pio_server.c 40.8 KB
Newer Older
Deike Kleberg's avatar
Deike Kleberg committed
1
2
/** @file ioServer.c
*/
3
4
5
6
#ifdef HAVE_CONFIG_H
#  include "config.h"
#endif

Deike Kleberg's avatar
Deike Kleberg committed
7
8
9
#include "pio_server.h"


10
#include <limits.h>
Deike Kleberg's avatar
Deike Kleberg committed
11
12
#include <stdlib.h>
#include <stdio.h>
13
14
15
16
17
18

#ifdef HAVE_PARALLEL_NC4
#include <core/ppm_combinatorics.h>
#include <core/ppm_rectilinear.h>
#include <ppm/ppm_uniform_partition.h>
#endif
19
#include <yaxt.h>
20

Deike Kleberg's avatar
Deike Kleberg committed
21
#include "cdi.h"
22
#include "cdipio.h"
23
#include "dmemory.h"
24
#include "namespace.h"
25
#include "taxis.h"
Deike Kleberg's avatar
Deike Kleberg committed
26
#include "pio.h"
Deike Kleberg's avatar
Deike Kleberg committed
27
#include "pio_comm.h"
28
#include "pio_interface.h"
Deike Kleberg's avatar
Deike Kleberg committed
29
#include "pio_rpc.h"
Deike Kleberg's avatar
Deike Kleberg committed
30
#include "pio_util.h"
31
#include "cdi_int.h"
32
33
34
#ifndef HAVE_NETCDF_PAR_H
#define MPI_INCLUDED
#endif
35
#include "pio_cdf_int.h"
36
#include "resource_handle.h"
37
#include "resource_unpack.h"
Thomas Jahns's avatar
Thomas Jahns committed
38
#include "stream_cdf.h"
Deike Kleberg's avatar
Deike Kleberg committed
39
#include "vlist_var.h"
40

41

42
43
44
static struct
{
  size_t size;
Thomas Jahns's avatar
Thomas Jahns committed
45
  unsigned char *buffer;
46
  int dictSize;
47
48
} *rxWin = NULL;

Thomas Jahns's avatar
Thomas Jahns committed
49
static MPI_Win getWin = MPI_WIN_NULL;
Thomas Jahns's avatar
Thomas Jahns committed
50
static MPI_Group groupModel = MPI_GROUP_NULL;
Deike Kleberg's avatar
Deike Kleberg committed
51

52
53
54
55
56
#ifdef HAVE_PARALLEL_NC4
/* prime factorization of number of pio collectors */
static uint32_t *pioPrimes;
static int numPioPrimes;
#endif
Deike Kleberg's avatar
Deike Kleberg committed
57

Deike Kleberg's avatar
Deike Kleberg committed
58
59
/************************************************************************/

60
static
Deike Kleberg's avatar
Deike Kleberg committed
61
62
void serverWinCleanup ()
{
63
64
  if (getWin != MPI_WIN_NULL)
    xmpi(MPI_Win_free(&getWin));
65
66
  if (rxWin)
    {
67
      free(rxWin[0].buffer);
68
      free(rxWin);
Deike Kleberg's avatar
Deike Kleberg committed
69
    }
70

71
  xdebug("%s", "cleaned up mpi_win");
Deike Kleberg's avatar
Deike Kleberg committed
72
}
73

Deike Kleberg's avatar
Deike Kleberg committed
74
 /************************************************************************/
75

76
77
static size_t
collDefBufferSizes()
Deike Kleberg's avatar
Deike Kleberg committed
78
{
79
  int nstreams, * streamIndexList, streamNo, vlistID, nvars, varID, iorank;
80
81
  int modelID;
  size_t sumGetBufferSizes = 0;
82
  int rankGlob = commInqRankGlob ();
Deike Kleberg's avatar
Deike Kleberg committed
83
  int nProcsModel = commInqNProcsModel ();
84
  int root = commInqRootGlob ();
Deike Kleberg's avatar
Deike Kleberg committed
85

86
  xassert(rxWin != NULL);
Deike Kleberg's avatar
Deike Kleberg committed
87

Deike Kleberg's avatar
Deike Kleberg committed
88
  nstreams = reshCountType ( &streamOps );
89
  streamIndexList = xmalloc((size_t)nstreams * sizeof (streamIndexList[0]));
90
  reshGetResHListOfType ( nstreams, streamIndexList, &streamOps );
Deike Kleberg's avatar
Deike Kleberg committed
91
92
  for ( streamNo = 0; streamNo < nstreams; streamNo++ )
    {
93
      // space required for data
94
      vlistID = streamInqVlist ( streamIndexList[streamNo] );
Deike Kleberg's avatar
Deike Kleberg committed
95
96
97
98
      nvars = vlistNvars ( vlistID );
      for ( varID = 0; varID < nvars; varID++ )
        {
          iorank = vlistInqVarIOrank ( vlistID, varID );
Deike Kleberg's avatar
Deike Kleberg committed
99
          xassert ( iorank != CDI_UNDEFID );
Deike Kleberg's avatar
Deike Kleberg committed
100
101
          if ( iorank == rankGlob )
            {
Deike Kleberg's avatar
Deike Kleberg committed
102
              for ( modelID = 0; modelID < nProcsModel; modelID++ )
103
                {
104
105
106
107
108
109
                  int decoChunk;
                  {
                    int varSize = vlistInqVarSize(vlistID, varID);
                    int nProcsModel = commInqNProcsModel();
                    decoChunk =
                      (int)ceilf(cdiPIOpartInflate_
110
111
                                 * (float)(varSize + nProcsModel - 1)
                                 / (float)nProcsModel);
112
                  }
Deike Kleberg's avatar
Deike Kleberg committed
113
                  xassert ( decoChunk > 0 );
114
                  rxWin[modelID].size += (size_t)decoChunk * sizeof (double)
115
116
117
118
                    /* re-align chunks to multiple of double size */
                    + sizeof (double) - 1
                    /* one header for data record, one for
                     * corresponding part descriptor*/
119
                    + 2 * sizeof (struct winHeaderEntry)
120
                    /* FIXME: heuristic for size of packed Xt_idxlist */
121
                    + sizeof (Xt_int) * (size_t)decoChunk * 3;
122
                  rxWin[modelID].dictSize += 2;
123
                }
Deike Kleberg's avatar
Deike Kleberg committed
124
            }
125
        }
Deike Kleberg's avatar
Deike Kleberg committed
126
127
      // space required for the 3 function calls streamOpen, streamDefVlist, streamClose 
      // once per stream and timestep for all collprocs only on the modelproc root
128
      rxWin[root].size += numRPCFuncs * sizeof (struct winHeaderEntry)
129
130
131
132
        /* serialized filename */
        + MAXDATAFILENAME
        /* data part of streamDefTimestep */
        + (2 * CDI_MAX_NAME + sizeof (taxis_t));
133
      rxWin[root].dictSize += numRPCFuncs;
Deike Kleberg's avatar
Deike Kleberg committed
134
    }
135
  free ( streamIndexList );
Deike Kleberg's avatar
Deike Kleberg committed
136
137

  for ( modelID = 0; modelID < nProcsModel; modelID++ )
138
    {
139
      /* account for size header */
140
      rxWin[modelID].dictSize += 1;
141
      rxWin[modelID].size += sizeof (struct winHeaderEntry);
142
143
144
      rxWin[modelID].size = roundUpToMultiple(rxWin[modelID].size,
                                              PIO_WIN_ALIGN);
      sumGetBufferSizes += (size_t)rxWin[modelID].size;
145
    }
Deike Kleberg's avatar
Deike Kleberg committed
146
  xassert ( sumGetBufferSizes <= MAXWINBUFFERSIZE );
147
  return sumGetBufferSizes;
Deike Kleberg's avatar
Deike Kleberg committed
148
}
149

Deike Kleberg's avatar
Deike Kleberg committed
150
 /************************************************************************/
151

152
153
154
static void
serverWinCreate(void)
{
Deike Kleberg's avatar
Deike Kleberg committed
155
  int ranks[1], modelID;
156
  MPI_Comm commCalc = commInqCommCalc ();
Deike Kleberg's avatar
Deike Kleberg committed
157
  MPI_Group groupCalc;
158
  int nProcsModel = commInqNProcsModel ();
159
160
161
  MPI_Info no_locks_info;
  xmpi(MPI_Info_create(&no_locks_info));
  xmpi(MPI_Info_set(no_locks_info, "no_locks", "true"));
Deike Kleberg's avatar
Deike Kleberg committed
162

163
  xmpi(MPI_Win_create(MPI_BOTTOM, 0, 1, no_locks_info, commCalc, &getWin));
Deike Kleberg's avatar
Deike Kleberg committed
164
165

  /* target group */
166
167
  ranks[0] = nProcsModel;
  xmpi ( MPI_Comm_group ( commCalc, &groupCalc ));
Deike Kleberg's avatar
Deike Kleberg committed
168
169
  xmpi ( MPI_Group_excl ( groupCalc, 1, ranks, &groupModel ));

170
  rxWin = xcalloc((size_t)nProcsModel, sizeof (rxWin[0]));
171
  size_t totalBufferSize = collDefBufferSizes();
Uwe Schulzweida's avatar
Uwe Schulzweida committed
172
  rxWin[0].buffer = (unsigned char*) xmalloc(totalBufferSize);
173
174
175
176
177
178
  size_t ofs = 0;
  for ( modelID = 1; modelID < nProcsModel; modelID++ )
    {
      ofs += rxWin[modelID - 1].size;
      rxWin[modelID].buffer = rxWin[0].buffer + ofs;
    }
Deike Kleberg's avatar
Deike Kleberg committed
179

180
181
  xmpi(MPI_Info_free(&no_locks_info));

182
  xdebug("%s", "created mpi_win, allocated getBuffer");
Deike Kleberg's avatar
Deike Kleberg committed
183
184
}

Deike Kleberg's avatar
Deike Kleberg committed
185
186
/************************************************************************/

187
static void
188
readFuncCall(struct winHeaderEntry *header)
Deike Kleberg's avatar
Deike Kleberg committed
189
190
{
  int root = commInqRootGlob ();
191
  int funcID = header->id;
192
  union funcArgs *funcArgs = &(header->specific.funcArgs);
Deike Kleberg's avatar
Deike Kleberg committed
193

194
  xassert(funcID >= MINFUNCID && funcID <= MAXFUNCID);
Deike Kleberg's avatar
Deike Kleberg committed
195
196
  switch ( funcID )
    {
197
198
    case STREAMCLOSE:
      {
199
        int streamID
200
          = namespaceAdaptKey2(funcArgs->streamChange.streamID);
201
202
203
204
        streamClose(streamID);
        xdebug("READ FUNCTION CALL FROM WIN:  %s, streamID=%d,"
               " closed stream",
               funcMap[(-1 - funcID)], streamID);
205
206
      }
      break;
Deike Kleberg's avatar
Deike Kleberg committed
207
    case STREAMOPEN:
208
      {
209
        size_t filenamesz = (size_t)funcArgs->newFile.fnamelen;
Deike Kleberg's avatar
Deike Kleberg committed
210
        xassert ( filenamesz > 0 && filenamesz < MAXDATAFILENAME );
211
        const char *filename
212
          = (const char *)(rxWin[root].buffer + header->offset);
213
        xassert(filename[filenamesz] == '\0');
214
        int filetype = funcArgs->newFile.filetype;
215
        int streamID = streamOpenWrite(filename, filetype);
216
        xassert(streamID != CDI_ELIBNAVAIL);
217
218
        xdebug("READ FUNCTION CALL FROM WIN:  %s, filenamesz=%zu,"
               " filename=%s, filetype=%d, OPENED STREAM %d",
219
               funcMap[(-1 - funcID)], filenamesz, filename,
220
               filetype, streamID);
221
      }
222
      break;
223
224
    case STREAMDEFVLIST:
      {
225
        int streamID
226
227
          = namespaceAdaptKey2(funcArgs->streamChange.streamID);
        int vlistID = namespaceAdaptKey2(funcArgs->streamChange.vlistID);
228
229
230
231
        streamDefVlist(streamID, vlistID);
        xdebug("READ FUNCTION CALL FROM WIN:  %s, streamID=%d,"
               " vlistID=%d, called streamDefVlist ().",
               funcMap[(-1 - funcID)], streamID, vlistID);
232
233
      }
      break;
234
235
236
    case STREAMDEFTIMESTEP:
      {
        MPI_Comm commCalc = commInqCommCalc ();
237
        int streamID = funcArgs->streamNewTimestep.streamID;
238
        int originNamespace = namespaceResHDecode(streamID).nsp;
239
240
241
        streamID = namespaceAdaptKey2(streamID);
        int oldTaxisID
          = vlistInqTaxis(streamInqVlist(streamID));
242
        int position = header->offset;
243
244
        int changedTaxisID
          = taxisUnpack((char *)rxWin[root].buffer, (int)rxWin[root].size,
245
                        &position, originNamespace, &commCalc, 0);
246
247
248
249
        taxis_t *oldTaxisPtr = taxisPtr(oldTaxisID);
        taxis_t *changedTaxisPtr = taxisPtr(changedTaxisID);
        ptaxisCopy(oldTaxisPtr, changedTaxisPtr);
        taxisDestroy(changedTaxisID);
250
        streamDefTimestep(streamID, funcArgs->streamNewTimestep.tsID);
251
252
      }
      break;
Deike Kleberg's avatar
Deike Kleberg committed
253
    default:
254
      xabort ( "REMOTE FUNCTIONCALL NOT IMPLEMENTED!" );
Deike Kleberg's avatar
Deike Kleberg committed
255
256
257
258
259
    }
}

/************************************************************************/

260
261
262
263
264
static void
resizeVarGatherBuf(int vlistID, int varID, double **buf, int *bufSize)
{
  int size = vlistInqVarSize(vlistID, varID);
  if (size <= *bufSize) ; else
265
    *buf = xrealloc(*buf, (size_t)(*bufSize = size) * sizeof (buf[0][0]));
266
267
268
269
270
271
272
}

static void
gatherArray(int root, int nProcsModel, int headerIdx,
            int vlistID,
            double *gatherBuf, int *nmiss)
{
273
274
  struct winHeaderEntry *winDict
    = (struct winHeaderEntry *)rxWin[root].buffer;
275
  int streamID = winDict[headerIdx].id;
276
  int varID = winDict[headerIdx].specific.dataRecord.varID;
277
  int varShape[3] = { 0, 0, 0 };
278
  cdiPioQueryVarDims(varShape, vlistID, varID);
279
280
281
282
283
  Xt_int varShapeXt[3];
  static const Xt_int origin[3] = { 0, 0, 0 };
  for (unsigned i = 0; i < 3; ++i)
    varShapeXt[i] = varShape[i];
  int varSize = varShape[0] * varShape[1] * varShape[2];
284
285
286
  struct Xt_offset_ext *partExts
    = xmalloc((size_t)nProcsModel * sizeof (partExts[0]));
  Xt_idxlist *part = xmalloc((size_t)nProcsModel * sizeof (part[0]));
287
288
  MPI_Comm commCalc = commInqCommCalc();
  {
289
    int nmiss_ = 0;
290
291
292
    for (int modelID = 0; modelID < nProcsModel; modelID++)
      {
        struct dataRecord *dataHeader
293
294
295
296
          = &((struct winHeaderEntry *)
              rxWin[modelID].buffer)[headerIdx].specific.dataRecord;
        int position =
          ((struct winHeaderEntry *)rxWin[modelID].buffer)[headerIdx + 1].offset;
297
298
299
        xassert(namespaceAdaptKey2(((struct winHeaderEntry *)
                                    rxWin[modelID].buffer)[headerIdx].id)
                == streamID
300
                && dataHeader->varID == varID
301
302
                && ((struct winHeaderEntry *)
                    rxWin[modelID].buffer)[headerIdx + 1].id == PARTDESCMARKER
303
304
                && position > 0
                && ((size_t)position
305
                    >= sizeof (struct winHeaderEntry) * (size_t)rxWin[modelID].dictSize)
306
307
308
309
                && ((size_t)position < rxWin[modelID].size));
        part[modelID] = xt_idxlist_unpack(rxWin[modelID].buffer,
                                          (int)rxWin[modelID].size,
                                          &position, commCalc);
310
311
312
313
314
        unsigned partSize = (unsigned)xt_idxlist_get_num_indices(part[modelID]);
        size_t charOfs = (size_t)((rxWin[modelID].buffer
                                   + ((struct winHeaderEntry *)
                                      rxWin[modelID].buffer)[headerIdx].offset)
                                  - rxWin[0].buffer);
315
316
        xassert(charOfs % sizeof (double) == 0
                && charOfs / sizeof (double) + partSize <= INT_MAX);
317
        int elemOfs = (int)(charOfs / sizeof (double));
318
        partExts[modelID].start = elemOfs;
319
        partExts[modelID].size = (int)partSize;
320
        partExts[modelID].stride = 1;
321
322
323
324
325
        nmiss_ += dataHeader->nmiss;
      }
    *nmiss = nmiss_;
  }
  Xt_idxlist srcList = xt_idxlist_collection_new(part, nProcsModel);
326
  for (int modelID = 0; modelID < nProcsModel; modelID++)
327
328
329
330
331
332
333
334
    xt_idxlist_delete(part[modelID]);
  free(part);
  Xt_xmap gatherXmap;
  {
    Xt_idxlist dstList
      = xt_idxsection_new(0, 3, varShapeXt, varShapeXt, origin);
    struct Xt_com_list full = { .list = dstList, .rank = 0 };
    gatherXmap = xt_xmap_intersection_new(1, &full, 1, &full, srcList, dstList,
335
                                          MPI_COMM_SELF);
336
337
338
339
    xt_idxlist_delete(dstList);
  }
  xt_idxlist_delete(srcList);

340
  struct Xt_offset_ext gatherExt = { .start = 0, .size = varSize, .stride = 1 };
341
  Xt_redist gatherRedist
342
343
    = xt_redist_p2p_ext_new(gatherXmap, nProcsModel, partExts, 1, &gatherExt,
                            MPI_DOUBLE);
344
  xt_xmap_delete(gatherXmap);
345
  xt_redist_s_exchange1(gatherRedist, rxWin[0].buffer, gatherBuf);
346
  free(partExts);
347
  xt_redist_delete(gatherRedist);
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
}

struct xyzDims
{
  int sizes[3];
};

static inline int
xyzGridSize(struct xyzDims dims)
{
  return dims.sizes[0] * dims.sizes[1] * dims.sizes[2];
}

#ifdef HAVE_PARALLEL_NC4
static void
363
queryVarBounds(struct PPM_extent varShape[3], int vlistID, int varID)
364
{
365
366
  varShape[0].first = 0;
  varShape[1].first = 0;
367
  varShape[2].first = 0;
368
  int sizes[3];
369
  cdiPioQueryVarDims(sizes, vlistID, varID);
370
371
  for (unsigned i = 0; i < 3; ++i)
    varShape[i].size = sizes[i];
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
}

/* compute distribution of collectors such that number of collectors
 * <= number of variable grid cells in each dimension */
static struct xyzDims
varDimsCollGridMatch(const struct PPM_extent varDims[3])
{
  xassert(PPM_extents_size(3, varDims) >= commInqSizeColl());
  struct xyzDims collGrid = { { 1, 1, 1 } };
  /* because of storage order, dividing dimension 3 first is preferred */
  for (int i = 0; i < numPioPrimes; ++i)
    {
      for (int dim = 2; dim >=0; --dim)
        if (collGrid.sizes[dim] * pioPrimes[i] <= varDims[dim].size)
          {
            collGrid.sizes[dim] *= pioPrimes[i];
            goto nextPrime;
          }
      /* no position found, retrack */
      xabort("Not yet implemented back-tracking needed.");
      nextPrime:
      ;
    }
  return collGrid;
}

static void
myVarPart(struct PPM_extent varShape[3], struct xyzDims collGrid,
          struct PPM_extent myPart[3])
{
  int32_t myCollGridCoord[3];
  {
    struct PPM_extent collGridShape[3];
    for (int i = 0; i < 3; ++i)
      {
        collGridShape[i].first = 0;
        collGridShape[i].size = collGrid.sizes[i];
      }
    PPM_lidx2rlcoord_e(3, collGridShape, commInqRankColl(), myCollGridCoord);
    xdebug("my coord: (%d, %d, %d)", myCollGridCoord[0], myCollGridCoord[1],
           myCollGridCoord[2]);
  }
  PPM_uniform_partition_nd(3, varShape, collGrid.sizes,
                           myCollGridCoord, myPart);
}
417
418
419
420
421
422
423
424
425
426
#elif defined (HAVE_LIBNETCDF)
/* needed for writing when some files are only written to by a single process */
/* cdiOpenFileMap(fileID) gives the writer process */
int cdiPioSerialOpenFileMap(int streamID)
{
  return stream_to_pointer(streamID)->ownerRank;
}
/* for load-balancing purposes, count number of files per process */
/* cdiOpenFileCounts[rank] gives number of open files rank has to himself */
static int *cdiSerialOpenFileCount = NULL;
Thomas Jahns's avatar
Thomas Jahns committed
427
428
429

static int
cdiPioNextOpenRank()
430
431
432
433
434
435
436
437
438
439
440
441
442
{
  xassert(cdiSerialOpenFileCount != NULL);
  int commCollSize = commInqSizeColl();
  int minRank = 0, minOpenCount = cdiSerialOpenFileCount[0];
  for (int i = 1; i < commCollSize; ++i)
    if (cdiSerialOpenFileCount[i] < minOpenCount)
      {
        minOpenCount = cdiSerialOpenFileCount[i];
        minRank = i;
      }
  return minRank;
}

443
444
static void
cdiPioOpenFileOnRank(int rank)
445
446
{
  xassert(cdiSerialOpenFileCount != NULL
447
          && (unsigned)rank < (unsigned)commInqSizeColl());
448
449
450
  ++(cdiSerialOpenFileCount[rank]);
}

451
452
static void
cdiPioCloseFileOnRank(int rank)
453
454
455
456
457
458
459
{
  xassert(cdiSerialOpenFileCount != NULL
          && rank >= 0 && rank < commInqSizeColl());
  xassert(cdiSerialOpenFileCount[rank] > 0);
  --(cdiSerialOpenFileCount[rank]);
}

460
461
462
463
464
465
466
467
468
469
static void
cdiPioServerCdfDefVars(stream_t *streamptr)
{
  int rank, rankOpen;
  if (commInqIOMode() == PIO_NONE
      || ((rank = commInqRankColl())
          == (rankOpen = cdiPioSerialOpenFileMap(streamptr->self))))
    cdfDefVars(streamptr);
}

470
471
#endif

472
473
474
475
476
477
struct streamMapping {
  int streamID, filetype;
  int firstHeaderIdx, lastHeaderIdx;
  int numVars, *varMap;
};

478
479
480
481
482
483
struct streamMap
{
  struct streamMapping *entries;
  int numEntries;
};

Thomas Jahns's avatar
Thomas Jahns committed
484
485
486
487
488
489
490
491
static int
smCmpStreamID(const void *a_, const void *b_)
{
  const struct streamMapping *a = a_, *b = b_;
  int streamIDa = a->streamID, streamIDb = b->streamID;
  return (streamIDa > streamIDb) - (streamIDa < streamIDb);
}

492
493
494
495
496
497
498
static inline int
inventorizeStream(struct streamMapping *streamMap, int numStreamIDs,
                  int *sizeStreamMap_, int streamID, int headerIdx)
{
  int sizeStreamMap = *sizeStreamMap_;
  if (numStreamIDs < sizeStreamMap) ; else
    {
499
500
501
      streamMap = xrealloc(streamMap,
                           (size_t)(sizeStreamMap *= 2)
                           * sizeof (streamMap[0]));
502
503
504
505
      *sizeStreamMap_ = sizeStreamMap;
    }
  streamMap[numStreamIDs].streamID = streamID;
  streamMap[numStreamIDs].firstHeaderIdx = headerIdx;
506
  streamMap[numStreamIDs].lastHeaderIdx = headerIdx;
507
508
509
510
511
512
513
514
515
  streamMap[numStreamIDs].numVars = -1;
  int filetype = streamInqFiletype(streamID);
  streamMap[numStreamIDs].filetype = filetype;
  if (filetype == FILETYPE_NC || filetype == FILETYPE_NC2
      || filetype == FILETYPE_NC4)
    {
      int vlistID = streamInqVlist(streamID);
      int nvars = vlistNvars(vlistID);
      streamMap[numStreamIDs].numVars = nvars;
516
517
      streamMap[numStreamIDs].varMap
        = xmalloc(sizeof (streamMap[numStreamIDs].varMap[0]) * (size_t)nvars);
518
519
520
521
522
523
      for (int i = 0; i < nvars; ++i)
        streamMap[numStreamIDs].varMap[i] = -1;
    }
  return numStreamIDs + 1;
}

524
525
526
527
528
529
530
531
532
533
static inline int
streamIsInList(struct streamMapping *streamMap, int numStreamIDs,
               int streamIDQuery)
{
  int p = 0;
  for (int i = 0; i < numStreamIDs; ++i)
    p |= streamMap[i].streamID == streamIDQuery;
  return p;
}

534
static struct streamMap
535
buildStreamMap(struct winHeaderEntry *winDict)
536
537
538
{
  int streamIDOld = CDI_UNDEFID;
  int oldStreamIdx = CDI_UNDEFID;
539
  int filetype = FILETYPE_UNDEF;
540
  int sizeStreamMap = 16;
541
542
  struct streamMapping *streamMap
    = xmalloc((size_t)sizeStreamMap * sizeof (streamMap[0]));
543
  int numDataEntries = winDict[0].specific.headerSize.numDataEntries;
544
  int numStreamIDs = 0;
545
  /* find streams written on this process */
546
547
548
  for (int headerIdx = 1; headerIdx < numDataEntries; headerIdx += 2)
    {
      int streamID
549
550
        = winDict[headerIdx].id
        = namespaceAdaptKey2(winDict[headerIdx].id);
551
552
553
554
555
556
557
558
559
560
561
      xassert(streamID > 0);
      if (streamID != streamIDOld)
        {
          for (int i = numStreamIDs - 1; i >= 0; --i)
            if ((streamIDOld = streamMap[i].streamID) == streamID)
              {
                oldStreamIdx = i;
                goto streamIDInventorized;
              }
          oldStreamIdx = numStreamIDs;
          streamIDOld = streamID;
562
563
          numStreamIDs = inventorizeStream(streamMap, numStreamIDs,
                                           &sizeStreamMap, streamID, headerIdx);
564
565
        }
      streamIDInventorized:
566
      filetype = streamMap[oldStreamIdx].filetype;
567
568
569
570
      streamMap[oldStreamIdx].lastHeaderIdx = headerIdx;
      if (filetype == FILETYPE_NC || filetype == FILETYPE_NC2
          || filetype == FILETYPE_NC4)
        {
571
          int varID = winDict[headerIdx].specific.dataRecord.varID;
572
573
574
          streamMap[oldStreamIdx].varMap[varID] = headerIdx;
        }
    }
575
576
577
578
  /* join with list of streams written to in total */
  {
    int *streamIDs, *streamIsWritten;
    int numTotalStreamIDs = streamSize();
Uwe Schulzweida's avatar
Uwe Schulzweida committed
579
    streamIDs = (int*) xmalloc(2 * sizeof (streamIDs[0]) * (size_t)numTotalStreamIDs);
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
    streamGetIndexList(numTotalStreamIDs, streamIDs);
    streamIsWritten = streamIDs + numTotalStreamIDs;
    for (int i = 0; i < numTotalStreamIDs; ++i)
      streamIsWritten[i] = streamIsInList(streamMap, numStreamIDs,
                                          streamIDs[i]);
    /* Find what streams are written to at all on any process */
    xmpi(MPI_Allreduce(MPI_IN_PLACE, streamIsWritten, numTotalStreamIDs,
                       MPI_INT, MPI_BOR, commInqCommColl()));
    /* append streams written to on other tasks to mapping */
    for (int i = 0; i < numTotalStreamIDs; ++i)
      if (streamIsWritten[i] && !streamIsInList(streamMap, numStreamIDs,
                                                streamIDs[i]))
        numStreamIDs = inventorizeStream(streamMap, numStreamIDs,
                                         &sizeStreamMap, streamIDs[i], -1);

    free(streamIDs);
  }
Thomas Jahns's avatar
Thomas Jahns committed
597
  /* sort written streams by streamID */
598
599
  streamMap = xrealloc(streamMap, sizeof (streamMap[0]) * (size_t)numStreamIDs);
  qsort(streamMap, (size_t)numStreamIDs, sizeof (streamMap[0]), smCmpStreamID);
600
601
602
  return (struct streamMap){ .entries = streamMap, .numEntries = numStreamIDs };
}

603
604
605
606
607
608
609
610
611
612
static void
writeGribStream(struct winHeaderEntry *winDict, struct streamMapping *mapping,
                double **data_, int *currentDataBufSize, int root,
                int nProcsModel)
{
  int streamID = mapping->streamID;
  int headerIdx, lastHeaderIdx = mapping->lastHeaderIdx;
  int vlistID = streamInqVlist(streamID);
  if (lastHeaderIdx < 0)
    {
613
      /* write zero bytes to trigger synchronization code in fileWrite */
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
      cdiPioFileWrite(streamInqFileID(streamID), NULL, 0,
                      streamInqCurTimestepID(streamID));
    }
  else
    for (headerIdx = mapping->firstHeaderIdx;
         headerIdx <= lastHeaderIdx;
         headerIdx += 2)
      if (streamID == winDict[headerIdx].id)
        {
          int varID = winDict[headerIdx].specific.dataRecord.varID;
          int size = vlistInqVarSize(vlistID, varID);
          int nmiss;
          resizeVarGatherBuf(vlistID, varID, data_, currentDataBufSize);
          double *data = *data_;
          gatherArray(root, nProcsModel, headerIdx,
                      vlistID, data, &nmiss);
          streamWriteVar(streamID, varID, data, nmiss);
          if ( ddebug > 2 )
            {
              char text[1024];
              sprintf(text, "streamID=%d, var[%d], size=%d",
                      streamID, varID, size);
              xprintArray(text, data, size, DATATYPE_FLT);
            }
        }
}
640

641
642
643
644
645
646
647
#ifdef HAVE_NETCDF4
static void
buildWrittenVars(struct streamMapping *mapping, int **varIsWritten_,
                 int myCollRank, MPI_Comm collComm)
{
  int nvars = mapping->numVars;
  int *varMap = mapping->varMap;
648
649
  int *varIsWritten = *varIsWritten_
    = xrealloc(*varIsWritten_, sizeof (*varIsWritten) * (size_t)nvars);
650
651
652
653
654
655
656
  for (int varID = 0; varID < nvars; ++varID)
    varIsWritten[varID] = ((varMap[varID] != -1)
                           ?myCollRank+1 : 0);
  xmpi(MPI_Allreduce(MPI_IN_PLACE, varIsWritten, nvars,
                     MPI_INT, MPI_BOR, collComm));
}
#endif
657

658
static void readGetBuffers()
Deike Kleberg's avatar
Deike Kleberg committed
659
{
660
  int nProcsModel = commInqNProcsModel ();
Deike Kleberg's avatar
Deike Kleberg committed
661
  int root        = commInqRootGlob ();
662
#ifdef HAVE_NETCDF4
663
  int myCollRank = commInqRankColl();
664
  MPI_Comm collComm = commInqCommColl();
665
#endif
666
  xdebug("%s", "START");
667

668
669
  struct winHeaderEntry *winDict
    = (struct winHeaderEntry *)rxWin[root].buffer;
670
  xassert(winDict[0].id == HEADERSIZEMARKER);
671
672
  {
    int dictSize = rxWin[root].dictSize,
673
      firstNonRPCEntry = dictSize - winDict[0].specific.headerSize.numRPCEntries - 1,
674
675
676
677
678
679
      headerIdx,
      numFuncCalls = 0;
    for (headerIdx = dictSize - 1;
         headerIdx > firstNonRPCEntry;
         --headerIdx)
      {
680
681
        xassert(winDict[headerIdx].id >= MINFUNCID
                && winDict[headerIdx].id <= MAXFUNCID);
682
        ++numFuncCalls;
683
        readFuncCall(winDict + headerIdx);
684
      }
685
    xassert(numFuncCalls == winDict[0].specific.headerSize.numRPCEntries);
686
  }
Thomas Jahns's avatar
Thomas Jahns committed
687
  /* build list of streams, data was transferred for */
688
  {
689
    struct streamMap map = buildStreamMap(winDict);
690
    double *data = NULL;
Thomas Jahns's avatar
Thomas Jahns committed
691
692
693
#ifdef HAVE_NETCDF4
    int *varIsWritten = NULL;
#endif
694
695
696
#if defined (HAVE_PARALLEL_NC4)
    double *writeBuf = NULL;
#endif
Thomas Jahns's avatar
Thomas Jahns committed
697
    int currentDataBufSize = 0;
698
    for (int streamIdx = 0; streamIdx < map.numEntries; ++streamIdx)
Thomas Jahns's avatar
Thomas Jahns committed
699
      {
700
        int streamID = map.entries[streamIdx].streamID;
Thomas Jahns's avatar
Thomas Jahns committed
701
        int vlistID = streamInqVlist(streamID);
702
        int filetype = map.entries[streamIdx].filetype;
Thomas Jahns's avatar
Thomas Jahns committed
703

704
        switch (filetype)
705
706
707
          {
          case FILETYPE_GRB:
          case FILETYPE_GRB2:
708
709
710
            writeGribStream(winDict, map.entries + streamIdx,
                            &data, &currentDataBufSize,
                            root, nProcsModel);
711
            break;
712
713
714
715
716
717
718
#ifdef HAVE_NETCDF4
          case FILETYPE_NC:
          case FILETYPE_NC2:
          case FILETYPE_NC4:
#ifdef HAVE_PARALLEL_NC4
            /* HAVE_PARALLE_NC4 implies having ScalES-PPM and yaxt */
            {
719
720
              int nvars = map.entries[streamIdx].numVars;
              int *varMap = map.entries[streamIdx].varMap;
721
722
              buildWrittenVars(map.entries + streamIdx, &varIsWritten,
                               myCollRank, collComm);
723
724
725
726
              for (int varID = 0; varID < nvars; ++varID)
                if (varIsWritten[varID])
                  {
                    struct PPM_extent varShape[3];
727
                    queryVarBounds(varShape, vlistID, varID);
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
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
785
786
787
788
789
790
791
792
793
794
795
                    struct xyzDims collGrid = varDimsCollGridMatch(varShape);
                    xdebug("writing varID %d with dimensions: "
                           "x=%d, y=%d, z=%d,\n"
                           "found distribution with dimensions:"
                           " x=%d, y=%d, z=%d.", varID,
                           varShape[0].size, varShape[1].size, varShape[2].size,
                           collGrid.sizes[0], collGrid.sizes[1],
                           collGrid.sizes[2]);
                    struct PPM_extent varChunk[3];
                    myVarPart(varShape, collGrid, varChunk);
                    int myChunk[3][2];
                    for (int i = 0; i < 3; ++i)
                      {
                        myChunk[i][0] = PPM_extent_start(varChunk[i]);
                        myChunk[i][1] = PPM_extent_end(varChunk[i]);
                      }
                    xdebug("Writing chunk { { %d, %d }, { %d, %d },"
                           " { %d, %d } }", myChunk[0][0], myChunk[0][1],
                           myChunk[1][0], myChunk[1][1], myChunk[2][0],
                           myChunk[2][1]);
                    Xt_int varSize[3];
                    for (int i = 0; i < 3; ++i)
                      varSize[2 - i] = varShape[i].size;
                    Xt_idxlist preRedistChunk, preWriteChunk;
                    /* prepare yaxt descriptor for current data
                       distribution after collect */
                    int nmiss;
                    if (varMap[varID] == -1)
                      {
                        preRedistChunk = xt_idxempty_new();
                        xdebug("%s", "I got none\n");
                      }
                    else
                      {
                        Xt_int preRedistStart[3] = { 0, 0, 0 };
                        preRedistChunk
                          = xt_idxsection_new(0, 3, varSize, varSize,
                                              preRedistStart);
                        resizeVarGatherBuf(vlistID, varID, &data,
                                           &currentDataBufSize);
                        int headerIdx = varMap[varID];
                        gatherArray(root, nProcsModel, headerIdx,
                                    vlistID, data, &nmiss);
                        xdebug("%s", "I got all\n");
                      }
                    MPI_Bcast(&nmiss, 1, MPI_INT, varIsWritten[varID] - 1,
                              collComm);
                    /* prepare yaxt descriptor for write chunk */
                    {
                      Xt_int preWriteChunkStart[3], preWriteChunkSize[3];
                      for (int i = 0; i < 3; ++i)
                        {
                          preWriteChunkStart[2 - i] = varChunk[i].first;
                          preWriteChunkSize[2 - i] = varChunk[i].size;
                        }
                      preWriteChunk = xt_idxsection_new(0, 3, varSize,
                                                        preWriteChunkSize,
                                                        preWriteChunkStart);
                    }
                    /* prepare redistribution */
                    {
                      Xt_xmap xmap = xt_xmap_all2all_new(preRedistChunk,
                                                         preWriteChunk,
                                                         collComm);
                      Xt_redist redist = xt_redist_p2p_new(xmap, MPI_DOUBLE);
                      xt_idxlist_delete(preRedistChunk);
                      xt_idxlist_delete(preWriteChunk);
                      xt_xmap_delete(xmap);
Uwe Schulzweida's avatar
Uwe Schulzweida committed
796
797
798
                      writeBuf = (double*) xrealloc(writeBuf,
                                                    sizeof (double)
                                                    * PPM_extents_size(3, varChunk));
799
                      xt_redist_s_exchange1(redist, data, writeBuf);
800
801
802
803
804
805
806
807
808
809
                      xt_redist_delete(redist);
                    }
                    /* write chunk */
                    streamWriteVarChunk(streamID, varID,
                                        (const int (*)[2])myChunk, writeBuf,
                                        nmiss);
                  }
            }
#else
            /* determine process which has stream open (writer) and
810
811
812
             * which has data for which variable (var owner)
             * three cases need to be distinguished */
            {
813
814
              int nvars = map.entries[streamIdx].numVars;
              int *varMap = map.entries[streamIdx].varMap;
815
816
              buildWrittenVars(map.entries + streamIdx, &varIsWritten,
                               myCollRank, collComm);
817
818
819
820
821
822
823
824
825
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
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
              int writerRank;
              if ((writerRank = cdiPioSerialOpenFileMap(streamID))
                  == myCollRank)
                {
                  for (int varID = 0; varID < nvars; ++varID)
                    if (varIsWritten[varID])
                      {
                        int nmiss;
                        int size = vlistInqVarSize(vlistID, varID);
                        resizeVarGatherBuf(vlistID, varID, &data,
                                           &currentDataBufSize);
                        int headerIdx = varMap[varID];
                        if (varIsWritten[varID] == myCollRank + 1)
                          {
                            /* this process has the full array and will
                             * write it */
                            xdebug("gathering varID=%d for direct writing",
                                   varID);
                            gatherArray(root, nProcsModel, headerIdx,
                                        vlistID, data, &nmiss);
                          }
                        else
                          {
                            /* another process has the array and will
                             * send it over */
                            MPI_Status stat;
                            xdebug("receiving varID=%d for writing from"
                                   " process %d",
                                   varID, varIsWritten[varID] - 1);
                            xmpiStat(MPI_Recv(&nmiss, 1, MPI_INT,
                                              varIsWritten[varID] - 1,
                                              COLLBUFNMISS,
                                              collComm, &stat), &stat);
                            xmpiStat(MPI_Recv(data, size, MPI_DOUBLE,
                                              varIsWritten[varID] - 1,
                                              COLLBUFTX,
                                              collComm, &stat), &stat);
                          }
                        streamWriteVar(streamID, varID, data, nmiss);
                      }
                }
              else
                for (int varID = 0; varID < nvars; ++varID)
                  if (varIsWritten[varID] == myCollRank + 1)
                    {
                      /* this process has the full array and another
                       * will write it */
                      int nmiss;
                      int size = vlistInqVarSize(vlistID, varID);
                      resizeVarGatherBuf(vlistID, varID, &data,
                                         &currentDataBufSize);
                      int headerIdx = varMap[varID];
                      gatherArray(root, nProcsModel, headerIdx,
                                  vlistID, data, &nmiss);
                      MPI_Request req;
                      MPI_Status stat;
                      xdebug("sending varID=%d for writing to"
                             " process %d",
                             varID, writerRank);
                      xmpi(MPI_Isend(&nmiss, 1, MPI_INT,
                                     writerRank, COLLBUFNMISS,
                                     collComm, &req));
                      xmpi(MPI_Send(data, size, MPI_DOUBLE,
                                    writerRank, COLLBUFTX,
                                    collComm));
                      xmpiStat(MPI_Wait(&req, &stat), &stat);
                    }
            }
885
886
887
#endif
            break;
#endif
888
889
890
          default:
            xabort("unhandled filetype in parallel I/O.");
          }
891
      }
Thomas Jahns's avatar
Thomas Jahns committed
892
893
#ifdef HAVE_NETCDF4
    free(varIsWritten);
Thomas Jahns's avatar
Thomas Jahns committed
894
895
896
#ifdef HAVE_PARALLEL_NC4
    free(writeBuf);
#endif
Thomas Jahns's avatar
Thomas Jahns committed
897
#endif
898
    free(map.entries);
Thomas Jahns's avatar
Thomas Jahns committed
899
    free(data);
900
  }
901
  xdebug("%s", "RETURN");
902
903
904
905
} 

/************************************************************************/

Deike Kleberg's avatar
Deike Kleberg committed
906

Thomas Jahns's avatar
Thomas Jahns committed
907
908
static
void clearModelWinBuffer(int modelID)
Deike Kleberg's avatar
Deike Kleberg committed
909
910
911
{
  int nProcsModel = commInqNProcsModel ();

Thomas Jahns's avatar
Thomas Jahns committed
912
913
914
915
  xassert((unsigned)modelID < (unsigned)nProcsModel &&
          rxWin != NULL && rxWin[modelID].buffer != NULL &&
          rxWin[modelID].size > 0 &&
          rxWin[modelID].size <= MAXWINBUFFERSIZE );
916
  memset(rxWin[modelID].buffer, 0, rxWin[modelID].size);
Deike Kleberg's avatar
Deike Kleberg committed
917
918
919
920
921
922
}


/************************************************************************/


923
static
924
void getTimeStepData()
Deike Kleberg's avatar
Deike Kleberg committed
925
{
926
  int modelID;
927
  char text[1024];
928
  int nProcsModel = commInqNProcsModel ();
Thomas Jahns's avatar
Thomas Jahns committed
929
930
  void *getWinBaseAddr;
  int attrFound;
931

932
  xdebug("%s", "START");
Deike Kleberg's avatar
Deike Kleberg committed
933

934
935
  for ( modelID = 0; modelID < nProcsModel; modelID++ )
    clearModelWinBuffer(modelID);
Deike Kleberg's avatar
Deike Kleberg committed
936
  // todo put in correct lbs and ubs
937
  xmpi(MPI_Win_start(groupModel, 0, getWin));
938
939
  xmpi(MPI_Win_get_attr(getWin, MPI_WIN_BASE, &getWinBaseAddr, &attrFound));
  xassert(attrFound);
Deike Kleberg's avatar
Deike Kleberg committed
940
941
  for ( modelID = 0; modelID < nProcsModel; modelID++ )
    {
942
      xdebug("modelID=%d, nProcsModel=%d, rxWin[%d].size=%zu,"
Thomas Jahns's avatar
Thomas Jahns committed
943
             " getWin=%p, sizeof(int)=%u",
944
             modelID, nProcsModel, modelID, rxWin[modelID].size,
Thomas Jahns's avatar
Thomas Jahns committed
945
             getWinBaseAddr, (unsigned)sizeof(int));
946
      /* FIXME: this needs to use MPI_PACK for portability */
947
      xmpi(MPI_Get(rxWin[modelID].buffer, (int)rxWin[modelID].size,
948
                   MPI_UNSIGNED_CHAR, modelID, 0,
949
                   (int)rxWin[modelID].size, MPI_UNSIGNED_CHAR, getWin));
Deike Kleberg's avatar
Deike Kleberg committed
950
    }
951
  xmpi ( MPI_Win_complete ( getWin ));
Deike Kleberg's avatar
Deike Kleberg committed
952

953
  if ( ddebug > 2 )
Deike Kleberg's avatar
Deike Kleberg committed
954
    for ( modelID = 0; modelID < nProcsModel; modelID++ )
955
      {
956
        sprintf(text, "rxWin[%d].size=%zu from PE%d rxWin[%d].buffer",
957
                modelID, rxWin[modelID].size, modelID, modelID);
958
        xprintArray(text, rxWin[modelID].buffer,
959
                    (int)(rxWin[modelID].size / sizeof (double)),
960
                    DATATYPE_FLT);
961
      }
962
963
  readGetBuffers();

964
  xdebug("%s", "RETURN");
Deike Kleberg's avatar
Deike Kleberg committed
965
}
Deike Kleberg's avatar
Deike Kleberg committed
966
967
968

/************************************************************************/

969
970
971
972
973
974
975
976
977
978
979
#if defined (HAVE_LIBNETCDF) && ! defined (HAVE_PARALLEL_NC4)
static int
cdiPioStreamCDFOpenWrap(const char *filename, const char *filemode,
                        int filetype, stream_t *streamptr,
                        int recordBufIsToBeCreated)
{
  switch (filetype)
    {
    case FILETYPE_NC4:
    case FILETYPE_NC4C:
      {
Thomas Jahns's avatar
Thomas Jahns committed
980
981
        /* Only needs initialization to shut up gcc */
        int rank = -1, fileID;
Thomas Jahns's avatar
Thomas Jahns committed
982
983
        int ioMode = commInqIOMode();
        if (ioMode == PIO_NONE
984
985
986
987
            || commInqRankColl() == (rank = cdiPioNextOpenRank()))
          fileID = cdiStreamOpenDefaultDelegate(filename, filemode, filetype,
                                                streamptr,
                                                recordBufIsToBeCreated);
988
989
        else
          streamptr->filetype = filetype;
Thomas Jahns's avatar
Thomas Jahns committed
990
        if (ioMode != PIO_NONE)
991
992
          xmpi(MPI_Bcast(&fileID, 1, MPI_INT, rank, commInqCommColl()));
        streamptr->ownerRank = rank;
993
        cdiPioOpenFileOnRank(rank);
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
        return fileID;
      }
    default:
      return cdiStreamOpenDefaultDelegate(filename, filemode, filetype,
                                          streamptr, recordBufIsToBeCreated);
    }
}

static void
cdiPioStreamCDFCloseWrap(stream_t *streamptr, int recordBufIsToBeDeleted)
{
  int fileID   = streamptr->fileID;
  int filetype = streamptr->filetype;
  if ( fileID == CDI_UNDEFID )
    Warning("File %s not open!", streamptr->filename);
  else
    switch (filetype)
      {
      case FILETYPE_NC:
      case FILETYPE_NC2:
      case FILETYPE_NC4:
      case FILETYPE_NC4C:
        {
1017
          int rank, rankOpen = cdiPioSerialOpenFileMap(streamptr->self);
1018
          if (commInqIOMode() == PIO_NONE
1019
              || ((rank = commInqRankColl()) == rankOpen))
1020
            cdiStreamCloseDefaultDelegate(streamptr, recordBufIsToBeDeleted);
1021
          cdiPioCloseFileOnRank(rankOpen);
1022
1023
1024
1025
1026
1027
          break;
        }
      default:
        cdiStreamCloseDefaultDelegate(streamptr, recordBufIsToBeDeleted);
      }
}
Thomas Jahns's avatar
Thomas Jahns committed
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038

static void
cdiPioCdfDefTimestep(stream_t *streamptr, int tsID)
{
  int rank, rankOpen, streamID = streamptr->self;
  if (commInqIOMode() == PIO_NONE
      || ((rank = commInqRankColl())
          == (rankOpen = cdiPioSerialOpenFileMap(streamID))))
    cdfDefTimestep(streamptr, tsID);
}

1039
1040
#endif

Deike Kleberg's avatar
Deike Kleberg committed
1041
1042
1043
1044
1045
1046
1047
1048
/**
  @brief is encapsulated in CDI library and run on I/O PEs.

  @param

  @return
*/

1049
void cdiPioServer(void (*postCommSetupActions)(void))
Deike Kleberg's avatar
Deike Kleberg committed
1050
{
Thomas Jahns's avatar
Thomas Jahns committed
1051
  int nProcsModel=commInqNProcsModel();
1052
1053
1054
  static int nfinished = 0;
  MPI_Comm commCalc;
  MPI_Status status;
Deike Kleberg's avatar
Deike Kleberg committed
1055

1056
  xdebug("%s", "START");
Deike Kleberg's avatar
Deike Kleberg committed
1057

1058
  cdiPioFileWritingInit(postCommSetupActions);
1059
1060
  if (commInqRankNode() == commInqSpecialRankNode())
    return;
1061
  commCalc = commInqCommCalc ();
1062
#ifdef HAVE_PARALLEL_NC4
1063
  cdiPioEnableNetCDFParAccess();
1064
1065
  numPioPrimes = PPM_prime_factorization_32((uint32_t)commInqSizeColl(),
                                            &pioPrimes);
1066
#elif defined (HAVE_LIBNETCDF)
1067
1068
  cdiSerialOpenFileCount = xcalloc(sizeof (cdiSerialOpenFileCount[0]),
                                   (size_t)commInqSizeColl());
1069
1070
1071
1072
1073
1074
1075
1076
  namespaceSwitchSet(NSSWITCH_STREAM_OPEN_BACKEND,
                     NSSW_FUNC(cdiPioStreamCDFOpenWrap));
  namespaceSwitchSet(NSSWITCH_STREAM_CLOSE_BACKEND,
                     NSSW_FUNC(cdiPioStreamCDFCloseWrap));
  namespaceSwitchSet(NSSWITCH_CDF_DEF_TIMESTEP,
                     NSSW_FUNC(cdiPioCdfDefTimestep));
  namespaceSwitchSet(NSSWITCH_CDF_STREAM_SETUP,