Homec4science
Diffusion lammps (efficient_neuronet) Browse

lammps/src/comm_tiled.cpp
76acb8caf15fefficient_neuronet

Last Change
Raw File

comm_tiled.cpp
View Options

/* ----------------------------------------------------------------------
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 <string.h>
#include "comm_tiled.h"
#include "comm_brick.h"
#include "atom.h"
#include "atom_vec.h"
#include "domain.h"
#include "force.h"
#include "pair.h"
#include "neighbor.h"
#include "modify.h"
#include "fix.h"
#include "compute.h"
#include "output.h"
#include "dump.h"
#include "memory.h"
#include "error.h"
using namespace LAMMPS_NS;
#define BUFFACTOR 1.5
#define BUFFACTOR 1.5
#define BUFMIN 1000
#define BUFEXTRA 1000
#define EPSILON 1.0e-6
#define DELTA_PROCS 16
enum{SINGLE,MULTI}; // same as in Comm
enum{LAYOUT_UNIFORM,LAYOUT_NONUNIFORM,LAYOUT_TILED}; // several files
/* ---------------------------------------------------------------------- */
CommTiled::CommTiled(LAMMPS *lmp) : Comm(lmp)
{
if (lmp->kokkos)
error->all(FLERR,"KOKKOS package does not yet support comm_style tiled");
style = 1;
layout = LAYOUT_UNIFORM;
pbc_flag = NULL;
init_buffers();
}
/* ---------------------------------------------------------------------- */
//IMPORTANT: we *MUST* pass "*oldcomm" to the Comm initializer here, as
// the code below *requires* that the (implicit) copy constructor
// for Comm is run and thus creating a shallow copy of "oldcomm".
// The call to Comm::copy_arrays() then converts the shallow copy
// into a deep copy of the class with the new layout.
//
CommTiled::CommTiled(LAMMPS *lmp, Comm *oldcomm) : Comm(*oldcomm)
{
if (lmp->kokkos)
error->all(FLERR,"KOKKOS package does not yet support comm_style tiled");
style = 1;
layout = oldcomm->layout;
Comm::copy_arrays(oldcomm);
init_buffers();
}
/* ---------------------------------------------------------------------- */
CommTiled::~CommTiled()
{
memory->destroy(buf_send);
memory->destroy(buf_recv);
memory->destroy(overlap);
deallocate_swap(nswap);
memory->sfree(rcbinfo);
}
/* ----------------------------------------------------------------------
initialize comm buffers and other data structs local to CommTiled
------------------------------------------------------------------------- */
void CommTiled::init_buffers()
{
// bufextra = max size of one exchanged atom
// = allowed overflow of sendbuf in exchange()
// atomvec, fix reset these 2 maxexchange values if needed
// only necessary if their size > BUFEXTRA
maxexchange = maxexchange_atom + maxexchange_fix;
bufextra = maxexchange + BUFEXTRA;
maxsend = BUFMIN;
memory->create(buf_send,maxsend+bufextra,"comm:buf_send");
maxrecv = BUFMIN;
memory->create(buf_recv,maxrecv,"comm:buf_recv");
maxoverlap = 0;
overlap = NULL;
nswap = 2 * domain->dimension;
allocate_swap(nswap);
rcbinfo = NULL;
}
/* ---------------------------------------------------------------------- */
void CommTiled::init()
{
Comm::init();
// temporary restrictions
if (triclinic)
error->all(FLERR,"Cannot yet use comm_style tiled with triclinic box");
if (mode == MULTI)
error->all(FLERR,"Cannot yet use comm_style tiled with multi-mode comm");
}
/* ----------------------------------------------------------------------
setup spatial-decomposition communication patterns
function of neighbor cutoff(s) & cutghostuser & current box size and tiling
------------------------------------------------------------------------- */
void CommTiled::setup()
{
int i,j,n;
// domain properties used in setup method and methods it calls
dimension = domain->dimension;
prd = domain->prd;
boxlo = domain->boxlo;
boxhi = domain->boxhi;
sublo = domain->sublo;
subhi = domain->subhi;
int *periodicity = domain->periodicity;
// set function pointers
if (layout != LAYOUT_TILED) {
box_drop = &CommTiled::box_drop_brick;
box_other = &CommTiled::box_other_brick;
box_touch = &CommTiled::box_touch_brick;
point_drop = &CommTiled::point_drop_brick;
} else {
box_drop = &CommTiled::box_drop_tiled;
box_other = &CommTiled::box_other_tiled;
box_touch = &CommTiled::box_touch_tiled;
point_drop = &CommTiled::point_drop_tiled;
}
// if RCB decomp exists and just changed, gather needed global RCB info
if (layout == LAYOUT_TILED) coord2proc_setup();
// set cutoff for comm forward and comm reverse
// check that cutoff < any periodic box length
double cut = MAX(neighbor->cutneighmax,cutghostuser);
cutghost[0] = cutghost[1] = cutghost[2] = cut;
if ((periodicity[0] && cut > prd[0]) ||
(periodicity[1] && cut > prd[1]) ||
(dimension == 3 && periodicity[2] && cut > prd[2]))
error->all(FLERR,"Communication cutoff for comm_style tiled "
"cannot exceed periodic box length");
// if cut = 0.0, set to epsilon to induce nearest neighbor comm
// this is b/c sendproc is used below to infer touching exchange procs
// exchange procs will be empty (leading to lost atoms) if sendproc = 0
// will reset sendproc/etc to 0 after exchange is setup, down below
int cutzero = 0;
if (cut == 0.0) {
cutzero = 1;
cut = MIN(prd[0],prd[1]);
if (dimension == 3) cut = MIN(cut,prd[2]);
cut *= EPSILON*EPSILON;
}
// setup forward/reverse communication
// loop over 6 swap directions
// determine which procs I will send to and receive from in each swap
// done by intersecting ghost box with all proc sub-boxes it overlaps
// sets nsendproc, nrecvproc, sendproc, recvproc
// sets sendother, recvother, sendself, pbc_flag, pbc, sendbox
// resets nprocmax
int noverlap1,indexme;
double lo1[3],hi1[3],lo2[3],hi2[3];
int one,two;
int iswap = 0;
for (int idim = 0; idim < dimension; idim++) {
for (int idir = 0; idir < 2; idir++) {
// one = first ghost box in same periodic image
// two = second ghost box wrapped across periodic boundary
// either may not exist
one = 1;
lo1[0] = sublo[0]; lo1[1] = sublo[1]; lo1[2] = sublo[2];
hi1[0] = subhi[0]; hi1[1] = subhi[1]; hi1[2] = subhi[2];
if (idir == 0) {
lo1[idim] = sublo[idim] - cut;
hi1[idim] = sublo[idim];
} else {
lo1[idim] = subhi[idim];
hi1[idim] = subhi[idim] + cut;
}
two = 0;
if (idir == 0 && periodicity[idim] && lo1[idim] < boxlo[idim]) two = 1;
if (idir == 1 && periodicity[idim] && hi1[idim] > boxhi[idim]) two = 1;
if (two) {
lo2[0] = sublo[0]; lo2[1] = sublo[1]; lo2[2] = sublo[2];
hi2[0] = subhi[0]; hi2[1] = subhi[1]; hi2[2] = subhi[2];
if (idir == 0) {
lo2[idim] = lo1[idim] + prd[idim];
hi2[idim] = boxhi[idim];
if (sublo[idim] == boxlo[idim]) one = 0;
} else {
lo2[idim] = boxlo[idim];
hi2[idim] = hi1[idim] - prd[idim];
if (subhi[idim] == boxhi[idim]) one = 0;
}
}
if (one) {
if (idir == 0) lo1[idim] = MAX(lo1[idim],boxlo[idim]);
else hi1[idim] = MIN(hi1[idim],boxhi[idim]);
if (lo1[idim] == hi1[idim]) one = 0;
}
// noverlap = # of overlaps of box1/2 with procs via box_drop()
// overlap = list of overlapping procs
// if overlap with self, indexme = index of me in list
indexme = -1;
noverlap = 0;
if (one) (this->*box_drop)(idim,lo1,hi1,indexme);
noverlap1 = noverlap;
if (two) (this->*box_drop)(idim,lo2,hi2,indexme);
// if self is in overlap list, move it to end of list
if (indexme >= 0) {
int tmp = overlap[noverlap-1];
overlap[noverlap-1] = overlap[indexme];
overlap[indexme] = tmp;
}
// reallocate 2nd dimensions of all send/recv arrays, based on noverlap
// # of sends of this swap = # of recvs of iswap +/- 1
if (noverlap > nprocmax[iswap]) {
int oldmax = nprocmax[iswap];
while (nprocmax[iswap] < noverlap) nprocmax[iswap] += DELTA_PROCS;
grow_swap_send(iswap,nprocmax[iswap],oldmax);
if (idir == 0) grow_swap_recv(iswap+1,nprocmax[iswap]);
else grow_swap_recv(iswap-1,nprocmax[iswap]);
}
// overlap how has list of noverlap procs
// includes PBC effects
if (noverlap && overlap[noverlap-1] == me) sendself[iswap] = 1;
else sendself[iswap] = 0;
if (noverlap && noverlap-sendself[iswap]) sendother[iswap] = 1;
else sendother[iswap] = 0;
nsendproc[iswap] = noverlap;
for (i = 0; i < noverlap; i++) sendproc[iswap][i] = overlap[i];
if (idir == 0) {
recvother[iswap+1] = sendother[iswap];
nrecvproc[iswap+1] = noverlap;
for (i = 0; i < noverlap; i++) recvproc[iswap+1][i] = overlap[i];
} else {
recvother[iswap-1] = sendother[iswap];
nrecvproc[iswap-1] = noverlap;
for (i = 0; i < noverlap; i++) recvproc[iswap-1][i] = overlap[i];
}
// compute sendbox for each of my sends
// obox = intersection of ghostbox with other proc's sub-domain
// sbox = what I need to send to other proc
// = sublo to MIN(sublo+cut,subhi) in idim, for idir = 0
// = MIN(subhi-cut,sublo) to subhi in idim, for idir = 1
// = obox in other 2 dims
// if sbox touches other proc's sub-box boundaries in lower dims,
// extend sbox in those lower dims to include ghost atoms
double oboxlo[3],oboxhi[3],sbox[6];
for (i = 0; i < noverlap; i++) {
pbc_flag[iswap][i] = 0;
pbc[iswap][i][0] = pbc[iswap][i][1] = pbc[iswap][i][2] =
pbc[iswap][i][3] = pbc[iswap][i][4] = pbc[iswap][i][5] = 0;
(this->*box_other)(idim,idir,overlap[i],oboxlo,oboxhi);
if (i < noverlap1) {
sbox[0] = MAX(oboxlo[0],lo1[0]);
sbox[1] = MAX(oboxlo[1],lo1[1]);
sbox[2] = MAX(oboxlo[2],lo1[2]);
sbox[3] = MIN(oboxhi[0],hi1[0]);
sbox[4] = MIN(oboxhi[1],hi1[1]);
sbox[5] = MIN(oboxhi[2],hi1[2]);
} else {
pbc_flag[iswap][i] = 1;
if (idir == 0) pbc[iswap][i][idim] = 1;
else pbc[iswap][i][idim] = -1;
sbox[0] = MAX(oboxlo[0],lo2[0]);
sbox[1] = MAX(oboxlo[1],lo2[1]);
sbox[2] = MAX(oboxlo[2],lo2[2]);
sbox[3] = MIN(oboxhi[0],hi2[0]);
sbox[4] = MIN(oboxhi[1],hi2[1]);
sbox[5] = MIN(oboxhi[2],hi2[2]);
}
if (idir == 0) {
sbox[idim] = sublo[idim];
if (i < noverlap1) sbox[3+idim] = MIN(sbox[3+idim]+cut,subhi[idim]);
else sbox[3+idim] = MIN(sbox[3+idim]-prd[idim]+cut,subhi[idim]);
} else {
if (i < noverlap1) sbox[idim] = MAX(sbox[idim]-cut,sublo[idim]);
else sbox[idim] = MAX(sbox[idim]+prd[idim]-cut,sublo[idim]);
sbox[3+idim] = subhi[idim];
}
if (idim >= 1) {
if (sbox[0] == oboxlo[0]) sbox[0] -= cut;
if (sbox[3] == oboxhi[0]) sbox[3] += cut;
}
if (idim == 2) {
if (sbox[1] == oboxlo[1]) sbox[1] -= cut;
if (sbox[4] == oboxhi[1]) sbox[4] += cut;
}
memcpy(sendbox[iswap][i],sbox,6*sizeof(double));
}
iswap++;
}
}
// setup exchange communication = subset of forward/reverse comm procs
// loop over dimensions
// determine which procs I will exchange with in each dimension
// subset of procs that touch my proc in forward/reverse comm
// sets nexchproc & exchproc, resets nexchprocmax
int proc;
for (int idim = 0; idim < dimension; idim++) {
// overlap = list of procs that touch my sub-box in idim
// proc can appear twice in list if touches in both directions
// 2nd add-to-list checks to insure each proc appears exactly once
noverlap = 0;
iswap = 2*idim;
n = nsendproc[iswap];
for (i = 0; i < n; i++) {
proc = sendproc[iswap][i];
if (proc == me) continue;
if ((this->*box_touch)(proc,idim,0)) {
if (noverlap == maxoverlap) {
maxoverlap += DELTA_PROCS;
memory->grow(overlap,maxoverlap,"comm:overlap");
}
overlap[noverlap++] = proc;
}
}
noverlap1 = noverlap;
iswap = 2*idim+1;
n = nsendproc[iswap];
MPI_Barrier(world);
for (i = 0; i < n; i++) {
proc = sendproc[iswap][i];
if (proc == me) continue;
if ((this->*box_touch)(proc,idim,1)) {
for (j = 0; j < noverlap1; j++)
if (overlap[j] == proc) break;
if (j < noverlap1) continue;
if (noverlap == maxoverlap) {
maxoverlap += DELTA_PROCS;
memory->grow(overlap,maxoverlap,"comm:overlap");
}
overlap[noverlap++] = proc;
}
}
MPI_Barrier(world);
// reallocate exchproc and exchnum if needed based on noverlap
if (noverlap > nexchprocmax[idim]) {
while (nexchprocmax[idim] < noverlap) nexchprocmax[idim] += DELTA_PROCS;
delete [] exchproc[idim];
exchproc[idim] = new int[nexchprocmax[idim]];
delete [] exchnum[idim];
exchnum[idim] = new int[nexchprocmax[idim]];
}
nexchproc[idim] = noverlap;
for (i = 0; i < noverlap; i++) exchproc[idim][i] = overlap[i];
}
// reset sendproc/etc to 0 if cut is really 0.0
if (cutzero) {
for (i = 0; i < nswap; i++) {
nsendproc[i] = nrecvproc[i] =
sendother[i] = recvother[i] = sendself[i] = 0;
}
}
// reallocate MPI Requests and Statuses as needed
int nmax = 0;
for (i = 0; i < nswap; i++) nmax = MAX(nmax,nprocmax[i]);
for (i = 0; i < dimension; i++) nmax = MAX(nmax,nexchprocmax[i]);
if (nmax > maxreqstat) {
maxreqstat = nmax;
delete [] requests;
requests = new MPI_Request[maxreqstat];
}
}
/* ----------------------------------------------------------------------
forward communication of atom coords every timestep
other per-atom attributes may also be sent via pack/unpack routines
------------------------------------------------------------------------- */
void CommTiled::forward_comm(int dummy)
{
int i,irecv,n,nsend,nrecv;
AtomVec *avec = atom->avec;
double **x = atom->x;
// exchange data with another set of procs in each swap
// post recvs from all procs except self
// send data to all procs except self
// copy data to self if sendself is set
// wait on all procs except self and unpack received data
// if comm_x_only set, exchange or copy directly to x, don't unpack
for (int iswap = 0; iswap < nswap; iswap++) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (comm_x_only) {
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Irecv(x[firstrecv[iswap][i]],size_forward_recv[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world,&requests[i]);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
n = avec->pack_comm(sendnum[iswap][i],sendlist[iswap][i],
buf_send,pbc_flag[iswap][i],pbc[iswap][i]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
avec->pack_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
x[firstrecv[iswap][nrecv]],pbc_flag[iswap][nsend],
pbc[iswap][nsend]);
}
if (recvother[iswap]) MPI_Waitall(nrecv,requests,MPI_STATUS_IGNORE);
} else if (ghost_velocity) {
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Irecv(&buf_recv[size_forward*forward_recv_offset[iswap][i]],
size_forward_recv[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world,&requests[i]);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
n = avec->pack_comm_vel(sendnum[iswap][i],sendlist[iswap][i],
buf_send,pbc_flag[iswap][i],pbc[iswap][i]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
avec->pack_comm_vel(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send,pbc_flag[iswap][nsend],pbc[iswap][nsend]);
avec->unpack_comm_vel(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
MPI_Waitany(nrecv,requests,&irecv,MPI_STATUS_IGNORE);
avec->unpack_comm_vel(recvnum[iswap][irecv],firstrecv[iswap][irecv],
&buf_recv[size_forward*
forward_recv_offset[iswap][irecv]]);
}
}
} else {
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Irecv(&buf_recv[size_forward*forward_recv_offset[iswap][i]],
size_forward_recv[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world,&requests[i]);
}
if (sendother[iswap]) {
for (i = 0; i < nsendproc[iswap]; i++) {
n = avec->pack_comm(sendnum[iswap][i],sendlist[iswap][i],
buf_send,pbc_flag[iswap][i],pbc[iswap][i]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
avec->pack_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send,pbc_flag[iswap][nsend],pbc[iswap][nsend]);
avec->unpack_comm(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
MPI_Waitany(nrecv,requests,&irecv,MPI_STATUS_IGNORE);
avec->unpack_comm(recvnum[iswap][irecv],firstrecv[iswap][irecv],
&buf_recv[size_forward*
forward_recv_offset[iswap][irecv]]);
}
}
}
}
}
/* ----------------------------------------------------------------------
reverse communication of forces on atoms every timestep
other per-atom attributes may also be sent via pack/unpack routines
------------------------------------------------------------------------- */
void CommTiled::reverse_comm()
{
int i,irecv,n,nsend,nrecv;
AtomVec *avec = atom->avec;
double **f = atom->f;
// exchange data with another set of procs in each swap
// post recvs from all procs except self
// send data to all procs except self
// copy data to self if sendself is set
// wait on all procs except self and unpack received data
// if comm_f_only set, exchange or copy directly from f, don't pack
for (int iswap = nswap-1; iswap >= 0; iswap--) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (comm_f_only) {
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
MPI_Irecv(&buf_recv[size_reverse*reverse_recv_offset[iswap][i]],
size_reverse_recv[iswap][i],MPI_DOUBLE,
sendproc[iswap][i],0,world,&requests[i]);
}
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Send(f[firstrecv[iswap][i]],size_reverse_send[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world);
}
if (sendself[iswap]) {
avec->unpack_reverse(sendnum[iswap][nsend],sendlist[iswap][nsend],
f[firstrecv[iswap][nrecv]]);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
MPI_Waitany(nsend,requests,&irecv,MPI_STATUS_IGNORE);
avec->unpack_reverse(sendnum[iswap][irecv],sendlist[iswap][irecv],
&buf_recv[size_reverse*
reverse_recv_offset[iswap][irecv]]);
}
}
} else {
if (sendother[iswap]) {
for (i = 0; i < nsend; i++)
MPI_Irecv(&buf_recv[size_reverse*reverse_recv_offset[iswap][i]],
size_reverse_recv[iswap][i],MPI_DOUBLE,
sendproc[iswap][i],0,world,&requests[i]);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
n = avec->pack_reverse(recvnum[iswap][i],firstrecv[iswap][i],
buf_send);
MPI_Send(buf_send,n,MPI_DOUBLE,recvproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
avec->pack_reverse(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
avec->unpack_reverse(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
MPI_Waitany(nsend,requests,&irecv,MPI_STATUS_IGNORE);
avec->unpack_reverse(sendnum[iswap][irecv],sendlist[iswap][irecv],
&buf_recv[size_reverse*
reverse_recv_offset[iswap][irecv]]);
}
}
}
}
}
/* ----------------------------------------------------------------------
exchange: move atoms to correct processors
atoms exchanged with procs that touch sub-box in each of 3 dims
send out atoms that have left my box, receive ones entering my box
atoms will be lost if not inside a touching proc's box
can happen if atom moves outside of non-periodic bounary
or if atom moves more than one proc away
this routine called before every reneighboring
for triclinic, atoms must be in lamda coords (0-1) before exchange is called
------------------------------------------------------------------------- */
void CommTiled::exchange()
{
int i,m,nexch,nsend,nrecv,nlocal,proc,offset;
double lo,hi,value;
double **x;
AtomVec *avec = atom->avec;
// clear global->local map for owned and ghost atoms
// b/c atoms migrate to new procs in exchange() and
// new ghosts are created in borders()
// map_set() is done at end of borders()
// clear ghost count and any ghost bonus data internal to AtomVec
if (map_style) atom->map_clear();
atom->nghost = 0;
atom->avec->clear_bonus();
// insure send buf is large enough for single atom
// bufextra = max size of one atom = allowed overflow of sendbuf
// fixes can change per-atom size requirement on-the-fly
int bufextra_old = bufextra;
maxexchange = maxexchange_atom + maxexchange_fix;
bufextra = maxexchange + BUFEXTRA;
if (bufextra > bufextra_old)
memory->grow(buf_send,maxsend+bufextra,"comm:buf_send");
// domain properties used in exchange method and methods it calls
// subbox bounds for orthogonal or triclinic
prd = domain->prd;
boxlo = domain->boxlo;
boxhi = domain->boxhi;
if (triclinic == 0) {
sublo = domain->sublo;
subhi = domain->subhi;
} else {
sublo = domain->sublo_lamda;
subhi = domain->subhi_lamda;
}
// loop over dimensions
dimension = domain->dimension;
for (int dim = 0; dim < dimension; dim++) {
// fill buffer with atoms leaving my box, using < and >=
// when atom is deleted, fill it in with last atom
x = atom->x;
lo = sublo[dim];
hi = subhi[dim];
nlocal = atom->nlocal;
i = nsend = 0;
while (i < nlocal) {
if (x[i][dim] < lo || x[i][dim] >= hi) {
if (nsend > maxsend) grow_send(nsend,1);
proc = (this->*point_drop)(dim,x[i]);
/*
// DEBUG:
// test if proc is not in exch list, means will lose atom
// could be that *should* lose atom
int flag = 0;
for (int k = 0; k < nexchproc[dim]; k++)
if (proc == exchproc[k]) flag = 1;
if (!flag)
printf("Losing exchange atom: dim %d me %d %proc %d: %g %g %g\n",
dim,me,proc,x[i][0],x[i][1],x[i][2]);
*/
if (proc != me) {
buf_send[nsend++] = proc;
nsend += avec->pack_exchange(i,&buf_send[nsend]);
avec->copy(nlocal-1,i,1);
nlocal--;
} else i++;
} else i++;
}
atom->nlocal = nlocal;
// send and recv atoms from neighbor procs that touch my sub-box in dim
// no send/recv with self
// send size of message first
// receiver may receive multiple messages, realloc buf_recv if needed
nexch = nexchproc[dim];
if (!nexch) continue;
for (m = 0; m < nexch; m++)
MPI_Irecv(&exchnum[dim][m],1,MPI_INT,
exchproc[dim][m],0,world,&requests[m]);
for (m = 0; m < nexch; m++)
MPI_Send(&nsend,1,MPI_INT,exchproc[dim][m],0,world);
MPI_Waitall(nexch,requests,MPI_STATUS_IGNORE);
nrecv = 0;
for (m = 0; m < nexch; m++) nrecv += exchnum[dim][m];
if (nrecv > maxrecv) grow_recv(nrecv);
offset = 0;
for (m = 0; m < nexch; m++) {
MPI_Irecv(&buf_recv[offset],exchnum[dim][m],
MPI_DOUBLE,exchproc[dim][m],0,world,&requests[m]);
offset += exchnum[dim][m];
}
for (m = 0; m < nexch; m++)
MPI_Send(buf_send,nsend,MPI_DOUBLE,exchproc[dim][m],0,world);
MPI_Waitall(nexch,requests,MPI_STATUS_IGNORE);
// check incoming atoms to see if I own it and they are in my box
// if so, add to my list
// box check is only for this dimension,
// atom may be passed to another proc in later dims
m = 0;
while (m < nrecv) {
proc = static_cast<int> (buf_recv[m++]);
if (proc == me) {
value = buf_recv[m+dim+1];
if (value >= lo && value < hi) {
m += avec->unpack_exchange(&buf_recv[m]);
continue;
}
}
m += static_cast<int> (buf_recv[m]);
}
}
if (atom->firstgroupname) atom->first_reorder();
}
/* ----------------------------------------------------------------------
borders: list nearby atoms to send to neighboring procs at every timestep
one list is created per swap/proc that will be made
as list is made, actually do communication
this does equivalent of a forward_comm(), so don't need to explicitly
call forward_comm() on reneighboring timestep
this routine is called before every reneighboring
for triclinic, atoms must be in lamda coords (0-1) before borders is called
------------------------------------------------------------------------- */
void CommTiled::borders()
{
int i,m,n,nlast,nsend,nrecv,ngroup,ncount,ncountall;
double xlo,xhi,ylo,yhi,zlo,zhi;
double *bbox;
double **x;
AtomVec *avec = atom->avec;
// send/recv max one = max # of atoms in single send/recv for any swap
// send/recv max all = max # of atoms in all sends/recvs within any swap
smaxone = smaxall = 0;
rmaxone = rmaxall = 0;
// loop over swaps in all dimensions
for (int iswap = 0; iswap < nswap; iswap++) {
// find atoms within sendboxes using >= and <
// hi test with ">" is important b/c don't want to send an atom
// in lower dim (on boundary) that a proc will recv again in higher dim
// for x-dim swaps, check owned atoms
// for yz-dim swaps, check owned and ghost atoms
// store sent atom indices in sendlist for use in future timesteps
// NOTE: assume SINGLE mode, add logic for MULTI mode later
x = atom->x;
if (iswap % 2 == 0) nlast = atom->nlocal + atom->nghost;
ncountall = 0;
for (m = 0; m < nsendproc[iswap]; m++) {
bbox = sendbox[iswap][m];
xlo = bbox[0]; ylo = bbox[1]; zlo = bbox[2];
xhi = bbox[3]; yhi = bbox[4]; zhi = bbox[5];
ncount = 0;
if (!bordergroup) {
for (i = 0; i < nlast; i++) {
if (x[i][0] >= xlo && x[i][0] < xhi &&
x[i][1] >= ylo && x[i][1] < yhi &&
x[i][2] >= zlo && x[i][2] < zhi) {
if (ncount == maxsendlist[iswap][m]) grow_list(iswap,m,ncount);
sendlist[iswap][m][ncount++] = i;
}
}
} else {
ngroup = atom->nfirst;
for (i = 0; i < ngroup; i++) {
if (x[i][0] >= xlo && x[i][0] < xhi &&
x[i][1] >= ylo && x[i][1] < yhi &&
x[i][2] >= zlo && x[i][2] < zhi) {
if (ncount == maxsendlist[iswap][m]) grow_list(iswap,m,ncount);
sendlist[iswap][m][ncount++] = i;
}
}
for (i = atom->nlocal; i < nlast; i++) {
if (x[i][0] >= xlo && x[i][0] < xhi &&
x[i][1] >= ylo && x[i][1] < yhi &&
x[i][2] >= zlo && x[i][2] < zhi) {
if (ncount == maxsendlist[iswap][m]) grow_list(iswap,m,ncount);
sendlist[iswap][m][ncount++] = i;
}
}
}
sendnum[iswap][m] = ncount;
smaxone = MAX(smaxone,ncount);
ncountall += ncount;
}
smaxall = MAX(smaxall,ncountall);
// send sendnum counts to procs who recv from me except self
// copy data to self if sendself is set
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (recvother[iswap])
for (m = 0; m < nrecv; m++)
MPI_Irecv(&recvnum[iswap][m],1,MPI_INT,
recvproc[iswap][m],0,world,&requests[m]);
if (sendother[iswap])
for (m = 0; m < nsend; m++)
MPI_Send(&sendnum[iswap][m],1,MPI_INT,sendproc[iswap][m],0,world);
if (sendself[iswap]) recvnum[iswap][nrecv] = sendnum[iswap][nsend];
if (recvother[iswap]) MPI_Waitall(nrecv,requests,MPI_STATUS_IGNORE);
// setup other per swap/proc values from sendnum and recvnum
for (m = 0; m < nsendproc[iswap]; m++) {
size_reverse_recv[iswap][m] = sendnum[iswap][m]*size_reverse;
if (m == 0) reverse_recv_offset[iswap][0] = 0;
else reverse_recv_offset[iswap][m] =
reverse_recv_offset[iswap][m-1] + sendnum[iswap][m-1];
}
ncountall = 0;
for (m = 0; m < nrecvproc[iswap]; m++) {
ncount = recvnum[iswap][m];
rmaxone = MAX(rmaxone,ncount);
ncountall += ncount;
size_forward_recv[iswap][m] = ncount*size_forward;
size_reverse_send[iswap][m] = ncount*size_reverse;
if (m == 0) {
firstrecv[iswap][0] = atom->nlocal + atom->nghost;
forward_recv_offset[iswap][0] = 0;
} else {
firstrecv[iswap][m] = firstrecv[iswap][m-1] + recvnum[iswap][m-1];
forward_recv_offset[iswap][m] =
forward_recv_offset[iswap][m-1] + recvnum[iswap][m-1];
}
}
rmaxall = MAX(rmaxall,ncountall);
// insure send/recv buffers are large enough for this border comm swap
if (smaxone*size_border > maxsend) grow_send(smaxone*size_border,0);
if (rmaxall*size_border > maxrecv) grow_recv(rmaxall*size_border);
// swap atoms with other procs using pack_border(), unpack_border()
// use Waitall() instead of Waitany() because calls to unpack_border()
// must increment per-atom arrays in ascending order
if (ghost_velocity) {
if (recvother[iswap]) {
for (m = 0; m < nrecv; m++)
MPI_Irecv(&buf_recv[size_border*forward_recv_offset[iswap][m]],
recvnum[iswap][m]*size_border,
MPI_DOUBLE,recvproc[iswap][m],0,world,&requests[m]);
}
if (sendother[iswap]) {
for (m = 0; m < nsend; m++) {
n = avec->pack_border_vel(sendnum[iswap][m],sendlist[iswap][m],
buf_send,pbc_flag[iswap][m],pbc[iswap][m]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][m],0,world);
}
}
if (sendself[iswap]) {
avec->pack_border_vel(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send,pbc_flag[iswap][nsend],
pbc[iswap][nsend]);
avec->unpack_border_vel(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
}
if (recvother[iswap]) {
MPI_Waitall(nrecv,requests,MPI_STATUS_IGNORE);
for (m = 0; m < nrecv; m++)
avec->unpack_border_vel(recvnum[iswap][m],firstrecv[iswap][m],
&buf_recv[size_border*
forward_recv_offset[iswap][m]]);
}
} else {
if (recvother[iswap]) {
for (m = 0; m < nrecv; m++)
MPI_Irecv(&buf_recv[size_border*forward_recv_offset[iswap][m]],
recvnum[iswap][m]*size_border,
MPI_DOUBLE,recvproc[iswap][m],0,world,&requests[m]);
}
if (sendother[iswap]) {
for (m = 0; m < nsend; m++) {
n = avec->pack_border(sendnum[iswap][m],sendlist[iswap][m],
buf_send,pbc_flag[iswap][m],pbc[iswap][m]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][m],0,world);
}
}
if (sendself[iswap]) {
avec->pack_border(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send,pbc_flag[iswap][nsend],pbc[iswap][nsend]);
avec->unpack_border(recvnum[iswap][nsend],firstrecv[iswap][nsend],
buf_send);
}
if (recvother[iswap]) {
MPI_Waitall(nrecv,requests,MPI_STATUS_IGNORE);
for (m = 0; m < nrecv; m++)
avec->unpack_border(recvnum[iswap][m],firstrecv[iswap][m],
&buf_recv[size_border*
forward_recv_offset[iswap][m]]);
}
}
// increment ghost atoms
n = nrecvproc[iswap];
if (n)
atom->nghost += forward_recv_offset[iswap][n-1] + recvnum[iswap][n-1];
}
// insure send/recv buffers are long enough for all forward & reverse comm
// send buf is for one forward or reverse sends to one proc
// recv buf is for all forward or reverse recvs in one swap
int max = MAX(maxforward*smaxone,maxreverse*rmaxone);
if (max > maxsend) grow_send(max,0);
max = MAX(maxforward*rmaxall,maxreverse*smaxall);
if (max > maxrecv) grow_recv(max);
// reset global->local map
if (map_style) atom->map_set();
}
/* ----------------------------------------------------------------------
forward communication invoked by a Pair
nsize used only to set recv buffer limit
------------------------------------------------------------------------- */
void CommTiled::forward_comm_pair(Pair *pair)
{
int i,irecv,n,nsend,nrecv;
int nsize = pair->comm_forward;
for (int iswap = 0; iswap < nswap; iswap++) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Irecv(&buf_recv[nsize*forward_recv_offset[iswap][i]],
nsize*recvnum[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world,&requests[i]);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
n = pair->pack_forward_comm(sendnum[iswap][i],sendlist[iswap][i],
buf_send,pbc_flag[iswap][i],pbc[iswap][i]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
pair->pack_forward_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send,pbc_flag[iswap][nsend],
pbc[iswap][nsend]);
pair->unpack_forward_comm(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
MPI_Waitany(nrecv,requests,&irecv,MPI_STATUS_IGNORE);
pair->unpack_forward_comm(recvnum[iswap][irecv],firstrecv[iswap][irecv],
&buf_recv[nsize*
forward_recv_offset[iswap][irecv]]);
}
}
}
}
/* ----------------------------------------------------------------------
reverse communication invoked by a Pair
nsize used only to set recv buffer limit
------------------------------------------------------------------------- */
void CommTiled::reverse_comm_pair(Pair *pair)
{
int i,irecv,n,nsend,nrecv;
int nsize = MAX(pair->comm_reverse,pair->comm_reverse_off);
for (int iswap = nswap-1; iswap >= 0; iswap--) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (sendother[iswap]) {
for (i = 0; i < nsend; i++)
MPI_Irecv(&buf_recv[nsize*reverse_recv_offset[iswap][i]],
nsize*sendnum[iswap][i],MPI_DOUBLE,
sendproc[iswap][i],0,world,&requests[i]);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
n = pair->pack_reverse_comm(recvnum[iswap][i],firstrecv[iswap][i],
buf_send);
MPI_Send(buf_send,n,MPI_DOUBLE,recvproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
pair->pack_reverse_comm(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
pair->unpack_reverse_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
MPI_Waitany(nsend,requests,&irecv,MPI_STATUS_IGNORE);
pair->unpack_reverse_comm(sendnum[iswap][irecv],sendlist[iswap][irecv],
&buf_recv[nsize*
reverse_recv_offset[iswap][irecv]]);
}
}
}
}
/* ----------------------------------------------------------------------
forward communication invoked by a Fix
size/nsize used only to set recv buffer limit
size = 0 (default) -> use comm_forward from Fix
size > 0 -> Fix passes max size per atom
the latter is only useful if Fix does several comm modes,
some are smaller than max stored in its comm_forward
------------------------------------------------------------------------- */
void CommTiled::forward_comm_fix(Fix *fix, int size)
{
int i,irecv,n,nsize,nsend,nrecv;
if (size) nsize = size;
else nsize = fix->comm_forward;
for (int iswap = 0; iswap < nswap; iswap++) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Irecv(&buf_recv[nsize*forward_recv_offset[iswap][i]],
nsize*recvnum[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world,&requests[i]);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
n = fix->pack_forward_comm(sendnum[iswap][i],sendlist[iswap][i],
buf_send,pbc_flag[iswap][i],pbc[iswap][i]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
fix->pack_forward_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send,pbc_flag[iswap][nsend],
pbc[iswap][nsend]);
fix->unpack_forward_comm(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
MPI_Waitany(nrecv,requests,&irecv,MPI_STATUS_IGNORE);
fix->unpack_forward_comm(recvnum[iswap][irecv],firstrecv[iswap][irecv],
&buf_recv[nsize*
forward_recv_offset[iswap][irecv]]);
}
}
}
}
/* ----------------------------------------------------------------------
reverse communication invoked by a Fix
size/nsize used only to set recv buffer limit
size = 0 (default) -> use comm_forward from Fix
size > 0 -> Fix passes max size per atom
the latter is only useful if Fix does several comm modes,
some are smaller than max stored in its comm_forward
------------------------------------------------------------------------- */
void CommTiled::reverse_comm_fix(Fix *fix, int size)
{
int i,irecv,n,nsize,nsend,nrecv;
if (size) nsize = size;
else nsize = fix->comm_reverse;
for (int iswap = nswap-1; iswap >= 0; iswap--) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (sendother[iswap]) {
for (i = 0; i < nsend; i++)
MPI_Irecv(&buf_recv[nsize*reverse_recv_offset[iswap][i]],
nsize*sendnum[iswap][i],MPI_DOUBLE,
sendproc[iswap][i],0,world,&requests[i]);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
n = fix->pack_reverse_comm(recvnum[iswap][i],firstrecv[iswap][i],
buf_send);
MPI_Send(buf_send,n,MPI_DOUBLE,recvproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
fix->pack_reverse_comm(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
fix->unpack_reverse_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
MPI_Waitany(nsend,requests,&irecv,MPI_STATUS_IGNORE);
fix->unpack_reverse_comm(sendnum[iswap][irecv],sendlist[iswap][irecv],
&buf_recv[nsize*
reverse_recv_offset[iswap][irecv]]);
}
}
}
}
/* ----------------------------------------------------------------------
reverse communication invoked by a Fix with variable size data
query fix for all pack sizes to insure buf_send is big enough
handshake sizes before irregular comm to insure buf_recv is big enough
NOTE: how to setup one big buf recv with correct offsets ??
------------------------------------------------------------------------- */
void CommTiled::reverse_comm_fix_variable(Fix *fix)
{
error->all(FLERR,"Reverse comm fix variable not yet supported by CommTiled");
}
/* ----------------------------------------------------------------------
forward communication invoked by a Compute
nsize used only to set recv buffer limit
------------------------------------------------------------------------- */
void CommTiled::forward_comm_compute(Compute *compute)
{
int i,irecv,n,nsend,nrecv;
int nsize = compute->comm_forward;
for (int iswap = 0; iswap < nswap; iswap++) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Irecv(&buf_recv[nsize*forward_recv_offset[iswap][i]],
nsize*recvnum[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world,&requests[i]);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
n = compute->pack_forward_comm(sendnum[iswap][i],sendlist[iswap][i],
buf_send,pbc_flag[iswap][i],
pbc[iswap][i]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
compute->pack_forward_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send,pbc_flag[iswap][nsend],
pbc[iswap][nsend]);
compute->unpack_forward_comm(recvnum[iswap][nrecv],
firstrecv[iswap][nrecv],buf_send);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
MPI_Waitany(nrecv,requests,&irecv,MPI_STATUS_IGNORE);
compute->
unpack_forward_comm(recvnum[iswap][irecv],firstrecv[iswap][irecv],
&buf_recv[nsize*
forward_recv_offset[iswap][irecv]]);
}
}
}
}
/* ----------------------------------------------------------------------
reverse communication invoked by a Compute
nsize used only to set recv buffer limit
------------------------------------------------------------------------- */
void CommTiled::reverse_comm_compute(Compute *compute)
{
int i,irecv,n,nsend,nrecv;
int nsize = compute->comm_reverse;
for (int iswap = nswap-1; iswap >= 0; iswap--) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (sendother[iswap]) {
for (i = 0; i < nsend; i++)
MPI_Irecv(&buf_recv[nsize*reverse_recv_offset[iswap][i]],
nsize*sendnum[iswap][i],MPI_DOUBLE,
sendproc[iswap][i],0,world,&requests[i]);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
n = compute->pack_reverse_comm(recvnum[iswap][i],firstrecv[iswap][i],
buf_send);
MPI_Send(buf_send,n,MPI_DOUBLE,recvproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
compute->pack_reverse_comm(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
compute->unpack_reverse_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
MPI_Waitany(nsend,requests,&irecv,MPI_STATUS_IGNORE);
compute->
unpack_reverse_comm(sendnum[iswap][irecv],sendlist[iswap][irecv],
&buf_recv[nsize*
reverse_recv_offset[iswap][irecv]]);
}
}
}
}
/* ----------------------------------------------------------------------
forward communication invoked by a Dump
nsize used only to set recv buffer limit
------------------------------------------------------------------------- */
void CommTiled::forward_comm_dump(Dump *dump)
{
int i,irecv,n,nsend,nrecv;
int nsize = dump->comm_forward;
for (int iswap = 0; iswap < nswap; iswap++) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Irecv(&buf_recv[nsize*forward_recv_offset[iswap][i]],
nsize*recvnum[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world,&requests[i]);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
n = dump->pack_forward_comm(sendnum[iswap][i],sendlist[iswap][i],
buf_send,pbc_flag[iswap][i],
pbc[iswap][i]);
MPI_Send(buf_send,n,MPI_DOUBLE,sendproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
dump->pack_forward_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send,pbc_flag[iswap][nsend],
pbc[iswap][nsend]);
dump->unpack_forward_comm(recvnum[iswap][nrecv],
firstrecv[iswap][nrecv],buf_send);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
MPI_Waitany(nrecv,requests,&irecv,MPI_STATUS_IGNORE);
dump->unpack_forward_comm(recvnum[iswap][irecv],firstrecv[iswap][irecv],
&buf_recv[nsize*
forward_recv_offset[iswap][irecv]]);
}
}
}
}
/* ----------------------------------------------------------------------
reverse communication invoked by a Dump
nsize used only to set recv buffer limit
------------------------------------------------------------------------- */
void CommTiled::reverse_comm_dump(Dump *dump)
{
int i,irecv,n,nsend,nrecv;
int nsize = dump->comm_reverse;
for (int iswap = nswap-1; iswap >= 0; iswap--) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
if (sendother[iswap]) {
for (i = 0; i < nsend; i++)
MPI_Irecv(&buf_recv[nsize*reverse_recv_offset[iswap][i]],
nsize*sendnum[iswap][i],MPI_DOUBLE,
sendproc[iswap][i],0,world,&requests[i]);
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
n = dump->pack_reverse_comm(recvnum[iswap][i],firstrecv[iswap][i],
buf_send);
MPI_Send(buf_send,n,MPI_DOUBLE,recvproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
dump->pack_reverse_comm(recvnum[iswap][nrecv],firstrecv[iswap][nrecv],
buf_send);
dump->unpack_reverse_comm(sendnum[iswap][nsend],sendlist[iswap][nsend],
buf_send);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
MPI_Waitany(nsend,requests,&irecv,MPI_STATUS_IGNORE);
dump->unpack_reverse_comm(sendnum[iswap][irecv],sendlist[iswap][irecv],
&buf_recv[nsize*
reverse_recv_offset[iswap][irecv]]);
}
}
}
}
/* ----------------------------------------------------------------------
forward communication of Nsize values in per-atom array
------------------------------------------------------------------------- */
void CommTiled::forward_comm_array(int nsize, double **array)
{
int i,j,k,m,iatom,last,irecv,nsend,nrecv;
// insure send/recv bufs are big enough for nsize
// based on smaxone/rmaxall from most recent borders() invocation
if (nsize > maxforward) {
maxforward = nsize;
if (maxforward*smaxone > maxsend) grow_send(maxforward*smaxone,0);
if (maxforward*rmaxall > maxrecv) grow_recv(maxforward*rmaxall);
}
for (int iswap = 0; iswap < nswap; iswap++) {
nsend = nsendproc[iswap] - sendself[iswap];
nrecv = nrecvproc[iswap] - sendself[iswap];
MPI_Barrier(world);
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++)
MPI_Irecv(&buf_recv[nsize*forward_recv_offset[iswap][i]],
nsize*recvnum[iswap][i],
MPI_DOUBLE,recvproc[iswap][i],0,world,&requests[i]);
}
if (sendother[iswap]) {
for (i = 0; i < nsend; i++) {
m = 0;
for (iatom = 0; iatom < sendnum[iswap][i]; iatom++) {
j = sendlist[iswap][i][iatom];
for (k = 0; k < nsize; k++)
buf_send[m++] = array[j][k];
}
MPI_Send(buf_send,nsize*sendnum[iswap][i],
MPI_DOUBLE,sendproc[iswap][i],0,world);
}
}
if (sendself[iswap]) {
m = 0;
for (iatom = 0; iatom < sendnum[iswap][nsend]; iatom++) {
j = sendlist[iswap][nsend][iatom];
for (k = 0; k < nsize; k++)
buf_send[m++] = array[j][k];
}
m = 0;
last = firstrecv[iswap][nrecv] + recvnum[iswap][nrecv];
for (iatom = firstrecv[iswap][nrecv]; iatom < last; iatom++)
for (k = 0; k < nsize; k++)
array[iatom][k] = buf_send[m++];
}
if (recvother[iswap]) {
for (i = 0; i < nrecv; i++) {
MPI_Waitany(nrecv,requests,&irecv,MPI_STATUS_IGNORE);
m = nsize*forward_recv_offset[iswap][irecv];
last = firstrecv[iswap][irecv] + recvnum[iswap][irecv];
for (iatom = firstrecv[iswap][irecv]; iatom < last; iatom++)
for (k = 0; k < nsize; k++)
array[iatom][k] = buf_recv[m++];
}
}
}
}
/* ----------------------------------------------------------------------
exchange info provided with all 6 stencil neighbors
NOTE: this method is currently not used
------------------------------------------------------------------------- */
int CommTiled::exchange_variable(int n, double *inbuf, double *&outbuf)
{
int nrecv = n;
return nrecv;
}
/* ----------------------------------------------------------------------
determine overlap list of Noverlap procs the lo/hi box overlaps
overlap = non-zero area in common between box and proc sub-domain
box is owned by me and extends in dim
------------------------------------------------------------------------- */
void CommTiled::box_drop_brick(int idim, double *lo, double *hi, int &indexme)
{
// NOTE: this is not triclinic compatible
// NOTE: these error messages are internal sanity checks
// should not occur, can be removed at some point
int index,dir;
if (hi[idim] == sublo[idim]) {
index = myloc[idim] - 1;
dir = -1;
} else if (lo[idim] == subhi[idim]) {
index = myloc[idim] + 1;
dir = 1;
} else if (hi[idim] == boxhi[idim]) {
index = procgrid[idim] - 1;
dir = -1;
} else if (lo[idim] == boxlo[idim]) {
index = 0;
dir = 1;
} else error->one(FLERR,"Comm tiled mis-match in box drop brick");
int other1,other2,proc;
double lower,upper;
double *split;
if (idim == 0) {
other1 = myloc[1]; other2 = myloc[2];
split = xsplit;
} else if (idim == 1) {
other1 = myloc[0]; other2 = myloc[2];
split = ysplit;
} else {
other1 = myloc[0]; other2 = myloc[1];
split = zsplit;
}
if (index < 0 || index > procgrid[idim])
error->one(FLERR,"Comm tiled invalid index in box drop brick");
while (1) {
lower = boxlo[idim] + prd[idim]*split[index];
if (index < procgrid[idim]-1)
upper = boxlo[idim] + prd[idim]*split[index+1];
else upper = boxhi[idim];
if (lower >= hi[idim] || upper <= lo[idim]) break;
if (idim == 0) proc = grid2proc[index][other1][other2];
else if (idim == 1) proc = grid2proc[other1][index][other2];
else proc = grid2proc[other1][other2][index];
if (noverlap == maxoverlap) {
maxoverlap += DELTA_PROCS;
memory->grow(overlap,maxoverlap,"comm:overlap");
}
if (proc == me) indexme = noverlap;
overlap[noverlap++] = proc;
index += dir;
if (index < 0 || index >= procgrid[idim]) break;
}
}
/* ----------------------------------------------------------------------
determine overlap list of Noverlap procs the lo/hi box overlaps
overlap = non-zero area in common between box and proc sub-domain
recursive method for traversing an RCB tree of cuts
no need to split lo/hi box as recurse b/c OK if box extends outside RCB box
------------------------------------------------------------------------- */
void CommTiled::box_drop_tiled(int idim, double *lo, double *hi, int &indexme)
{
box_drop_tiled_recurse(lo,hi,0,nprocs-1,indexme);
}
void CommTiled::box_drop_tiled_recurse(double *lo, double *hi,
int proclower, int procupper,
int &indexme)
{
// end recursion when partition is a single proc
// add proc to overlap list
if (proclower == procupper) {
if (noverlap == maxoverlap) {
maxoverlap += DELTA_PROCS;
memory->grow(overlap,maxoverlap,"comm:overlap");
}
if (proclower == me) indexme = noverlap;
overlap[noverlap++] = proclower;
return;
}
// drop box on each side of cut it extends beyond
// use > and < criteria so does not include a box it only touches
// procmid = 1st processor in upper half of partition
// = location in tree that stores this cut
// dim = 0,1,2 dimension of cut
// cut = position of cut
int procmid = proclower + (procupper - proclower) / 2 + 1;
int idim = rcbinfo[procmid].dim;
double cut = boxlo[idim] + prd[idim]*rcbinfo[procmid].cutfrac;
if (lo[idim] < cut)
box_drop_tiled_recurse(lo,hi,proclower,procmid-1,indexme);
if (hi[idim] > cut)
box_drop_tiled_recurse(lo,hi,procmid,procupper,indexme);
}
/* ----------------------------------------------------------------------
return other box owned by proc as lo/hi corner pts
------------------------------------------------------------------------- */
void CommTiled::box_other_brick(int idim, int idir,
int proc, double *lo, double *hi)
{
lo[0] = sublo[0]; lo[1] = sublo[1]; lo[2] = sublo[2];
hi[0] = subhi[0]; hi[1] = subhi[1]; hi[2] = subhi[2];
int other1,other2,oproc;
double *split;
if (idim == 0) {
other1 = myloc[1]; other2 = myloc[2];
split = xsplit;
} else if (idim == 1) {
other1 = myloc[0]; other2 = myloc[2];
split = ysplit;
} else {
other1 = myloc[0]; other2 = myloc[1];
split = zsplit;
}
int dir = -1;
if (idir) dir = 1;
int index = myloc[idim];
int n = procgrid[idim];
for (int i = 0; i < n; i++) {
index += dir;
if (index < 0) index = n-1;
else if (index >= n) index = 0;
if (idim == 0) oproc = grid2proc[index][other1][other2];
else if (idim == 1) oproc = grid2proc[other1][index][other2];
else oproc = grid2proc[other1][other2][index];
if (proc == oproc) {
lo[idim] = boxlo[idim] + prd[idim]*split[index];
if (split[index+1] < 1.0)
hi[idim] = boxlo[idim] + prd[idim]*split[index+1];
else hi[idim] = boxhi[idim];
return;
}
}
}
/* ----------------------------------------------------------------------
return other box owned by proc as lo/hi corner pts
------------------------------------------------------------------------- */
void CommTiled::box_other_tiled(int idim, int idir,
int proc, double *lo, double *hi)
{
double (*split)[2] = rcbinfo[proc].mysplit;
lo[0] = boxlo[0] + prd[0]*split[0][0];
if (split[0][1] < 1.0) hi[0] = boxlo[0] + prd[0]*split[0][1];
else hi[0] = boxhi[0];
lo[1] = boxlo[1] + prd[1]*split[1][0];
if (split[1][1] < 1.0) hi[1] = boxlo[1] + prd[1]*split[1][1];
else hi[1] = boxhi[1];
lo[2] = boxlo[2] + prd[2]*split[2][0];
if (split[2][1] < 1.0) hi[2] = boxlo[2] + prd[2]*split[2][1];
else hi[2] = boxhi[2];
}
/* ----------------------------------------------------------------------
return 1 if proc's box touches me, else 0
procneigh stores 6 procs that touch me
------------------------------------------------------------------------- */
int CommTiled::box_touch_brick(int proc, int idim, int idir)
{
if (procneigh[idim][idir] == proc) return 1;
return 0;
}
/* ----------------------------------------------------------------------
return 1 if proc's box touches me, else 0
------------------------------------------------------------------------- */
int CommTiled::box_touch_tiled(int proc, int idim, int idir)
{
// sending to left
// only touches if proc hi = my lo, or if proc hi = boxhi and my lo = boxlo
if (idir == 0) {
if (rcbinfo[proc].mysplit[idim][1] == rcbinfo[me].mysplit[idim][0])
return 1;
else if (rcbinfo[proc].mysplit[idim][1] == 1.0 &&
rcbinfo[me].mysplit[idim][0] == 0.0)
return 1;
// sending to right
// only touches if proc lo = my hi, or if proc lo = boxlo and my hi = boxhi
} else {
if (rcbinfo[proc].mysplit[idim][0] == rcbinfo[me].mysplit[idim][1])
return 1;
else if (rcbinfo[proc].mysplit[idim][0] == 0.0 &&
rcbinfo[me].mysplit[idim][1] == 1.0)
return 1;
}
return 0;
}
/* ----------------------------------------------------------------------
------------------------------------------------------------------------- */
int CommTiled::point_drop_brick(int idim, double *x)
{
if (closer_subbox_edge(idim,x)) return procneigh[idim][1];
return procneigh[idim][0];
}
/* ----------------------------------------------------------------------
determine overlap list of Noverlap procs the lo/hi box overlaps
overlap = non-zero area in common between box and proc sub-domain
recursive method for traversing an RCB tree of cuts
no need to split lo/hi box as recurse b/c OK if box extends outside RCB box
------------------------------------------------------------------------- */
int CommTiled::point_drop_tiled(int idim, double *x)
{
double xnew[3];
xnew[0] = x[0]; xnew[1] = x[1]; xnew[2] = x[2];
if (idim == 0) {
if (xnew[1] < sublo[1] || xnew[1] > subhi[1]) {
if (closer_subbox_edge(1,x)) xnew[1] = subhi[1];
else xnew[1] = sublo[1];
}
}
if (idim <= 1) {
if (xnew[2] < sublo[2] || xnew[2] > subhi[2]) {
if (closer_subbox_edge(2,x)) xnew[2] = subhi[2];
else xnew[2] = sublo[2];
}
}
int proc = point_drop_tiled_recurse(xnew,0,nprocs-1);
if (proc == me) return me;
if (idim == 0) {
int done = 1;
if (rcbinfo[proc].mysplit[1][0] == rcbinfo[me].mysplit[1][1]) {
xnew[1] -= EPSILON * (subhi[1]-sublo[1]);
done = 0;
}
if (rcbinfo[proc].mysplit[2][0] == rcbinfo[me].mysplit[2][1]) {
xnew[2] -= EPSILON * (subhi[2]-sublo[2]);
done = 0;
}
if (!done) {
proc = point_drop_tiled_recurse(xnew,0,nprocs-1);
done = 1;
if (rcbinfo[proc].mysplit[1][0] == rcbinfo[me].mysplit[1][1]) {
xnew[1] -= EPSILON * (subhi[1]-sublo[1]);
done = 0;
}
if (rcbinfo[proc].mysplit[2][0] == rcbinfo[me].mysplit[2][1]) {
xnew[2] -= EPSILON * (subhi[2]-sublo[2]);
done = 0;
}
if (!done) proc = point_drop_tiled_recurse(xnew,0,nprocs-1);
}
} else if (idim == 1) {
if (rcbinfo[proc].mysplit[2][0] == rcbinfo[me].mysplit[2][1]) {
xnew[2] -= EPSILON * (subhi[2]-sublo[2]);
proc = point_drop_tiled_recurse(xnew,0,nprocs-1);
}
}
return proc;
}
/* ----------------------------------------------------------------------
recursive point drop thru RCB tree
------------------------------------------------------------------------- */
int CommTiled::point_drop_tiled_recurse(double *x,
int proclower, int procupper)
{
// end recursion when partition is a single proc
// return proc
if (proclower == procupper) return proclower;
// drop point on side of cut it is on
// use < criterion so point is not on high edge of proc sub-domain
// procmid = 1st processor in upper half of partition
// = location in tree that stores this cut
// dim = 0,1,2 dimension of cut
// cut = position of cut
int procmid = proclower + (procupper - proclower) / 2 + 1;
int idim = rcbinfo[procmid].dim;
double cut = boxlo[idim] + prd[idim]*rcbinfo[procmid].cutfrac;
if (x[idim] < cut) return point_drop_tiled_recurse(x,proclower,procmid-1);
else return point_drop_tiled_recurse(x,procmid,procupper);
}
/* ----------------------------------------------------------------------
assume x[idim] is outside subbox bounds in same dim
------------------------------------------------------------------------- */
int CommTiled::closer_subbox_edge(int idim, double *x)
{
double deltalo,deltahi;
if (sublo[idim] == boxlo[idim])
deltalo = fabs(x[idim]-prd[idim] - sublo[idim]);
else deltalo = fabs(x[idim] - sublo[idim]);
if (subhi[idim] == boxhi[idim])
deltahi = fabs(x[idim]+prd[idim] - subhi[idim]);
else deltahi = fabs(x[idim] - subhi[idim]);
if (deltalo < deltahi) return 0;
return 1;
}
/* ----------------------------------------------------------------------
if RCB decomp exists and just changed, gather needed global RCB info
------------------------------------------------------------------------- */
void CommTiled::coord2proc_setup()
{
if (!rcbnew) return;
if (!rcbinfo)
rcbinfo = (RCBinfo *)
memory->smalloc(nprocs*sizeof(RCBinfo),"comm:rcbinfo");
rcbnew = 0;
RCBinfo rcbone;
memcpy(&rcbone.mysplit[0][0],&mysplit[0][0],6*sizeof(double));
rcbone.cutfrac = rcbcutfrac;
rcbone.dim = rcbcutdim;
MPI_Allgather(&rcbone,sizeof(RCBinfo),MPI_CHAR,
rcbinfo,sizeof(RCBinfo),MPI_CHAR,world);
}
/* ----------------------------------------------------------------------
determine which proc owns atom with coord x[3] based on current decomp
x will be in box (orthogonal) or lamda coords (triclinic)
if layout = UNIFORM or NONUNIFORM, invoke parent method
if layout = TILED, use point_drop_recurse()
return owning proc ID, ignore igx,igy,igz
------------------------------------------------------------------------- */
int CommTiled::coord2proc(double *x, int &igx, int &igy, int &igz)
{
if (layout != LAYOUT_TILED) return Comm::coord2proc(x,igx,igy,igz);
return point_drop_tiled_recurse(x,0,nprocs-1);
}
/* ----------------------------------------------------------------------
realloc the size of the send buffer as needed with BUFFACTOR and bufextra
if flag = 1, realloc
if flag = 0, don't need to realloc with copy, just free/malloc
------------------------------------------------------------------------- */
void CommTiled::grow_send(int n, int flag)
{
maxsend = static_cast<int> (BUFFACTOR * n);
if (flag)
memory->grow(buf_send,maxsend+bufextra,"comm:buf_send");
else {
memory->destroy(buf_send);
memory->create(buf_send,maxsend+bufextra,"comm:buf_send");
}
}
/* ----------------------------------------------------------------------
free/malloc the size of the recv buffer as needed with BUFFACTOR
------------------------------------------------------------------------- */
void CommTiled::grow_recv(int n)
{
maxrecv = static_cast<int> (BUFFACTOR * n);
memory->destroy(buf_recv);
memory->create(buf_recv,maxrecv,"comm:buf_recv");
}
/* ----------------------------------------------------------------------
realloc the size of the iswap sendlist as needed with BUFFACTOR
------------------------------------------------------------------------- */
void CommTiled::grow_list(int iswap, int iwhich, int n)
{
maxsendlist[iswap][iwhich] = static_cast<int> (BUFFACTOR * n);
memory->grow(sendlist[iswap][iwhich],maxsendlist[iswap][iwhich],
"comm:sendlist[i][j]");
}
/* ----------------------------------------------------------------------
allocation of swap info
------------------------------------------------------------------------- */
void CommTiled::allocate_swap(int n)
{
nsendproc = new int[n];
nrecvproc = new int[n];
sendother = new int[n];
recvother = new int[n];
sendself = new int[n];
nprocmax = new int[n];
sendproc = new int*[n];
recvproc = new int*[n];
sendnum = new int*[n];
recvnum = new int*[n];
size_forward_recv = new int*[n];
firstrecv = new int*[n];
size_reverse_send = new int*[n];
size_reverse_recv = new int*[n];
forward_recv_offset = new int*[n];
reverse_recv_offset = new int*[n];
pbc_flag = new int*[n];
pbc = new int**[n];
sendbox = new double**[n];
maxsendlist = new int*[n];
sendlist = new int**[n];
for (int i = 0; i < n; i++) {
sendproc[i] = recvproc[i] = NULL;
sendnum[i] = recvnum[i] = NULL;
size_forward_recv[i] = firstrecv[i] = NULL;
size_reverse_send[i] = size_reverse_recv[i] = NULL;
forward_recv_offset[i] = reverse_recv_offset[i] = NULL;
pbc_flag[i] = NULL;
pbc[i] = NULL;
sendbox[i] = NULL;
maxsendlist[i] = NULL;
sendlist[i] = NULL;
}
maxreqstat = 0;
requests = NULL;
for (int i = 0; i < n; i++) {
nprocmax[i] = DELTA_PROCS;
grow_swap_send(i,DELTA_PROCS,0);
grow_swap_recv(i,DELTA_PROCS);
}
nexchproc = new int[n/2];
nexchprocmax = new int[n/2];
exchproc = new int*[n/2];
exchnum = new int*[n/2];
for (int i = 0; i < n/2; i++) {
nexchprocmax[i] = DELTA_PROCS;
exchproc[i] = new int[DELTA_PROCS];
exchnum[i] = new int[DELTA_PROCS];
}
}
/* ----------------------------------------------------------------------
grow info for swap I, to allow for N procs to communicate with
ditto for complementary recv for swap I+1 or I-1, as invoked by caller
------------------------------------------------------------------------- */
void CommTiled::grow_swap_send(int i, int n, int nold)
{
delete [] sendproc[i];
sendproc[i] = new int[n];
delete [] sendnum[i];
sendnum[i] = new int[n];
delete [] size_reverse_recv[i];
size_reverse_recv[i] = new int[n];
delete [] reverse_recv_offset[i];
reverse_recv_offset[i] = new int[n];
delete [] pbc_flag[i];
pbc_flag[i] = new int[n];
memory->destroy(pbc[i]);
memory->create(pbc[i],n,6,"comm:pbc_flag");
memory->destroy(sendbox[i]);
memory->create(sendbox[i],n,6,"comm:sendbox");
delete [] maxsendlist[i];
maxsendlist[i] = new int[n];
for (int j = 0; j < nold; j++) memory->destroy(sendlist[i][j]);
delete [] sendlist[i];
sendlist[i] = new int*[n];
for (int j = 0; j < n; j++) {
maxsendlist[i][j] = BUFMIN;
memory->create(sendlist[i][j],BUFMIN,"comm:sendlist[i][j]");
}
}
void CommTiled::grow_swap_recv(int i, int n)
{
delete [] recvproc[i];
recvproc[i] = new int[n];
delete [] recvnum[i];
recvnum[i] = new int[n];
delete [] size_forward_recv[i];
size_forward_recv[i] = new int[n];
delete [] firstrecv[i];
firstrecv[i] = new int[n];
delete [] forward_recv_offset[i];
forward_recv_offset[i] = new int[n];
delete [] size_reverse_send[i];
size_reverse_send[i] = new int[n];
}
/* ----------------------------------------------------------------------
deallocate swap info
------------------------------------------------------------------------- */
void CommTiled::deallocate_swap(int n)
{
delete [] nsendproc;
delete [] nrecvproc;
delete [] sendother;
delete [] recvother;
delete [] sendself;
for (int i = 0; i < n; i++) {
delete [] sendproc[i];
delete [] recvproc[i];
delete [] sendnum[i];
delete [] recvnum[i];
delete [] size_forward_recv[i];
delete [] firstrecv[i];
delete [] size_reverse_send[i];
delete [] size_reverse_recv[i];
delete [] forward_recv_offset[i];
delete [] reverse_recv_offset[i];
delete [] pbc_flag[i];
memory->destroy(pbc[i]);
memory->destroy(sendbox[i]);
delete [] maxsendlist[i];
for (int j = 0; j < nprocmax[i]; j++) memory->destroy(sendlist[i][j]);
delete [] sendlist[i];
}
delete [] sendproc;
delete [] recvproc;
delete [] sendnum;
delete [] recvnum;
delete [] size_forward_recv;
delete [] firstrecv;
delete [] size_reverse_send;
delete [] size_reverse_recv;
delete [] forward_recv_offset;
delete [] reverse_recv_offset;
delete [] pbc_flag;
delete [] pbc;
delete [] sendbox;
delete [] maxsendlist;
delete [] sendlist;
delete [] requests;
delete [] nprocmax;
delete [] nexchproc;
delete [] nexchprocmax;
for (int i = 0; i < n/2; i++) {
delete [] exchproc[i];
delete [] exchnum[i];
}
delete [] exchproc;
delete [] exchnum;
}
/* ----------------------------------------------------------------------
return # of bytes of allocated memory
------------------------------------------------------------------------- */
bigint CommTiled::memory_usage()
{
bigint bytes = 0;
return bytes;
}
c4science · Help