void ParFUM_createComm(int meshid, int dim, MPI_Comm comm)
{
int rank, comm_size;
ParFUM_desharing(meshid);
ParFUM_deghosting(meshid);
//MPI_Barrier(MPI_COMM_WORLD);
MPI_Comm_rank(comm, &rank);
MPI_Comm_size(comm, &comm_size);
if (rank==0) CkPrintf("Recreating shared nodes...\n");
ParFUM_recreateSharedNodes(meshid, dim, comm);
//MPI_Barrier(MPI_COMM_WORLD);
if (rank==0) CkPrintf("Generating global node numbers...\n");
ParFUM_generateGlobalNodeNumbers(meshid, comm);
FEM_Mesh *mesh = (FEM_chunk::get("ParFUM_recreateSharedNodes"))->lookup(meshid,"ParFUM_recreateSharedNodes");
//MPI_Barrier(MPI_COMM_WORLD);
if (rank==0) CkPrintf("Gathering ghost data...\n");
struct ghostdata *gdata;
if(rank == 0){
gdata = gatherGhosts();
}else{
gdata = new ghostdata;
}
MPI_Bcast_pup(*gdata,0,comm);
if (rank==0) CkPrintf("Making ghosts...\n");
makeGhosts(mesh,comm,0,gdata->numLayers,gdata->layers);
//MPI_Barrier(MPI_COMM_WORLD);
delete gdata;
}
int FEM_master_parallel_part(int fem_mesh,int masterRank,FEM_Comm_t comm_context){
const char *caller="FEM_Create_connmsa";
FEMAPI(caller);
FEMchunk *c=FEMchunk::get(caller);
FEM_Mesh *m=c->lookup(fem_mesh,caller);
m->setAbsoluteGlobalno();
int nelem = m->nElems();
int numChunks;
MPI_Comm_size((MPI_Comm)comm_context,&numChunks);
printf("master -> number of elements %d \n",nelem);
DEBUG(m->print(0));
/*load the connectivity information into the eptr and
eind datastructure. It will be read by the other slave
elements and used to call parmetis*/
MSA1DINT eptr(nelem,numChunks);
MSA1DINT eind(nelem*10,numChunks);
/*
after the msa array has been created and loaded with connectivity data
tell the slaves about the msa array
*/
struct conndata data;
data.nelem = nelem;
data.nnode = m->node.size();
data.arr1 = eptr;
data.arr2 = eind;
MPI_Bcast_pup(data,masterRank,(MPI_Comm)comm_context);
eptr.enroll(numChunks);
eind.enroll(numChunks);
int indcount=0,ptrcount=0;
for(int t=0;t<m->elem.size();t++){
if(m->elem.has(t)){
FEM_Elem &k=m->elem[t];
for(int e=0;e<k.size();e++){
eptr.set(ptrcount)=indcount;
ptrcount++;
for(int n=0;n<k.getNodesPer();n++){
eind.set(indcount)=k.getConn(e,n);
indcount++;
}
}
}
}
eptr.set(ptrcount) = indcount;
printf("master -> ptrcount %d indcount %d \n",ptrcount,indcount);
/*
break up the mesh such that each chunk gets the same number of elements
and the nodes corresponding to those elements. However this is not the partition.
This is just distributing the data, so that when partition is done using parmetis
all the requests for data do not go to chunk 0. Instead after partition each chunk
can send the element and node data to the chunks that will need it
*/
FEM_Mesh *mesh_array=FEM_break_mesh(m,ptrcount,numChunks);
/*
Send the broken up meshes to the different chunks.
*/
sendBrokenMeshes(mesh_array,comm_context);
delete [] mesh_array;
FEM_Mesh mypiece;
MPI_Recv_pup(mypiece,masterRank,MESH_CHUNK_TAG,(MPI_Comm)comm_context);
/*
call parmetis
*/
struct partconndata *partdata = FEM_call_parmetis(data,comm_context);
printf("done with parmetis \n");
/*
Set up a msa to store the partitions to which a node belongs.
A node can belong to multiple partitions.
*/
int totalNodes = m->node.size();
MSA1DINTLIST nodepart(totalNodes,numChunks);
MPI_Bcast_pup(nodepart,masterRank,(MPI_Comm)comm_context);
nodepart.enroll(numChunks);
FEM_write_nodepart(nodepart,partdata);
/*
Set up a msa to store the nodes that belong to a partition
*/
MSA1DNODELIST part2node(numChunks,numChunks);
MPI_Bcast_pup(part2node,masterRank,(MPI_Comm)comm_context);
part2node.enroll(numChunks);
FEM_write_part2node(nodepart,part2node,partdata,(MPI_Comm)comm_context);
/*
Set up a msa to store the elements that belong to a partition
*/
MSA1DINTLIST part2elem(numChunks,numChunks);
MPI_Bcast_pup(part2elem,masterRank,(MPI_Comm)comm_context);
part2elem.enroll(numChunks);
FEM_write_part2elem(part2elem,partdata,(MPI_Comm)comm_context);
/*
Get the list of elements and nodes that belong to this partition
*/
NodeList lnodes = part2node.get(masterRank);
IntList lelems = part2elem.get(masterRank);
/*
Build an MSA of FEM_Mesh, with each index containing the mesh for that chunk
*/
MSA1DFEMMESH part2mesh(numChunks,numChunks);
MPI_Bcast_pup(part2mesh,masterRank,(MPI_Comm)comm_context);
part2mesh.enroll(numChunks);
FEM_write_part2mesh(part2mesh,partdata, &data,nodepart,numChunks,masterRank,&mypiece);
/*
Get your mesh consisting of elements and nodes out of the mesh MSA
*/
MeshElem me = part2mesh.get(masterRank);
printf("[%d] Number of elements in my partitioned mesh %d number of nodes %d \n",masterRank,me.m->nElems(),me.m->node.size());
addIDXLists(me.m,lnodes,masterRank);
/*
Broadcast the user data to all the meshes
*/
DEBUG(printf("[%d] Length of udata vector in master %d \n",masterRank,m->udata.size()));
MPI_Bcast_pup(m->udata,masterRank,(MPI_Comm)comm_context);
me.m->udata = m->udata;
delete partdata;
/*
collect the ghost data and send it to all the chunks.
*/
struct ghostdata *gdata = gatherGhosts();
printf("[%d] number of ghost layers %d \n",masterRank,gdata->numLayers);
MPI_Bcast_pup(*gdata,masterRank,(MPI_Comm)comm_context);
/*
make ghosts for this mesh
*/
makeGhosts(me.m,(MPI_Comm)comm_context,masterRank,gdata->numLayers,gdata->layers);
delete gdata;
me.m->becomeGetting();
FEMchunk *chunk = FEMchunk::get("FEM_Mesh_Parallel_broadcast");
int tempMeshNo = chunk->meshes.put(me.m);
int new_mesh = FEM_Mesh_copy(tempMeshNo);
FEM_Mesh *nmesh = c->lookup(new_mesh,"master_parallel_broadcast");
DEBUG(printf("[%d] Length of udata vector in master new_mesh %d \n",masterRank,nmesh->udata.size()));
return new_mesh;
};
int FEM_slave_parallel_part(int fem_mesh,int masterRank,FEM_Comm_t comm_context){
int myRank;
MPI_Comm_rank((MPI_Comm)comm_context,&myRank);
int numChunks;
MPI_Comm_size((MPI_Comm)comm_context,&numChunks);
/*Receive the name of the msa arrays that contain the
connectivity information*/
struct conndata data;
MPI_Bcast_pup(data,masterRank,(MPI_Comm)comm_context);
data.arr1.enroll(numChunks);
data.arr2.enroll(numChunks);
DEBUG(printf("Recv -> %d \n",data.nelem));
/*
Receive the broken up mesh from the masterRank.
These will be used later to give each partitioned mesh
its elements and data.
*/
FEM_Mesh mypiece;
MPI_Recv_pup(mypiece,masterRank,MESH_CHUNK_TAG,(MPI_Comm)comm_context);
/*
call parmetis and get the resuts back from it
*/
struct partconndata *partdata = FEM_call_parmetis(data,comm_context);
/*
write to the msa that contains the partitions to which a node belongs
*/
MSA1DINTLIST nodepart;
MPI_Bcast_pup(nodepart,masterRank,(MPI_Comm)comm_context);
nodepart.enroll(numChunks);
FEM_write_nodepart(nodepart,partdata);
/*
write to the msa that stores the nodes that belong to each partition
*/
MSA1DNODELIST part2node;
MPI_Bcast_pup(part2node,masterRank,(MPI_Comm)comm_context);
part2node.enroll(numChunks);
FEM_write_part2node(nodepart,part2node,partdata,(MPI_Comm)comm_context);
/*
write the msa that stores the elements that belong to each partition
*/
MSA1DINTLIST part2elem;
MPI_Bcast_pup(part2elem,masterRank,(MPI_Comm)comm_context);
part2elem.enroll(numChunks);
FEM_write_part2elem(part2elem,partdata,(MPI_Comm)comm_context);
/*
Get the list of elements and nodes that belong to this partition
*/
NodeList lnodes = part2node.get(myRank);
IntList lelems = part2elem.get(myRank);
/*
Get the FEM msa and write the different mesh
*/
MSA1DFEMMESH part2mesh;
MPI_Bcast_pup(part2mesh,masterRank,(MPI_Comm)comm_context);
part2mesh.enroll(numChunks);
FEM_write_part2mesh(part2mesh,partdata,&data,nodepart,numChunks,myRank,&mypiece);
/*
Get your mesh consisting of elements and nodes out of the mesh MSA
*/
MeshElem me = part2mesh.get(myRank);
printf("[%d] Number of elements in my partitioned mesh %d number of nodes %d \n",myRank,me.m->nElems(),me.m->node.size());
addIDXLists(me.m,lnodes,myRank);
/*
Receive the user data from master
*/
MPI_Bcast_pup(me.m->udata,masterRank,(MPI_Comm)comm_context);
DEBUG(printf("[%d] Length of udata vector %d \n",myRank,me.m->udata.size()));
delete partdata;
struct ghostdata *gdata = new ghostdata;
MPI_Bcast_pup(*gdata,masterRank,(MPI_Comm)comm_context);
printf("[%d] number of ghost layers %d \n",myRank,gdata->numLayers);
/*
make ghosts
*/
makeGhosts(me.m,(MPI_Comm )comm_context,masterRank,gdata->numLayers,gdata->layers);
me.m->becomeGetting();
FEMchunk *chunk = FEMchunk::get("FEM_Mesh_Parallel_broadcast");
int tempMeshNo = chunk->meshes.put(me.m);
int new_mesh = FEM_Mesh_copy(tempMeshNo);
delete gdata;
return new_mesh;
}
int FEM_master_parallel_part(int fem_mesh,int masterRank,FEM_Comm_t comm_context){
const char *caller="FEM_Create_connmsa";
FEMAPI(caller);
FEM_chunk *c=FEM_chunk::get(caller);
FEM_Mesh *m=c->lookup(fem_mesh,caller);
m->setAbsoluteGlobalno();
int nelem = m->nElems();
int numChunks;
MPI_Comm_size((MPI_Comm)comm_context,&numChunks);
printf("master -> number of elements %d \n",nelem);
DEBUG(m->print(0));
/*load the connectivity information into the eptr and
eind datastructure. It will be read by the other slave
elements and used to call parmetis*/
MSA1DINT eptrMSA(nelem,numChunks);
MSA1DINT eindMSA(nelem*10,numChunks);
/*
after the msa array has been created and loaded with connectivity data
tell the slaves about the msa array
*/
struct conndata data;
data.nelem = nelem;
data.nnode = m->node.size();
data.arr1 = eptrMSA;
data.arr2 = eindMSA;
MPI_Bcast_pup(data,masterRank,(MPI_Comm)comm_context);
eptrMSA.enroll(numChunks);
eindMSA.enroll(numChunks);
MSA1DINT::Write wPtr = eptrMSA.getInitialWrite();
MSA1DINT::Write wInd = eindMSA.getInitialWrite();
int indcount=0,ptrcount=0;
for(int t=0;t<m->elem.size();t++){
if(m->elem.has(t)){
FEM_Elem &k=m->elem[t];
for(int e=0;e<k.size();e++){
wPtr.set(ptrcount)=indcount;
ptrcount++;
for(int n=0;n<k.getNodesPer();n++){
wInd.set(indcount)=k.getConn(e,n);
indcount++;
}
}
}
}
wPtr.set(ptrcount) = indcount;
printf("master -> ptrcount %d indcount %d sizeof(MSA1DINT) %d sizeof(MSA1DINTLIST) %d memory %d\n",ptrcount,indcount,sizeof(MSA1DINT),sizeof(MSA1DINTLIST),CmiMemoryUsage());
/*
break up the mesh such that each chunk gets the same number of elements
and the nodes corresponding to those elements. However this is not the partition.
This is just distributing the data, so that when partition is done using parmetis
all the requests for data do not go to chunk 0. Instead after partition each chunk
can send the element and node data to the chunks that will need it
*/
FEM_Mesh *mesh_array=FEM_break_mesh(m,ptrcount,numChunks);
/*
Send the broken up meshes to the different chunks.
*/
sendBrokenMeshes(mesh_array,comm_context);
delete [] mesh_array;
FEM_Mesh mypiece;
MPI_Recv_pup(mypiece,masterRank,MESH_CHUNK_TAG,(MPI_Comm)comm_context);
/*
call parmetis
*/
double parStartTime = CkWallTimer();
MSA1DINT::Read rPtr = wPtr.syncToRead();
MSA1DINT::Read rInd = wInd.syncToRead();
printf("starting FEM_call_parmetis \n");
struct partconndata *partdata = FEM_call_parmetis(data.nelem, rPtr, rInd, comm_context);
printf("done with parmetis %d FEM_Mesh %d in %.6lf \n",CmiMemoryUsage(),sizeof(FEM_Mesh),CkWallTimer()-parStartTime);
double dataArrangeStartTime = CkWallTimer();
/*
Set up a msa to store the partitions to which a node belongs.
A node can belong to multiple partitions.
*/
int totalNodes = m->node.size();
MSA1DINTLIST nodepart(totalNodes,numChunks);
MPI_Bcast_pup(nodepart,masterRank,(MPI_Comm)comm_context);
nodepart.enroll(numChunks);
MSA1DINTLIST::Accum nodepartAcc = nodepart.getInitialAccum();
FEM_write_nodepart(nodepartAcc,partdata,(MPI_Comm)comm_context);
printf("Creating mapping of node to partition took %.6lf\n",CkWallTimer()-dataArrangeStartTime);
dataArrangeStartTime = CkWallTimer();
MSA1DINTLIST::Read nodepartRead = nodepartAcc.syncToRead();
/*
Set up a msa to store the nodes that belong to a partition
*/
MSA1DNODELIST part2node(numChunks,numChunks);
MPI_Bcast_pup(part2node,masterRank,(MPI_Comm)comm_context);
part2node.enroll(numChunks);
MSA1DNODELIST::Accum part2nodeAcc = part2node.getInitialAccum();
FEM_write_part2node(nodepartRead, part2nodeAcc, partdata, (MPI_Comm)comm_context);
/*
Get the list of elements and nodes that belong to this partition
*/
MSA1DNODELIST::Read rPart2node = part2nodeAcc.syncToRead();
NodeList lnodes = rPart2node.get(masterRank);
lnodes.uniquify();
// IntList lelems = part2elem.get(masterRank);
printf("Creating mapping of partition to node took %.6lf\n",CkWallTimer()-dataArrangeStartTime);
printf("Time spent doing +=ElemList %.6lf \n",elemlistaccTime);
dataArrangeStartTime = CkWallTimer();
/*
Build an MSA of FEM_Mesh, with each index containing the mesh for that chunk
*/
MSA1DFEMMESH part2mesh(numChunks,numChunks);
MPI_Bcast_pup(part2mesh,masterRank,(MPI_Comm)comm_context);
part2mesh.enroll(numChunks);
MSA1DFEMMESH::Accum aPart2mesh = part2mesh.getInitialAccum();
FEM_write_part2mesh(aPart2mesh,partdata, &data,nodepartRead,numChunks,masterRank,&mypiece);
/*
Get your mesh consisting of elements and nodes out of the mesh MSA
*/
MSA1DFEMMESH::Read rPart2mesh = aPart2mesh.syncToRead();
MeshElem me = rPart2mesh.get(masterRank);
//printf("[%d] Number of elements in my partitioned mesh %d number of nodes %d \n",masterRank,me.m->nElems(),me.m->node.size());
DEBUG(printf("[%d] Memory usage on vp 0 close to max %d \n",CkMyPe(),CmiMemoryUsage()));
//Free up the eptr and eind MSA arrays stored in data
delete &rPtr;
delete &rInd;
data.arr1.FreeMem();
data.arr2.FreeMem();
nodepart.FreeMem();
DEBUG(printf("[%d] Memory usage on vp 0 after FreeMem %d \n",CkMyPe(),CmiMemoryUsage()));
addIDXLists(me.m,lnodes,masterRank);
part2node.FreeMem();
DEBUG(printf("[%d] Memory usage on vp 0 after addIDXL %d \n",CkMyPe(),CmiMemoryUsage()));
/*
Broadcast the user data to all the meshes
*/
DEBUG(printf("[%d] Length of udata vector in master %d \n",masterRank,m->udata.size()));
MPI_Bcast_pup(m->udata,masterRank,(MPI_Comm)comm_context);
me.m->udata = m->udata;
delete partdata;
printf("[%d] Data Arrangement took %.6lf \n",masterRank,CkWallTimer()-dataArrangeStartTime);
/*
collect the ghost data and send it to all the chunks.
*/
struct ghostdata *gdata = gatherGhosts();
DEBUG(printf("[%d] number of ghost layers %d \n",masterRank,gdata->numLayers));
MPI_Bcast_pup(*gdata,masterRank,(MPI_Comm)comm_context);
/*
make ghosts for this mesh
*/
printf("[%d] Starting to generate number of ghost layers %d \n",masterRank,gdata->numLayers);
double _startTime = CkWallTimer();
makeGhosts(me.m,(MPI_Comm)comm_context,masterRank,gdata->numLayers,gdata->layers);
delete gdata;
printf("[%d] Ghost generation took %.6lf \n",masterRank,CkWallTimer()-_startTime);
me.m->becomeGetting();
FEM_chunk *chunk = FEM_chunk::get("FEM_Mesh_Parallel_broadcast");
int tempMeshNo = chunk->meshes.put(me.m);
int new_mesh = FEM_Mesh_copy(tempMeshNo);
FEM_Mesh *nmesh = c->lookup(new_mesh,"master_parallel_broadcast");
DEBUG(printf("[%d] Length of udata vector in master new_mesh %d \n",masterRank,nmesh->udata.size()));
part2mesh.FreeMem();
printf("[%d] Max Memory usage on vp 0 at end of parallel partition %d \n",CkMyPe(),CmiMaxMemoryUsage());
return new_mesh;
}
