fdurastante
/
amg4psblas
mirror of https://github.com/sfilippone/amg4psblas.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Removed one critical region from PARALLEL_PROCESS_EXPOSED_VERTEX_B

Browse Source
omp-walther
StefanoPetrilli 3 years ago
parent 5bc4f2a080
commit e2ca97ca47
1 changed files with 56 additions and 30 deletions
Show Stats Download Patch File Download Diff File

78
amgprec/impl/aggregator/algoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP.cpp
Unescape Escape View File

@ -211,13 +211,17 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
vector <MilanLongInt> GMate; //Proportional to the number of ghost vertices
MilanLongInt S;
MilanLongInt privateMyCard = 0;
staticQueue U, privateU;
staticQueue U, privateU, privateQLocalVtx, privateQGhostVtx, privateQMsgType, privateQOwner;
/*
staticQueue privateReqQLocalVtx, privateReqQGhostVtx, privateReqQMsgType, privateReqQOwner;
*/
bool isEmpty;
#ifdef TIME_TRACKER
double Ghost2LocalInitialization = MPI_Wtime();
#endif
#pragma omp parallel private(insertMe, k, u, w, v, k1, adj1, adj2, adj11, adj12, heaviestEdgeWt, ghostOwner, privateU, privateMyCard, isEmpty) firstprivate(StartIndex, EndIndex) default(shared) num_threads(4)
#pragma omp parallel private(insertMe, k, u, w, v, k1, adj1, adj2, adj11, adj12, heaviestEdgeWt, ghostOwner, privateU, privateMyCard, isEmpty, privateQLocalVtx, privateQGhostVtx, privateQMsgType, privateQOwner /*, privateReqQLocalVtx, privateReqQGhostVtx, privateReqQMsgType, privateReqQOwner*/) firstprivate(StartIndex, EndIndex) default(shared) num_threads(4)
{
// TODO comments about the reduction
@ -291,21 +295,18 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
double verGhostPtrInitialization = MPI_Wtime();
#endif
}
/*
* OMP verGhostPtrInitialization
*
* schedule(static) assign to each thread an huge chunk
* it is used in this case to reduce the overhead of chunk assignment
* and to reduce false sharing
* Not parallelizable
*/
#pragma omp for nowait schedule(static)
for (i = 0; i < numGhostVertices; i++) { //O(|Ghost Vertices|)
verGhostPtr[i + 1] = verGhostPtr[i] + Counter[i];
#ifdef PRINT_DEBUG_INFO_
cout<<verGhostPtr[i]<<"\t"; fflush(stdout);
#endif
}
} // End of single region
#ifdef TIME_TRACKER
verGhostPtrInitialization = MPI_Wtime() - verGhostPtrInitialization;
@ -461,6 +462,30 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
* in parallel.
*/
MilanLongInt size = numGhostEdges; //TODO how can I decide a meaningfull size?
//Fail messages
privateQLocalVtx.~staticQueue();
privateQGhostVtx.~staticQueue();
privateQMsgType.~staticQueue();
privateQOwner.~staticQueue();
//Request messages
/*
privateReqQLocalVtx.~staticQueue();
privateReqQGhostVtx.~staticQueue();
privateReqQMsgType.~staticQueue();
privateReqQOwner.~staticQueue();
*/
new(&privateQLocalVtx) staticQueue(size);
new(&privateQGhostVtx) staticQueue(size);
new(&privateQMsgType) staticQueue(size);
new(&privateQOwner) staticQueue(size);
/*
new(&privateReqQLocalVtx) staticQueue(size);
new(&privateReqQGhostVtx) staticQueue(size);
new(&privateReqQMsgType) staticQueue(size);
new(&privateReqQOwner) staticQueue(size);
*/
#pragma omp for reduction(+: msgInd, NumMessagesBundled, myCard, PCounter[:numProcs]) schedule(static)
for ( v=0; v < NLVer; v++ )
{
@ -499,18 +524,12 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
myCard++;
if ((w < StartIndex) || (w > EndIndex)) { //w is a ghost vertex
//Build the Message Packet:
//Message[0] = v+StartIndex; //LOCAL
//Message[1] = w; //GHOST
//Message[2] = REQUEST; //TYPE
//Send a Request (Asynchronous)
#ifdef PRINT_DEBUG_INFO_
cout<<"\n("<<myRank<<")Sending a request message (291):";
cout<<"\n("<<myRank<<")Local is: "<<v+StartIndex<<" Ghost is "<<w<<" Owner is: "<< findOwnerOfGhost(w, verDistance, myRank, numProcs) <<endl;
fflush(stdout);
#endif
/* MPI_Bsend(&Message[0], 3, MPI_INT, inputSubGraph.findOwner(w),
ComputeTag, comm);*/
msgInd++;
NumMessagesBundled++;
ghostOwner = findOwnerOfGhost(w, verDistance, myRank, numProcs);
@ -522,9 +541,6 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
QLocalVtx.push_back(v + StartIndex);
QGhostVtx.push_back(w);
QMsgType.push_back(REQUEST);
//ghostOwner = inputSubGraph.findOwner(w);
assert(ghostOwner != -1);
assert(ghostOwner != myRank);
QOwner.push_back(ghostOwner);
if (candidateMate[NLVer + Ghost2LocalMap[w]] == v + StartIndex) {
@ -601,13 +617,10 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
assert(ghostOwner != -1);
assert(ghostOwner != myRank);
PCounter[ghostOwner]++;
#pragma omp critical
{
QLocalVtx.push_back(v + StartIndex);
QGhostVtx.push_back(w);
QMsgType.push_back(FAILURE);
QOwner.push_back(ghostOwner);
}
privateQLocalVtx.push_back(v + StartIndex);
privateQGhostVtx.push_back(w);
privateQMsgType.push_back(FAILURE);
privateQOwner.push_back(ghostOwner);
} //End of if(GHOST)
} //End of for loop
@ -615,6 +628,17 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
//End: PARALLEL_PROCESS_EXPOSED_VERTEX_B(v)
} //End of for ( v=0; v < NLVer; v++ )
#pragma omp critical(privateMsg)
{
while (!privateQLocalVtx.empty())
{
QLocalVtx.push_back(privateQLocalVtx.pop_front());
QGhostVtx.push_back(privateQGhostVtx.pop_front());
QMsgType.push_back(privateQMsgType.pop_front());
QOwner.push_back(privateQOwner.pop_front());
}
}
tempCounter.clear(); //Do not need this any more
@ -626,7 +650,9 @@ void dalgoDistEdgeApproxDomEdgesLinearSearchMesgBndlSmallMateCMP(
/////////////////////////// PROCESS MATCHED VERTICES //////////////////////////////
///////////////////////////////////////////////////////////////////////////////////
privateU.~staticQueue();
new(&privateU) staticQueue(1000); //TODO how can I put a meaningfull size?
isEmpty = false;
#ifdef COUNT_LOCAL_VERTEX

Write Preview
Loading…
Cancel
Save