org.ejml.EjmlParameters Maven / Gradle / Ivy

/*
 * Copyright (c) 2009-2017, 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 org.ejml;


/**
 * This is a list of parameters that are used across the code.  To tune performance
 * for a particular system change these values.
 *
 * @author Peter Abeles
 */
public class EjmlParameters {

    public static final float TOL32 = 1e-4f;
    public static final double TOL64 = 1e-8;


    /**
     * Used to adjust which algorithms are used.  Often there is a trade off between memory usage
     * and speed.
     */
    public static MemoryUsage MEMORY = MemoryUsage.FASTER;

    /**
     * 
     * In modern computers there are high speed memory caches.  It is assumed that a square
     * block with this width can be contained entirely in one of those caches.  Settings this
     * value too large can have a dramatic effect on performance in some situations.  Setting
     * it too low results in a less dramatic performance hit.  The optimal value is dependent
     * on the computer's memory architecture.
     * 
     */
    // See design notes
    public static int BLOCK_WIDTH = 60;
    public static int BLOCK_WIDTH_CHOL = 20;

    /**
     * Number of elements in a block.
     */
    public static int BLOCK_SIZE = BLOCK_WIDTH*BLOCK_WIDTH;

    public static int TRANSPOSE_SWITCH = 375;

    /**
     * At what point does it switch from a small matrix multiply to the reorder version.
     */
    public static int MULT_COLUMN_SWITCH = 15;
    public static int MULT_TRANAB_COLUMN_SWITCH = 40;
    public static int MULT_INNER_SWITCH = 100;

    public static int CMULT_COLUMN_SWITCH = 7;
    public static int CMULT_TRANAB_COLUMN_SWITCH = 20;
    
    /**
     * 
     * At which point should it switch to the block cholesky algorithm.
     * 
     * 
     * In benchmarks  the basic actually performed slightly better at 1000
     * but in JVM 1.6 it some times get stuck in a mode where the basic version was very slow
     * in that case the block performed much better.
     * 
     */
    public static int SWITCH_BLOCK64_CHOLESKY = 1000;

    public static int SWITCH_BLOCK64_QR = 1500;

    public static enum MemoryUsage
    {
        /**
         * Use lower memory algorithm while not totally sacrificing speed.
         */
        LOW_MEMORY,
        /**
         * Always favor faster algorithms even if they use more memory.
         */
        FASTER

    }
}