public final void sumOverThreadsAndMPI() {
for (int ThreadNo = 0; ThreadNo < NumberofThreads; ThreadNo++) {
TotalNumberofPoints += NumberofPoints[ThreadNo];
}
// Note - parallel for
try {
forallChunked(
0,
SALSAUtility.ThreadCount - 1,
(threadIndex) -> {
int beginindex = ParallelArrayRanges[threadIndex].getStartIndex();
int indexlength = ParallelArrayRanges[threadIndex].getLength();
for (int ArrayLoop = beginindex; ArrayLoop < beginindex + indexlength; ArrayLoop++) {
TotalVectorSum[ArrayLoop] = 0.0;
for (int ThreadNo = 0; ThreadNo < NumberofThreads; ThreadNo++) {
TotalVectorSum[ArrayLoop] += VectorSum[ThreadNo][ArrayLoop];
}
}
});
} catch (SuspendableException e) {
SALSAUtility.printAndThrowRuntimeException(e.getMessage());
}
if (SALSAUtility.MPI_Size > 1) {
SALSAUtility.StartSubTimer(SALSAUtility.MPIREDUCETiming1);
// Note - MPI Call - Allreduce - double - sum
TotalNumberofPoints = SALSAUtility.mpiOps.allReduce(TotalNumberofPoints, MPI.SUM);
// Note - MPI Call - Allreduce - double[] - sum
SALSAUtility.mpiOps.allReduce(TotalVectorSum, MPI.SUM);
SALSAUtility.StopSubTimer(SALSAUtility.MPIREDUCETiming1);
}
}
public final void sumOverThreadsAndMPI() {
SALSAUtility.StartSubTimer(SALSAUtility.ThreadTiming);
// Note - parallel for
try {
forallChunked(
0,
SALSAUtility.ThreadCount - 1,
(threadIndex) -> {
int beginindex = ParallelArrayRanges[threadIndex].getStartIndex();
int indexlength = ParallelArrayRanges[threadIndex].getLength();
for (int ArrayLoop = beginindex; ArrayLoop < beginindex + indexlength; ArrayLoop++) {
double tmp = 0.0;
for (int ThreadNo = 0; ThreadNo < NumberofThreads; ThreadNo++) {
tmp += VectorSum[ThreadNo][ArrayLoop];
}
TotalVectorSum[ArrayLoop] = tmp;
}
});
} catch (SuspendableException e) {
SALSAUtility.printAndThrowRuntimeException(e.getMessage());
}
SALSAUtility.StopSubTimer(SALSAUtility.ThreadTiming);
if (SALSAUtility.MPI_Size > 1) {
SALSAUtility.StartSubTimer(SALSAUtility.MPIREDUCETiming1);
// Note - MPI Call - Allreduce - double - sum
TotalNumberofPoints = SALSAUtility.mpiOps.allReduce(TotalNumberofPoints, MPI.SUM);
int bigsize = TotalVectorSum.length;
if (bigsize <= 4096) {
// Note - MPI Call - Allreduce - double[] - sum
SALSAUtility.mpiOps.allReduce(TotalVectorSum, MPI.SUM);
} else {
double[] buffer = new double[4096];
int start = 0;
while (start < bigsize) {
int whatsLeft = Math.min(bigsize - start, 4096);
System.arraycopy(TotalVectorSum, start, buffer, 0, whatsLeft);
// Note - MPI Call - Allreduce - double[] - sum
SALSAUtility.mpiOps.allReduce(buffer, MPI.SUM);
System.arraycopy(buffer, 0, TotalVectorSum, start, whatsLeft);
start += whatsLeft;
}
}
SALSAUtility.StopSubTimer(SALSAUtility.MPIREDUCETiming1);
}
}
