blob: 55554240b9c2ae98ccfc6b87634299e5c8b17246 [file] [log] [blame]
/*
* 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.sysds.runtime.instructions.fed;
import java.util.concurrent.Future;
import org.apache.sysds.common.Types.ExecType;
import org.apache.sysds.hops.fedplanner.FTypes.FType;
import org.apache.sysds.runtime.DMLRuntimeException;
import org.apache.sysds.runtime.controlprogram.caching.MatrixObject;
import org.apache.sysds.runtime.controlprogram.context.ExecutionContext;
import org.apache.sysds.runtime.controlprogram.federated.FederatedRange;
import org.apache.sysds.runtime.controlprogram.federated.FederatedRequest;
import org.apache.sysds.runtime.controlprogram.federated.FederatedRequest.RequestType;
import org.apache.sysds.runtime.controlprogram.federated.FederatedResponse;
import org.apache.sysds.runtime.controlprogram.federated.FederationMap;
import org.apache.sysds.runtime.controlprogram.federated.FederationUtils;
import org.apache.sysds.runtime.instructions.InstructionUtils;
import org.apache.sysds.runtime.instructions.cp.AggregateUnaryCPInstruction;
import org.apache.sysds.runtime.instructions.cp.CPOperand;
import org.apache.sysds.runtime.instructions.cp.ScalarObject;
import org.apache.sysds.runtime.instructions.spark.AggregateUnarySPInstruction;
import org.apache.sysds.runtime.matrix.operators.AggregateUnaryOperator;
import org.apache.sysds.runtime.matrix.operators.Operator;
import org.apache.sysds.runtime.meta.DataCharacteristics;
import org.apache.sysds.runtime.meta.MatrixCharacteristics;
public class AggregateUnaryFEDInstruction extends UnaryFEDInstruction {
private AggregateUnaryFEDInstruction(AggregateUnaryOperator auop,
CPOperand in, CPOperand out, String opcode, String istr, FederatedOutput fedOut)
{
super(FEDType.AggregateUnary, auop, in, out, opcode, istr, fedOut);
}
protected AggregateUnaryFEDInstruction(Operator op,
CPOperand in1, CPOperand in2, CPOperand out, String opcode, String istr, FederatedOutput fedOut)
{
super(FEDType.AggregateUnary, op, in1, in2, out, opcode, istr, fedOut);
}
protected AggregateUnaryFEDInstruction(Operator op,
CPOperand in1, CPOperand in2, CPOperand out, String opcode, String istr)
{
super(FEDType.AggregateUnary, op, in1, in2, out, opcode, istr);
}
protected AggregateUnaryFEDInstruction(Operator op, CPOperand in1,
CPOperand in2, CPOperand in3, CPOperand out, String opcode, String istr)
{
super(FEDType.AggregateUnary, op, in1, in2, in3, out, opcode, istr);
}
public static AggregateUnaryFEDInstruction parseInstruction(AggregateUnaryCPInstruction instr) {
return new AggregateUnaryFEDInstruction(instr.getOperator(), instr.input1, instr.input2, instr.input3,
instr.output, instr.getOpcode(), instr.getInstructionString());
}
public static AggregateUnaryFEDInstruction parseInstruction(AggregateUnarySPInstruction instr) {
// TODO: during processing the NONE-flag of AggregateUnarySPInstruction (SparkAggType) will be removed, making the instruction unparseable
return new AggregateUnaryFEDInstruction(instr.getOperator(), instr.input1, instr.input2, instr.input3,
instr.output, instr.getOpcode(), instr.getInstructionString());
}
public static AggregateUnaryFEDInstruction parseInstruction(String str) {
String[] parts = InstructionUtils.getInstructionPartsWithValueType(str);
String opcode = parts[0];
CPOperand in1 = new CPOperand(parts[1]);
CPOperand out = new CPOperand(parts[2]);
AggregateUnaryOperator aggun = null;
if(opcode.equalsIgnoreCase("uarimax") || opcode.equalsIgnoreCase("uarimin"))
if(InstructionUtils.getExecType(str) == ExecType.SPARK)
aggun = InstructionUtils.parseAggregateUnaryRowIndexOperator(opcode, 1, 1);
else
aggun = InstructionUtils.parseAggregateUnaryRowIndexOperator(opcode, Integer.parseInt(parts[4]), 1);
else
aggun = InstructionUtils.parseBasicAggregateUnaryOperator(opcode);
FederatedOutput fedOut = null;
if ( parts.length == 5 && !parts[4].equals("uarimin") && !parts[4].equals("uarimax") )
fedOut = FederatedOutput.valueOf(parts[4]);
else
fedOut = FederatedOutput.valueOf(parts[5]);
return new AggregateUnaryFEDInstruction(aggun, in1, out, opcode, str, fedOut);
}
@Override
public void processInstruction(ExecutionContext ec) {
if (getOpcode().contains("var")) {
processVar(ec);
} else {
processDefault(ec);
}
}
private void processDefault(ExecutionContext ec){
AggregateUnaryOperator aop = (AggregateUnaryOperator) _optr;
MatrixObject in = ec.getMatrixObject(input1);
if ( !in.isFederated() )
throw new DMLRuntimeException("Input is not federated " + input1);
FederationMap map = in.getFedMapping();
if ( map == null )
throw new DMLRuntimeException("Input federation map is null for input " + input1);
if((instOpcode.equalsIgnoreCase("uarimax") || instOpcode.equalsIgnoreCase("uarimin")) && in.isFederated(FType.COL))
instString = InstructionUtils.replaceOperand(instString, 5, "2");
// create federated commands for aggregation
// (by default obtain output, even though unnecessary row aggregates)
if ( _fedOut.isForcedFederated() )
if(instString.startsWith("SPARK"))
processFederatedSPOutput(map, in, ec, aop);
else
processFederatedOutput(map, in, ec);
else {
if(instString.startsWith("SPARK"))
processGetSPOutput(map, in, ec, aop);
else
processGetOutput(map, aop, ec, in);
}
}
/**
* Sends federated request with instruction without retrieving the result from the workers.
* @param map federation map of the input
* @param in input matrix object
* @param ec execution context
*/
private void processFederatedOutput(FederationMap map, MatrixObject in, ExecutionContext ec){
if ( output.isScalar() )
throw new DMLRuntimeException("Output of FED instruction, " + output.toString()
+ ", is a scalar and the output is set to be federated. Scalars cannot be federated. ");
FederatedRequest fr1 = FederationUtils.callInstruction(instString, output,
new CPOperand[]{input1}, new long[]{in.getFedMapping().getID()}, true);
map.execute(getTID(), true, fr1);
MatrixObject out = ec.getMatrixObject(output);
deriveNewOutputFedMapping(in, out, fr1);
}
/**
* Set output fed mapping based on federated partitioning and aggregation type.
* @param in matrix object from which fed partitioning originates from
* @param out matrix object holding the dimensions of the instruction output
* @param fr1 federated request holding the instruction execution call
*/
private void deriveNewOutputFedMapping(MatrixObject in, MatrixObject out, FederatedRequest fr1){
//Get agg type
//if ( !(instOpcode.equals("uack+") || instOpcode.equals("uark+")) )
// throw new DMLRuntimeException("Operation " + instOpcode + " is unknown to FOUT processing");
boolean isColAgg = ((AggregateUnaryOperator) _optr).isColAggregate();
//Get partition type
FType inFtype = in.getFedMapping().getType();
//Get fedmap from in
FederationMap inputFedMapCopy = in.getFedMapping().copyWithNewID(fr1.getID());
//if partition type is row and aggregation type is row
// then get row dim split from input and use as row dimension and get col dimension from output col dimension
// and set FType to ROW
if ( inFtype.isRowPartitioned() && !isColAgg ){
for ( FederatedRange range : inputFedMapCopy.getFederatedRanges() )
range.setEndDim(1,out.getNumColumns());
inputFedMapCopy.setType(FType.ROW);
}
//if partition type is row and aggregation type is col
// then get row and col dimension from out and use those dimensions for both federated workers
// and set FType to PART
//if partition type is col and aggregation type is row
// then set row and col dimension from out and use those dimensions for both federated workers
// and set FType to PART
if ( (inFtype.isRowPartitioned() && isColAgg) || (inFtype.isColPartitioned() && !isColAgg) ){
/*for ( FederatedRange range : inputFedMapCopy.getFederatedRanges() ){
range.setBeginDim(0,0);
range.setBeginDim(1,0);
range.setEndDim(0,out.getNumRows());
range.setEndDim(1,out.getNumColumns());
}
inputFedMapCopy.setType(FType.PART);*/
throw new DMLRuntimeException("PART output not supported");
}
//if partition type is col and aggregation type is col
// then set row dimension to output and col dimension to in col split
// and set FType to COL
if ( inFtype.isColPartitioned() && isColAgg ){
for ( FederatedRange range : inputFedMapCopy.getFederatedRanges() )
range.setEndDim(0,out.getNumRows());
inputFedMapCopy.setType(FType.COL);
}
//set out fedmap in the end
out.setFedMapping(inputFedMapCopy);
}
/**
* Sends federated request with instruction and retrieves the result from the workers.
* @param map federation map of input
* @param aggUOptr aggregate unary operator of the instruction
* @param ec execution context
* @param in input matrix object
*/
private void processGetOutput(FederationMap map, AggregateUnaryOperator aggUOptr, ExecutionContext ec, MatrixObject in){
FederatedRequest fr1 = FederationUtils.callInstruction(instString, output,
new CPOperand[]{input1}, new long[]{in.getFedMapping().getID()}, true);
FederatedRequest fr2 = new FederatedRequest(RequestType.GET_VAR, fr1.getID());
//execute federated commands and cleanups
Future<FederatedResponse>[] tmp = map.execute(getTID(), fr1, fr2);
if( output.isScalar() )
ec.setVariable(output.getName(), FederationUtils.aggScalar(aggUOptr, tmp, map));
else
ec.setMatrixOutput(output.getName(), FederationUtils.aggMatrix(aggUOptr, tmp, map));
}
private void processVar(ExecutionContext ec){
if ( _fedOut.isForcedFederated() ){
throw new DMLRuntimeException("Output of " + toString() + " should not be federated "
+ "since the instruction requires consolidation of partial results to be computed.");
}
boolean isSpark = instString.startsWith("SPARK");
AggregateUnaryOperator aop = (AggregateUnaryOperator) _optr;
MatrixObject in = ec.getMatrixObject(input1);
FederationMap map = in.getFedMapping();
long id = FederationUtils.getNextFedDataID();
FederatedRequest tmpRequest = null;
if(isSpark) {
if ( output.isScalar() ) {
ScalarObject scalarOut = ec.getScalarInput(output);
tmpRequest = map.broadcast(scalarOut);
id = tmpRequest.getID();
}
else {
if((map.getType() == FType.COL && aop.isColAggregate()) || (map.getType() == FType.ROW && aop.isRowAggregate()))
tmpRequest = new FederatedRequest(RequestType.PUT_VAR, id, new MatrixCharacteristics(-1, -1), in.getDataType());
else {
DataCharacteristics dc = ec.getDataCharacteristics(output.getName());
tmpRequest = new FederatedRequest(RequestType.PUT_VAR, id, dc, in.getDataType());
}
}
}
// federated ranges mean for variance
Future<FederatedResponse>[] meanTmp = null;
if (getOpcode().contains("var")) {
String meanInstr = instString.replace(getOpcode(), getOpcode().replace("var", "mean"));
//create federated commands for aggregation
FederatedRequest meanFr1 = FederationUtils.callInstruction(meanInstr, output, id,
new CPOperand[]{input1}, new long[]{in.getFedMapping().getID()}, isSpark ? ExecType.SPARK : ExecType.CP, isSpark);
FederatedRequest meanFr2 = new FederatedRequest(RequestType.GET_VAR, meanFr1.getID());
meanTmp = map.execute(getTID(), true, isSpark ?
new FederatedRequest[] {tmpRequest, meanFr1, meanFr2} :
new FederatedRequest[] {meanFr1, meanFr2});
}
//create federated commands for aggregation
FederatedRequest fr1 = FederationUtils.callInstruction(instString, output, id,
new CPOperand[]{input1}, new long[]{in.getFedMapping().getID()}, isSpark ? ExecType.SPARK : ExecType.CP, isSpark);
FederatedRequest fr2 = new FederatedRequest(RequestType.GET_VAR, fr1.getID());
//execute federated commands and cleanups
Future<FederatedResponse>[] tmp = map.execute(getTID(), true, isSpark ?
new FederatedRequest[] {tmpRequest, fr1, fr2} :
new FederatedRequest[] { fr1, fr2});
if( output.isScalar() )
ec.setVariable(output.getName(), FederationUtils.aggScalar(aop, tmp, meanTmp, map));
else
ec.setMatrixOutput(output.getName(), FederationUtils.aggMatrix(aop, tmp, meanTmp, map));
}
private void processFederatedSPOutput(FederationMap map, MatrixObject in, ExecutionContext ec, AggregateUnaryOperator aop) {
DataCharacteristics dc = ec.getDataCharacteristics(output.getName());
FederatedRequest fr1;
long id = FederationUtils.getNextFedDataID();
if((map.getType() == FType.COL && aop.isColAggregate()) ||
(map.getType() == FType.ROW && aop.isRowAggregate()))
fr1 = new FederatedRequest(RequestType.PUT_VAR, id, new MatrixCharacteristics(-1, -1), in.getDataType());
else
fr1 = new FederatedRequest(RequestType.PUT_VAR, id, dc, in.getDataType());
FederatedRequest fr2 = FederationUtils.callInstruction(instString, output, id,
new CPOperand[]{input1}, new long[]{in.getFedMapping().getID()}, ExecType.SPARK, true);
map.execute(getTID(), true, fr1, fr2);
// derive new fed mapping for output
MatrixObject out = ec.getMatrixObject(output);
out.setFedMapping(in.getFedMapping().copyWithNewID(fr2.getID()));
}
private void processGetSPOutput(FederationMap map, MatrixObject in, ExecutionContext ec, AggregateUnaryOperator aop) {
DataCharacteristics dc = ec.getDataCharacteristics(output.getName());
FederatedRequest fr1;
long id = FederationUtils.getNextFedDataID();
if ( output.isScalar() ) {
ScalarObject scalarOut = ec.getScalarInput(output);
fr1 = map.broadcast(scalarOut);
id = fr1.getID();
}
else {
if((map.getType() == FType.COL && aop.isColAggregate()) || (map.getType() == FType.ROW && aop.isRowAggregate()))
fr1 = new FederatedRequest(RequestType.PUT_VAR, id, new MatrixCharacteristics(-1, -1), in.getDataType());
else
fr1 = new FederatedRequest(RequestType.PUT_VAR, id, dc, in.getDataType());
}
FederatedRequest fr2 = FederationUtils.callInstruction(instString, output, id,
new CPOperand[]{input1}, new long[]{in.getFedMapping().getID()}, ExecType.SPARK, true);
FederatedRequest fr3 = new FederatedRequest(RequestType.GET_VAR, fr2.getID());
//execute federated commands and cleanups
Future<FederatedResponse>[] tmp = map.execute(getTID(), fr1, fr2, fr3);
if( output.isScalar() )
ec.setVariable(output.getName(), FederationUtils.aggScalar(aop, tmp, map));
else
ec.setMatrixOutput(output.getName(), FederationUtils.aggMatrix(aop, tmp, map));
}
}