All Downloads are FREE. Search and download functionalities are using the official Maven repository.

pabeles.concurrency.ConcurrencyOps Maven / Gradle / Ivy

Go to download

A fast and easy to use dense and sparse matrix linear algebra library written in Java.

There is a newer version: 0.43.1
Show newest version
/*
 * Copyright (c) 2021, Peter Abeles. All Rights Reserved.
 *
 * This file is part of Efficient Java Matrix Library (EJML).
 *
 * Licensed 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 pabeles.concurrency;

import java.util.concurrent.ExecutionException;
import java.util.concurrent.ForkJoinPool;
import java.util.function.IntConsumer;
import java.util.stream.IntStream;

/**
 * Location of controls for turning on and off concurrent (i.e. threaded) algorithms.
 *
 * -Djava.util.concurrent.ForkJoinPool.common.parallelism=16
 *
 * @author Peter Abeles
 */
public class ConcurrencyOps {

    // Custom thread pool for streams so that the number of threads can be controlled
    private static ForkJoinPool pool = new ForkJoinPool();

    /**
     * Changes the maximum number of threads available in the thread pool
     *
     * @param maxThreads Maximum number of threads. If less than 1 it will be forced to be one
     */
    public static void setMaxThreads( int maxThreads ) {
        pool = new ForkJoinPool(Math.max(1, maxThreads));
    }

    /**
     * Returns the maximum number of threads which can be run at once in this pool
     */
    public static int getMaxThreads() {
        return pool.getParallelism();
    }

    /**
     * Concurrent for loop. Each loop with spawn as a thread up to the maximum number of threads.
     *
     * @param start starting value, inclusive
     * @param endExclusive ending value, exclusive
     * @param consumer The consumer
     */
    public static void loopFor( int start, int endExclusive, IntConsumer consumer ) {
        try {
            pool.submit(() -> IntStream.range(start, endExclusive).parallel().forEach(consumer)).get();
        } catch (InterruptedException | ExecutionException e) {
            e.printStackTrace();
        }
    }

    /**
     * Concurrent for loop. Each loop with spawn as a thread up to the maximum number of threads.
     *
     * @param start starting value, inclusive
     * @param endExclusive ending value, exclusive
     * @param step fixed sized step for each iteration
     * @param consumer The consumer
     */
    public static void loopFor( int start, int endExclusive, int step, IntConsumer consumer ) {
        if (step <= 0)
            throw new IllegalArgumentException("Step must be a positive number.");
        if (start >= endExclusive)
            return;
        try {
            int range = endExclusive - start;
            int iterations = range/step + ((range%step == 0) ? 0 : 1);
            pool.submit(() -> IntStream.range(0, iterations).parallel().forEach(i -> consumer.accept(start + i*step))).get();
        } catch (InterruptedException | ExecutionException e) {
            e.printStackTrace();
        }
    }

    /**
     * Concurrent for loop. Each loop with spawn as a thread up to the maximum number of threads.
     *
     * @param start starting value, inclusive
     * @param endExclusive ending value, exclusive
     * @param step fixed sized step for each iteration
     * @param consumer The consumer
     */
    public static 
    void loopFor( int start, int endExclusive, int step, GrowArray workspace, IntObjectConsumer consumer ) {
        if (step <= 0)
            throw new IllegalArgumentException("Step must be a positive number.");
        if (start >= endExclusive)
            return;
        try {
            pool.submit(new IntObjectTask<>(start, endExclusive, step, pool.getParallelism(), -1, workspace, consumer)).get();
        } catch (InterruptedException | ExecutionException e) {
            e.printStackTrace();
        }
    }

    /**
     * Automatically breaks the problem up into blocks based on the number of threads available. It is assumed
     * that there is some cost associated with processing a block and the number of blocks is minimized.
     *
     * Examples:
     * 
    *
  • Given a range of 0 to 100, and minBlock is 5, and 10 threads. Blocks will be size 10.
  • *
  • Given a range of 0 to 100, and minBlock is 20, and 10 threads. Blocks will be size 20.
  • *
  • Given a range of 0 to 100, and minBlock is 15, and 10 threads. Blocks will be size 16 and 20.
  • *
  • Given a range of 0 to 100, and minBlock is 80, and 10 threads. Blocks will be size 100.
  • *
* * @param start First index, inclusive * @param endExclusive Last index, exclusive * @param minBlock Minimum size of a block * @param consumer The consumer */ public static void loopBlocks( int start, int endExclusive, int minBlock, IntRangeConsumer consumer ) { final ForkJoinPool pool = ConcurrencyOps.pool; int numThreads = pool.getParallelism(); int range = endExclusive - start; if (range == 0) // nothing to do here! return; if (range < 0) throw new IllegalArgumentException("end must be more than start. " + start + " -> " + endExclusive); int block = selectBlockSize(range, minBlock, numThreads); try { pool.submit(new IntRangeTask(start, endExclusive, block, consumer)).get(); } catch (InterruptedException | ExecutionException e) { e.printStackTrace(); } } static int selectBlockSize( int range, int minBlock, int numThreads ) { // attempt to split the load between each thread equally int block = Math.max(minBlock, range/numThreads); // now attempt to make each block the same size int N = Math.max(1, range/block); return range/N; } /** * Splits the range of values up into blocks. It's assumed the cost to process a block is small so * more can be created. * * @param start First index, inclusive * @param endExclusive Last index, exclusive * @param consumer The consumer */ public static void loopBlocks( int start, int endExclusive, IntRangeConsumer consumer ) { final ForkJoinPool pool = ConcurrencyOps.pool; int numThreads = pool.getParallelism(); int range = endExclusive - start; if (range == 0) // nothing to do here! return; if (range < 0) throw new IllegalArgumentException("end must be more than start. " + start + " -> " + endExclusive); // Did some experimentation here. Gave it more threads than were needed or exactly what was needed // exactly seemed to do better in the test cases int blockSize = Math.max(1, range/numThreads); try { pool.submit(new IntRangeTask(start, endExclusive, blockSize, consumer)).get(); } catch (InterruptedException | ExecutionException e) { throw new RuntimeException(e); } } /** * Splits the range of values up into blocks. For each block workspace data will be declared using a * {@link GrowArray} and passed on. This workspace can be used to collect results and combine later on * * @param start First index, inclusive * @param endExclusive Last index, exclusive * @param consumer The consumer */ public static void loopBlocks( int start, int endExclusive, GrowArray workspace, IntRangeObjectConsumer consumer ) { final ForkJoinPool pool = ConcurrencyOps.pool; int numThreads = pool.getParallelism(); int range = endExclusive - start; if (range == 0) // nothing to do here! return; if (range < 0) throw new IllegalArgumentException("end must be more than start. " + start + " -> " + endExclusive); // Did some experimentation here. Gave it more threads than were needed or exactly what was needed // exactly seemed to do better in the test cases int blockSize = Math.max(1, range/numThreads); runLoopBlocks(start, endExclusive, workspace, consumer, pool, blockSize); } /** * Splits the range of values up into blocks. For each block workspace data will be declared using a * {@link GrowArray} and passed on. This workspace can be used to collect results and combine later on * * @param start First index, inclusive * @param endExclusive Last index, exclusive * @param minBlock Minimum size of a block * @param consumer The consumer */ public static void loopBlocks( int start, int endExclusive, int minBlock, GrowArray workspace, IntRangeObjectConsumer consumer ) { final ForkJoinPool pool = ConcurrencyOps.pool; int numThreads = pool.getParallelism(); int range = endExclusive - start; if (range == 0) // nothing to do here! return; if (range < 0) throw new IllegalArgumentException("end must be more than start. " + start + " -> " + endExclusive); int blockSize = selectBlockSize(range, minBlock, numThreads); runLoopBlocks(start, endExclusive, workspace, consumer, pool, blockSize); } private static void runLoopBlocks( int start, int endExclusive, GrowArray workspace, IntRangeObjectConsumer consumer, ForkJoinPool pool, int blockSize ) { workspace.reset(); try { pool.submit(new IntRangeObjectTask<>(start, endExclusive, blockSize, workspace, consumer)).get(); } catch (InterruptedException | ExecutionException e) { throw new RuntimeException(e); } } /** * Computes sums up the results using the specified primitive type * * @param start First index, inclusive * @param endExclusive Last index, exclusive * @param type Primtive data type, e.g. int.class, float.class, double.class * @param producer Given an integer input produce a Number output * @return The sum */ public static Number sum( int start, int endExclusive, Class type, IntProducerNumber producer ) { try { return pool.submit(new IntOperatorTask.Sum(start, endExclusive, type, producer)).get(); } catch (InterruptedException | ExecutionException e) { throw new RuntimeException(e); } } /** * Computes the maximum value * * @param start First index, inclusive * @param endExclusive Last index, exclusive * @param type Primtive data type, e.g. int.class, float.class, double.class * @param producer Given an integer input produce a Number output * @return The sum */ public static Number max( int start, int endExclusive, Class type, IntProducerNumber producer ) { try { return pool.submit(new IntOperatorTask.Max(start, endExclusive, type, producer)).get(); } catch (InterruptedException | ExecutionException e) { throw new RuntimeException(e); } } /** * Computes the maximum value * * @param start First index, inclusive * @param endExclusive Last index, exclusive * @param type Primtive data type, e.g. int.class, float.class, double.class * @param producer Given an integer input produce a Number output * @return The sum */ public static Number min( int start, int endExclusive, Class type, IntProducerNumber producer ) { try { return pool.submit(new IntOperatorTask.Min(start, endExclusive, type, producer)).get(); } catch (InterruptedException | ExecutionException e) { throw new RuntimeException(e); } } /** Returns the thread pool. */ public static ForkJoinPool getThreadPool() { return pool; } public interface NewInstance { D newInstance(); } public interface Reset { void reset( D data ); } }




© 2015 - 2024 Weber Informatics LLC | Privacy Policy