org.apache.sysml.runtime.instructions.mr.PMMJMRInstruction 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.mr;
import java.util.ArrayList;
import org.apache.sysml.hops.OptimizerUtils;
import org.apache.sysml.lops.PMMJ.CacheType;
import org.apache.sysml.runtime.DMLRuntimeException;
import org.apache.sysml.runtime.instructions.InstructionUtils;
import org.apache.sysml.runtime.matrix.data.MatrixBlock;
import org.apache.sysml.runtime.matrix.data.MatrixValue;
import org.apache.sysml.runtime.matrix.mapred.CachedValueMap;
import org.apache.sysml.runtime.matrix.mapred.DistributedCacheInput;
import org.apache.sysml.runtime.matrix.mapred.IndexedMatrixValue;
import org.apache.sysml.runtime.matrix.mapred.MRBaseForCommonInstructions;
import org.apache.sysml.runtime.matrix.operators.Operator;
import org.apache.sysml.runtime.util.UtilFunctions;
public class PMMJMRInstruction extends BinaryMRInstructionBase implements IDistributedCacheConsumer
{
private long _rlen = -1;
private boolean _outputEmptyBlocks = true;
public PMMJMRInstruction(Operator op, byte in1, byte in2, byte out, long nrow, CacheType ctype, boolean outputEmpty, String istr)
{
super(op, in1, in2, out);
instString = istr;
_rlen = nrow;
_outputEmptyBlocks = outputEmpty;
//NOTE: cache type only used by distributed cache input
}
public long getNumRows() {
return _rlen;
}
public boolean getOutputEmptyBlocks() {
return _outputEmptyBlocks;
}
public static PMMJMRInstruction parseInstruction ( String str )
throws DMLRuntimeException
{
InstructionUtils.checkNumFields ( str, 6 );
String[] parts = InstructionUtils.getInstructionParts(str);
String opcode = parts[0];
byte in1 = Byte.parseByte(parts[1]);
byte in2 = Byte.parseByte(parts[2]);
long nrow = UtilFunctions.toLong(Double.parseDouble(parts[3]));
byte out = Byte.parseByte(parts[4]);
CacheType ctype = CacheType.valueOf(parts[5]);
boolean outputEmpty = Boolean.parseBoolean(parts[6]);
if(!opcode.equalsIgnoreCase("pmm"))
throw new DMLRuntimeException("Unknown opcode while parsing an PmmMRInstruction: " + str);
return new PMMJMRInstruction(new Operator(true), in1, in2, out, nrow, ctype, outputEmpty, str);
}
@Override
public void processInstruction(Class valueClass,
CachedValueMap cachedValues, IndexedMatrixValue tempValue,
IndexedMatrixValue zeroInput, int blockRowFactor, int blockColFactor)
throws DMLRuntimeException
{
//get both matrix inputs (left side always permutation)
DistributedCacheInput dcInput = MRBaseForCommonInstructions.dcValues.get(input1);
IndexedMatrixValue in2 = cachedValues.getFirst(input2);
IndexedMatrixValue in1 = dcInput.getDataBlock((int)in2.getIndexes().getRowIndex(), 1);
MatrixBlock mb1 = (MatrixBlock)in1.getValue();
MatrixBlock mb2 = (MatrixBlock)in2.getValue();
//compute target block indexes
long minPos = UtilFunctions.toLong( mb1.minNonZero() );
long maxPos = UtilFunctions.toLong( mb1.max() );
long rowIX1 = (minPos-1)/blockRowFactor+1;
long rowIX2 = (maxPos-1)/blockRowFactor+1;
boolean multipleOuts = (rowIX1 != rowIX2);
if( minPos >= 1 ) //at least one row selected
{
//output sparsity estimate
double spmb1 = OptimizerUtils.getSparsity(mb1.getNumRows(), 1, mb1.getNonZeros());
long estnnz = (long) (spmb1 * mb2.getNonZeros());
boolean sparse = MatrixBlock.evalSparseFormatInMemory(blockRowFactor, mb2.getNumColumns(), estnnz);
//compute and allocate output blocks
IndexedMatrixValue out1 = cachedValues.holdPlace(output, valueClass);
IndexedMatrixValue out2 = multipleOuts ? cachedValues.holdPlace(output, valueClass) : null;
out1.getValue().reset(blockRowFactor, mb2.getNumColumns(), sparse);
if( out2 != null )
out2.getValue().reset(UtilFunctions.computeBlockSize(_rlen, rowIX2, blockRowFactor), mb2.getNumColumns(), sparse);
//compute core matrix permutation (assumes that out1 has default blocksize,
//hence we do a meta data correction afterwards)
mb1.permutationMatrixMultOperations(mb2, out1.getValue(), (out2!=null)?out2.getValue():null);
((MatrixBlock)out1.getValue()).setNumRows(UtilFunctions.computeBlockSize(_rlen, rowIX1, blockRowFactor));
out1.getIndexes().setIndexes(rowIX1, in2.getIndexes().getColumnIndex());
if( out2 != null )
out2.getIndexes().setIndexes(rowIX2, in2.getIndexes().getColumnIndex());
//empty block output filter (enabled by compiler consumer operation is in CP)
if( !_outputEmptyBlocks && out1.getValue().isEmpty()
&& (out2==null || out2.getValue().isEmpty() ) )
{
cachedValues.remove(output);
}
}
}
@Override //IDistributedCacheConsumer
public boolean isDistCacheOnlyIndex( String inst, byte index )
{
return (index==input1 && index!=input2);
}
@Override //IDistributedCacheConsumer
public void addDistCacheIndex( String inst, ArrayList indexes )
{
indexes.add(input1);
}
}