Example usage for com.google.common.collect Range canonical

Introduction

In this page you can find the example usage for com.google.common.collect Range canonical.

Prototype

public Range<C> canonical(DiscreteDomain<C> domain) 

Source Link

Document

Returns the canonical form of this range in the given domain.

Usage

From source file:edu.mit.streamjit.impl.compiler2.SubsetBiasAllocationStrategy.java

@Override
public void allocateGroup(ActorGroup group, Range<Integer> iterations, List<Core> cores, Configuration config) {
    int id = group.id();
    int numCores = config.getParameter("Group" + id + "CoreCount", Configuration.IntParameter.class).getValue();
    Configuration.PermutationParameter<Integer> coreOrderParam = config.getParameter("Group" + id + "CoreOrder",
            Configuration.PermutationParameter.class, Integer.class);
    ImmutableList<? extends Integer> coreOrder = coreOrderParam.getUniverse();
    int rawBiasCount = config.getParameter("Group" + id + "BiasCount", Configuration.IntParameter.class)
            .getValue();//from www  .j  ava2s.  c o  m
    int biasCount = Math.min(rawBiasCount, numCores - 1);
    float bias = config.getParameter("Group" + id + "Bias", Configuration.FloatParameter.class).getValue();

    try {
        List<Core> subset = new ArrayList<>(numCores);
        for (int i = 0; i < coreOrder.size() && subset.size() < numCores; ++i)
            if (coreOrder.get(i) < cores.size())
                subset.add(cores.get(coreOrder.get(i)));
        List<Core> biasSubset = new ArrayList<>(biasCount);
        while (biasSubset.size() < biasCount)
            biasSubset.add(subset.remove(0));

        float deficitFraction = biasCount * (1 - bias) / numCores, surplusFraction = 1 - deficitFraction;
        assert deficitFraction >= 0 && surplusFraction >= 0 : String.format("%d %d %f -> %f %f", numCores,
                biasCount, bias, deficitFraction, surplusFraction);
        iterations = iterations.canonical(DiscreteDomain.integers());
        int totalIterations = iterations.upperEndpoint() - iterations.lowerEndpoint();
        int biasIterations = (int) (totalIterations * deficitFraction);
        //We pass a null config to ensure we don't interfere with the other strategy.
        if (biasCount > 0)
            new FullDataParallelAllocationStrategy(biasCount).allocateGroup(group,
                    Range.closedOpen(iterations.lowerEndpoint(), iterations.lowerEndpoint() + biasIterations),
                    biasSubset, null);
        if (numCores - biasCount > 0)
            new FullDataParallelAllocationStrategy(numCores - biasCount).allocateGroup(group,
                    Range.closedOpen(iterations.lowerEndpoint() + biasIterations, iterations.upperEndpoint()),
                    subset, null);
    } catch (Exception ex) {
        StringBuilder sb = new StringBuilder();
        sb.append(iterations).append(" of ").append(group).append("\n");
        sb.append(String.format("numCores %d, raw biasCount %d, biasCount %d, bias %f, order %s", numCores,
                rawBiasCount, biasCount, bias, coreOrder));
        throw new RuntimeException(sb.toString(), ex);
    }
}

---

From source file:edu.mit.streamjit.impl.compiler2.SubsetBiasAverageAllocationStrategy.java

@Override
public void allocateGroup(ActorGroup group, Range<Integer> iterations, List<Core> cores, Configuration config) {
    int numCores = 0, biasCount = 0;
    List<ImmutableList<? extends Integer>> coreOrders = new ArrayList<>();
    float bias = 0;
    for (Actor a : group.actors()) {
        int id = a.id();
        numCores += config.getParameter("Group" + id + "CoreCount", Configuration.IntParameter.class)
                .getValue();/*w w w.j  a va2 s.  c  om*/
        Configuration.PermutationParameter<Integer> coreOrderParam = config.getParameter(
                "Group" + id + "CoreOrder", Configuration.PermutationParameter.class, Integer.class);
        coreOrders.add(coreOrderParam.getUniverse());
        int ourBiasCount = config.getParameter("Group" + id + "BiasCount", Configuration.IntParameter.class)
                .getValue();
        biasCount += Math.min(ourBiasCount, numCores - 1);
        bias += config.getParameter("Group" + id + "Bias", Configuration.FloatParameter.class).getValue();
    }
    numCores = IntMath.divide(numCores, group.actors().size(), RoundingMode.CEILING);
    biasCount = IntMath.divide(biasCount, group.actors().size(), RoundingMode.FLOOR);
    bias /= group.actors().size();
    //Transpose coreOrders.
    List<Integer> coreOrder = new ArrayList<>();
    for (int i = 0; i < coreOrders.get(0).size(); ++i)
        for (int j = 0; j < coreOrders.size(); ++j)
            coreOrder.add(coreOrders.get(j).get(i));
    //Remove duplicates preserving order.
    coreOrder = new ArrayList<>(new LinkedHashSet<>(coreOrder));

    List<Core> subset = new ArrayList<>(numCores);
    for (int i = 0; i < coreOrder.size() && subset.size() < numCores; ++i)
        if (coreOrder.get(i) < cores.size())
            subset.add(cores.get(coreOrder.get(i)));
    List<Core> biasSubset = new ArrayList<>(biasCount);
    while (biasSubset.size() < biasCount)
        biasSubset.add(subset.remove(0));

    float deficitFraction = biasCount * (1 - bias) / numCores, surplusFraction = 1 - deficitFraction;
    assert deficitFraction >= 0 && surplusFraction >= 0 : String.format("%d %d %f -> %f %f", numCores,
            biasCount, bias, deficitFraction, surplusFraction);
    iterations = iterations.canonical(DiscreteDomain.integers());
    int totalIterations = iterations.upperEndpoint() - iterations.lowerEndpoint();
    int biasIterations = (int) (totalIterations * deficitFraction);
    //We pass a null config to ensure we don't interfere with the other strategy.
    if (biasCount > 0)
        new FullDataParallelAllocationStrategy(biasCount).allocateGroup(group,
                Range.closedOpen(iterations.lowerEndpoint(), iterations.lowerEndpoint() + biasIterations),
                biasSubset, null);
    if (numCores - biasCount > 0)
        new FullDataParallelAllocationStrategy(numCores - biasCount).allocateGroup(group,
                Range.closedOpen(iterations.lowerEndpoint() + biasIterations, iterations.upperEndpoint()),
                subset, null);
}

---

• «
• 1
• 2
• »