SIONlib  1.7.2
Scalable I/O library for parallel access to task-local files
sion_mpi_gen.c
Go to the documentation of this file.
1 /****************************************************************************
2 ** SIONLIB http://www.fz-juelich.de/jsc/sionlib **
3 *****************************************************************************
4 ** Copyright (c) 2008-2018 **
5 ** Forschungszentrum Juelich, Juelich Supercomputing Centre **
6 ** **
7 ** See the file COPYRIGHT in the package base directory for details **
8 ****************************************************************************/
15 #define _XOPEN_SOURCE 700
16 
17 #include <stdlib.h>
18 #include <stdio.h>
19 #include <stdarg.h>
20 #include <string.h>
21 #include <time.h>
22 
23 #include <sys/time.h>
24 
25 #include <sys/types.h>
26 #include <fcntl.h>
27 
28 #include <unistd.h>
29 
30 #include "mpi.h"
31 
32 #include "sion.h"
33 #include "sion_debug.h"
34 #include "sion_error_handler.h"
35 #include "sion_internal.h"
36 #include "sion_fd.h"
37 #include "sion_filedesc.h"
38 #include "sion_printts.h"
39 #include "sion_flags.h"
40 
41 #ifdef SION_MPI
42 
43 #include "sion_generic.h"
44 
45 #include "sion_mpi.h"
46 #include "sion_mpi_internal_gen.h"
47 
48 #include "sion_mpi_cb_gen.h"
49 
50 int _sion_mpi_api_aid = -1;
51 
85 int sion_paropen_mpi(const char* fname,
86  const char* file_mode,
87  int* numFiles,
88  MPI_Comm gComm,
89  const MPI_Comm* lComm,
90  sion_int64* chunksize,
91  sion_int32* fsblksize,
92  int* globalrank,
93  FILE** fileptr,
94  char** newfname
95  )
96 {
97  int rc = SION_NOT_SUCCESS, sid = SION_ID_UNDEF;
98  int filenumber, gtasks, gRank, lRank, lSize;
99 
100  _mpi_api_commdata *gen_gcomm;
101 
102  _sion_flags_store* flags_store = NULL;
103 
104  MPI_Comm_size(gComm, &gtasks);
105  MPI_Comm_rank(gComm, &gRank);
106 
107  DPRINTFP((1, "sion_paropen_mpi", gRank, "enter parallel open of file %s\n", fname));
108 
109  /* check parameters */
110  if (lComm == NULL) {
111  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mpi: No lComm variable given"));
112  }
113  if (numFiles == NULL) {
114  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mpi: No numFiles variable given"));
115  }
116 
117  flags_store = _sion_parse_flags(file_mode);
118  /* parse file mode */
119  if ( ! flags_store ) {
120  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mpi: could not parse file mode in %s, aborting ...\n", file_mode));
121  }
122 
123  /* register callbacks for generic interface */
124  if(_sion_mpi_api_aid<0) _sion_mpi_api_aid=_sion_register_callbacks_mpi();
125 
126 
127  if (flags_store->mask&_SION_FMODE_WRITE) {
128  /* file mode WRITE */
129 
130  if (*numFiles <= 0) {
131  /* lComm contains local communicator */
132 
133  rc = _sion_get_info_from_splitted_comm_mpi(gComm, *lComm, numFiles, &filenumber, &lRank, &lSize);
134  if(rc != SION_SUCCESS) {
135  _sion_flags_destroy_store(&flags_store);
136  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mpi: error in _sion_get_info_from_splitted_comm_mpi"));
137  }
138  DPRINTFP((1, "sion_paropen_mpi", gRank, "%d local communicators found\n", *numFiles));
139 
140  } else {
141  /* number of files is given */
142 
143  rc = _sion_gen_info_from_gcomm_mpi(*numFiles, gComm, &filenumber, &lRank, &lSize);
144  if(rc != SION_SUCCESS) {
145  _sion_flags_destroy_store(&flags_store);
146  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mpi: error in _sion_gen_info_from_gcomm_mpi"));
147  }
148  DPRINTFP((1, "sion_paropen_mpi", gRank, "Global communicator divided in %d local communicators\n", *numFiles));
149  }
150 
151  /* overwrite globalrank set by user, necessary for multi-file support */
152  *globalrank = gRank;
153 
154  } else if (flags_store->mask&_SION_FMODE_READ) {
155  /* file mode READ */
156  /* nothing to do info will be returned by generic paropen */
157 
158  /* set to gRank, current rank in global communicator, this is
159  different to older versions of SIONlib, where globalrank comes
160  from file in read case */
161  *globalrank = gRank;
162 
163 
164  if(! (flags_store->mask&_SION_FMODE_BUDDY) ) {
165  lRank=lSize=-1; /* will be set by sion_generic_paropen */
166  } else {
167  /* lvomm must be given for buddy checkpointing */
168  rc = _sion_get_info_from_splitted_comm_mpi(gComm, *lComm, numFiles, &filenumber, &lRank, &lSize);
169  if(rc != SION_SUCCESS) return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mpi: error in _sion_get_info_from_splitted_comm_mpi"));
170  DPRINTFP((1, "sion_paropen_mpi", gRank, "%d local communicators found\n", *numFiles));
171  }
172 
173  } else {
174 
175  _sion_flags_destroy_store(&flags_store);
176  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mpi: unknown file mode"));
177  }
178 
179  /* create generic communicator container */
180  gen_gcomm = (_mpi_api_commdata *) malloc(sizeof(_mpi_api_commdata));
181  if (gen_gcomm == NULL) {
182  _sion_flags_destroy_store(&flags_store);
183  return(_sion_errorprint(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"cannot allocate mpi internal data structure of size %lu (_mpi_api_commdata), aborting ...\n",
184  (unsigned long) sizeof(_mpi_api_commdata)));
185  }
186  gen_gcomm->comm=gComm;
187  gen_gcomm->commset=1;
188  gen_gcomm->local=0;
189  gen_gcomm->rank=gRank;
190  gen_gcomm->size=gtasks;
191  gen_gcomm->lcommgroup=NULL;
192 
193  _sion_flags_destroy_store(&flags_store);
194 
195  DPRINTFP((1, "sion_paropen_mpi", gRank, "enter parallel open of %d files (current name %s) in %s mode\n", *numFiles, fname, file_mode));
196  sid = sion_generic_paropen(_sion_mpi_api_aid, fname, file_mode, chunksize, fsblksize, gen_gcomm, gRank, gtasks, &filenumber, numFiles, &lRank, &lSize,
197  fileptr, newfname);
198  DPRINTFP((1, "sion_paropen_mpi", gRank, "leave parallel open of %d files in %s mode #tasks=%d sid=%d\n", *numFiles, file_mode, lSize, sid));
199 
200  /* test return code from internal open */
201  if ( sid == SION_ID_NOT_VALID ) {
202  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mpi: invalid return code from internal open %d", rc));
203  }
204 
205  DPRINTFP((1, "sion_paropen_mpi", gRank, "leave parallel open of file %s sid=%d\n", fname, sid));
206 
207  return (sid);
208 }
209 
210 
221 int sion_parclose_mpi(int sid)
222 {
223  int rc = 0;
224 
225  DPRINTFP((1, "sion_parclose_mpi", _SION_DEFAULT_RANK, "enter parallel close of sid %d\n", sid));
226 
227  rc = sion_generic_parclose(sid);
228 
229  DPRINTFP((1, "sion_parclose_mpi", _SION_DEFAULT_RANK, "leave parallel close of sid %d rc=%d\n", sid, rc));
230 
231  return (rc);
232 }
233 
234 int sion_parreinit_mpi( int sid,
235  sion_int64 chunksize )
236 {
237  int rc = 0;
238 
239  DPRINTFP((1, "sion_parreinit_mpi", _SION_DEFAULT_RANK, "enter parallel reinit of sid %d\n", sid));
240 
241  rc = sion_generic_parreinit(sid, chunksize);
242 
243  DPRINTFP((1, "sion_parreinit_mpi", _SION_DEFAULT_RANK, "leave parallel reinit of sid %d rc=%d\n", sid, rc));
244 
245  return (rc);
246 }
247 
248 int sion_paropen_mapped_mpi( char *fname,
249  const char *file_mode,
250  int *numFiles,
251  MPI_Comm gComm,
252  int *nlocaltasks,
253  int **globalranks,
254  sion_int64 **chunksizes,
255  int **mapping_filenrs,
256  int **mapping_lranks,
257  sion_int32 *fsblksize,
258  FILE **fileptr) {
259 
260  int sid=SION_ID_UNDEF;
261  int gtasks, gRank;
262  char *lprefix;
263  _mpi_api_commdata *gen_gcomm;
264  _sion_flags_store* flags_store = NULL;
265 
266  MPI_Comm_size(gComm, &gtasks);
267  MPI_Comm_rank(gComm, &gRank);
268 
269  DPRINTFP((1, "sion_paropen_mapped_mpi", gRank, "enter parallel open of file %s\n", fname));
270 
271  /* check parameters */
272  if (numFiles == NULL) {
273  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mapped_mpi: No numFiles variable given"));
274  }
275 
276  lprefix = calloc(SION_FILENAME_LENGTH,1);
277  if (lprefix == NULL) {
278  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mapped_mpi: cannot allocate temporary memory of size %lu (lprefix), aborting ...\n", (unsigned long) SION_FILENAME_LENGTH));
279  }
280 
281  flags_store = _sion_parse_flags(file_mode);
282  if ( ! flags_store ) {
283  free(lprefix);
284  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mapped_mpi: could not parse file mode in %s, aborting ...\n", file_mode));
285  }
286 
287  /* register callbacks for generic interface */
288  if(_sion_mpi_api_aid<0) _sion_mpi_api_aid=_sion_register_callbacks_mpi();
289 
290  if (flags_store->mask&_SION_FMODE_WRITE) {
291  /* file mode WRITE */
292 
293  if (*numFiles <= 0) {
294  _sion_flags_destroy_store(&flags_store);
295  free(lprefix);
296  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mapped_mpi: numFiles variable <= 0 not allowed for mapped files in write mode"));
297  }
298 
299  /* prefix must be used in generic open function */
300  strcpy(lprefix, fname);
301 
302  }
303  else if (flags_store->mask&_SION_FMODE_READ) {
304  /* file mode READ */
305  /* nothing to do here so far, filenumbers and mapping will be determined by in generic routine */
306 
307  } else {
308  _sion_flags_destroy_store(&flags_store);
309  free(lprefix);
310  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mapped_mpi: unknown file mode"));
311  }
312  _sion_flags_destroy_store(&flags_store);
313 
314  /* create generic communicator container */
315  gen_gcomm = (_mpi_api_commdata *) malloc(sizeof(_mpi_api_commdata));
316  if (gen_gcomm == NULL) {
317  free(lprefix);
318  return(_sion_errorprint(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"cannot allocate mpi internal data structure of size %lu (_mpi_api_commdata), aborting ...\n",
319  (unsigned long) sizeof(_mpi_api_commdata)));
320  }
321  gen_gcomm->comm=gComm;
322  gen_gcomm->commset=1;
323  gen_gcomm->local=0;
324  gen_gcomm->rank=gRank;
325  gen_gcomm->size=gtasks;
326  gen_gcomm->lcommgroup=NULL;
327 
328 
329  DPRINTFP((1, "sion_paropen_mapped_mpi", gRank, "enter parallel open of %d files (current name %s) in %s mode (sid=%d)\n", *numFiles, fname, file_mode, sid));
330  sid=sion_generic_paropen_mapped(_sion_mpi_api_aid, fname, file_mode, numFiles, gen_gcomm, gRank, gtasks, nlocaltasks, globalranks, chunksizes,
331  mapping_filenrs, mapping_lranks, fsblksize, fileptr);
332  DPRINTFP((1, "sion_paropen_mapped_mpi", gRank, "leave parallel open of %d files in %s mode #tasks=%d sid=%d\n", *numFiles, file_mode, *nlocaltasks, sid));
333 
334  /* test return code from internal open */
335  if ( sid == SION_ID_NOT_VALID ) {
336  free(lprefix);
337  return(_sion_errorprint_mpi(SION_ID_NOT_VALID,_SION_ERROR_RETURN,"sion_paropen_mapped_mpi: invalid return code from internal open %d", sid));
338  }
339 
340  if(lprefix) free(lprefix);
341  DPRINTFP((1, "sion_paropen_mapped_mpi", gRank, "leave parallel open of file %s sid=%d\n", fname, sid));
342 
343 
344  return(sid);
345 }
346 
347 int sion_parclose_mapped_mpi( int sid ) {
348  int rc = 0;
349 
350  DPRINTFP((1, "sion_parclose_mapped_mpi", _SION_DEFAULT_RANK, "enter parallel close of sid %d\n", sid));
351 
352  rc = sion_generic_parclose_mapped(sid);
353 
354  DPRINTFP((1, "sion_parclose_mapped_mpi", _SION_DEFAULT_RANK, "leave parallel close of sid %d rc=%d\n", sid, rc));
355 
356  return(rc);
357 }
358 
359 /* end of ifdef MPI */
360 #endif
int sion_generic_paropen(int aid, const char *fname, const char *file_mode, sion_int64 *chunksize, sion_int32 *fsblksize, void *gcommgroup, int grank, int gsize, int *filenumber, int *numfiles, const int *lrank, const int *lsize, FILE **fileptr, char **newfname)
Open a sion file a generic interface.
Definition: sion_generic.c:351
int sion_parclose_mpi(int sid)
Close a sion file using MPI.
Definition: sion_mpi_gen.c:221
int sion_paropen_mpi(const char *fname, const char *file_mode, int *numFiles, MPI_Comm gComm, const MPI_Comm *lComm, sion_int64 *chunksize, sion_int32 *fsblksize, int *globalrank, FILE **fileptr, char **newfname)
Open a sion file using MPI.
Definition: sion_mpi_gen.c:85
_sion_flags_store * _sion_parse_flags(const char *flags)
Parse flags and return a flags store with key value pairs.
Definition: sion_flags.c:291
int _sion_gen_info_from_gcomm_mpi(int numFiles, MPI_Comm gComm, int *filenumber, int *lrank, int *lsize)
Splits a Communicator in numfiles different communicators.
Sion Time Stamp Header.