lammps/examples/COUPLE/library/irregular.cpp

394 lines
10 KiB
C++

#include "stdio.h"
#include "stdlib.h"
#include "string.h"
#include "irregular.h"
#include "memory.h"
#include "error.h"
#define MAX(A,B) ((A) > (B)) ? (A) : (B)
enum{UNSET,SET};
enum{NONE,SAME,VARYING};
/* ---------------------------------------------------------------------- */
Irregular::Irregular(MPI_Comm caller)
{
comm = caller;
MPI_Comm_rank(comm,&me);
MPI_Comm_size(comm,&nprocs);
memory = new Memory(comm);
error = new Error(comm);
init();
patternflag = UNSET;
sizestyle = NONE;
}
/* ---------------------------------------------------------------------- */
Irregular::~Irregular()
{
delete memory;
delete error;
deallocate();
}
/* ----------------------------------------------------------------------
n = # of datums contributed by this proc
proclist = which proc each datum is to be sent to
------------------------------------------------------------------------- */
void Irregular::pattern(int n, int *proclist)
{
// free any previous irregular info
deallocate();
init();
patternflag = SET;
sizestyle = NONE;
ndatumsend = n;
// list = 1 for procs I send to, including self
// nrecv = # of messages I receive, not including self
// self = 0 if no data for self, 1 if there is
int *list = new int[nprocs];
int *counts = new int[nprocs];
for (int i = 0; i < nprocs; i++) {
list[i] = 0;
counts[i] = 1;
}
for (int i = 0; i < n; i++) list[proclist[i]] = 1;
MPI_Reduce_scatter(list,&nrecv,counts,MPI_INT,MPI_SUM,comm);
self = 0;
if (list[me]) self = 1;
if (self) nrecv--;
// storage for recv info, not including self
recvproc = new int[nrecv];
recvcount = new int[nrecv];
recvsize = new int[nrecv];
request = new MPI_Request[nrecv];
status = new MPI_Status[nrecv];
// list = # of datums to send to each proc, including self
// nsend = # of messages I send, not including self
for (int i = 0; i < nprocs; i++) list[i] = 0;
for (int i = 0; i < n; i++) list[proclist[i]]++;
nsend = 0;
for (int i = 0; i < nprocs; i++) if (list[i] > 0) nsend++;
if (self) nsend--;
// storage for send info, including self
sendproc = new int[nsend+self];
sendcount = new int[nsend+self];
sendsize = new int[nsend+self];
sendindices = (int *) memory->smalloc(n*sizeof(int),"sendindices");
// setup sendprocs and sendcounts, including self
// each proc begins with iproc > me, and continues until iproc = me
// list ends up with pointer to which send that proc is associated with
int iproc = me;
int isend = 0;
for (int i = 0; i < nprocs; i++) {
iproc++;
if (iproc == nprocs) iproc = 0;
if (list[iproc] > 0) {
sendproc[isend] = iproc;
sendcount[isend] = list[iproc];
list[iproc] = isend;
isend++;
}
}
// post all receives for datum counts
for (int i = 0; i < nrecv; i++)
MPI_Irecv(&recvcount[i],1,MPI_INT,MPI_ANY_SOURCE,0,comm,&request[i]);
// barrier to insure receives are posted
MPI_Barrier(comm);
// send each datum count, packing buf with needed datums
for (int i = 0; i < nsend; i++)
MPI_Send(&sendcount[i],1,MPI_INT,sendproc[i],0,comm);
// insure all MPI_ANY_SOURCE messages are received
// set recvproc
if (nrecv) MPI_Waitall(nrecv,request,status);
for (int i = 0; i < nrecv; i++) recvproc[i] = status[i].MPI_SOURCE;
// ndatumrecv = total datums received, including self
ndatumrecv = 0;
for (int i = 0; i < nrecv; i++)
ndatumrecv += recvcount[i];
if (self) ndatumrecv += sendcount[nsend];
// setup sendindices, including self
// counts = offset into sendindices for each proc I send to
// let sc0 = sendcount[0], sc1 = sendcount[1], etc
// sendindices[0:sc0-1] = indices of datums in 1st message
// sendindices[sc0:sc0+sc1-1] = indices of datums in 2nd message, etc
counts[0] = 0;
for (int i = 1; i < nsend+self; i++)
counts[i] = counts[i-1] + sendcount[i-1];
for (int i = 0; i < n; i++) {
isend = list[proclist[i]];
sendindices[counts[isend]++] = i;
}
// clean up
delete [] counts;
delete [] list;
}
/* ----------------------------------------------------------------------
n = size of each received datum
return total size in bytes of received data on this proc
------------------------------------------------------------------------- */
int Irregular::size(int n)
{
if (patternflag == UNSET) error->all("Cannot size without pattern");
sizestyle = SAME;
nsize = n;
nsendmax = 0;
for (int i = 0; i < nsend+self; i++) {
sendsize[i] = nsize * sendcount[i];
if (i < nsend) nsendmax = MAX(nsendmax,sendsize[i]);
}
for (int i = 0; i < nrecv; i++) recvsize[i] = nsize * recvcount[i];
nbytesrecv = nsize * ndatumrecv;
return nbytesrecv;
}
/* ----------------------------------------------------------------------
slength,rlength = size of each datum to send and recv
soffset = offset into eventual buffer of send data for each datum
soffset can be NULL, in which case will build sendoffset from slength
return total size in bytes of received data on this proc
------------------------------------------------------------------------- */
int Irregular::size(int *slength, int *soffset, int *rlength)
{
if (patternflag == UNSET) error->all("Cannot size without pattern");
sizestyle = VARYING;
// store local copy of pointers to send lengths/offsets
// if soffset not provided, create local copy from slength
sendsizedatum = slength;
if (soffset == NULL) {
sendoffsetflag = 1;
sendoffset = (int *) memory->smalloc(ndatumsend*sizeof(int),"sendoffset");
if (ndatumsend) sendoffset[0] = 0;
for (int i = 1; i < ndatumsend; i++)
sendoffset[i] = sendoffset[i-1] + sendsizedatum[i-1];
} else {
if (sendoffsetflag) memory->sfree(sendoffset);
sendoffsetflag = 0;
sendoffset = soffset;
}
nsendmax = 0;
int m = 0;
for (int i = 0; i < nsend+self; i++) {
sendsize[i] = 0;
for (int j = 0; j < sendcount[i]; j++)
sendsize[i] += sendsizedatum[sendindices[m++]];
if (i < nsend) nsendmax = MAX(nsendmax,sendsize[i]);
}
nbytesrecv = 0;
m = 0;
for (int i = 0; i < nrecv; i++) {
recvsize[i] = 0;
for (int j = 0; j < recvcount[i]; j++) recvsize[i] += rlength[m++];
nbytesrecv += recvsize[i];
}
if (self) nbytesrecv += sendsize[nsend];
return nbytesrecv;
}
/* ----------------------------------------------------------------------
wrapper on 2 versions of exchange
------------------------------------------------------------------------- */
void Irregular::exchange(char *sendbuf, char *recvbuf)
{
if (sizestyle == SAME) exchange_same(sendbuf,recvbuf);
else if (sizestyle == VARYING) exchange_varying(sendbuf,recvbuf);
else error->all("Irregular size was not set");
}
/* ----------------------------------------------------------------------
sendbuf = data to send
recvbuf = buffer to recv all data into
requires nsize,nsendmax,recvsize,sendsize be setup by size(int)
------------------------------------------------------------------------- */
void Irregular::exchange_same(char *sendbuf, char *recvbuf)
{
// post all receives
int recvoffset = 0;
for (int irecv = 0; irecv < nrecv; irecv++) {
MPI_Irecv(&recvbuf[recvoffset],recvsize[irecv],MPI_BYTE,
recvproc[irecv],0,comm,&request[irecv]);
recvoffset += recvsize[irecv];
}
// malloc buf for largest send
char *buf = (char *) memory->smalloc(nsendmax,"buf");
// barrier to insure receives are posted
MPI_Barrier(comm);
// send each message, packing buf with needed datums
int m = 0;
for (int isend = 0; isend < nsend; isend++) {
int bufoffset = 0;
for (int i = 0; i < sendcount[isend]; i++) {
memcpy(&buf[bufoffset],&sendbuf[nsize*sendindices[m++]],nsize);
bufoffset += nsize;
}
MPI_Send(buf,sendsize[isend],MPI_BYTE,sendproc[isend],0,comm);
}
// copy self data directly from sendbuf to recvbuf
if (self)
for (int i = 0; i < sendcount[nsend]; i++) {
memcpy(&recvbuf[recvoffset],&sendbuf[nsize*sendindices[m++]],nsize);
recvoffset += nsize;
}
// free send buffer
memory->sfree(buf);
// wait on all incoming messages
if (nrecv) MPI_Waitall(nrecv,request,status);
}
/* ----------------------------------------------------------------------
sendbuf = data to send
recvbuf = buffer to recv all data into
requires nsendmax,recvsize,sendsize,sendoffset,sendsizedatum
be setup by size(int *, int *)
------------------------------------------------------------------------- */
void Irregular::exchange_varying(char *sendbuf, char *recvbuf)
{
// post all receives
int recvoffset = 0;
for (int irecv = 0; irecv < nrecv; irecv++) {
MPI_Irecv(&recvbuf[recvoffset],recvsize[irecv],MPI_BYTE,
recvproc[irecv],0,comm,&request[irecv]);
recvoffset += recvsize[irecv];
}
// malloc buf for largest send
char *buf = (char *) memory->smalloc(nsendmax,"buf");
// barrier to insure receives are posted
MPI_Barrier(comm);
// send each message, packing buf with needed datums
int index;
int m = 0;
for (int isend = 0; isend < nsend; isend++) {
int bufoffset = 0;
for (int i = 0; i < sendcount[isend]; i++) {
index = sendindices[m++];
memcpy(&buf[bufoffset],&sendbuf[sendoffset[index]],sendsizedatum[index]);
bufoffset += sendsizedatum[index];
}
MPI_Send(buf,sendsize[isend],MPI_BYTE,sendproc[isend],0,comm);
}
// copy self data directly from sendbuf to recvbuf
if (self)
for (int i = 0; i < sendcount[nsend]; i++) {
index = sendindices[m++];
memcpy(&recvbuf[recvoffset],&sendbuf[sendoffset[index]],
sendsizedatum[index]);
recvoffset += sendsizedatum[index];
}
// free send buffer
memory->sfree(buf);
// wait on all incoming messages
if (nrecv) MPI_Waitall(nrecv,request,status);
}
/* ---------------------------------------------------------------------- */
void Irregular::init()
{
sendoffsetflag = 0;
sendproc = sendcount = sendsize = sendindices = NULL;
sendoffset = NULL;
recvproc = recvcount = recvsize = NULL;
request = NULL;
status = NULL;
}
/* ---------------------------------------------------------------------- */
void Irregular::deallocate()
{
delete [] sendproc;
delete [] sendcount;
delete [] sendsize;
memory->sfree(sendindices);
if (sendoffsetflag) memory->sfree(sendoffset);
delete [] recvproc;
delete [] recvcount;
delete [] recvsize;
delete [] request;
delete [] status;
}