1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506
|
/* ----------------------------------------------------------------------
LAMMPS - Large-scale Atomic/Molecular Massively Parallel Simulator
http://lammps.sandia.gov, Sandia National Laboratories
Steve Plimpton, sjplimp@sandia.gov
Copyright (2003) Sandia Corporation. Under the terms of Contract
DE-AC04-94AL85000 with Sandia Corporation, the U.S. Government retains
certain rights in this software. This software is distributed under
the GNU General Public License.
See the README file in the top-level LAMMPS directory.
------------------------------------------------------------------------- */
#include "mpi.h"
#include "string.h"
#include "comm.h"
#include "universe.h"
#include "atom.h"
#include "force.h"
#include "domain.h"
#include "group.h"
#include "procmap.h"
#include "memory.h"
#include "error.h"
using namespace LAMMPS_NS;
enum{SINGLE,MULTI}; // same as in Comm sub-styles
enum{MULTIPLE}; // same as in ProcMap
enum{ONELEVEL,TWOLEVEL,NUMA,CUSTOM};
enum{CART,CARTREORDER,XYZ};
/* ---------------------------------------------------------------------- */
Comm::Comm(LAMMPS *lmp) : Pointers(lmp)
{
MPI_Comm_rank(world,&me);
MPI_Comm_size(world,&nprocs);
mode = 0;
bordergroup = 0;
cutghostuser = 0.0;
ghost_velocity = 0;
user_procgrid[0] = user_procgrid[1] = user_procgrid[2] = 0;
coregrid[0] = coregrid[1] = coregrid[2] = 1;
gridflag = ONELEVEL;
mapflag = CART;
customfile = NULL;
recv_from_partition = send_to_partition = -1;
otherflag = 0;
outfile = NULL;
xsplit = ysplit = zsplit = NULL;
}
/* ---------------------------------------------------------------------- */
Comm::~Comm()
{
memory->destroy(xsplit);
memory->destroy(ysplit);
memory->destroy(zsplit);
delete [] customfile;
delete [] outfile;
}
/* ----------------------------------------------------------------------
modify communication params
invoked from input script by comm_modify command
------------------------------------------------------------------------- */
void Comm::modify_params(int narg, char **arg)
{
if (narg < 1) error->all(FLERR,"Illegal comm_modify command");
int iarg = 0;
while (iarg < narg) {
if (strcmp(arg[iarg],"mode") == 0) {
if (iarg+2 > narg) error->all(FLERR,"Illegal comm_modify command");
if (strcmp(arg[iarg+1],"single") == 0) mode = SINGLE;
else if (strcmp(arg[iarg+1],"multi") == 0) mode = MULTI;
else error->all(FLERR,"Illegal comm_modify command");
iarg += 2;
} else if (strcmp(arg[iarg],"group") == 0) {
if (iarg+2 > narg) error->all(FLERR,"Illegal comm_modify command");
bordergroup = group->find(arg[iarg+1]);
if (bordergroup < 0)
error->all(FLERR,"Invalid group in comm_modify command");
if (bordergroup && (atom->firstgroupname == NULL ||
strcmp(arg[iarg+1],atom->firstgroupname) != 0))
error->all(FLERR,"Comm_modify group != atom_modify first group");
iarg += 2;
} else if (strcmp(arg[iarg],"cutoff") == 0) {
if (iarg+2 > narg) error->all(FLERR,"Illegal comm_modify command");
cutghostuser = force->numeric(FLERR,arg[iarg+1]);
if (cutghostuser < 0.0)
error->all(FLERR,"Invalid cutoff in comm_modify command");
iarg += 2;
} else if (strcmp(arg[iarg],"vel") == 0) {
if (iarg+2 > narg) error->all(FLERR,"Illegal comm_modify command");
if (strcmp(arg[iarg+1],"yes") == 0) ghost_velocity = 1;
else if (strcmp(arg[iarg+1],"no") == 0) ghost_velocity = 0;
else error->all(FLERR,"Illegal comm_modify command");
iarg += 2;
} else error->all(FLERR,"Illegal comm_modify command");
}
}
/* ----------------------------------------------------------------------
set dimensions for 3d grid of processors, and associated flags
invoked from input script by processors command
------------------------------------------------------------------------- */
void Comm::set_processors(int narg, char **arg)
{
if (narg < 3) error->all(FLERR,"Illegal processors command");
if (strcmp(arg[0],"*") == 0) user_procgrid[0] = 0;
else user_procgrid[0] = force->inumeric(FLERR,arg[0]);
if (strcmp(arg[1],"*") == 0) user_procgrid[1] = 0;
else user_procgrid[1] = force->inumeric(FLERR,arg[1]);
if (strcmp(arg[2],"*") == 0) user_procgrid[2] = 0;
else user_procgrid[2] = force->inumeric(FLERR,arg[2]);
if (user_procgrid[0] < 0 || user_procgrid[1] < 0 || user_procgrid[2] < 0)
error->all(FLERR,"Illegal processors command");
int p = user_procgrid[0]*user_procgrid[1]*user_procgrid[2];
if (p && p != nprocs)
error->all(FLERR,"Specified processors != physical processors");
int iarg = 3;
while (iarg < narg) {
if (strcmp(arg[iarg],"grid") == 0) {
if (iarg+2 > narg) error->all(FLERR,"Illegal processors command");
if (strcmp(arg[iarg+1],"onelevel") == 0) {
gridflag = ONELEVEL;
} else if (strcmp(arg[iarg+1],"twolevel") == 0) {
if (iarg+6 > narg) error->all(FLERR,"Illegal processors command");
gridflag = TWOLEVEL;
ncores = force->inumeric(FLERR,arg[iarg+2]);
if (strcmp(arg[iarg+3],"*") == 0) user_coregrid[0] = 0;
else user_coregrid[0] = force->inumeric(FLERR,arg[iarg+3]);
if (strcmp(arg[iarg+4],"*") == 0) user_coregrid[1] = 0;
else user_coregrid[1] = force->inumeric(FLERR,arg[iarg+4]);
if (strcmp(arg[iarg+5],"*") == 0) user_coregrid[2] = 0;
else user_coregrid[2] = force->inumeric(FLERR,arg[iarg+5]);
if (ncores <= 0 || user_coregrid[0] < 0 ||
user_coregrid[1] < 0 || user_coregrid[2] < 0)
error->all(FLERR,"Illegal processors command");
iarg += 4;
} else if (strcmp(arg[iarg+1],"numa") == 0) {
gridflag = NUMA;
} else if (strcmp(arg[iarg],"custom") == 0) {
if (iarg+3 > narg) error->all(FLERR,"Illegal processors command");
gridflag = CUSTOM;
delete [] customfile;
int n = strlen(arg[iarg+2]) + 1;
customfile = new char[n];
strcpy(customfile,arg[iarg+2]);
iarg += 1;
} else error->all(FLERR,"Illegal processors command");
iarg += 2;
} else if (strcmp(arg[iarg],"map") == 0) {
if (iarg+2 > narg) error->all(FLERR,"Illegal processors command");
if (strcmp(arg[iarg+1],"cart") == 0) mapflag = CART;
else if (strcmp(arg[iarg+1],"cart/reorder") == 0) mapflag = CARTREORDER;
else if (strcmp(arg[iarg+1],"xyz") == 0 ||
strcmp(arg[iarg+1],"xzy") == 0 ||
strcmp(arg[iarg+1],"yxz") == 0 ||
strcmp(arg[iarg+1],"yzx") == 0 ||
strcmp(arg[iarg+1],"zxy") == 0 ||
strcmp(arg[iarg+1],"zyx") == 0) {
mapflag = XYZ;
strcpy(xyz,arg[iarg+1]);
} else error->all(FLERR,"Illegal processors command");
iarg += 2;
} else if (strcmp(arg[iarg],"part") == 0) {
if (iarg+4 > narg) error->all(FLERR,"Illegal processors command");
if (universe->nworlds == 1)
error->all(FLERR,
"Cannot use processors part command "
"without using partitions");
int isend = force->inumeric(FLERR,arg[iarg+1]);
int irecv = force->inumeric(FLERR,arg[iarg+2]);
if (isend < 1 || isend > universe->nworlds ||
irecv < 1 || irecv > universe->nworlds || isend == irecv)
error->all(FLERR,"Invalid partitions in processors part command");
if (isend-1 == universe->iworld) {
if (send_to_partition >= 0)
error->all(FLERR,
"Sending partition in processors part command "
"is already a sender");
send_to_partition = irecv-1;
}
if (irecv-1 == universe->iworld) {
if (recv_from_partition >= 0)
error->all(FLERR,
"Receiving partition in processors part command "
"is already a receiver");
recv_from_partition = isend-1;
}
// only receiver has otherflag dependency
if (strcmp(arg[iarg+3],"multiple") == 0) {
if (universe->iworld == irecv-1) {
otherflag = 1;
other_style = MULTIPLE;
}
} else error->all(FLERR,"Illegal processors command");
iarg += 4;
} else if (strcmp(arg[iarg],"file") == 0) {
if (iarg+2 > narg) error->all(FLERR,"Illegal processors command");
delete [] outfile;
int n = strlen(arg[iarg+1]) + 1;
outfile = new char[n];
strcpy(outfile,arg[iarg+1]);
iarg += 2;
} else error->all(FLERR,"Illegal processors command");
}
// error checks
if (gridflag == NUMA && mapflag != CART)
error->all(FLERR,"Processors grid numa and map style are incompatible");
if (otherflag && (gridflag == NUMA || gridflag == CUSTOM))
error->all(FLERR,
"Processors part option and grid style are incompatible");
}
/* ----------------------------------------------------------------------
create a 3d grid of procs based on Nprocs and box size & shape
map processors to grid, setup xyz split for a uniform grid
------------------------------------------------------------------------- */
void Comm::set_proc_grid(int outflag)
{
// recv 3d proc grid of another partition if my 3d grid depends on it
if (recv_from_partition >= 0) {
MPI_Status status;
if (me == 0) {
MPI_Recv(other_procgrid,3,MPI_INT,
universe->root_proc[recv_from_partition],0,
universe->uworld,&status);
MPI_Recv(other_coregrid,3,MPI_INT,
universe->root_proc[recv_from_partition],0,
universe->uworld,&status);
}
MPI_Bcast(other_procgrid,3,MPI_INT,0,world);
MPI_Bcast(other_coregrid,3,MPI_INT,0,world);
}
// create ProcMap class to create 3d grid and map procs to it
ProcMap *pmap = new ProcMap(lmp);
// create 3d grid of processors
// produces procgrid and coregrid (if relevant)
if (gridflag == ONELEVEL) {
pmap->onelevel_grid(nprocs,user_procgrid,procgrid,
otherflag,other_style,other_procgrid,other_coregrid);
} else if (gridflag == TWOLEVEL) {
pmap->twolevel_grid(nprocs,user_procgrid,procgrid,
ncores,user_coregrid,coregrid,
otherflag,other_style,other_procgrid,other_coregrid);
} else if (gridflag == NUMA) {
pmap->numa_grid(nprocs,user_procgrid,procgrid,coregrid);
} else if (gridflag == CUSTOM) {
pmap->custom_grid(customfile,nprocs,user_procgrid,procgrid);
}
// error check on procgrid
// should not be necessary due to ProcMap
if (procgrid[0]*procgrid[1]*procgrid[2] != nprocs)
error->all(FLERR,"Bad grid of processors");
if (domain->dimension == 2 && procgrid[2] != 1)
error->all(FLERR,"Processor count in z must be 1 for 2d simulation");
// grid2proc[i][j][k] = proc that owns i,j,k location in 3d grid
if (grid2proc) memory->destroy(grid2proc);
memory->create(grid2proc,procgrid[0],procgrid[1],procgrid[2],
"comm:grid2proc");
// map processor IDs to 3d processor grid
// produces myloc, procneigh, grid2proc
if (gridflag == ONELEVEL) {
if (mapflag == CART)
pmap->cart_map(0,procgrid,myloc,procneigh,grid2proc);
else if (mapflag == CARTREORDER)
pmap->cart_map(1,procgrid,myloc,procneigh,grid2proc);
else if (mapflag == XYZ)
pmap->xyz_map(xyz,procgrid,myloc,procneigh,grid2proc);
} else if (gridflag == TWOLEVEL) {
if (mapflag == CART)
pmap->cart_map(0,procgrid,ncores,coregrid,myloc,procneigh,grid2proc);
else if (mapflag == CARTREORDER)
pmap->cart_map(1,procgrid,ncores,coregrid,myloc,procneigh,grid2proc);
else if (mapflag == XYZ)
pmap->xyz_map(xyz,procgrid,ncores,coregrid,myloc,procneigh,grid2proc);
} else if (gridflag == NUMA) {
pmap->numa_map(0,coregrid,myloc,procneigh,grid2proc);
} else if (gridflag == CUSTOM) {
pmap->custom_map(procgrid,myloc,procneigh,grid2proc);
}
// print 3d grid info to screen and logfile
if (outflag && me == 0) {
if (screen) {
fprintf(screen," %d by %d by %d MPI processor grid\n",
procgrid[0],procgrid[1],procgrid[2]);
if (gridflag == NUMA || gridflag == TWOLEVEL)
fprintf(screen," %d by %d by %d core grid within node\n",
coregrid[0],coregrid[1],coregrid[2]);
}
if (logfile) {
fprintf(logfile," %d by %d by %d MPI processor grid\n",
procgrid[0],procgrid[1],procgrid[2]);
if (gridflag == NUMA || gridflag == TWOLEVEL)
fprintf(logfile," %d by %d by %d core grid within node\n",
coregrid[0],coregrid[1],coregrid[2]);
}
}
// print 3d grid details to outfile
if (outfile) pmap->output(outfile,procgrid,grid2proc);
// free ProcMap class
delete pmap;
// set xsplit,ysplit,zsplit for uniform spacings
memory->destroy(xsplit);
memory->destroy(ysplit);
memory->destroy(zsplit);
memory->create(xsplit,procgrid[0]+1,"comm:xsplit");
memory->create(ysplit,procgrid[1]+1,"comm:ysplit");
memory->create(zsplit,procgrid[2]+1,"comm:zsplit");
for (int i = 0; i < procgrid[0]; i++) xsplit[i] = i * 1.0/procgrid[0];
for (int i = 0; i < procgrid[1]; i++) ysplit[i] = i * 1.0/procgrid[1];
for (int i = 0; i < procgrid[2]; i++) zsplit[i] = i * 1.0/procgrid[2];
xsplit[procgrid[0]] = ysplit[procgrid[1]] = zsplit[procgrid[2]] = 1.0;
// set lamda box params after procs are assigned
// only set once unless load-balancing occurs
if (domain->triclinic) domain->set_lamda_box();
// send my 3d proc grid to another partition if requested
if (send_to_partition >= 0) {
if (me == 0) {
MPI_Send(procgrid,3,MPI_INT,
universe->root_proc[send_to_partition],0,
universe->uworld);
MPI_Send(coregrid,3,MPI_INT,
universe->root_proc[send_to_partition],0,
universe->uworld);
}
}
}
/* ----------------------------------------------------------------------
communicate inbuf around full ring of processors with messtag
nbytes = size of inbuf = n datums * nper bytes
callback() is invoked to allow caller to process/update each proc's inbuf
if self=1 (default), then callback() is invoked on final iteration
using original inbuf, which may have been updated
for non-NULL outbuf, final updated inbuf is copied to it
outbuf = inbuf is OK
------------------------------------------------------------------------- */
void Comm::ring(int n, int nper, void *inbuf, int messtag,
void (*callback)(int, char *), void *outbuf, int self)
{
MPI_Request request;
MPI_Status status;
int nbytes = n*nper;
int maxbytes;
MPI_Allreduce(&nbytes,&maxbytes,1,MPI_INT,MPI_MAX,world);
char *buf,*bufcopy;
memory->create(buf,maxbytes,"comm:buf");
memory->create(bufcopy,maxbytes,"comm:bufcopy");
memcpy(buf,inbuf,nbytes);
int next = me + 1;
int prev = me - 1;
if (next == nprocs) next = 0;
if (prev < 0) prev = nprocs - 1;
for (int loop = 0; loop < nprocs; loop++) {
if (me != next) {
MPI_Irecv(bufcopy,maxbytes,MPI_CHAR,prev,messtag,world,&request);
MPI_Send(buf,nbytes,MPI_CHAR,next,messtag,world);
MPI_Wait(&request,&status);
MPI_Get_count(&status,MPI_CHAR,&nbytes);
memcpy(buf,bufcopy,nbytes);
}
if (self || loop < nprocs-1) callback(nbytes/nper,buf);
}
if (outbuf) memcpy(outbuf,buf,nbytes);
memory->destroy(buf);
memory->destroy(bufcopy);
}
/* ----------------------------------------------------------------------
proc 0 reads Nlines from file into buf and bcasts buf to all procs
caller allocates buf to max size needed
each line is terminated by newline, even if last line in file is not
return 0 if successful, 1 if get EOF error before read is complete
------------------------------------------------------------------------- */
int Comm::read_lines_from_file(FILE *fp, int nlines, int maxline, char *buf)
{
int m;
if (me == 0) {
m = 0;
for (int i = 0; i < nlines; i++) {
if (!fgets(&buf[m],maxline,fp)) {
m = 0;
break;
}
m += strlen(&buf[m]);
}
if (m) {
if (buf[m-1] != '\n') strcpy(&buf[m++],"\n");
m++;
}
}
MPI_Bcast(&m,1,MPI_INT,0,world);
if (m == 0) return 1;
MPI_Bcast(buf,m,MPI_CHAR,0,world);
return 0;
}
/* ----------------------------------------------------------------------
proc 0 reads Nlines from file into buf and bcasts buf to all procs
caller allocates buf to max size needed
each line is terminated by newline, even if last line in file is not
return 0 if successful, 1 if get EOF error before read is complete
------------------------------------------------------------------------- */
int Comm::read_lines_from_file_universe(FILE *fp, int nlines, int maxline,
char *buf)
{
int m;
int me_universe = universe->me;
MPI_Comm uworld = universe->uworld;
if (me_universe == 0) {
m = 0;
for (int i = 0; i < nlines; i++) {
if (!fgets(&buf[m],maxline,fp)) {
m = 0;
break;
}
m += strlen(&buf[m]);
}
if (m) {
if (buf[m-1] != '\n') strcpy(&buf[m++],"\n");
m++;
}
}
MPI_Bcast(&m,1,MPI_INT,0,uworld);
if (m == 0) return 1;
MPI_Bcast(buf,m,MPI_CHAR,0,uworld);
return 0;
}
|