org.apache.sysml.runtime.instructions.spark.data.PartitionedMatrixBlock Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of systemml Show documentation
Show all versions of systemml Show documentation
Declarative Machine Learning
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing,
* software distributed under the License is distributed on an
* "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
* KIND, either express or implied. See the License for the
* specific language governing permissions and limitations
* under the License.
*/
package org.apache.sysml.runtime.instructions.spark.data;
import java.io.DataInput;
import java.io.DataOutput;
import java.io.Externalizable;
import java.io.IOException;
import java.io.ObjectInput;
import java.io.ObjectInputStream;
import java.io.ObjectOutput;
import java.io.ObjectOutputStream;
import java.util.ArrayList;
import scala.Tuple2;
import org.apache.sysml.runtime.DMLRuntimeException;
import org.apache.sysml.runtime.DMLUnsupportedOperationException;
import org.apache.sysml.runtime.instructions.spark.utils.SparkUtils;
import org.apache.sysml.runtime.matrix.data.MatrixBlock;
import org.apache.sysml.runtime.matrix.data.MatrixIndexes;
import org.apache.sysml.runtime.matrix.data.OperationsOnMatrixValues;
import org.apache.sysml.runtime.matrix.mapred.IndexedMatrixValue;
import org.apache.sysml.runtime.util.FastBufferedDataInputStream;
import org.apache.sysml.runtime.util.FastBufferedDataOutputStream;
import org.apache.sysml.runtime.util.IndexRange;
/**
* The main purpose of this class is to provide a handle for partitioned matrix blocks, to be used
* as broadcasts. Distributed tasks require block-partitioned broadcasts but a lazy partitioning per
* task would create instance-local copies and hence replicate broadcast variables which are shared
* by all tasks within an executor.
*
*/
public class PartitionedMatrixBlock implements Externalizable
{
private static final long serialVersionUID = -5706923809800365593L;
private MatrixBlock[] _partBlocks = null;
private int _rlen = -1;
private int _clen = -1;
private int _brlen = -1;
private int _bclen = -1;
private int _offset = 0;
public PartitionedMatrixBlock() {
//do nothing (required for Externalizable)
}
public PartitionedMatrixBlock(MatrixBlock mb, int brlen, int bclen)
{
//get the input matrix block
int rlen = mb.getNumRows();
int clen = mb.getNumColumns();
//partitioning input broadcast
_rlen = rlen;
_clen = clen;
_brlen = brlen;
_bclen = bclen;
int nrblks = getNumRowBlocks();
int ncblks = getNumColumnBlocks();
_partBlocks = new MatrixBlock[nrblks * ncblks];
try
{
for( int i=0, ix=0; i nrblks || colIndex <= 0 || colIndex > ncblks ) {
throw new DMLRuntimeException("Block indexes ["+rowIndex+","+colIndex+"] out of range ["+nrblks+","+ncblks+"]");
}
//get the requested matrix block
int rix = rowIndex - 1;
int cix = colIndex - 1;
int ix = rix*ncblks+cix - _offset;
return _partBlocks[ ix ];
}
/**
*
* @param rowIndex
* @param colIndex
* @param mb
* @throws DMLRuntimeException
*/
public void setMatrixBlock(int rowIndex, int colIndex, MatrixBlock mb)
throws DMLRuntimeException
{
//check for valid block index
int nrblks = getNumRowBlocks();
int ncblks = getNumColumnBlocks();
if( rowIndex <= 0 || rowIndex > nrblks || colIndex <= 0 || colIndex > ncblks ) {
throw new DMLRuntimeException("Block indexes ["+rowIndex+","+colIndex+"] out of range ["+nrblks+","+ncblks+"]");
}
//get the requested matrix block
int rix = rowIndex - 1;
int cix = colIndex - 1;
int ix = rix*ncblks+cix - _offset;
_partBlocks[ ix ] = mb;
}
/**
*
* @return
*/
public long estimateSizeInMemory()
{
long ret = 24; //header
ret += 32; //block array
if( _partBlocks != null )
for( MatrixBlock mb : _partBlocks )
ret += mb.estimateSizeInMemory();
return ret;
}
/**
*
* @return
*/
public long estimateSizeOnDisk()
{
long ret = 24; //header
if( _partBlocks != null )
for( MatrixBlock mb : _partBlocks )
ret += mb.estimateSizeOnDisk();
return ret;
}
/**
*
* @param offset
* @param numBlks
* @return
*/
public PartitionedMatrixBlock createPartition( int offset, int numBlks )
{
PartitionedMatrixBlock ret = new PartitionedMatrixBlock();
ret._rlen = _rlen;
ret._clen = _clen;
ret._brlen = _brlen;
ret._bclen = _bclen;
ret._partBlocks = new MatrixBlock[numBlks];
ret._offset = offset;
System.arraycopy(_partBlocks, offset, ret._partBlocks, 0, numBlks);
return ret;
}
/**
* Utility for slice operations over partitioned matrices, where the index range can cover
* multiple blocks. The result is always a single result matrix block. All semantics are
* equivalent to the core matrix block slice operations.
*
* @param rl
* @param ru
* @param cl
* @param cu
* @param matrixBlock
* @return
* @throws DMLUnsupportedOperationException
* @throws DMLRuntimeException
*/
public MatrixBlock sliceOperations(long rl, long ru, long cl, long cu, MatrixBlock matrixBlock)
throws DMLRuntimeException, DMLUnsupportedOperationException
{
int lrl = (int) rl;
int lru = (int) ru;
int lcl = (int) cl;
int lcu = (int) cu;
ArrayList> allBlks = new ArrayList>();
int start_iix = (lrl-1)/_brlen+1;
int end_iix = (lru-1)/_brlen+1;
int start_jix = (lcl-1)/_bclen+1;
int end_jix = (lcu-1)/_bclen+1;
for( int iix = start_iix; iix <= end_iix; iix++ )
for(int jix = start_jix; jix <= end_jix; jix++)
{
MatrixBlock in = getMatrixBlock(iix, jix);
IndexedMatrixValue imv = new IndexedMatrixValue(new MatrixIndexes(iix, jix), in);
ArrayList outlist = new ArrayList();
IndexRange ixrange = new IndexRange(rl, ru, cl, cu);
OperationsOnMatrixValues.performSlice(imv, ixrange, _brlen, _bclen, outlist);
allBlks.addAll(SparkUtils.fromIndexedMatrixBlock(outlist));
}
if(allBlks.size() == 1) {
return allBlks.get(0)._2;
}
else {
//allocate output matrix
MatrixBlock ret = new MatrixBlock(lru-lrl+1, lcu-lcl+1, false);
for(Tuple2 kv : allBlks) {
ret.merge(kv._2, false);
}
return ret;
}
}
/**
* Redirects the default java serialization via externalizable to our default
* hadoop writable serialization for efficient broadcast deserialization.
*
* @param is
* @throws IOException
*/
public void readExternal(ObjectInput is)
throws IOException
{
DataInput dis = is;
if( is instanceof ObjectInputStream ) {
//fast deserialize of dense/sparse blocks
ObjectInputStream ois = (ObjectInputStream)is;
dis = new FastBufferedDataInputStream(ois);
}
readHeaderAndPayload(dis);
}
/**
* Redirects the default java serialization via externalizable to our default
* hadoop writable serialization for efficient broadcast serialization.
*
* @param is
* @throws IOException
*/
public void writeExternal(ObjectOutput os)
throws IOException
{
if( os instanceof ObjectOutputStream ) {
//fast serialize of dense/sparse blocks
ObjectOutputStream oos = (ObjectOutputStream)os;
FastBufferedDataOutputStream fos = new FastBufferedDataOutputStream(oos);
writeHeaderAndPayload(fos);
fos.flush();
}
else {
//default serialize (general case)
writeHeaderAndPayload(os);
}
}
/**
*
* @param dos
* @throws IOException
*/
private void writeHeaderAndPayload(DataOutput dos)
throws IOException
{
dos.writeInt(_rlen);
dos.writeInt(_clen);
dos.writeInt(_brlen);
dos.writeInt(_bclen);
dos.writeInt(_offset);
dos.writeInt(_partBlocks.length);
for( MatrixBlock mb : _partBlocks )
mb.write(dos);
}
/**
*
* @param din
* @throws IOException
*/
private void readHeaderAndPayload(DataInput dis)
throws IOException
{
_rlen = dis.readInt();
_clen = dis.readInt();
_brlen = dis.readInt();
_bclen = dis.readInt();
_offset = dis.readInt();
int len = dis.readInt();
_partBlocks = new MatrixBlock[len];
for( int i=0; i