    (rreq_)->kind = MPID_REQUEST_RECV;				\
    (rreq_)->comm = NULL;					\
365
    (rreq_)->greq_fns = NULL;                                   \
366
    MPID_cc_set(&(rreq_)->cc, 1);                               \
367
368
369
    (rreq_)->cc_ptr		   = &(rreq_)->cc;              \
    (rreq_)->status.MPI_ERROR	   = MPI_SUCCESS;               \
    (rreq_)->status.cancelled	   = FALSE;                     \
370
    (rreq_)->partner_request   = NULL;                          \
371
372
373
374
375
    (rreq_)->dev.state = 0;                                     \
    (rreq_)->dev.cancel_pending = FALSE;                        \
    (rreq_)->dev.datatype_ptr = NULL;                           \
    (rreq_)->dev.segment_ptr = NULL;                            \
    (rreq_)->dev.iov_offset   = 0;                              \
376
377
    (rreq_)->dev.OnDataAvail	   = NULL;                      \
    (rreq_)->dev.OnFinal	   = NULL;                      \
378
379
380
     MPIDI_CH3_REQUEST_INIT(rreq_);\
}

381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
/* creates a new, trivially complete recv request that is suitable for
 * returning when a user passed MPI_PROC_NULL */
#define MPIDI_Request_create_null_rreq(rreq_, mpi_errno_, FAIL_)           \
    do {                                                                   \
        (rreq_) = MPID_Request_create();                                   \
        if ((rreq_) != NULL) {                                             \
            MPIU_Object_set_ref((rreq_), 1);                               \
            /* MT FIXME should these be handled by MPID_Request_create? */ \
            MPID_cc_set(&(rreq_)->cc, 0);                                  \
            (rreq_)->kind = MPID_REQUEST_RECV;                             \
            MPIR_Status_set_procnull(&(rreq_)->status);                    \
        }                                                                  \
        else {                                                             \
            MPIU_ERR_SETANDJUMP(mpi_errno_,MPI_ERR_OTHER,"**nomemreq");    \
        }                                                                  \
    } while (0)

398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
#define MPIDI_REQUEST_MSG_MASK (0x3 << MPIDI_REQUEST_MSG_SHIFT)
#define MPIDI_REQUEST_MSG_SHIFT 0
#define MPIDI_REQUEST_NO_MSG 0
#define MPIDI_REQUEST_EAGER_MSG 1
#define MPIDI_REQUEST_RNDV_MSG 2
#define MPIDI_REQUEST_SELF_MSG 3

#define MPIDI_Request_get_msg_type(req_)				\
(((req_)->dev.state & MPIDI_REQUEST_MSG_MASK) >> MPIDI_REQUEST_MSG_SHIFT)

#define MPIDI_Request_set_msg_type(req_, msgtype_)			\
{									\
    (req_)->dev.state &= ~MPIDI_REQUEST_MSG_MASK;			\
    (req_)->dev.state |= ((msgtype_) << MPIDI_REQUEST_MSG_SHIFT) & MPIDI_REQUEST_MSG_MASK;\
}

#define MPIDI_REQUEST_SRBUF_MASK (0x1 << MPIDI_REQUEST_SRBUF_SHIFT)
#define MPIDI_REQUEST_SRBUF_SHIFT 2

#define MPIDI_Request_get_srbuf_flag(req_)					\
(((req_)->dev.state & MPIDI_REQUEST_SRBUF_MASK) >> MPIDI_REQUEST_SRBUF_SHIFT)

#define MPIDI_Request_set_srbuf_flag(req_, flag_)			\
{									\
    (req_)->dev.state &= ~MPIDI_REQUEST_SRBUF_MASK;			\
    (req_)->dev.state |= ((flag_) << MPIDI_REQUEST_SRBUF_SHIFT) & MPIDI_REQUEST_SRBUF_MASK;	\
}

#define MPIDI_REQUEST_SYNC_SEND_MASK (0x1 << MPIDI_REQUEST_SYNC_SEND_SHIFT)
#define MPIDI_REQUEST_SYNC_SEND_SHIFT 3

#define MPIDI_Request_get_sync_send_flag(req_)						\
(((req_)->dev.state & MPIDI_REQUEST_SYNC_SEND_MASK) >> MPIDI_REQUEST_SYNC_SEND_SHIFT)

#define MPIDI_Request_set_sync_send_flag(req_, flag_)			\
{									\
    (req_)->dev.state &= ~MPIDI_REQUEST_SYNC_SEND_MASK;			\
    (req_)->dev.state |= ((flag_) << MPIDI_REQUEST_SYNC_SEND_SHIFT) & MPIDI_REQUEST_SYNC_SEND_MASK;\
}

#define MPIDI_REQUEST_TYPE_MASK (0xF << MPIDI_REQUEST_TYPE_SHIFT)
#define MPIDI_REQUEST_TYPE_SHIFT 4
#define MPIDI_REQUEST_TYPE_RECV 0
#define MPIDI_REQUEST_TYPE_SEND 1
#define MPIDI_REQUEST_TYPE_RSEND 2
#define MPIDI_REQUEST_TYPE_SSEND 3
/* We need a BSEND type for persistent bsends (see mpid_startall.c) */
#define MPIDI_REQUEST_TYPE_BSEND 4
#define MPIDI_REQUEST_TYPE_PUT_RESP 5
#define MPIDI_REQUEST_TYPE_GET_RESP 6
#define MPIDI_REQUEST_TYPE_ACCUM_RESP 7
#define MPIDI_REQUEST_TYPE_PUT_RESP_DERIVED_DT 8
#define MPIDI_REQUEST_TYPE_GET_RESP_DERIVED_DT 9
#define MPIDI_REQUEST_TYPE_ACCUM_RESP_DERIVED_DT 10
#define MPIDI_REQUEST_TYPE_PT_SINGLE_PUT 11
#define MPIDI_REQUEST_TYPE_PT_SINGLE_ACCUM 12


#define MPIDI_Request_get_type(req_)						\
(((req_)->dev.state & MPIDI_REQUEST_TYPE_MASK) >> MPIDI_REQUEST_TYPE_SHIFT)

#define MPIDI_Request_set_type(req_, type_)				\
{									\
    (req_)->dev.state &= ~MPIDI_REQUEST_TYPE_MASK;			\
    (req_)->dev.state |= ((type_) << MPIDI_REQUEST_TYPE_SHIFT) & MPIDI_REQUEST_TYPE_MASK;\
}

/* NOTE: Request updates may require atomic ops (critical sections) if
   a fine-grain thread-sync model is used. */
#define MPIDI_Request_cancel_pending(req_, flag_)	\
{							\
    *(flag_) = (req_)->dev.cancel_pending;		\
    (req_)->dev.cancel_pending = TRUE;			\
}

473
474
475
476
477
478
479
480
481
482
483
/* the following two macros were formerly a single confusing macro with side
   effects named MPIDI_Request_recv_pending() */
#define MPIDI_Request_check_pending(req_, recv_pending_)   \
    do {                                                   \
        *(recv_pending_) = (req_)->dev.recv_pending_count; \
    } while (0)

#define MPIDI_Request_decr_pending(req_)    \
    do {                                    \
        --(req_)->dev.recv_pending_count;   \
    } while (0)
484
485
486
487
488
489
490
491
492

/* MPIDI_Request_fetch_and_clear_rts_sreq() - atomically fetch current 
   partner RTS sreq and nullify partner request */
#define MPIDI_Request_fetch_and_clear_rts_sreq(sreq_, rts_sreq_)	\
    {									\
    	*(rts_sreq_) = (sreq_)->partner_request;			\
    	(sreq_)->partner_request = NULL;				\
    }

493
/* FIXME: We've moved to allow finer-grain critical sections... */
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
/* Note: In the current implementation, the mpid_xsend.c routines that
   make use of MPIDI_VC_FAI_send_seqnum are all protected by the 
   SINGLE_CS_ENTER/EXIT macros, so all uses of this macro are 
   alreay within a critical section when needed.  If/when we move to
   a finer-grain model, we'll need to examine whether this requires
   a separate lock. */
#if defined(MPID_USE_SEQUENCE_NUMBERS)
#   define MPIDI_Request_set_seqnum(req_, seqnum_)	\
    {							\
    	(req_)->dev.seqnum = (seqnum_);			\
    }
#   define MPIDI_VC_FAI_send_seqnum(vc_, seqnum_out_)	\
    {							\
	(seqnum_out_) = (vc_)->seqnum_send++;		\
    }
#   define MPIDI_Pkt_set_seqnum(pkt_, seqnum_)	\
    {						\
    	(pkt_)->seqnum = (seqnum_);		\
    }
#   define MPIDI_VC_Init_seqnum_send(vc_)	\
    {						\
    	(vc_)->seqnum_send = 0;			\
    }
#else
#   define MPIDI_Request_set_seqnum(req_, seqnum_)
#   define MPIDI_VC_FAI_send_seqnum(vc_, seqnum_out_)
#   define MPIDI_Pkt_set_seqnum(pkt_, seqnum_)
#   define MPIDI_VC_Init_seqnum_send(vc_)
#endif


/*-------------------
  END REQUEST SECTION
  -------------------*/


/*------------------
  BEGIN COMM SECTION
  ------------------*/
533
534
#define MPIDI_Comm_get_vc(comm_, rank_, vcp_) *(vcp_) = (comm_)->vcr[(rank_)]

535
536
537
538
539
540
541
542
543
544
#ifdef USE_MPIDI_DBG_PRINT_VC
void MPIDI_DBG_PrintVC(MPIDI_VC_t *vc);
void MPIDI_DBG_PrintVCState2(MPIDI_VC_t *vc, MPIDI_VC_State_t new_state);
void MPIDI_DBG_PrintVCState(MPIDI_VC_t *vc);
#else
#define MPIDI_DBG_PrintVC(vc)
#define MPIDI_DBG_PrintVCState2(vc, new_state)
#define MPIDI_DBG_PrintVCState(vc)
#endif

545
546
547
548
#define MPIDI_Comm_get_vc_set_active(comm_, rank_, vcp_) do {           \
        *(vcp_) = (comm_)->vcr[(rank_)];                                \
        if ((*(vcp_))->state == MPIDI_VC_STATE_INACTIVE)                \
        {                                                               \
549
            MPIDI_DBG_PrintVCState2(*(vcp_), MPIDI_VC_STATE_ACTIVE);     \
550
551
552
553
            MPIDI_CHANGE_VC_STATE((*(vcp_)), ACTIVE);                   \
        }                                                               \
    } while(0)

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
/*----------------
  END COMM SECTION
  ----------------*/


/*--------------------
  BEGIN PACKET SECTION
  --------------------*/
#if !defined(MPICH_DEBUG_MEMINIT)
#   define MPIDI_Pkt_init(pkt_, type_)		\
    {						\
	(pkt_)->type = (type_);			\
    }
#else
#   define MPIDI_Pkt_init(pkt_, type_)				\
    {								\
	memset((void *) (pkt_), 0xfc, sizeof(MPIDI_CH3_Pkt_t));	\
	(pkt_)->type = (type_);					\
    }
#endif

/*------------------
  END PACKET SECTION
  ------------------*/


/*---------------------------
  BEGIN PROCESS GROUP SECTION
  ---------------------------*/
/* FIXME: Determine which of these functions should be exported to all of 
   the MPICH routines and which are internal to the device implementation */
typedef int (*MPIDI_PG_Compare_ids_fn_t)(void * id1, void * id2);
typedef int (*MPIDI_PG_Destroy_fn_t)(MPIDI_PG_t * pg);

int MPIDI_PG_Init( int *, char ***, 
		   MPIDI_PG_Compare_ids_fn_t, MPIDI_PG_Destroy_fn_t);
int MPIDI_PG_Finalize(void);
int MPIDI_PG_Create(int vct_sz, void * pg_id, MPIDI_PG_t ** ppg);
int MPIDI_PG_Destroy(MPIDI_PG_t * pg);
int MPIDI_PG_Find(void * id, MPIDI_PG_t ** pgp);
int MPIDI_PG_Id_compare(void *id1, void *id2);
595
596
597
598
599
600
601
602
603

/* Always use the MPIDI_PG_iterator type, never its expansion.  Otherwise it
   will be difficult to make any changes later. */
typedef MPIDI_PG_t * MPIDI_PG_iterator;
/* 'iter' is similar to 'saveptr' in strtok_r */
int MPIDI_PG_Get_iterator(MPIDI_PG_iterator *iter);
int MPIDI_PG_Has_next(MPIDI_PG_iterator *iter);
int MPIDI_PG_Get_next(MPIDI_PG_iterator *iter, MPIDI_PG_t **pgp);

604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
int MPIDI_PG_Close_VCs( void );

int MPIDI_PG_InitConnKVS( MPIDI_PG_t * );
int MPIDI_PG_GetConnKVSname( char ** );
int MPIDI_PG_InitConnString( MPIDI_PG_t * );
int MPIDI_PG_GetConnString( MPIDI_PG_t *, int, char *, int );
int MPIDI_PG_Dup_vcr( MPIDI_PG_t *, int, struct MPIDI_VC ** );
int MPIDI_PG_Get_size(MPIDI_PG_t * pg);
void MPIDI_PG_IdToNum( MPIDI_PG_t *, int * );
int MPIU_PG_Printall( FILE * );
int MPIDI_PG_CheckForSingleton( void );

/* CH3_PG_Init allows the channel to pre-initialize the process group */
int MPIDI_CH3_PG_Init( MPIDI_PG_t * );

#define MPIDI_PG_add_ref(pg_)			\
620
do {                                            \
621
    MPIU_Object_add_ref(pg_);			\
622
} while (0)
623
#define MPIDI_PG_release_ref(pg_, inuse_)	\
624
do {                                            \
625
    MPIU_Object_release_ref(pg_, inuse_);	\
626
} while (0)
627

628
629
#define MPIDI_PG_Get_vc(pg_, rank_, vcp_) *(vcp_) = &(pg_)->vct[rank_]

630
631
632
633
#define MPIDI_PG_Get_vc_set_active(pg_, rank_, vcp_)  do {              \
        *(vcp_) = &(pg_)->vct[rank_];                                   \
        if ((*(vcp_))->state == MPIDI_VC_STATE_INACTIVE)                \
        {                                                               \
634
            MPIDI_DBG_PrintVCState2(*(vcp_), MPIDI_VC_STATE_ACTIVE);     \
635
636
637
            MPIDI_CHANGE_VC_STATE((*(vcp_)), ACTIVE);                   \
        }                                                               \
    } while(0)
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661

#define MPIDI_PG_Get_size(pg_) ((pg_)->size)

#ifdef MPIDI_DEV_IMPLEMENTS_KVS
int MPIDI_PG_To_string(MPIDI_PG_t *pg_ptr, char **str_ptr, int *);
int MPIDI_PG_Create_from_string(const char * str, MPIDI_PG_t ** pg_pptr, 
				int *flag);
#endif
/*-------------------------
  END PROCESS GROUP SECTION
  -------------------------*/


/*--------------------------------
  BEGIN VIRTUAL CONNECTION SECTION
  --------------------------------*/
/*E
  MPIDI_VC_State - States for a virtual connection.
 
  Notes:
  A closed connection is placed into 'STATE_INACTIVE'. (is this true?)
 E*/
typedef enum MPIDI_VC_State
{
662
663
664
665
666
667
668
669
    MPIDI_VC_STATE_INACTIVE=1,      /* Comm either hasn't started or has completed. */
    MPIDI_VC_STATE_ACTIVE,          /* Comm has started and hasn't completed */
    MPIDI_VC_STATE_LOCAL_CLOSE,     /* Local side has initiated close protocol */
    MPIDI_VC_STATE_REMOTE_CLOSE,    /* Remote side has initiated close protocol */
    MPIDI_VC_STATE_CLOSE_ACKED,     /* Both have initiated close, we have acknowledged remote side */
    MPIDI_VC_STATE_CLOSED,          /* Both have initiated close, both have acked */
    MPIDI_VC_STATE_INACTIVE_CLOSED, /* INACTIVE VCs are moved to this state in Finalize */
    MPIDI_VC_STATE_MORIBUND         /* Abnormally terminated, there may be unsent/unreceived msgs */
670
671
672
673
} MPIDI_VC_State_t;

struct MPID_Comm;

674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
#ifdef ENABLE_COMM_OVERRIDES
typedef struct MPIDI_Comm_ops
{
    /* Overriding calls in case of matching-capable interfaces */
    int (*recv_posted)(struct MPIDI_VC *vc, struct MPID_Request *req);
    
    int (*send)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		int dest, int tag, MPID_Comm *comm, int context_offset,
		struct MPID_Request **request);
    int (*rsend)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		 int dest, int tag, MPID_Comm *comm, int context_offset,
		 struct MPID_Request **request);
    int (*ssend)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		 int dest, int tag, MPID_Comm *comm, int context_offset,
		 struct MPID_Request **request );
    int (*isend)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		 int dest, int tag, MPID_Comm *comm, int context_offset,
		 struct MPID_Request **request );
    int (*irsend)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		  int dest, int tag, MPID_Comm *comm, int context_offset,
		  struct MPID_Request **request );
    int (*issend)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		  int dest, int tag, MPID_Comm *comm, int context_offset,
		  struct MPID_Request **request );
    
    int (*send_init)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		     int dest, int tag, MPID_Comm *comm, int context_offset,
		     struct MPID_Request **request );
    int (*bsend_init)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		      int dest, int tag, MPID_Comm *comm, int context_offset,
		      struct MPID_Request **request);
    int (*rsend_init)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		      int dest, int tag, MPID_Comm *comm, int context_offset,
		      struct MPID_Request **request );
    int (*ssend_init)(struct MPIDI_VC *vc, const void *buf, int count, MPI_Datatype datatype,
		      int dest, int tag, MPID_Comm *comm, int context_offset,
		      struct MPID_Request **request );
    int (*startall)(struct MPIDI_VC *vc, int count,  struct MPID_Request *requests[]);
    
    int (*cancel_send)(struct MPIDI_VC *vc,  struct MPID_Request *sreq);
    int (*cancel_recv)(struct MPIDI_VC *vc,  struct MPID_Request *rreq);
715
716
717
718
719
    
    int (*probe)(struct MPIDI_VC *vc,  int source, int tag, MPID_Comm *comm, int context_offset,
		                  MPI_Status *status);
    int (*iprobe)(struct MPIDI_VC *vc,  int source, int tag, MPID_Comm *comm, int context_offset,
		  int *flag, MPI_Status *status);
720
721
    int (*improbe)(struct MPIDI_VC *vc,  int source, int tag, MPID_Comm *comm, int context_offset,
                   int *flag, MPID_Request **message, MPI_Status *status);
722
} MPIDI_Comm_ops_t;
723
724
725

extern int (*MPIDI_Anysource_iprobe_fn)(int tag, MPID_Comm * comm, int context_offset, int *flag,
                                        MPI_Status * status);
726
727
728
extern int (*MPIDI_Anysource_improbe_fn)(int tag, MPID_Comm * comm, int context_offset,
                                         int *flag, MPID_Request **message,
                                         MPI_Status * status);
729
730
#endif

731
732
733
734
735
736
737
738
739
typedef struct MPIDI_VC
{
    /* XXX - need better comment */
    /* MPIU_Object fields.  MPIDI_VC_t objects are not allocated using the 
       MPIU_Object system, but we do use the associated
       reference counting routines.  The handle value is required 
       when debugging objects (the handle kind is used in reporting
       on changes to the object).
    */
740
    MPIU_OBJECT_HEADER; /* adds handle and ref_count fields */
741
742
743
744
745
746
747
748
749
750
751
752

    /* state of the VC */
    MPIDI_VC_State_t state;

    /* Process group to which this VC belongs */
    struct MPIDI_PG * pg;

    /* Rank of the process in that process group associated with this VC */
    int pg_rank;

    /* Local process ID */
    int lpid;
753

754
755
756
    /* The node id of this process, used for topologically aware collectives. */
    MPID_Node_id_t node_id;

757
758
    /* port name tag */ 
    int port_name_tag; /* added to handle dynamic process mgmt */
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
    
#if defined(MPID_USE_SEQUENCE_NUMBERS)
    /* Sequence number of the next packet to be sent */
    MPID_Seqnum_t seqnum_send;
#endif
    
#if defined(MPIDI_CH3_MSGS_UNORDERED)
    /* Sequence number of the next packet we expect to receive */
    MPID_Seqnum_t seqnum_recv;

    /* Queue for holding packets received out of order.  NOTE: the CH3 device 
       only orders packets.  Handling of out-of-order data
       is the responsibility of the channel. */
    MPIDI_CH3_Pkt_send_container_t * msg_reorder_queue;
#endif

    /* rendezvous function pointers.  Called to send a rendevous
       message or when one is matched */
    int (* rndvSend_fn)( struct MPID_Request **sreq_p, const void * buf, int count, 
                         MPI_Datatype datatype, int dt_contig, MPIDI_msg_sz_t data_sz, 
                         MPI_Aint dt_true_lb, int rank, int tag,
                         struct MPID_Comm * comm, int context_offset );
    int (* rndvRecv_fn)( struct MPIDI_VC * vc, struct MPID_Request *rreq );

    /* eager message threshold */
    int eager_max_msg_sz;
785
786
787
    /* eager message threshold for ready sends.  -1 means there's no limit */
    int ready_eager_max_msg_sz;
 
788
789
790
791
792
793
    /* noncontiguous send function pointer.  Called to send a
       noncontiguous message.  Caller must initialize
       sreq->dev.segment, _first and _size.  Contiguous messages are
       called directly from CH3 and cannot be overridden. */
    int (* sendNoncontig_fn)( struct MPIDI_VC *vc, struct MPID_Request *sreq,
			      void *header, MPIDI_msg_sz_t hdr_sz );
794

795
#ifdef ENABLE_COMM_OVERRIDES
796
    MPIDI_Comm_ops_t *comm_ops;
797
798
#endif

799
800
801
802
803
    /* Rather than have each channel define its own fields for the 
       channel-specific data, we provide a fixed-sized scratchpad.  Currently,
       this has a very generous size, though this may shrink later (a channel
       can always allocate storage and hang it off of the end).  This 
       is necessary to allow dynamic loading of channels at MPI_Init time. */
804
805
806
    /* The ssm channel needed a *huge* space for the VC.  But now that it's
       gone, we need to determine the appropriate smaller size to use instead. */
    /* Note also that for dynamically-loaded channels, the VCs must all be the
807
808
809
       same size, so MPIDI_CH3_VC_SIZE should not be overridden when building
       multiple channels that will be used together */
#ifndef MPIDI_CH3_VC_SIZE
810
#define MPIDI_CH3_VC_SIZE 256
811
#endif
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
    int32_t channel_private[MPIDI_CH3_VC_SIZE];
# if defined(MPIDI_CH3_VC_DECL)
    MPIDI_CH3_VC_DECL
# endif
}
MPIDI_VC_t;

typedef enum MPIDI_VC_Event
{
    MPIDI_VC_EVENT_TERMINATED
}
MPIDI_VC_Event_t;

#ifndef HAVE_MPIDI_VCRT
#define HAVE_MPIDI_VCRT
typedef struct MPIDI_VCRT * MPID_VCRT;
typedef struct MPIDI_VC * MPID_VCR;
#endif

831
832
/* number of VCs that are in MORIBUND state */
extern int MPIDI_Failed_vc_count;
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848

/* Initialize a new VC */
int MPIDI_VC_Init( MPIDI_VC_t *, MPIDI_PG_t *, int );

#if defined(MPIDI_CH3_MSGS_UNORDERED)
#   define MPIDI_VC_Init_seqnum_recv(vc_);	\
    {						\
    	(vc_)->seqnum_recv = 0;			\
    	(vc_)->msg_reorder_queue = NULL;	\
    }
#else
#   define MPIDI_VC_Init_seqnum_recv(vc_);
#endif


#define MPIDI_VC_add_ref( _vc )                                 \
849
    do { MPIU_Object_add_ref( _vc ); } while (0)
850
851

#define MPIDI_VC_release_ref( _vc, _inuse ) \
852
    do { MPIU_Object_release_ref( _vc, _inuse ); } while (0)
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
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
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
961
962
963
964
965
966
967
968
969
970

/*------------------------------
  END VIRTUAL CONNECTION SECTION
  ------------------------------*/


/*---------------------------------
  BEGIN SEND/RECEIVE BUFFER SECTION
  ---------------------------------*/
#if !defined(MPIDI_CH3U_Offsetof)
#    define MPIDI_CH3U_Offsetof(struct_, field_) ((MPI_Aint) &((struct_*)0)->field_)
#endif

#if !defined(MPIDI_CH3U_SRBuf_size)
#    define MPIDI_CH3U_SRBuf_size (256 * 1024)
#endif

typedef struct __MPIDI_CH3U_SRBuf_element {
    /* Keep the buffer at the top to help keep the memory alignment */
    char   buf[MPIDI_CH3U_SRBuf_size];
    struct __MPIDI_CH3U_SRBuf_element * next;
} MPIDI_CH3U_SRBuf_element_t;

extern MPIDI_CH3U_SRBuf_element_t * MPIDI_CH3U_SRBuf_pool;

#if !defined (MPIDI_CH3U_SRBuf_get)
#   define MPIDI_CH3U_SRBuf_get(req_)                                   \
    {                                                                   \
        MPIDI_CH3U_SRBuf_element_t * tmp;                               \
        if (!MPIDI_CH3U_SRBuf_pool) {                                   \
             MPIDI_CH3U_SRBuf_pool =                                    \
                MPIU_Malloc(sizeof(MPIDI_CH3U_SRBuf_element_t));        \
            MPIDI_CH3U_SRBuf_pool->next = NULL;                         \
        }                                                               \
        tmp = MPIDI_CH3U_SRBuf_pool;                                    \
        MPIDI_CH3U_SRBuf_pool = MPIDI_CH3U_SRBuf_pool->next;            \
        tmp->next = NULL;                                               \
        (req_)->dev.tmpbuf = tmp->buf;                                  \
    }
#endif

#if !defined (MPIDI_CH3U_SRBuf_free)
#   define MPIDI_CH3U_SRBuf_free(req_)                                  \
    {                                                                   \
        MPIDI_CH3U_SRBuf_element_t * tmp;                               \
        MPIU_Assert(MPIDI_Request_get_srbuf_flag(req_));                \
        MPIDI_Request_set_srbuf_flag((req_), FALSE);                    \
        tmp = (MPIDI_CH3U_SRBuf_element_t *) (((MPI_Aint) ((req_)->dev.tmpbuf)) - \
               ((MPI_Aint) MPIDI_CH3U_Offsetof(MPIDI_CH3U_SRBuf_element_t, buf))); \
        tmp->next = MPIDI_CH3U_SRBuf_pool;                              \
        MPIDI_CH3U_SRBuf_pool = tmp;                                    \
    }
#endif

#if !defined(MPIDI_CH3U_SRBuf_alloc)
#   define MPIDI_CH3U_SRBuf_alloc(req_, size_)				\
    {									\
        MPIDI_CH3U_SRBuf_get(req_);                                     \
 	if ((req_)->dev.tmpbuf != NULL)					\
 	{								\
 	    (req_)->dev.tmpbuf_sz = MPIDI_CH3U_SRBuf_size;		\
 	    MPIDI_Request_set_srbuf_flag((req_), TRUE);			\
 	}								\
 	else								\
 	{								\
 	    (req_)->dev.tmpbuf_sz = 0;					\
 	}								\
    }
#endif
/*-------------------------------
  END SEND/RECEIVE BUFFER SECTION
  -------------------------------*/

/*----------------------------
  BEGIN DEBUGGING TOOL SECTION
  ----------------------------*/

/* If there is no support for dynamic processes, there will be no
   channel-specific connection state */
#ifdef USE_DBG_LOGGING

#ifdef MPIDI_CH3_HAS_NO_DYNAMIC_PROCESS
#define MPIDI_CH3_VC_GetStateString( _c ) "none"
#else
/* FIXME: This duplicates a value in util/sock/ch3usock.h */
const char *MPIDI_CH3_VC_GetStateString(struct MPIDI_VC *);
const char *MPIDI_CH3_VC_SockGetStateString(struct MPIDI_VC *);
#endif

/* These tw routines are in mpidi_pg.c and are used to print the 
   connection string (which is attached to a process group) */
int MPIDI_PrintConnStr( const char *file, int line, 
			const char *label, const char *str );
int MPIDI_PrintConnStrToFile( FILE *fd, const char *file, int line, 
			      const char *label, const char *str );
#endif

/* These macros simplify and unify the debugging of changes in the
   connection state 

   MPIU_DBG_VCSTATECHANGE(vc,newstate) - use when changing the state
   of a VC

   MPIU_DBG_VCCHSTATECHANGE(vc,newstate) - use when changing the state
   of the channel-specific part of the vc (e.g., vc->ch.state)

   MPIU_DBG_CONNSTATECHANGE(vc,conn,newstate ) - use when changing the
   state of a conn.  vc may be null

   MPIU_DBG_CONNSTATECHANGEMSG(vc,conn,newstate,msg ) - use when changing the
   state of a conn.  vc may be null.  Like CONNSTATECHANGE, but allows
   an additional message

   MPIU_DBG_PKT(conn,pkt,msg) - print out a short description of an
   packet being sent/received on the designated connection, prefixed with
   msg.

*/
971
#define MPIU_DBG_VCSTATECHANGE(_vc,_newstate) do { \
972
973
974
     MPIU_DBG_MSG_FMT(CH3_CONNECT,TYPICAL,(MPIU_DBG_FDEST, \
     "vc=%p: Setting state (vc) from %s to %s, vcchstate is %s", \
                 _vc, MPIDI_VC_GetStateString((_vc)->state), \
975
                 #_newstate, MPIDI_CH3_VC_GetStateString( (_vc) ))); \
976
} while (0)
977
978
979
980

#define MPIU_DBG_VCCHSTATECHANGE(_vc,_newstate) \
     MPIU_DBG_MSG_FMT(CH3_CONNECT,TYPICAL,(MPIU_DBG_FDEST, \
     "vc=%p: Setting state (ch) from %s to %s, vc state is %s", \
981
           _vc, MPIDI_CH3_VC_GetStateString((_vc)), \
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
           #_newstate, MPIDI_VC_GetStateString( (_vc)->state )) )

#define MPIU_DBG_CONNSTATECHANGE(_vc,_conn,_newstate) \
     MPIU_DBG_MSG_FMT(CH3_CONNECT,TYPICAL,(MPIU_DBG_FDEST, \
     "vc=%p,conn=%p: Setting state (conn) from %s to %s, vcstate = %s", \
             _vc, _conn, \
             MPIDI_Conn_GetStateString((_conn)->state), #_newstate, \
             _vc ? MPIDI_VC_GetStateString((_vc)->state) : "<no vc>" ))

#define MPIU_DBG_CONNSTATECHANGE_MSG(_vc,_conn,_newstate,_msg) \
     MPIU_DBG_MSG_FMT(CH3_CONNECT,TYPICAL,(MPIU_DBG_FDEST, \
     "vc=%p,conn=%p: Setting conn state from %s to %s, vcstate = %s %s", \
             _vc, _conn, \
             MPIDI_Conn_GetStateString((_conn)->state), #_newstate, \
             _vc ? MPIDI_VC_GetStateString((_vc)->state) : "<no vc>", _msg ))
#define MPIU_DBG_VCUSE(_vc,_msg) \
     MPIU_DBG_MSG_FMT(CH3_CONNECT,TYPICAL,(MPIU_DBG_FDEST,\
      "vc=%p: Using vc for %s", _vc, _msg ))
#define MPIU_DBG_PKT(_conn,_pkt,_msg) \
     MPIU_DBG_MSG_FMT(CH3_OTHER,TYPICAL,(MPIU_DBG_FDEST,\
     "conn=%p: %s %s", _conn, _msg, MPIDI_Pkt_GetDescString( _pkt ) ))

const char *MPIDI_Pkt_GetDescString( MPIDI_CH3_Pkt_t *pkt );

/* These macros help trace communication headers */
#define MPIU_DBG_MSGPKT(_vc,_tag,_contextid,_dest,_size,_kind)	\
    MPIU_DBG_MSG_FMT(CH3_MSG,TYPICAL,(MPIU_DBG_FDEST,\
		      "%s: vc=%p, tag=%d, context=%d, dest=%d, datasz=" MPIDI_MSG_SZ_FMT,\
		      _kind,_vc,_tag,_contextid,_dest,_size) )

/* FIXME: Switch this to use the common debug code */
void MPIDI_dbg_printf(int, char *, char *, ...);
void MPIDI_err_printf(char *, char *, ...);

/* FIXME: This does not belong here */
1017
1018
#ifdef USE_MPIDI_DBG_PRINT_VC
extern char *MPIDI_DBG_parent_str;
1019
1020
1021
1022
1023
#endif

#if defined(MPICH_DBG_OUTPUT)
#define MPIDI_DBG_PRINTF(e_)				\
{                                               	\
1024
    if (MPIU_dbg_state != MPIU_DBG_STATE_NONE)		\
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
    {							\
	MPIDI_dbg_printf e_;				\
    }							\
}
#else
#   define MPIDI_DBG_PRINTF(e)
#endif

#define MPIDI_ERR_PRINTF(e) MPIDI_err_printf e

#if defined(HAVE_CPP_VARARGS)
#   define MPIDI_dbg_printf(level, func, fmt, args...)			\
    {									\
    	MPIU_dbglog_printf("[%d] %s(): " fmt "\n", MPIR_Process.comm_world->rank, func, ## args);	\
    }
#   define MPIDI_err_printf(func, fmt, args...)				\
    {									\
    	MPIU_Error_printf("[%d] ERROR - %s(): " fmt "\n", MPIR_Process.comm_world->rank, func, ## args);	\
    	fflush(stdout);							\
    }
#endif

/* This is used to quote a name in a definition (see FUNCNAME/FCNAME below) */
#define MPIDI_QUOTE(A) MPIDI_QUOTE2(A)
#define MPIDI_QUOTE2(A) #A

#ifdef MPICH_DBG_OUTPUT
    void MPIDI_DBG_Print_packet(MPIDI_CH3_Pkt_t *pkt);
#else
#   define MPIDI_DBG_Print_packet(a)
#endif

/* Given a state, return the string for this state (VC's and connections) */
const char * MPIDI_VC_GetStateString(int);
/*--------------------------
  END DEBUGGING TOOL SECTION
  --------------------------*/


/* Prototypes for internal device routines */
int MPIDI_Isend_self(const void *, int, MPI_Datatype, int, int, MPID_Comm *, 
		     int, int, MPID_Request **);

/*--------------------------
  BEGIN MPI PORT SECTION 
  --------------------------*/
/* These are the default functions */
int MPIDI_Comm_connect(const char *, MPID_Info *, int, MPID_Comm *, MPID_Comm **);
int MPIDI_Comm_accept(const char *, MPID_Info *, int, MPID_Comm *, MPID_Comm **);

1075
int MPIDI_Comm_spawn_multiple(int, char **, char ***, const int *, MPID_Info **,
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
			      int, MPID_Comm *, MPID_Comm **, int *);


/* This structure defines a module that handles the routines that 
   work with MPI port names */
typedef struct MPIDI_Port_Ops {
    int (*OpenPort)( MPID_Info *, char * );
    int (*ClosePort)( const char * );
    int (*CommAccept)( const char *, MPID_Info *, int, MPID_Comm *, 
		       MPID_Comm ** );
    int (*CommConnect)( const char *, MPID_Info *, int, MPID_Comm *, 
			MPID_Comm ** );
} MPIDI_PortFns;
#define MPIDI_PORTFNS_VERSION 1
int MPIDI_CH3_PortFnsInit( MPIDI_PortFns * );

/* Utility routines provided in src/ch3u_port.c for working with connection
   queues */
int MPIDI_CH3I_Acceptq_enqueue(MPIDI_VC_t * vc, int port_name_tag);
int MPIDI_CH3I_Acceptq_dequeue(MPIDI_VC_t ** vc, int port_name_tag);
#ifdef MPIDI_CH3_CHANNEL_AVOIDS_SELECT
int MPIDI_CH3_Complete_Acceptq_dequeue(MPIDI_VC_t * vc);
#else
#define MPIDI_CH3_Complete_Acceptq_dequeue(vc)  MPI_SUCCESS
#endif
/*--------------------------
  END MPI PORT SECTION 
  --------------------------*/

/* part of mpid_vc.c, this routine completes any pending operations 
   on a communicator */
int MPIDI_CH3U_Comm_FinishPending( MPID_Comm * );

#define MPIDI_MAX_KVS_VALUE_LEN    4096

/* ------------------------------------------------------------------------- */
/* mpirma.h (in src/mpi/rma?) */
/* ------------------------------------------------------------------------- */

1115
1116
1117
1118
1119
1120
1121
1122
/* The Win_fns table contains pointers to the channel's implementation of the
 * RMA window creation routines.  The channel must provide the init function,
 * which can optionally override any defaults already set by CH3.
 */

typedef struct MPIDI_CH3U_Win_fns_s {
    int (*create)(void *, MPI_Aint, int, MPID_Info *, MPID_Comm *, MPID_Win **);
    int (*allocate)(MPI_Aint, int, MPID_Info *, MPID_Comm *, void *, MPID_Win **);
1123
    int (*allocate_shared)(MPI_Aint, int, MPID_Info *, MPID_Comm *, void **, MPID_Win **);
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
    int (*create_dynamic)(MPID_Info *, MPID_Comm *, MPID_Win **);
} MPIDI_CH3U_Win_fns_t;

extern MPIDI_CH3U_Win_fns_t MPIDI_CH3U_Win_fns;

/* CH3 and Channel window functions initializers */
int MPIDI_Win_fns_init(MPIDI_CH3U_Win_fns_t *win_fns);
int MPIDI_CH3_Win_fns_init(MPIDI_CH3U_Win_fns_t *win_fns);

/* Default window creation functions provided by CH3 */
int MPIDI_CH3U_Win_create(void *, MPI_Aint, int, MPID_Info *, MPID_Comm *,
                         MPID_Win **);
int MPIDI_CH3U_Win_allocate(MPI_Aint size, int disp_unit, MPID_Info *info,
                           MPID_Comm *comm, void *baseptr, MPID_Win **win);
1138
1139
int MPIDI_CH3U_Win_allocate_shared(MPI_Aint size, int disp_unit, MPID_Info *info_ptr,
                                   MPID_Comm *comm_ptr, void **baseptr, MPID_Win **win_ptr);
1140
1141
1142
1143
1144
1145
1146
1147
1148
int MPIDI_CH3U_Win_create_dynamic(MPID_Info *info, MPID_Comm *comm, MPID_Win **win);


/* MPI RMA Utility functions */

int MPIDI_CH3U_Win_create_gather(void *, MPI_Aint, int, MPID_Info *, MPID_Comm *,
                                 MPID_Win **);


1149
/* MPI-2 RMA Routines */
1150
1151

int MPIDI_Win_create(void *, MPI_Aint, int, MPID_Info *, MPID_Comm *,
1152
1153
1154
                     MPID_Win **);
int MPIDI_Win_free(MPID_Win **); 

1155
int MPIDI_Put(const void *, int, MPI_Datatype, int, MPI_Aint, int,
1156
              MPI_Datatype, MPID_Win *); 
1157
int MPIDI_Get(void *, int, MPI_Datatype, int, MPI_Aint, int,
1158
              MPI_Datatype, MPID_Win *);
1159
int MPIDI_Accumulate(const void *, int, MPI_Datatype, int, MPI_Aint, int,
1160
1161
1162
                     MPI_Datatype, MPI_Op, MPID_Win *);

int MPIDI_Win_fence(int, MPID_Win *);
1163
1164
int MPIDI_Win_post(MPID_Group *group_ptr, int assert, MPID_Win *win_ptr);
int MPIDI_Win_start(MPID_Group *group_ptr, int assert, MPID_Win *win_ptr);
1165
1166
1167
1168
int MPIDI_Win_test(MPID_Win *win_ptr, int *flag);
int MPIDI_Win_wait(MPID_Win *win_ptr);
int MPIDI_Win_complete(MPID_Win *win_ptr);

1169
1170
int MPIDI_Win_lock(int lock_type, int dest, int assert, MPID_Win *win_ptr);
int MPIDI_Win_unlock(int dest, MPID_Win *win_ptr);
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180

/* MPI-3 RMA Routines */

int MPIDI_Win_allocate(MPI_Aint size, int disp_unit, MPID_Info *info,
                       MPID_Comm *comm, void *baseptr, MPID_Win **win);
int MPIDI_Win_allocate_shared(MPI_Aint size, MPID_Info *info_ptr, MPID_Comm *comm_ptr,
                              void **baseptr, MPID_Win **win_ptr);
int MPIDI_Win_create_dynamic(MPID_Info *info, MPID_Comm *comm, MPID_Win **win);
int MPIDI_Win_attach(MPID_Win *win, void *base, MPI_Aint size);
int MPIDI_Win_detach(MPID_Win *win, const void *base);
1181
int MPIDI_Win_shared_query(MPID_Win *win_ptr, int rank, MPI_Aint *size, int *disp_unit, void *base);
1182

1183
1184
1185
int MPIDI_Win_set_info(MPID_Win *win, MPID_Info *info);
int MPIDI_Win_get_info(MPID_Win *win, MPID_Info **info_used);

1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
int MPIDI_Get_accumulate(const void *origin_addr, int origin_count,
                         MPI_Datatype origin_datatype, void *result_addr, int result_count,
                         MPI_Datatype result_datatype, int target_rank, MPI_Aint target_disp,
                         int target_count, MPI_Datatype target_datatype, MPI_Op op, MPID_Win *win);
int MPIDI_Fetch_and_op(const void *origin_addr, void *result_addr,
                       MPI_Datatype datatype, int target_rank, MPI_Aint target_disp,
                       MPI_Op op, MPID_Win *win);
int MPIDI_Compare_and_swap(const void *origin_addr, const void *compare_addr,
                           void *result_addr, MPI_Datatype datatype, int target_rank,
                           MPI_Aint target_disp, MPID_Win *win);
int MPIDI_Rput(const void *origin_addr, int origin_count,
               MPI_Datatype origin_datatype, int target_rank, MPI_Aint target_disp,
               int target_count, MPI_Datatype target_datatype, MPID_Win *win,
1199
               MPID_Request **request);
1200
1201
1202
int MPIDI_Rget(void *origin_addr, int origin_count,
               MPI_Datatype origin_datatype, int target_rank, MPI_Aint target_disp,
               int target_count, MPI_Datatype target_datatype, MPID_Win *win,
1203
               MPID_Request **request);
1204
1205
1206
int MPIDI_Raccumulate(const void *origin_addr, int origin_count,
                      MPI_Datatype origin_datatype, int target_rank, MPI_Aint target_disp,
                      int target_count, MPI_Datatype target_datatype, MPI_Op op, MPID_Win *win,
1207
                      MPID_Request **request);
1208
1209
1210
1211
int MPIDI_Rget_accumulate(const void *origin_addr, int origin_count,
                          MPI_Datatype origin_datatype, void *result_addr, int result_count,
                          MPI_Datatype result_datatype, int target_rank, MPI_Aint target_disp,
                          int target_count, MPI_Datatype target_datatype, MPI_Op op, MPID_Win *win,
1212
                          MPID_Request **request);
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222

int MPIDI_Win_lock_all(int assert, MPID_Win *win);
int MPIDI_Win_unlock_all(MPID_Win *win);
int MPIDI_Win_flush(int rank, MPID_Win *win);
int MPIDI_Win_flush_all(MPID_Win *win);
int MPIDI_Win_flush_local(int rank, MPID_Win *win);
int MPIDI_Win_flush_local_all(MPID_Win *win);
int MPIDI_Win_sync(MPID_Win *win);


1223
1224
1225
1226
1227
1228
void *MPIDI_Alloc_mem(size_t size, MPID_Info *info_ptr);
int MPIDI_Free_mem(void *ptr);

/* internal */
int MPIDI_CH3I_Release_lock(MPID_Win * win_ptr);
int MPIDI_CH3I_Try_acquire_win_lock(MPID_Win * win_ptr, int requested_lock);
1229
1230
int MPIDI_CH3I_Send_lock_granted_pkt(MPIDI_VC_t * vc, MPID_Win *win_ptr, int source_win_hdl);
int MPIDI_CH3I_Send_pt_rma_done_pkt(MPIDI_VC_t * vc, MPID_Win *win_ptr, int source_win_hdl);
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269

#define MPIDI_CH3I_DATATYPE_IS_PREDEFINED(type, predefined) \
    if ((HANDLE_GET_KIND(type) == HANDLE_KIND_BUILTIN) || \
        (type == MPI_FLOAT_INT) || (type == MPI_DOUBLE_INT) || \
        (type == MPI_LONG_INT) || (type == MPI_SHORT_INT) || \
	(type == MPI_LONG_DOUBLE_INT)) \
        predefined = 1; \
    else predefined = 0;

int MPIDI_CH3I_Progress_finalize(void);

/*@
  MPIDI_CH3_Progress_signal_completion - Inform the progress engine that a 
  pending request has completed.

  IMPLEMENTORS:
  In a single-threaded environment, this routine can be implemented by
  incrementing a request completion counter.  In a
  multi-threaded environment, the request completion counter must be atomically
  incremented, and any threaded blocking in the
  progress engine must be woken up when a request is completed.

  Notes on the implementation:

  This code is designed to support one particular model of thread-safety.
  It is common to many of the channels and was moved into this file because
  the MPIDI_CH3_Progress_signal_completion reference is used by the 
  macro the implements MPID_Request_set_completed.  Note that there is 
  a function version of MPID_Request_set_completed for use by greq_complete.c

@*/

/*
 * MPIDI_CH3_Progress_signal_completion() is used to notify the progress
 * engine that a completion has occurred.  The multi-threaded version will need
 * to wake up any (and all) threads blocking in MPIDI_CH3_Progress().
 */

/* This allows the channel to define an alternate to the 
1270
   completion counter.  */
1271
#ifndef MPIDI_CH3I_INCR_PROGRESS_COMPLETION_COUNT
1272
1273
#define MPIDI_CH3I_INCR_PROGRESS_COMPLETION_COUNT                                \
    do {                                                                         \
1274
        MPIU_THREAD_CS_ENTER(COMPLETION,);                                       \
1275
1276
1277
1278
        ++MPIDI_CH3I_progress_completion_count;                                  \
        MPIU_DBG_MSG_D(CH3_PROGRESS,VERBOSE,                                     \
                     "just incremented MPIDI_CH3I_progress_completion_count=%d", \
                     MPIDI_CH3I_progress_completion_count);                      \
1279
        MPIU_THREAD_CS_EXIT(COMPLETION,);                                        \
1280
    } while (0)
1281
1282
#endif

1283

1284
1285
1286
/* The following is part of an implementation of a control of a 
   resource shared among threads - it needs to be managed more 
   explicitly as such as shared resource */
1287
1288
1289
1290
1291
1292
#ifndef MPICH_IS_THREADED
#   define MPIDI_CH3_Progress_signal_completion()	\
    {							\
       MPIDI_CH3I_INCR_PROGRESS_COMPLETION_COUNT;		\
    }
#else
1293
    /* TODO these decls should probably move into each channel as appropriate */
1294
1295
1296
    extern volatile int MPIDI_CH3I_progress_blocked;
    extern volatile int MPIDI_CH3I_progress_wakeup_signalled;

1297
1298
1299
1300
1301
1302
1303
1304
1305
/* This allows the channel to hook the MPIDI_CH3_Progress_signal_completion
 * macro when it is necessary to wake up some part of the progress engine from a
 * blocking operation.  Currently ch3:sock uses it, ch3:nemesis does not. */
/* MT alternative implementations of this macro are responsible for providing any
 * synchronization (acquiring MPIDCOMM, etc) */
#ifndef MPIDI_CH3I_PROGRESS_WAKEUP
# define MPIDI_CH3I_PROGRESS_WAKEUP do {/*do nothing*/} while(0)
#endif

1306
    void MPIDI_CH3I_Progress_wakeup(void);
1307
1308
1309
1310
1311
    /* MT TODO profiling is needed here.  We currently protect the completion
     * counter with the COMPLETION critical section, which could be a source of
     * contention.  It should be possible to peform these updates atomically via
     * OPA instead, but the additional complexity should be justified by
     * profiling evidence.  [goodell@ 2010-06-29] */
1312
#   define MPIDI_CH3_Progress_signal_completion()			\
1313
    do {                                                                \
1314
        MPIDI_CH3I_INCR_PROGRESS_COMPLETION_COUNT;                      \
1315
1316
        MPIDI_CH3I_PROGRESS_WAKEUP;                                     \
    } while (0)
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
#endif

/* Function that may be used to provide business card info */
int MPIDI_CH3I_BCInit( char **bc_val_p, int *val_max_sz_p);
/* Function to free the storage allocated by MPIDI_CH3I_BCInit */
int MPIDI_CH3I_BCFree( char *publish_bc );

/* Inform the process group of our connection information string (business
   card) */
int MPIDI_PG_SetConnInfo( int rank, const char *connString );

1328
1329
1330
/* Fill in the node_id information for each VC in the given PG. */
int MPIDI_Populate_vc_node_ids(MPIDI_PG_t *pg, int our_pg_rank);

1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
/* NOTE: Channel function prototypes are in mpidi_ch3_post.h since some of the 
   macros require their declarations. */

/* FIXME: These should be defined only when these particular utility
   packages are used.  Best would be to keep these prototypes in the
   related util/xxx directories, and either copy them into an include
   directory used only for builds or add (yet another) include path */
/* from util/sock */
int MPIDI_VC_InitSock( MPIDI_VC_t *);
int MPIDI_CH3I_Connect_to_root_sock(const char *, MPIDI_VC_t **);


int MPIDI_CH3I_VC_post_sockconnect(MPIDI_VC_t * );

/* FIXME: Where should this go? */

/* Used internally to broadcast process groups belonging to peercomm to
 all processes in comm*/
int MPID_PG_BCast( MPID_Comm *peercomm_p, MPID_Comm *comm_p, int root );

/* Channel defintitions */
/*@
  MPIDI_CH3_iStartMsg - A non-blocking request to send a CH3 packet.  A r
  equest object is allocated only if the send could not be completed 
  immediately.

  Input Parameters:
+ vc - virtual connection to send the message over
. pkt - pointer to a MPIDI_CH3_Pkt_t structure containing the substructure to 
  be sent
- pkt_sz - size of the packet substucture

  Output Parameters:
. sreq_ptr - send request or NULL if the send completed immediately

  Return value:
  An mpi error code.
  
  NOTE:
  The packet structure may be allocated on the stack.

  IMPLEMETORS:
  If the send can not be completed immediately, the CH3 packet structure must 
  be stored internally until the request is complete.
  
  If the send completes immediately, the channel implementation should return 
  NULL.
@*/
int MPIDI_CH3_iStartMsg(MPIDI_VC_t * vc, void * pkt, MPIDI_msg_sz_t pkt_sz, 
			MPID_Request **sreq_ptr);


/*@
  MPIDI_CH3_iStartMsgv - A non-blocking request to send a CH3 packet and 
  associated data.  A request object is allocated only if
  the send could not be completed immediately.

  Input Parameters:
+ vc - virtual connection to send the message over
. iov - a vector of a structure contains a buffer pointer and length
- iov_n - number of elements in the vector

  Output Parameters:
. sreq_ptr - send request or NULL if the send completed immediately

  Return value:
  An mpi error code.
  
  NOTE:
  The first element in the vector must point to the packet structure.   The 
  packet structure and the vector may be allocated on
  the stack.

  IMPLEMENTORS:
  If the send can not be completed immediately, the CH3 packet structure and 
  the vector must be stored internally until the
  request is complete.
  
  If the send completes immediately, the channel implementation should return 
  NULL.
@*/
int MPIDI_CH3_iStartMsgv(MPIDI_VC_t * vc, MPID_IOV * iov, int iov_n, 
			 MPID_Request **sreq_ptr);


/*@
  MPIDI_CH3_iSend - A non-blocking request to send a CH3 packet using an 
  existing request object.  When the send is complete
  the channel implementation will call the OnDataAvail routine in the
  request, if any (if not, the channel implementation will mark the 
  request as complete).

  Input Parameters:
+ vc - virtual connection over which to send the CH3 packet
. sreq - pointer to the send request object
. pkt - pointer to a MPIDI_CH3_Pkt_t structure containing the substructure to 
  be sent
- pkt_sz - size of the packet substucture

  Return value:
  An mpi error code.
  
  NOTE:
  The packet structure may be allocated on the stack.

  IMPLEMETORS:
  If the send can not be completed immediately, the packet structure must be 
  stored internally until the request is complete.

  If the send completes immediately, the channel implementation still must 
  invoke the OnDataAvail routine in the request, if any; otherwise, is 
  must set the request as complete.
@*/
int MPIDI_CH3_iSend(MPIDI_VC_t * vc, MPID_Request * sreq, void * pkt, 
		    MPIDI_msg_sz_t pkt_sz);


/*@
  MPIDI_CH3_iSendv - A non-blocking request to send a CH3 packet and 
  associated data using an existing request object.  When
  the send is complete the channel implementation will call the
  OnDataAvail routine in the request, if any.

  Input Parameters:
+ vc - virtual connection over which to send the CH3 packet and data
. sreq - pointer to the send request object
. iov - a vector of a structure contains a buffer pointer and length
- iov_n - number of elements in the vector

  Return value:
  An mpi error code.
  
  NOTE:
  The first element in the vector must point to the packet structure.   The 
  packet structure and the vector may be allocated on
  the stack.

  IMPLEMENTORS:
  If the send can not be completed immediately, the packet structure and the 
  vector must be stored internally until the request is
  complete.

  If the send completes immediately, the channel implementation still must 
  call the OnDataAvail routine in the request, if any.
@*/
int MPIDI_CH3_iSendv(MPIDI_VC_t * vc, MPID_Request * sreq, MPID_IOV * iov, 
		     int iov_n);

/*@
  MPIDI_CH3_Connection_terminate - terminate the underlying connection 
  associated with the specified VC

  Input Parameters:
. vc - virtual connection

  Return value:
  An MPI error code
@*/
int MPIDI_CH3_Connection_terminate(MPIDI_VC_t * vc);

/* MPIDI_CH3_Connect_to_root (really connect to peer) - channel routine
   for connecting to a process through a port, used in implementing
   MPID_Comm_connect and accept */
int MPIDI_CH3_Connect_to_root(const char *, MPIDI_VC_t **);

/*
 * Channel utility prototypes
 */
1499
int MPIDI_CH3U_Recvq_init(void);
1500
1501
int MPIDI_CH3U_Recvq_FU(int, int, int, MPI_Status * );
MPID_Request * MPIDI_CH3U_Recvq_FDU(MPI_Request, MPIDI_Message_match *);
1502
1503
MPID_Request * MPIDI_CH3U_Recvq_FDU_matchonly(int source, int tag, int context_id, MPID_Comm *comm,
                                   int *foundp);
1504
1505
1506
MPID_Request * MPIDI_CH3U_Recvq_FDU_or_AEP(int source, int tag, 
                                          int context_id, MPID_Comm *comm, void *user_buf,
                                           int user_count, MPI_Datatype datatype, int * foundp);
1507
1508
1509
int MPIDI_CH3U_Recvq_DP(MPID_Request * rreq);
MPID_Request * MPIDI_CH3U_Recvq_FDP_or_AEU(MPIDI_Message_match * match, 
					   int * found);
1510
int MPIDI_CH3U_Recvq_count_unexp(void);
1511
int MPIDI_CH3U_Complete_posted_with_error(MPIDI_VC_t *vc);
1512
int MPIDI_CH3U_Complete_disabled_anysources(void);
1513

1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530

int MPIDI_CH3U_Request_load_send_iov(MPID_Request * const sreq, 
				     MPID_IOV * const iov, int * const iov_n);
int MPIDI_CH3U_Request_load_recv_iov(MPID_Request * const rreq);
int MPIDI_CH3U_Request_unpack_uebuf(MPID_Request * rreq);
int MPIDI_CH3U_Request_unpack_srbuf(MPID_Request * rreq);

void MPIDI_CH3U_Buffer_copy(const void * const sbuf, int scount, 
			    MPI_Datatype sdt, int * smpi_errno,
			    void * const rbuf, int rcount, MPI_Datatype rdt, 
			    MPIDI_msg_sz_t * rdata_sz, int * rmpi_errno);
int MPIDI_CH3U_Post_data_receive(int found, MPID_Request ** rreqp);
int MPIDI_CH3U_Post_data_receive_found(MPID_Request * rreqp);
int MPIDI_CH3U_Post_data_receive_unexpected(MPID_Request * rreqp);
int MPIDI_CH3U_Receive_data_found(MPID_Request *rreq, char *buf, MPIDI_msg_sz_t *buflen, int *complete);
int MPIDI_CH3U_Receive_data_unexpected(MPID_Request * rreq, char *buf, MPIDI_msg_sz_t *buflen, int *complete);

1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
/* Initialization routine for ch3u_comm.c */
int MPIDI_CH3I_Comm_init(void);

int MPIDI_CH3I_Comm_handle_failed_procs(MPID_Group *new_failed_procs);

/* The functions below allow channels to register functions to be
   called immediately after a communicator has been created, and
   immediately before a communicator is to be destroyed.
 */
int MPIDI_CH3U_Comm_register_create_hook(int (*hook_fn)(struct MPID_Comm *, void *), void *param);
int MPIDI_CH3U_Comm_register_destroy_hook(int (*hook_fn)(struct MPID_Comm *, void *), void *param);

1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
/* FIXME: This is a macro! */
#ifndef MPIDI_CH3_Request_add_ref
/*@
  MPIDI_CH3_Request_add_ref - Increment the reference count associated with a
  request object

  Input Parameters:
. req - pointer to the request object
@*/
void MPIDI_CH3_Request_add_ref(MPID_Request * req);
#endif

/*@
  MPIDI_CH3_GetParentPort - obtain the port name associated with the parent

  Output Parameters:
.  parent_port_name - the port name associated with the parent communicator

  Return value:
  A MPI error code.
  
  NOTE:
  'MPIDI_CH3_GetParentPort' should only be called if the initialization
  (in the current implementation, done with the static function 
  'InitPGFromPMI' in 'mpid_init.c') has determined that this process
  in fact has a parent.
@*/
int MPIDI_CH3_GetParentPort(char ** parent_port_name);

/*@
   MPIDI_CH3_FreeParentPort - This routine frees the storage associated with
   a parent port (allocted with MPIDH_CH3_GetParentPort).

  @*/
void MPIDI_CH3_FreeParentPort( void );

/*E
  MPIDI_CH3_Abort - Abort this process.

  Input Parameters:
+ exit_code - exit code to be returned by the process
- error_msg - error message to print

  Return value:
  This function should not return.

  Notes:
  This routine is used only if the channel defines
  'MPIDI_CH3_IMPLEMENTS_ABORT'.  This allows the channel to handle 
  aborting processes, particularly when the channel does not use the standard
  PMI interface.
E*/
int MPIDI_CH3_Abort(int exit_code, char * error_msg);

/* FIXME: Move these prototypes into header files in the appropriate 
   util directories  */
/* added by brad.  upcalls for MPIDI_CH3_Init that contain code which could be 
   executed by two or more channels */
int MPIDI_CH3U_Init_sock(int has_parent, MPIDI_PG_t * pg_p, int pg_rank,
                         char **bc_val_p, int *val_max_sz_p);

/* added by brad.  business card related global and functions */
/* FIXME: Make these part of the channel support headers */
#define MAX_HOST_DESCRIPTION_LEN 256
int MPIDI_CH3U_Get_business_card_sock(int myRank, 
				      char **bc_val_p, int *val_max_sz_p);

int MPIDI_CH3_Get_business_card(int myRank, char *value, int length);

/*
 * Channel upcall prototypes
 */

/*E
  MPIDI_CH3U_Handle_recv_pkt- Handle a freshly received CH3 packet.

  Input Parameters:
+ vc - virtual connection over which the packet was received
- pkt - pointer to the CH3 packet

  Output Parameter:
. rreqp - receive request defining data to be received; may be NULL

  NOTE:
  Multiple threads may not simultaneously call this routine with the same 
  virtual connection.  This constraint eliminates the
  need to lock the VC and thus improves performance.  If simultaneous upcalls 
  for a single VC are a possible, then the calling
  routine must serialize the calls (perhaps by locking the VC).  Special 
  consideration may need to be given to packet ordering
  if the channel has made guarantees about ordering.
E*/
int MPIDI_CH3U_Handle_recv_pkt(MPIDI_VC_t * vc, MPIDI_CH3_Pkt_t * pkt, 
			       MPIDI_msg_sz_t *buflen, MPID_Request ** rreqp);

/*@
  MPIDI_CH3U_Handle_recv_req - Process a receive request for which all of the 
  data has been received (and copied) into the
  buffers described by the request's IOV.

  Input Parameters:
+ vc - virtual connection over which the data was received
- rreq - pointer to the receive request object

  Output Parameter:
. complete - data transfer for the request has completed
@*/
int MPIDI_CH3U_Handle_recv_req(MPIDI_VC_t * vc, MPID_Request * rreq, 
			       int * complete);

/* Handle_send_req invokes the action (method/function) when data 
   becomes available.  It is an obsolete routine; the completion 
   function should be invoked directly.  */
int MPIDI_CH3U_Handle_send_req(MPIDI_VC_t * vc, MPID_Request * sreq, 
			       int *complete);

int MPIDI_CH3U_Handle_connection(MPIDI_VC_t * vc, MPIDI_VC_Event_t event);

int MPIDI_CH3U_VC_SendClose( MPIDI_VC_t *vc, int rank );
int MPIDI_CH3U_VC_WaitForClose( void );
#ifdef MPIDI_CH3_HAS_CHANNEL_CLOSE
int MPIDI_CH3_Channel_close( void );
#else
#define MPIDI_CH3_Channel_close( )   MPI_SUCCESS
#endif
1668
1669
1670
/* MPIDI_CH3U_Check_for_failed_procs() reads PMI_dead_processes key
   and marks VCs to those processes as failed */
int MPIDI_CH3U_Check_for_failed_procs(void);
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786

/*@
  MPIDI_CH3_Pre_init - Allows the channel to initialize before PMI_init is 
  called, and allows the
  channel to optionally set the rank, size, and whether this process has a 
  parent.

  Output Parameters:
+ setvals - boolean value that is true if this function set has_parent, rank, 
  and size
. has_parent - boolean value that is true if this MPI job was spawned by 
  another set of MPI processes
. rank - rank of this process in the process group
- size - number of processes in the process group

  Return value:
  A MPI error code.

  Notes:
  This function is optional, and is used only when HAVE_CH3_PRE_INIT is 
  defined.  It is called by CH3 before PMI_Init.  If the function sets setvals 
  to TRUE, CH3 will not use PMI to get the rank,  size, etc.
@*/
int MPIDI_CH3_Pre_init (int *setvals, int *has_parent, int *rank, int *size);

/*@
  MPIDI_CH3_Init - Initialize the channel implementation.

  Input Parameters:
+ has_parent - boolean value that is true if this MPI job was spawned by 
  another set of MPI processes
. pg_ptr - the new process group representing MPI_COMM_WORLD
- pg_rank - my rank in the process group

  Return value:
  A MPI error code.

  Notes:
  MPID_Init has called 'PMI_Init' and created the process group structure 
  before this routine is called.
@*/
int MPIDI_CH3_Init(int has_parent, MPIDI_PG_t *pg_ptr, int pg_rank );

/*@
  MPIDI_CH3_Finalize - Shutdown the channel implementation.

  Return value:
  A MPI error class.
@*/
int MPIDI_CH3_Finalize(void);

/*@
  MPIDI_CH3_VC_Init - Perform channel-specific initialization of a VC

  Input Parameter:
. vc - Virtual connection to initialize
  @*/
int MPIDI_CH3_VC_Init( struct MPIDI_VC *vc );

/*@
   MPIDI_CH3_PG_Destroy - Perform any channel-specific actions when freeing 
   a process group 

    Input Parameter:
.   pg - Process group on which to act
@*/
int MPIDI_CH3_PG_Destroy( struct MPIDI_PG *pg );

/*@ MPIDI_CH3_VC_Destroy - Perform and channel-specific actions when freeing a
    virtual connection.

    Input Parameter:
.   vc - Virtual connection on which to act
@*/
int MPIDI_CH3_VC_Destroy( struct MPIDI_VC *vc );

/*@ MPIDI_CH3_InitComplete - Perform any channel-specific initialization 
  actions after MPID_Init but before MPI_Init (or MPI_Initthread) returns
  @*/
int MPIDI_CH3_InitCompleted( void );

#ifdef MPIDI_CH3_HASIMPL_HEADER
#include "mpidi_ch3_mpid.h"
#endif
/* Routines in support of ch3 */

/* Routine to return the tag associated with a port */
int MPIDI_GetTagFromPort( const char *, int * );

/* Here are the packet handlers */
int MPIDI_CH3_PktHandler_EagerSend( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				   MPIDI_msg_sz_t *, MPID_Request ** );
#ifdef USE_EAGER_SHORT
int MPIDI_CH3_PktHandler_EagerShortSend( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					 MPIDI_msg_sz_t *, MPID_Request ** );
#endif
int MPIDI_CH3_PktHandler_ReadySend( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				    MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_EagerSyncSend( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_EagerSyncAck( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				       MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_RndvReqToSend( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_RndvClrToSend( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_RndvSend( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				   MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_CancelSendReq( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_CancelSendResp( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					 MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_Put( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
			      MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_Accumulate( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				     MPIDI_msg_sz_t *, MPID_Request ** );
1787
1788
int MPIDI_CH3_PktHandler_Accumulate_Immed( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				     MPIDI_msg_sz_t *, MPID_Request ** );
1789
1790
1791
1792
int MPIDI_CH3_PktHandler_CAS( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
                              MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_CASResp( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
                                  MPIDI_msg_sz_t *, MPID_Request ** );
1793
1794
1795
1796
int MPIDI_CH3_PktHandler_FOP( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
                              MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_FOPResp( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
                                  MPIDI_msg_sz_t *, MPID_Request ** );
1797
1798
int MPIDI_CH3_PktHandler_Get_AccumResp( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
                                        MPIDI_msg_sz_t *, MPID_Request ** );
1799
1800
1801
1802
1803
1804
1805
1806
int MPIDI_CH3_PktHandler_Get( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
			      MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_GetResp( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				 MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_Lock( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
			      MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_LockGranted( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				      MPIDI_msg_sz_t *, MPID_Request ** );
1807
1808
int MPIDI_CH3_PktHandler_Unlock( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *,
                                 MPIDI_msg_sz_t *, MPID_Request ** );
1809
1810
int MPIDI_CH3_PktHandler_Flush( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *,
                                MPIDI_msg_sz_t *, MPID_Request ** );
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
int MPIDI_CH3_PktHandler_PtRMADone( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				    MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_LockPutUnlock( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_LockAccumUnlock( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					  MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_LockGetUnlock( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
					MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_FlowCntlUpdate( MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt,
					 MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_Close( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *, 
				MPIDI_msg_sz_t *, MPID_Request ** );
int MPIDI_CH3_PktHandler_EndCH3( MPIDI_VC_t *, MPIDI_CH3_Pkt_t *,
				 MPIDI_msg_sz_t *, MPID_Request ** );

/* PktHandler function:
   vc  (INPUT) -- vc on which the packet was received
   pkt (INPUT) -- pointer to packet header at beginning of receive buffer
   buflen (I/O) -- IN: number of bytes received into receive buffer
                   OUT: number of bytes processed by the handler function
   req (OUTPUT) -- NULL, if the whole message has been processed by the handler
                   function, otherwise, pointer to the receive request for this
                   message.  The IOV will be set describing where the rest of the
                   message should be received.
*/
typedef int MPIDI_CH3_PktHandler_Fcn(MPIDI_VC_t *vc, MPIDI_CH3_Pkt_t *pkt,
				     MPIDI_msg_sz_t *buflen, MPID_Request **req );
int MPIDI_CH3_PktHandler_Init( MPIDI_CH3_PktHandler_Fcn *[], int );

#ifdef MPICH_DBG_OUTPUT
int MPIDI_CH3_PktPrint_CancelSendReq( FILE *, MPIDI_CH3_Pkt_t * );
int MPIDI_CH3_PktPrint_CancelSendResp( FILE *, MPIDI_CH3_Pkt_t * );
int MPIDI_CH3_PktPrint_EagerSend( FILE *, MPIDI_CH3_Pkt_t * );
int MPIDI_CH3_PktPrint_ReadySend( FILE *, MPIDI_CH3_Pkt_t * );
int MPIDI_CH3_PktPrint_RndvReqToSend( FILE *, MPIDI_CH3_Pkt_t * );
int MPIDI_CH3_PktPrint_RndvClrToSend( FILE *, MPIDI_CH3_Pkt_t * );
int MPIDI_CH3_PktPrint_RndvSend( FILE *, MPIDI_CH3_Pkt_t * );
int MPIDI_CH3_PktPrint_EagerSyncSend( FILE *fp, MPIDI_CH3_Pkt_t *pkt );
int MPIDI_CH3_PktPrint_EagerSyncAck( FILE *fp, MPIDI_CH3_Pkt_t *pkt );
#endif

/* Routines to create packets (used in implementing MPI communications */
int MPIDI_CH3_EagerNoncontigSend( MPID_Request **, MPIDI_CH3_Pkt_type_t, 
				  const void *, int, 
				  MPI_Datatype, MPIDI_msg_sz_t, int, int, MPID_Comm *, 
				  int );
int MPIDI_CH3_EagerContigSend( MPID_Request **, MPIDI_CH3_Pkt_type_t, 
			       const void *, MPIDI_msg_sz_t, int, 
			       int, MPID_Comm *, int );
int MPIDI_CH3_EagerContigShortSend( MPID_Request **, MPIDI_CH3_Pkt_type_t, 
				    const void *, MPIDI_msg_sz_t, 
				    int, int, MPID_Comm *, int );
int MPIDI_CH3_EagerContigIsend( MPID_Request **, MPIDI_CH3_Pkt_type_t, 
				const void *, MPIDI_msg_sz_t, int, 
				int, MPID_Comm *, int );


int MPIDI_CH3_RndvSend( MPID_Request **, const void *, int, MPI_Datatype, 
			int, MPIDI_msg_sz_t, MPI_Aint, int, int, MPID_Comm *, int );

int MPIDI_CH3_EagerSyncNoncontigSend( MPID_Request **, const void *, int, 
				      MPI_Datatype, MPIDI_msg_sz_t, int, MPI_Aint,
				      int, int, MPID_Comm *, int );
int MPIDI_CH3_EagerSyncZero(MPID_Request **, int, int, MPID_Comm *, int );

1876
1877
int MPIDI_CH3_SendNoncontig_iov( struct MPIDI_VC *vc, struct MPID_Request *sreq,
                                 void *header, MPIDI_msg_sz_t hdr_sz );
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902

/* Routines to ack packets, called in the receive routines when a 
   message is matched */
int MPIDI_CH3_EagerSyncAck( MPIDI_VC_t *, MPID_Request * );
int MPIDI_CH3_RecvFromSelf( MPID_Request *, void *, int, MPI_Datatype );
int MPIDI_CH3_RecvRndv( MPIDI_VC_t *, MPID_Request * );

/* Handler routines to continuing after an IOV is processed (assigned to the
   OnDataAvail field in the device part of a request) */
int MPIDI_CH3_ReqHandler_RecvComplete( MPIDI_VC_t *, MPID_Request *, int * );
int MPIDI_CH3_ReqHandler_UnpackUEBufComplete( MPIDI_VC_t *, MPID_Request *,
					      int * );
int MPIDI_CH3_ReqHandler_ReloadIOV( MPIDI_VC_t *, MPID_Request *, int * );

int MPIDI_CH3_ReqHandler_UnpackSRBufReloadIOV( MPIDI_VC_t *, MPID_Request *,
					       int * );
int MPIDI_CH3_ReqHandler_UnpackSRBufComplete( MPIDI_VC_t *, MPID_Request *,
					      int * );
int MPIDI_CH3_ReqHandler_PutRespDerivedDTComplete( MPIDI_VC_t *, 
						   MPID_Request *, int * );
int MPIDI_CH3_ReqHandler_PutAccumRespComplete( MPIDI_VC_t *, MPID_Request *,
					       int * );
int MPIDI_CH3_ReqHandler_AccumRespDerivedDTComplete( MPIDI_VC_t *, 
						     MPID_Request *,
						     int * );
1903
1904
1905
int MPIDI_CH3_ReqHandler_GetAccumRespComplete( MPIDI_VC_t *vc, 
                                               MPID_Request *rreq, 
                                               int *complete );
1906
1907
1908
1909
int MPIDI_CH3_ReqHandler_SinglePutAccumComplete( MPIDI_VC_t *, MPID_Request *,
						 int * );
int MPIDI_CH3_ReqHandler_GetRespDerivedDTComplete( MPIDI_VC_t *, 
						   MPID_Request *, int * );
1910
int MPIDI_CH3_ReqHandler_FOPComplete( MPIDI_VC_t *, MPID_Request *, int * );
1911
1912
1913
1914
1915
1916

/* Send Handlers */
int MPIDI_CH3_ReqHandler_SendReloadIOV( MPIDI_VC_t *vc, MPID_Request *sreq, 
					int *complete );
int MPIDI_CH3_ReqHandler_GetSendRespComplete( MPIDI_VC_t *, MPID_Request *,
					      int * );
1917
1918
1919
1920
/* Thread Support */
#ifdef MPICH_IS_THREADED
#if MPIU_THREAD_GRANULARITY == MPIU_THREAD_GRANULARITY_GLOBAL
/* There is a single, global lock, held for the duration of an MPI call */
1921
1922
#define MPIU_THREAD_CS_ENTER_CH3COMM(context_)
#define MPIU_THREAD_CS_EXIT_CH3COMM(context_)
1923

1924
1925
1926
#define MPIU_THREAD_CS_ENTER_LMT(_context)
#define MPIU_THREAD_CS_EXIT_LMT(_context)

1927
#elif MPIU_THREAD_GRANULARITY == MPIU_THREAD_GRANULARITY_PER_OBJECT
1928

1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
#define MPIU_THREAD_CS_ENTER_POBJ_MUTEX(mutex_p_)                                                           \
    do {                                                                                                    \
        MPIU_DBG_MSG_P(THREAD,VERBOSE,"attempting to ENTER per-object CS, mutex=%s", MPIU_QUOTE(mutex_p_)); \
        /* FIXME do we need nest checking here? the existing macros won't work unmodified...*/              \
        MPID_Thread_mutex_lock(mutex_p_);                                                                   \
    } while (0)
#define MPIU_THREAD_CS_EXIT_POBJ_MUTEX(mutex_p_)                                                            \
    do {                                                                                                    \
        MPIU_DBG_MSG_P(THREAD,VERBOSE,"attempting to EXIT per-object CS, mutex=%s", MPIU_QUOTE(mutex_p_));  \
        /* FIXME do we need nest checking here? the existing macros won't work unmodified...*/              \
        MPID_Thread_mutex_unlock(mutex_p_);                                                                 \
    } while (0)

1942
/* There is a per object lock */
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
#define MPIU_THREAD_CS_ENTER_CH3COMM(context_)                      \
    do {                                                            \
        if (MPIU_ISTHREADED)                                        \
            MPIU_THREAD_CS_ENTER_POBJ_MUTEX(&context_->pobj_mutex); \
    } while (0)
#define MPIU_THREAD_CS_EXIT_CH3COMM(context_)                       \
    do {                                                            \
        if (MPIU_ISTHREADED)                                        \
            MPIU_THREAD_CS_EXIT_POBJ_MUTEX(&context_->pobj_mutex);  \
    } while (0)
1953

1954
1955
1956
1957
1958
/* MT FIXME making LMT into MPIDCOMM for now because of overwhelming deadlock
 * issues */
#define MPIU_THREAD_CS_ENTER_LMT(context_) MPIU_THREAD_CS_ENTER_MPIDCOMM(context_)
#define MPIU_THREAD_CS_EXIT_LMT(context_)  MPIU_THREAD_CS_EXIT_MPIDCOMM(context_)

1959
1960
#elif MPIU_THREAD_GRANULARITY == MPIU_THREAD_GRANULARITY_SINGLE
/* No thread support, make all operations a no-op */
1961
1962
/* FIXME incomplete? or already handled by the upper level? */
/* FIXME does it make sense to have (MPICH_IS_THREADED && _GRANULARITY==_SINGLE) ? */
1963
1964
1965
1966
1967
1968
1969

#else
#error Unrecognized thread granularity
#endif
#else

#endif /* MPICH_IS_THREADED */
1970
1971
1972

#endif /* !defined(MPICH_MPIDIMPL_H_INCLUDED) */