Search in sources :

Example 1 with AppendGInstruction

use of org.apache.sysml.runtime.instructions.mr.AppendGInstruction in project incubator-systemml by apache.

the class MRBaseForCommonInstructions method processOneInstruction.

protected void processOneInstruction(MRInstruction ins, Class<? extends MatrixValue> valueClass, CachedValueMap cachedValues, IndexedMatrixValue tempValue, IndexedMatrixValue zeroInput) throws DMLRuntimeException {
    if (ins instanceof AggregateBinaryInstruction) {
        byte input = ((AggregateBinaryInstruction) ins).input1;
        MatrixCharacteristics dim = dimensions.get(input);
        if (dim == null)
            throw new DMLRuntimeException("dimension for instruction " + ins + "  is unset!!!");
        ins.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dim.getRowsPerBlock(), dim.getColsPerBlock());
    } else if (ins instanceof ZeroOutInstruction || ins instanceof AggregateUnaryInstruction || ins instanceof RangeBasedReIndexInstruction || ins instanceof CumulativeSplitInstruction) {
        byte input = ((UnaryMRInstructionBase) ins).input;
        MatrixCharacteristics dim = dimensions.get(input);
        if (dim == null)
            throw new DMLRuntimeException("dimension for instruction " + ins + "  is unset!!!");
        if (ins instanceof CumulativeAggregateInstruction)
            ((CumulativeAggregateInstruction) ins).setMatrixCharacteristics(dim);
        if (ins instanceof CumulativeSplitInstruction)
            ((CumulativeSplitInstruction) ins).setMatrixCharacteristics(dim);
        ins.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dim.getRowsPerBlock(), dim.getColsPerBlock());
    } else if (ins instanceof ReorgInstruction) {
        ReorgInstruction rinst = (ReorgInstruction) ins;
        byte input = rinst.input;
        MatrixCharacteristics dim = dimensions.get(input);
        if (dim == null)
            throw new DMLRuntimeException("dimension for instruction " + ins + "  is unset!!!");
        rinst.setInputMatrixCharacteristics(dim);
        //MMCJMRMapper does not output empty blocks, no need to generate
        rinst.setOutputEmptyBlocks(!(this instanceof MMCJMRMapper));
        ins.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dim.getRowsPerBlock(), dim.getColsPerBlock());
    } else if (ins instanceof MatrixReshapeMRInstruction) {
        MatrixReshapeMRInstruction mrins = (MatrixReshapeMRInstruction) ins;
        byte input = mrins.input;
        byte output = mrins.output;
        MatrixCharacteristics dimIn = dimensions.get(input);
        MatrixCharacteristics dimOut = dimensions.get(output);
        if (dimIn == null || dimOut == null)
            throw new DMLRuntimeException("dimension for instruction " + ins + "  is unset!!!");
        mrins.setMatrixCharacteristics(dimIn, dimOut);
        mrins.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dimIn.getRowsPerBlock(), dimIn.getColsPerBlock());
    } else if (ins instanceof AppendMInstruction) {
        byte input = ((AppendMInstruction) ins).input1;
        MatrixCharacteristics dim = dimensions.get(input);
        if (dim == null)
            throw new DMLRuntimeException("dimension for instruction " + ins + "  is unset!!!");
        ins.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dim.getRowsPerBlock(), dim.getColsPerBlock());
    } else if (ins instanceof BinaryMInstruction || ins instanceof RemoveEmptyMRInstruction) {
        byte input = ((BinaryMRInstructionBase) ins).input1;
        MatrixCharacteristics dim = dimensions.get(input);
        if (dim == null)
            throw new DMLRuntimeException("dimension for instruction " + ins + "  is unset!!!");
        ins.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dim.getRowsPerBlock(), dim.getColsPerBlock());
    } else if (ins instanceof AppendGInstruction) {
        AppendGInstruction arinst = ((AppendGInstruction) ins);
        byte input = arinst.input1;
        MatrixCharacteristics dimIn = dimensions.get(input);
        if (dimIn == null)
            throw new DMLRuntimeException("Dimensions for instruction " + arinst + "  is unset!!!");
        arinst.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dimIn.getRowsPerBlock(), dimIn.getColsPerBlock());
    } else if (ins instanceof UnaryMRInstructionBase) {
        UnaryMRInstructionBase rinst = (UnaryMRInstructionBase) ins;
        MatrixCharacteristics dimIn = dimensions.get(rinst.input);
        if (dimIn == null)
            throw new DMLRuntimeException("Dimensions for instruction " + rinst + "  is unset!!!");
        rinst.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dimIn.getRowsPerBlock(), dimIn.getColsPerBlock());
    } else if (ins instanceof BinaryMRInstructionBase) {
        BinaryMRInstructionBase rinst = (BinaryMRInstructionBase) ins;
        MatrixCharacteristics dimIn = dimensions.get(rinst.input1);
        if (//not set for all
        dimIn != null)
            rinst.processInstruction(valueClass, cachedValues, tempValue, zeroInput, dimIn.getRowsPerBlock(), dimIn.getColsPerBlock());
        else
            ins.processInstruction(valueClass, cachedValues, tempValue, zeroInput, -1, -1);
    } else
        ins.processInstruction(valueClass, cachedValues, tempValue, zeroInput, -1, -1);
//System.out.println(ins.getMRInstructionType()+" in "+time.stop());
}
Also used : BinaryMRInstructionBase(org.apache.sysml.runtime.instructions.mr.BinaryMRInstructionBase) AppendGInstruction(org.apache.sysml.runtime.instructions.mr.AppendGInstruction) AggregateUnaryInstruction(org.apache.sysml.runtime.instructions.mr.AggregateUnaryInstruction) RangeBasedReIndexInstruction(org.apache.sysml.runtime.instructions.mr.RangeBasedReIndexInstruction) MatrixCharacteristics(org.apache.sysml.runtime.matrix.MatrixCharacteristics) DMLRuntimeException(org.apache.sysml.runtime.DMLRuntimeException) MatrixReshapeMRInstruction(org.apache.sysml.runtime.instructions.mr.MatrixReshapeMRInstruction) RemoveEmptyMRInstruction(org.apache.sysml.runtime.instructions.mr.RemoveEmptyMRInstruction) AppendMInstruction(org.apache.sysml.runtime.instructions.mr.AppendMInstruction) ZeroOutInstruction(org.apache.sysml.runtime.instructions.mr.ZeroOutInstruction) ReorgInstruction(org.apache.sysml.runtime.instructions.mr.ReorgInstruction) CumulativeAggregateInstruction(org.apache.sysml.runtime.instructions.mr.CumulativeAggregateInstruction) AggregateBinaryInstruction(org.apache.sysml.runtime.instructions.mr.AggregateBinaryInstruction) UnaryMRInstructionBase(org.apache.sysml.runtime.instructions.mr.UnaryMRInstructionBase) CumulativeSplitInstruction(org.apache.sysml.runtime.instructions.mr.CumulativeSplitInstruction) BinaryMInstruction(org.apache.sysml.runtime.instructions.mr.BinaryMInstruction)

Example 2 with AppendGInstruction

use of org.apache.sysml.runtime.instructions.mr.AppendGInstruction in project incubator-systemml by apache.

the class MRJobConfiguration method computeMatrixCharacteristics.

/**
	 * NOTE: this method needs to be in-sync with MRBaseForCommonInstructions.processOneInstruction,
	 * otherwise, the latter will potentially fail with missing dimension information.
	 * 
	 * @param job job configuration
	 * @param inputIndexes array of byte indexes
	 * @param dataGenInstructions data gen instructions as a string
	 * @param instructionsInMapper instruction in mapper as a string
	 * @param reblockInstructions reblock instructions as a string
	 * @param aggInstructionsInReducer aggregate instructions in reducer as a string
	 * @param aggBinInstructions binary aggregate instructions as a string
	 * @param otherInstructionsInReducer other instructions in reducer as a string
	 * @param resultIndexes array of byte result indexes
	 * @param mapOutputIndexes set of map output indexes
	 * @param forMMCJ ?
	 * @return reducer groups
	 * @throws DMLRuntimeException if DMLRuntimeException occurs
	 */
public static MatrixChar_N_ReducerGroups computeMatrixCharacteristics(JobConf job, byte[] inputIndexes, String dataGenInstructions, String instructionsInMapper, String reblockInstructions, String aggInstructionsInReducer, String aggBinInstructions, String otherInstructionsInReducer, byte[] resultIndexes, HashSet<Byte> mapOutputIndexes, boolean forMMCJ) throws DMLRuntimeException {
    HashSet<Byte> intermediateMatrixIndexes = new HashSet<Byte>();
    HashMap<Byte, MatrixCharacteristics> dims = new HashMap<Byte, MatrixCharacteristics>();
    for (byte i : inputIndexes) {
        MatrixCharacteristics dim = new MatrixCharacteristics(getNumRows(job, i), getNumColumns(job, i), getNumRowsPerBlock(job, i), getNumColumnsPerBlock(job, i), getNumNonZero(job, i));
        dims.put(i, dim);
    }
    DataGenMRInstruction[] dataGenIns = null;
    dataGenIns = MRInstructionParser.parseDataGenInstructions(dataGenInstructions);
    if (dataGenIns != null) {
        for (DataGenMRInstruction ins : dataGenIns) {
            MatrixCharacteristics.computeDimension(dims, ins);
        }
    }
    MRInstruction[] insMapper = MRInstructionParser.parseMixedInstructions(instructionsInMapper);
    if (insMapper != null) {
        for (MRInstruction ins : insMapper) {
            MatrixCharacteristics.computeDimension(dims, ins);
            if (ins instanceof UnaryMRInstructionBase) {
                UnaryMRInstructionBase tempIns = (UnaryMRInstructionBase) ins;
                setIntermediateMatrixCharactristics(job, tempIns.input, dims.get(tempIns.input));
                intermediateMatrixIndexes.add(tempIns.input);
            } else if (ins instanceof AppendMInstruction) {
                AppendMInstruction tempIns = (AppendMInstruction) ins;
                setIntermediateMatrixCharactristics(job, tempIns.input1, dims.get(tempIns.input1));
                intermediateMatrixIndexes.add(tempIns.input1);
            } else if (ins instanceof AppendGInstruction) {
                AppendGInstruction tempIns = (AppendGInstruction) ins;
                setIntermediateMatrixCharactristics(job, tempIns.input1, dims.get(tempIns.input1));
                intermediateMatrixIndexes.add(tempIns.input1);
            } else if (ins instanceof BinaryMInstruction) {
                BinaryMInstruction tempIns = (BinaryMInstruction) ins;
                setIntermediateMatrixCharactristics(job, tempIns.input1, dims.get(tempIns.input1));
                intermediateMatrixIndexes.add(tempIns.input1);
            } else if (ins instanceof AggregateBinaryInstruction) {
                AggregateBinaryInstruction tempIns = (AggregateBinaryInstruction) ins;
                setIntermediateMatrixCharactristics(job, tempIns.input1, dims.get(tempIns.input1));
                //TODO
                intermediateMatrixIndexes.add(tempIns.input1);
            } else if (ins instanceof MapMultChainInstruction) {
                MapMultChainInstruction tempIns = (MapMultChainInstruction) ins;
                setIntermediateMatrixCharactristics(job, tempIns.getInput1(), dims.get(tempIns.getInput2()));
                intermediateMatrixIndexes.add(tempIns.getInput1());
            } else if (ins instanceof PMMJMRInstruction) {
                PMMJMRInstruction tempIns = (PMMJMRInstruction) ins;
                setIntermediateMatrixCharactristics(job, tempIns.input2, dims.get(tempIns.input2));
                intermediateMatrixIndexes.add(tempIns.input2);
            }
        }
    }
    ReblockInstruction[] reblockIns = MRInstructionParser.parseReblockInstructions(reblockInstructions);
    if (reblockIns != null) {
        for (ReblockInstruction ins : reblockIns) {
            MatrixCharacteristics.computeDimension(dims, ins);
            setMatrixCharactristicsForReblock(job, ins.output, dims.get(ins.output));
        }
    }
    Instruction[] aggIns = MRInstructionParser.parseAggregateInstructions(aggInstructionsInReducer);
    if (aggIns != null) {
        for (Instruction ins : aggIns) {
            MatrixCharacteristics.computeDimension(dims, (MRInstruction) ins);
            // if instruction's output is not in resultIndexes, then add its dimensions to jobconf
            MRInstruction mrins = (MRInstruction) ins;
            boolean found = false;
            for (byte b : resultIndexes) {
                if (b == mrins.output) {
                    found = true;
                    break;
                }
            }
            if (!found) {
                setIntermediateMatrixCharactristics(job, mrins.output, dims.get(mrins.output));
                intermediateMatrixIndexes.add(mrins.output);
            }
        }
    }
    long numReduceGroups = 0;
    AggregateBinaryInstruction[] aggBinIns = getAggregateBinaryInstructions(job);
    if (aggBinIns != null) {
        for (AggregateBinaryInstruction ins : aggBinIns) {
            MatrixCharacteristics dim1 = dims.get(ins.input1);
            MatrixCharacteristics dim2 = dims.get(ins.input2);
            setMatrixCharactristicsForBinAgg(job, ins.input1, dim1);
            setMatrixCharactristicsForBinAgg(job, ins.input2, dim2);
            MatrixCharacteristics.computeDimension(dims, ins);
            if (//there will be only one aggbin operation for MMCJ
            forMMCJ)
                numReduceGroups = (long) Math.ceil((double) dim1.getCols() / (double) dim1.getColsPerBlock());
        }
    }
    if (!forMMCJ) {
        //store the skylines
        ArrayList<Long> xs = new ArrayList<Long>(mapOutputIndexes.size());
        ArrayList<Long> ys = new ArrayList<Long>(mapOutputIndexes.size());
        for (byte idx : mapOutputIndexes) {
            MatrixCharacteristics dim = dims.get(idx);
            long x = (long) Math.ceil((double) dim.getRows() / (double) dim.getRowsPerBlock());
            long y = (long) Math.ceil((double) dim.getCols() / (double) dim.getColsPerBlock());
            int i = 0;
            boolean toadd = true;
            while (i < xs.size()) {
                if ((x >= xs.get(i) && y > ys.get(i)) || (x > xs.get(i) && y >= ys.get(i))) {
                    //remove any included x's and y's
                    xs.remove(i);
                    ys.remove(i);
                } else if (//if included in others, stop
                x <= xs.get(i) && y <= ys.get(i)) {
                    toadd = false;
                    break;
                } else
                    i++;
            }
            if (toadd) {
                xs.add(x);
                ys.add(y);
            }
        }
        //sort by x
        TreeMap<Long, Long> map = new TreeMap<Long, Long>();
        for (int i = 0; i < xs.size(); i++) map.put(xs.get(i), ys.get(i));
        numReduceGroups = 0;
        //compute area
        long prev = 0;
        for (Entry<Long, Long> e : map.entrySet()) {
            numReduceGroups += (e.getKey() - prev) * e.getValue();
            prev = e.getKey();
        }
    }
    MRInstruction[] insReducer = MRInstructionParser.parseMixedInstructions(otherInstructionsInReducer);
    if (insReducer != null) {
        for (MRInstruction ins : insReducer) {
            MatrixCharacteristics.computeDimension(dims, ins);
            if (ins instanceof UnaryMRInstructionBase) {
                UnaryMRInstructionBase tempIns = (UnaryMRInstructionBase) ins;
                setIntermediateMatrixCharactristics(job, tempIns.input, dims.get(tempIns.input));
                intermediateMatrixIndexes.add(tempIns.input);
            } else if (ins instanceof RemoveEmptyMRInstruction) {
                RemoveEmptyMRInstruction tempIns = (RemoveEmptyMRInstruction) ins;
                setIntermediateMatrixCharactristics(job, tempIns.input1, dims.get(tempIns.input1));
                intermediateMatrixIndexes.add(tempIns.input1);
            }
            // if instruction's output is not in resultIndexes, then add its dimensions to jobconf
            boolean found = false;
            for (byte b : resultIndexes) {
                if (b == ins.output) {
                    found = true;
                    break;
                }
            }
            if (!found) {
                setIntermediateMatrixCharactristics(job, ins.output, dims.get(ins.output));
                intermediateMatrixIndexes.add(ins.output);
            }
        }
    }
    setIntermediateMatrixIndexes(job, intermediateMatrixIndexes);
    for (byte tag : mapOutputIndexes) setMatrixCharactristicsForMapperOutput(job, tag, dims.get(tag));
    MatrixCharacteristics[] stats = new MatrixCharacteristics[resultIndexes.length];
    MatrixCharacteristics resultDims;
    for (int i = 0; i < resultIndexes.length; i++) {
        resultDims = dims.get(resultIndexes[i]);
        stats[i] = resultDims;
        setMatrixCharactristicsForOutput(job, resultIndexes[i], stats[i]);
    }
    return new MatrixChar_N_ReducerGroups(stats, numReduceGroups);
}
Also used : HashMap(java.util.HashMap) ArrayList(java.util.ArrayList) DataGenMRInstruction(org.apache.sysml.runtime.instructions.mr.DataGenMRInstruction) ReblockInstruction(org.apache.sysml.runtime.instructions.mr.ReblockInstruction) CSVReblockInstruction(org.apache.sysml.runtime.instructions.mr.CSVReblockInstruction) CSVWriteInstruction(org.apache.sysml.runtime.instructions.mr.CSVWriteInstruction) ReblockInstruction(org.apache.sysml.runtime.instructions.mr.ReblockInstruction) BinaryMInstruction(org.apache.sysml.runtime.instructions.mr.BinaryMInstruction) GroupedAggregateInstruction(org.apache.sysml.runtime.instructions.mr.GroupedAggregateInstruction) AggregateInstruction(org.apache.sysml.runtime.instructions.mr.AggregateInstruction) Instruction(org.apache.sysml.runtime.instructions.Instruction) PMMJMRInstruction(org.apache.sysml.runtime.instructions.mr.PMMJMRInstruction) DataGenMRInstruction(org.apache.sysml.runtime.instructions.mr.DataGenMRInstruction) AppendMInstruction(org.apache.sysml.runtime.instructions.mr.AppendMInstruction) AggregateBinaryInstruction(org.apache.sysml.runtime.instructions.mr.AggregateBinaryInstruction) CM_N_COVInstruction(org.apache.sysml.runtime.instructions.mr.CM_N_COVInstruction) MapMultChainInstruction(org.apache.sysml.runtime.instructions.mr.MapMultChainInstruction) MRInstruction(org.apache.sysml.runtime.instructions.mr.MRInstruction) CSVReblockInstruction(org.apache.sysml.runtime.instructions.mr.CSVReblockInstruction) AppendGInstruction(org.apache.sysml.runtime.instructions.mr.AppendGInstruction) RemoveEmptyMRInstruction(org.apache.sysml.runtime.instructions.mr.RemoveEmptyMRInstruction) PMMJMRInstruction(org.apache.sysml.runtime.instructions.mr.PMMJMRInstruction) RemoveEmptyMRInstruction(org.apache.sysml.runtime.instructions.mr.RemoveEmptyMRInstruction) UnaryMRInstructionBase(org.apache.sysml.runtime.instructions.mr.UnaryMRInstructionBase) MapMultChainInstruction(org.apache.sysml.runtime.instructions.mr.MapMultChainInstruction) PMMJMRInstruction(org.apache.sysml.runtime.instructions.mr.PMMJMRInstruction) DataGenMRInstruction(org.apache.sysml.runtime.instructions.mr.DataGenMRInstruction) MRInstruction(org.apache.sysml.runtime.instructions.mr.MRInstruction) RemoveEmptyMRInstruction(org.apache.sysml.runtime.instructions.mr.RemoveEmptyMRInstruction) HashSet(java.util.HashSet) AppendGInstruction(org.apache.sysml.runtime.instructions.mr.AppendGInstruction) TreeMap(java.util.TreeMap) MatrixCharacteristics(org.apache.sysml.runtime.matrix.MatrixCharacteristics) AppendMInstruction(org.apache.sysml.runtime.instructions.mr.AppendMInstruction) AggregateBinaryInstruction(org.apache.sysml.runtime.instructions.mr.AggregateBinaryInstruction) BinaryMInstruction(org.apache.sysml.runtime.instructions.mr.BinaryMInstruction)

Aggregations

AggregateBinaryInstruction (org.apache.sysml.runtime.instructions.mr.AggregateBinaryInstruction)2 AppendGInstruction (org.apache.sysml.runtime.instructions.mr.AppendGInstruction)2 AppendMInstruction (org.apache.sysml.runtime.instructions.mr.AppendMInstruction)2 BinaryMInstruction (org.apache.sysml.runtime.instructions.mr.BinaryMInstruction)2 RemoveEmptyMRInstruction (org.apache.sysml.runtime.instructions.mr.RemoveEmptyMRInstruction)2 UnaryMRInstructionBase (org.apache.sysml.runtime.instructions.mr.UnaryMRInstructionBase)2 MatrixCharacteristics (org.apache.sysml.runtime.matrix.MatrixCharacteristics)2 ArrayList (java.util.ArrayList)1 HashMap (java.util.HashMap)1 HashSet (java.util.HashSet)1 TreeMap (java.util.TreeMap)1 DMLRuntimeException (org.apache.sysml.runtime.DMLRuntimeException)1 Instruction (org.apache.sysml.runtime.instructions.Instruction)1 AggregateInstruction (org.apache.sysml.runtime.instructions.mr.AggregateInstruction)1 AggregateUnaryInstruction (org.apache.sysml.runtime.instructions.mr.AggregateUnaryInstruction)1 BinaryMRInstructionBase (org.apache.sysml.runtime.instructions.mr.BinaryMRInstructionBase)1 CM_N_COVInstruction (org.apache.sysml.runtime.instructions.mr.CM_N_COVInstruction)1 CSVReblockInstruction (org.apache.sysml.runtime.instructions.mr.CSVReblockInstruction)1 CSVWriteInstruction (org.apache.sysml.runtime.instructions.mr.CSVWriteInstruction)1 CumulativeAggregateInstruction (org.apache.sysml.runtime.instructions.mr.CumulativeAggregateInstruction)1