no.uib.cipr.matrix.distributed.CollectiveCommunications Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of mtj Show documentation
Show all versions of mtj Show documentation
A comprehensive collection of matrix data structures, linear solvers, least squares methods,
eigenvalue, and singular value decompositions.
/*
* Copyright (C) 2003-2006 Bjørn-Ove Heimsund
*
* This file is part of MTJ.
*
* This library is free software; you can redistribute it and/or modify it
* under the terms of the GNU Lesser General Public License as published by the
* Free Software Foundation; either version 2.1 of the License, or (at your
* option) any later version.
*
* This library is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License
* for more details.
*
* You should have received a copy of the GNU Lesser General Public License
* along with this library; if not, write to the Free Software Foundation,
* Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
package no.uib.cipr.matrix.distributed;
import java.lang.reflect.Array;
import java.util.ArrayList;
import java.util.List;
import java.util.concurrent.BrokenBarrierException;
import java.util.concurrent.CyclicBarrier;
import java.util.concurrent.Exchanger;
import no.uib.cipr.matrix.distributed.Communicator.SendRecv;
/**
* Collective communications between all the participating threads.
*
* @deprecated the no.uib.cipr.matrix.distributed package has been deprecated because
* of a number of hard to fix concurrency bugs. It is distributed only for backwards compatibility,
* but is not recommended. The utility of this package is questionable, as it does not allow
* distribution of computation between JVMs or across a network. For many people, distributed
* computing of multiple matrices can be achieved at a user-level through the
* JPPF Framework.
* Users who need to deal with few very large matrices may wish to implement their own storage classes
* and solvers using JPPF, but this will not be supported directly in matrix-toolkits-java.
*/
@Deprecated
public class CollectiveCommunications {
/**
* Size of the whole collective
*/
final int size;
/**
* Exchanges pipes between the threads. Only used for setup of the
* point-to-point communications
*/
private final List>> ex;
/**
* Sets up a collective of the given size
*
* @param size
* Number of members of the collective
*/
public CollectiveCommunications(int size) {
if (size < 1)
throw new IllegalArgumentException("size < 1");
this.size = size;
barrier = new CyclicBarrier(size);
broadcast = new Broadcast();
gather = new Gather();
scatter = new Scatter();
allGather = new AllGather();
allToAll = new AllToAll();
reduce = new Reduce();
ex = new ArrayList>>();
for (int i = 0; i < size; ++i) {
List> iex = new ArrayList>();
// Get from symmetry
for (int j = 0; j < i; ++j)
iex.add(ex.get(j).get(i));
// Main diagonal (ignored)
iex.add(null);
// New entries
for (int j = i + 1; j < size; ++j)
iex.add(new Exchanger());
ex.add(iex);
}
}
/**
* Gets the size of the collective
*/
public int size() {
return size;
}
/**
* Creates a communicator for point-to-point data-exchange. This method
* should be called by all the threads in the collective
*
* @param rank
* Rank of the communicator
*/
public Communicator createCommunicator(int rank) {
if (rank < 0 || rank >= size)
throw new IllegalArgumentException("rank < 0 || rank >= size");
return new Communicator(rank, ex.get(rank), this);
}
static void await(CyclicBarrier barrier) {
try {
barrier.await();
} catch (InterruptedException e) {
throw new RuntimeException(e);
} catch (BrokenBarrierException e) {
throw new RuntimeException(e);
}
}
/**
* Blocks the caller until all threads have called
*/
void barrier() {
await(barrier);
}
private final CyclicBarrier barrier;
void broadcast(Object buffer, int root, int rank) {
broadcast.buffer[rank] = buffer;
if (rank == root)
broadcast.root = root;
await(broadcast.barrier);
}
private final Broadcast broadcast;
private class Broadcast implements Runnable {
final CyclicBarrier barrier = new CyclicBarrier(size, this);
int root;
final Object[] buffer = new Object[size];
public void run() {
int length = Array.getLength(buffer[root]);
for (int i = 0; i < size; ++i)
System.arraycopy(buffer[root], 0, buffer[i], 0, length);
}
}
void gather(Object sendbuf, Object[] recvbuf, int root, int rank) {
gather.setSend(sendbuf, rank);
if (rank == root)
gather.recvbuf = recvbuf;
await(gather.barrier);
}
private final Gather gather;
private class Gather implements Runnable {
CyclicBarrier barrier = new CyclicBarrier(size, this);
Object[] recvbuf = new Object[size], sendbuf = new Object[size];
private final int[] length = new int[size];
public void setSend(Object sendbuf, int rank) {
this.sendbuf[rank] = sendbuf;
length[rank] = Array.getLength(sendbuf);
}
public void run() {
for (int i = 0; i < size; ++i)
System.arraycopy(sendbuf[i], 0, recvbuf[i], 0, length[i]);
}
}
void scatter(Object[] sendbuf, Object recvbuf, int root, int rank) {
scatter.setRecv(recvbuf, rank);
if (rank == root)
scatter.sendbuf = sendbuf;
await(scatter.barrier);
}
private final Scatter scatter;
private class Scatter implements Runnable {
final CyclicBarrier barrier = new CyclicBarrier(size, this);
Object[] sendbuf = new Object[size], recvbuf = new Object[size];
private final int[] length = new int[size];
public void setRecv(Object recvbuf, int rank) {
this.recvbuf[rank] = recvbuf;
length[rank] = Array.getLength(recvbuf);
}
public void run() {
for (int i = 0; i < size; ++i)
System.arraycopy(sendbuf[i], 0, recvbuf[i], 0, length[i]);
}
}
void allGather(Object sendbuf, Object[] recvbuf, int rank) {
allGather.setSendRecv(sendbuf, recvbuf, rank);
await(allGather.barrier);
}
private final AllGather allGather;
private class AllGather implements Runnable {
final CyclicBarrier barrier = new CyclicBarrier(size, this);
private final Object sendbuf[] = new Object[size],
recvbuf[][] = new Object[size][size];
private final int[] length = new int[size];
public void setSendRecv(Object send, Object[] recv, int rank) {
sendbuf[rank] = send;
recvbuf[rank] = recv;
length[rank] = Array.getLength(send);
}
public void run() {
for (int i = 0; i < size; ++i)
for (int j = 0; j < size; ++j)
System
.arraycopy(sendbuf[i], 0, recvbuf[j][i], 0,
length[i]);
}
}
void allToAll(Object[] sendbuf, Object[] recvbuf, int rank) {
allToAll.setSendRecv(sendbuf, recvbuf, rank);
await(allToAll.barrier);
}
private final AllToAll allToAll;
private class AllToAll implements Runnable {
final CyclicBarrier barrier = new CyclicBarrier(size, this);
private final Object[][] sendbuf = new Object[size][size],
recvbuf = new Object[size][size];
private final int[][] length = new int[size][size];
public void setSendRecv(Object[] send, Object[] recv, int rank) {
sendbuf[rank] = send;
recvbuf[rank] = recv;
for (int i = 0; i < size; ++i)
length[rank][i] = Array.getLength(send[i]);
}
public void run() {
for (int i = 0; i < size; ++i)
for (int j = 0; j < size; ++j)
System.arraycopy(sendbuf[i][j], 0, recvbuf[j][i], 0,
length[i][j]);
}
}
void reduce(Object sendbuf, Object recvbuf, Reduction op, int root, int rank) {
reduce.sendbuf[rank] = sendbuf;
if (rank == root) {
reduce.op = op;
reduce.recvbuf = recvbuf;
}
await(reduce.barrier);
}
private final Reduce reduce;
private class Reduce implements Runnable {
CyclicBarrier barrier = new CyclicBarrier(size, this);
Reduction op;
Object sendbuf[] = new Object[size], recvbuf;
public void run() {
op.init(recvbuf);
for (int i = 0; i < size; ++i)
op.op(recvbuf, sendbuf[i]);
}
}
void allReduce(Object sendbuf, Object recvbuf, Reduction op, int rank) {
reduce(sendbuf, recvbuf, op, 0, rank);
broadcast(recvbuf, 0, rank);
}
}