/*

   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.batik.ext.awt;

import java.awt.Color;
import java.awt.PaintContext;
import java.awt.Rectangle;
import java.awt.RenderingHints;
import java.awt.color.ColorSpace;
import java.awt.geom.AffineTransform;
import java.awt.geom.NoninvertibleTransformException;
import java.awt.geom.Rectangle2D;
import java.awt.image.ColorModel;
import java.awt.image.DataBuffer;
import java.awt.image.DataBufferInt;
import java.awt.image.DirectColorModel;
import java.awt.image.Raster;
import java.awt.image.SinglePixelPackedSampleModel;
import java.awt.image.WritableRaster;
import java.lang.ref.WeakReference;

import org.apache.batik.ext.awt.image.GraphicsUtil;

This is the superclass for all PaintContexts which use a multiple color
gradient to fill in their raster. It provides the actual color interpolation
functionality.  Subclasses only have to deal with using the gradient to fill
pixels in a raster.
Author:
Nicholas Talian, Vincent Hardy, Jim Graham, Jerry Evans, Vincent Hardy
Version:
$Id: MultipleGradientPaintContext.java 1804130 2017-08-04 14:41:11Z ssteiner $/**
 * This is the superclass for all PaintContexts which use a multiple color
 * gradient to fill in their raster. It provides the actual color interpolation
 * functionality.  Subclasses only have to deal with using the gradient to fill
 * pixels in a raster.
 *
 * @author Nicholas Talian, Vincent Hardy, Jim Graham, Jerry Evans
 * @author <a href="mailto:vincent.hardy@eng.sun.com">Vincent Hardy</a>
 * @version $Id: MultipleGradientPaintContext.java 1804130 2017-08-04 14:41:11Z ssteiner $
 */
abstract class MultipleGradientPaintContext implements PaintContext {

    protected static final boolean DEBUG = false;

    
The color model data is generated in (always un premult).
/**
     * The color model data is generated in (always un premult).
     */
    protected ColorModel dataModel;
    
PaintContext's output ColorModel ARGB if colors are not all
opaque, RGB otherwise.  Linear and premult are matched to
output ColorModel.
/**
     * PaintContext's output ColorModel ARGB if colors are not all
     * opaque, RGB otherwise.  Linear and premult are matched to
     * output ColorModel.
     */
    protected ColorModel model;

    
Color model used if gradient colors are all opaque /** Color model used if gradient colors are all opaque */
    private static ColorModel lrgbmodel_NA = new DirectColorModel
        (ColorSpace.getInstance(ColorSpace.CS_LINEAR_RGB),
         24, 0xff0000, 0xFF00, 0xFF, 0x0,
         false, DataBuffer.TYPE_INT);

    private static ColorModel srgbmodel_NA = new DirectColorModel
        (ColorSpace.getInstance(ColorSpace.CS_sRGB),
         24, 0xff0000, 0xFF00, 0xFF, 0x0,
         false, DataBuffer.TYPE_INT);

    
Color model used if some gradient colors are transparent /** Color model used if some gradient colors are transparent */
    private static ColorModel lrgbmodel_A = new DirectColorModel
        (ColorSpace.getInstance(ColorSpace.CS_LINEAR_RGB),
         32, 0xff0000, 0xFF00, 0xFF, 0xFF000000,
         false, DataBuffer.TYPE_INT);

    private static ColorModel srgbmodel_A = new DirectColorModel
        (ColorSpace.getInstance(ColorSpace.CS_sRGB),
         32, 0xff0000, 0xFF00, 0xFF, 0xFF000000,
         false, DataBuffer.TYPE_INT);

     
The cached colorModel /** The cached colorModel */
    protected static ColorModel cachedModel;

    
The cached raster, which is reusable among instances /** The cached raster, which is reusable among instances */
    protected static WeakReference cached;

    
Raster is reused whenever possible /** Raster is reused whenever possible */
    protected WritableRaster saved;

    
The method to use when painting out of the gradient bounds. /** The method to use when painting out of the gradient bounds. */
    protected MultipleGradientPaint.CycleMethodEnum cycleMethod;

    
The colorSpace in which to perform the interpolation /** The colorSpace in which to perform the interpolation */
    protected MultipleGradientPaint.ColorSpaceEnum colorSpace;

    
Elements of the inverse transform matrix. /** Elements of the inverse transform matrix. */
    protected float a00, a01, a10, a11, a02, a12;

    
This boolean specifies wether we are in simple lookup mode, where an
input value between 0 and 1 may be used to directly index into a single
array of gradient colors.  If this boolean value is false, then we have
to use a 2-step process where we have to determine which gradient array
we fall into, then determine the index into that array.
/** This boolean specifies wether we are in simple lookup mode, where an
     * input value between 0 and 1 may be used to directly index into a single
     * array of gradient colors.  If this boolean value is false, then we have
     * to use a 2-step process where we have to determine which gradient array
     * we fall into, then determine the index into that array.
     */
    protected boolean isSimpleLookup = true;

    
This boolean indicates if the gradient appears to have sudden
 discontinuities in it, this may be because of multiple stops
 at the same location or use of the REPEATE mode.
/** This boolean indicates if the gradient appears to have sudden
     *  discontinuities in it, this may be because of multiple stops
     *  at the same location or use of the REPEATE mode.
     */
    protected boolean hasDiscontinuity = false;

    
Size of gradients array for scaling the 0-1 index when looking up
 colors the fast way.  /** Size of gradients array for scaling the 0-1 index when looking up
     *  colors the fast way.  */
    protected int fastGradientArraySize;

    
Array which contains the interpolated color values for each interval,
used by calculateSingleArrayGradient().  It is protected for possible
direct access by subclasses.
/**
     * Array which contains the interpolated color values for each interval,
     * used by calculateSingleArrayGradient().  It is protected for possible
     * direct access by subclasses.
     */
    protected int[] gradient;

    
Array of gradient arrays, one array for each interval.  Used by
 calculateMultipleArrayGradient().
/** Array of gradient arrays, one array for each interval.  Used by
     *  calculateMultipleArrayGradient().
     */
    protected int[][] gradients;

    
This holds the blend of all colors in the gradient.
 we use this at extreamly low resolutions to ensure we
 get a decent blend of the colors.
/** This holds the blend of all colors in the gradient.
     *  we use this at extreamly low resolutions to ensure we
     *  get a decent blend of the colors.
     */
    protected int gradientAverage;

    
This holds the color to use when we are off the bottom of the
gradient /** This holds the color to use when we are off the bottom of the
     * gradient */
    protected int gradientUnderflow;

    
This holds the color to use when we are off the top of the
gradient /** This holds the color to use when we are off the top of the
     * gradient */
    protected int gradientOverflow;

    
Length of the 2D slow lookup gradients array. /** Length of the 2D slow lookup gradients array. */
    protected int gradientsLength;

    
Normalized intervals array /** Normalized intervals array */
    protected float[] normalizedIntervals;

    
fractions array /** fractions array */
    protected float[] fractions;

    
Used to determine if gradient colors are all opaque /** Used to determine if gradient colors are all opaque */
    private int transparencyTest;

    
Colorspace conversion lookup tables /** Colorspace conversion lookup tables */
    private static final int[] SRGBtoLinearRGB = new int[256];
    private static final int[] LinearRGBtoSRGB = new int[256];

    //build the tables
    static{
        for (int k = 0; k < 256; k++) {
            SRGBtoLinearRGB[k] = convertSRGBtoLinearRGB(k);
            LinearRGBtoSRGB[k] = convertLinearRGBtoSRGB(k);
        }
    }

    
Constant number of max colors between any 2 arbitrary colors.
Used for creating and indexing gradients arrays.
/** Constant number of max colors between any 2 arbitrary colors.
     * Used for creating and indexing gradients arrays.
     */
    protected static final int GRADIENT_SIZE = 256;
    protected static final int GRADIENT_SIZE_INDEX = GRADIENT_SIZE -1;

    
Maximum length of the fast single-array.  If the estimated array size
is greater than this, switch over to the slow lookup method.
No particular reason for choosing this number, but it seems to provide
satisfactory performance for the common case (fast lookup).
/** Maximum length of the fast single-array.  If the estimated array size
     * is greater than this, switch over to the slow lookup method.
     * No particular reason for choosing this number, but it seems to provide
     * satisfactory performance for the common case (fast lookup).
     */
    private static final int MAX_GRADIENT_ARRAY_SIZE = 5000;

   
Constructor for superclass. Does some initialization, but leaves most
of the heavy-duty math for calculateGradient(), so the subclass may do
some other manipulation beforehand if necessary.  This is not possible
if this computation is done in the superclass constructor which always
gets called first.
/** Constructor for superclass. Does some initialization, but leaves most
    * of the heavy-duty math for calculateGradient(), so the subclass may do
    * some other manipulation beforehand if necessary.  This is not possible
    * if this computation is done in the superclass constructor which always
    * gets called first.
    **/
    protected MultipleGradientPaintContext(ColorModel cm,
                                        Rectangle deviceBounds,
                                        Rectangle2D userBounds,
                                        AffineTransform t,
                                        RenderingHints hints,
                                        float[] fractions,
                                        Color[] colors,
                                        MultipleGradientPaint.CycleMethodEnum
                                        cycleMethod,
                                        MultipleGradientPaint.ColorSpaceEnum
                                        colorSpace)
        throws NoninvertibleTransformException
    {
        //We have to deal with the cases where the 1st gradient stop is not
        //equal to 0 and/or the last gradient stop is not equal to 1.
        //In both cases, create a new point and replicate the previous
        //extreme point's color.

        boolean fixFirst = false;
        boolean fixLast = false;
        int len = fractions.length;

        //if the first gradient stop is not equal to zero, fix this condition
        if (fractions[0] != 0f) {
            fixFirst = true;
            len++;
        }

        //if the last gradient stop is not equal to one, fix this condition
        if (fractions[fractions.length - 1] != 1.0f) {
            fixLast = true;
            len++;
        }

        for (int i=0; i<fractions.length-1; i++)
            if (fractions[i] == fractions[i+1])
                len--;

        this.fractions      = new float[len];
        Color [] loColors   = new Color[len-1];
        Color [] hiColors   = new Color[len-1];
        normalizedIntervals = new float[len-1];

        gradientUnderflow = colors[0].getRGB();
        gradientOverflow  = colors[colors.length-1].getRGB();

        int idx = 0;
        if (fixFirst) {
            this.fractions[0] = 0;
            loColors[0] = colors[0];
            hiColors[0] = colors[0];
            normalizedIntervals[0] = fractions[0];
            idx++;
        }

        for (int i=0; i<fractions.length-1; i++) {
            if (fractions[i] == fractions[i+1]) {
                // System.out.println("EQ Fracts");
                if (!colors[i].equals(colors[i+1])) {
                    hasDiscontinuity = true;
                }
                continue;
            }
            this.fractions[idx] = fractions[i];
            loColors[idx] = colors[i];
            hiColors[idx] = colors[i+1];
            normalizedIntervals[idx] = fractions[i+1]-fractions[i];
            idx++;
        }

        this.fractions[idx] = fractions[fractions.length-1];

        if (fixLast) {
            loColors[idx] = hiColors[idx] = colors[colors.length-1];
            normalizedIntervals[idx] = 1-fractions[fractions.length-1];
            idx++;
            this.fractions[idx] = 1;
        }

        // The inverse transform is needed to from device to user space.
        // Get all the components of the inverse transform matrix.
        AffineTransform tInv = t.createInverse();

        double[] m = new double[6];
        tInv.getMatrix(m);
        a00 = (float)m[0];
        a10 = (float)m[1];
        a01 = (float)m[2];
        a11 = (float)m[3];
        a02 = (float)m[4];
        a12 = (float)m[5];

        //copy some flags
        this.cycleMethod = cycleMethod;
        this.colorSpace = colorSpace;

        // Setup an example Model, we may refine it later.
        if (cm.getColorSpace() == lrgbmodel_A.getColorSpace())
            dataModel = lrgbmodel_A;
        else if (cm.getColorSpace() == srgbmodel_A.getColorSpace())
            dataModel = srgbmodel_A;
        else
            throw new IllegalArgumentException
                ("Unsupported ColorSpace for interpolation");

        calculateGradientFractions(loColors, hiColors);

        model = GraphicsUtil.coerceColorModel(dataModel,
                                              cm.isAlphaPremultiplied());
    }


    
This function is the meat of this class.  It calculates an array of
gradient colors based on an array of fractions and color values at those
fractions.
/** This function is the meat of this class.  It calculates an array of
     * gradient colors based on an array of fractions and color values at those
     * fractions.
     */
    protected final void calculateGradientFractions
        (Color []loColors, Color []hiColors) {

        //if interpolation should occur in Linear RGB space, convert the
        //colors using the lookup table
        if (colorSpace == LinearGradientPaint.LINEAR_RGB) {
            int[] workTbl = SRGBtoLinearRGB; // local is cheaper

            for (int i = 0; i < loColors.length; i++) {

                loColors[i] = interpolateColor( workTbl, loColors[ i ] );

                hiColors[i] = interpolateColor( workTbl, hiColors[ i ] );

            }
        }

        //initialize to be fully opaque for ANDing with colors
        transparencyTest = 0xff000000;
        if (cycleMethod == MultipleGradientPaint.NO_CYCLE) {
            // Include overflow and underflow colors in transparency
            // test.
            transparencyTest &= gradientUnderflow;
            transparencyTest &= gradientOverflow;
        }

        //array of interpolation arrays
        gradients = new int[fractions.length - 1][];
        gradientsLength = gradients.length;

        // TODO ??? whats going on here
        // ??? the following comments and the name Imin suggest, that we search for something small
        // ??? but the for-loop actually looks for the LARGEST value

        // Find smallest interval
        int n = normalizedIntervals.length;

        float Imin = 1;
        float[] workTbl = normalizedIntervals;   // local is cheaper
        for(int i = 0; i < n; i++) {
            // ??? find the LARGEST value in normalizedIntervals
            Imin = (Imin > workTbl[i]) ? workTbl[i] : Imin;
        }

        //estimate the size of the entire gradients array.
        //This is to prevent a tiny interval from causing the size of array to
        //explode.  If the estimated size is too large, break to using
        //seperate arrays for each interval, and using an indexing scheme at
        //look-up time.
        int estimatedSize = 0;

        if (Imin == 0) {
            estimatedSize = Integer.MAX_VALUE;
            hasDiscontinuity = true;
        } else {
            for (float aWorkTbl : workTbl) {
                estimatedSize += (aWorkTbl / Imin) * GRADIENT_SIZE;
            }
        }


        if (estimatedSize > MAX_GRADIENT_ARRAY_SIZE) {
            //slow method
            calculateMultipleArrayGradient(loColors, hiColors);
            if ((cycleMethod == MultipleGradientPaint.REPEAT) &&
                (gradients[0][0] !=
                 gradients[gradients.length-1][GRADIENT_SIZE_INDEX]))
                hasDiscontinuity = true;
        } else {
            //fast method
            calculateSingleArrayGradient(loColors, hiColors, Imin);
            if ((cycleMethod == MultipleGradientPaint.REPEAT) &&
                (gradient[0] != gradient[fastGradientArraySize]))
                hasDiscontinuity = true;
        }

        // Use the most 'economical' model (no alpha).
        if((transparencyTest >>> 24) == 0xff) {
            if (dataModel.getColorSpace() == lrgbmodel_NA.getColorSpace())
                dataModel = lrgbmodel_NA;
            else if (dataModel.getColorSpace() == srgbmodel_NA.getColorSpace())
                dataModel = srgbmodel_NA;
            model = dataModel;
        }
    }

    
We assume, that we always generate valid colors. When this is valid, we can compose the
color-value by ourselves and use the faster Color-ctor, which does not check the incoming values.
Params:
workTbl – typically SRGBtoLinearRGB
inColor – the color to interpolate
Returns:
the interpolated color/**
     * We assume, that we always generate valid colors. When this is valid, we can compose the
     * color-value by ourselves and use the faster Color-ctor, which does not check the incoming values.
     *
     * @param workTbl typically SRGBtoLinearRGB
     * @param inColor the color to interpolate
     * @return the interpolated color
     */
    private static Color interpolateColor( int[] workTbl, Color inColor ) {

        int oldColor = inColor.getRGB();

        int newColorValue =
                  (( workTbl[ (oldColor >> 24 ) & 0xff ] & 0xff ) << 24 ) |
                  (( workTbl[ (oldColor >> 16 ) & 0xff ] & 0xff ) << 16 ) |
                  (( workTbl[ (oldColor >>  8 ) & 0xff ] & 0xff ) <<  8 ) |
                  (( workTbl[ (oldColor       ) & 0xff ] & 0xff ));

        return new Color( newColorValue, true );
    }

    
FAST LOOKUP METHOD
This method calculates the gradient color values and places them in a
single int array, gradient[].  It does this by allocating space for
each interval based on its size relative to the smallest interval in
the array.  The smallest interval is allocated 255 interpolated values
(the maximum number of unique in-between colors in a 24 bit color
system), and all other intervals are allocated
size = (255 * the ratio of their size to the smallest interval).
This scheme expedites a speedy retrieval because the colors are
distributed along the array according to their user-specified
distribution.  All that is needed is a relative index from 0 to 1.
The only problem with this method is that the possibility exists for
the array size to balloon in the case where there is a
disproportionately small gradient interval.  In this case the other
intervals will be allocated huge space, but much of that data is
redundant.  We thus need to use the space conserving scheme below.
Params:
Imin – the size of the smallest interval/**
     * FAST LOOKUP METHOD
     *
     * This method calculates the gradient color values and places them in a
     * single int array, gradient[].  It does this by allocating space for
     * each interval based on its size relative to the smallest interval in
     * the array.  The smallest interval is allocated 255 interpolated values
     * (the maximum number of unique in-between colors in a 24 bit color
     * system), and all other intervals are allocated
     * size = (255 * the ratio of their size to the smallest interval).
     *
     * This scheme expedites a speedy retrieval because the colors are
     * distributed along the array according to their user-specified
     * distribution.  All that is needed is a relative index from 0 to 1.
     *
     * The only problem with this method is that the possibility exists for
     * the array size to balloon in the case where there is a
     * disproportionately small gradient interval.  In this case the other
     * intervals will be allocated huge space, but much of that data is
     * redundant.  We thus need to use the space conserving scheme below.
     *
     * @param Imin the size of the smallest interval
     *
     */
    private void calculateSingleArrayGradient
        (Color [] loColors, Color [] hiColors, float Imin) {

        //set the flag so we know later it is a non-simple lookup
        isSimpleLookup = true;

        int gradientsTot = 1; //the eventual size of the single array

        // These are fixed point 8.16 (start with 0.5)
        int aveA = 0x008000;
        int aveR = 0x008000;
        int aveG = 0x008000;
        int aveB = 0x008000;

        //for every interval (transition between 2 colors)
        for(int i=0; i < gradients.length; i++){

            //create an array whose size is based on the ratio to the
            //smallest interval.
            int nGradients = (int)((normalizedIntervals[i]/Imin)*255f);
            gradientsTot += nGradients;
            gradients[i] = new int[nGradients];

            //the the 2 colors (keyframes) to interpolate between
            int rgb1 = loColors[i].getRGB();
            int rgb2 = hiColors[i].getRGB();

            //fill this array with the colors in between rgb1 and rgb2
            interpolate(rgb1, rgb2, gradients[i]);

            // Calculate Average of two colors...
            int argb = gradients[i][GRADIENT_SIZE/2];
            float norm = normalizedIntervals[i];
            aveA += (int)(((argb>> 8)&0xFF0000)*norm);
            aveR += (int)(((argb    )&0xFF0000)*norm);
            aveG += (int)(((argb<< 8)&0xFF0000)*norm);
            aveB += (int)(((argb<<16)&0xFF0000)*norm);

            //if the colors are opaque, transparency should still be 0xff000000
            transparencyTest &=  rgb1 & rgb2;
        }

        gradientAverage = (((aveA & 0xFF0000)<< 8) |
                           ((aveR & 0xFF0000)    ) |
                           ((aveG & 0xFF0000)>> 8) |
                           ((aveB & 0xFF0000)>>16));

        // Put all gradients in a single array
        gradient = new int[gradientsTot];
        int curOffset = 0;
        for (int[] gradient1 : gradients) {
            System.arraycopy(gradient1, 0, gradient,
                    curOffset, gradient1.length);
            curOffset += gradient1.length;
        }
        gradient[gradient.length-1] = hiColors[hiColors.length-1].getRGB();

        //if interpolation occurred in Linear RGB space, convert the
        //gradients back to SRGB using the lookup table
        if (colorSpace == LinearGradientPaint.LINEAR_RGB) {
            if (dataModel.getColorSpace() ==
                ColorSpace.getInstance(ColorSpace.CS_sRGB)) {
                for (int i = 0; i < gradient.length; i++) {
                    gradient[i] =
                        convertEntireColorLinearRGBtoSRGB(gradient[i]);
                }
                gradientAverage =
                    convertEntireColorLinearRGBtoSRGB(gradientAverage);
            }
        } else {
            if (dataModel.getColorSpace() ==
                ColorSpace.getInstance(ColorSpace.CS_LINEAR_RGB)) {
                for (int i = 0; i < gradient.length; i++) {
                    gradient[i] =
                        convertEntireColorSRGBtoLinearRGB(gradient[i]);
                }
                gradientAverage =
                    convertEntireColorSRGBtoLinearRGB(gradientAverage);
            }
        }

        fastGradientArraySize = gradient.length - 1;
    }


    
SLOW LOOKUP METHOD
This method calculates the gradient color values for each interval and
places each into its own 255 size array.  The arrays are stored in
gradients[][].  (255 is used because this is the maximum number of
unique colors between 2 arbitrary colors in a 24 bit color system)
This method uses the minimum amount of space (only 255 * number of
intervals), but it aggravates the lookup procedure, because now we
have to find out which interval to select, then calculate the index
within that interval.  This causes a significant performance hit,
because it requires this calculation be done for every point in
the rendering loop.
For those of you who are interested, this is a classic example of the
time-space tradeoff.
/**
     * SLOW LOOKUP METHOD
     *
     * This method calculates the gradient color values for each interval and
     * places each into its own 255 size array.  The arrays are stored in
     * gradients[][].  (255 is used because this is the maximum number of
     * unique colors between 2 arbitrary colors in a 24 bit color system)
     *
     * This method uses the minimum amount of space (only 255 * number of
     * intervals), but it aggravates the lookup procedure, because now we
     * have to find out which interval to select, then calculate the index
     * within that interval.  This causes a significant performance hit,
     * because it requires this calculation be done for every point in
     * the rendering loop.
     *
     * For those of you who are interested, this is a classic example of the
     * time-space tradeoff.
     *
     */
    private void calculateMultipleArrayGradient
        (Color [] loColors, Color [] hiColors) {

        //set the flag so we know later it is a non-simple lookup
        isSimpleLookup = false;

        int rgb1; //2 colors to interpolate
        int rgb2;

        // These are fixed point 8.16 (start with 0.5)
        int aveA = 0x008000;
        int aveR = 0x008000;
        int aveG = 0x008000;
        int aveB = 0x008000;

        //for every interval (transition between 2 colors)
        for(int i=0; i < gradients.length; i++){

            // This interval will never actually be used (zero size)
            if (normalizedIntervals[i] == 0)
                continue;

            //create an array of the maximum theoretical size for each interval
            gradients[i] = new int[GRADIENT_SIZE];

            //get the the 2 colors
            rgb1 = loColors[i].getRGB();
            rgb2 = hiColors[i].getRGB();

            //fill this array with the colors in between rgb1 and rgb2
            interpolate(rgb1, rgb2, gradients[i]);

            // Calculate Average of two colors...
            int argb = gradients[i][GRADIENT_SIZE/2];
            float norm = normalizedIntervals[i];
            aveA += (int)(((argb>> 8)&0xFF0000)*norm);
            aveR += (int)(((argb    )&0xFF0000)*norm);
            aveG += (int)(((argb<< 8)&0xFF0000)*norm);
            aveB += (int)(((argb<<16)&0xFF0000)*norm);

            //if the colors are opaque, transparency should still be 0xff000000
            transparencyTest &= rgb1;
            transparencyTest &= rgb2;
        }

        gradientAverage = (((aveA & 0xFF0000)<< 8) |
                           ((aveR & 0xFF0000)    ) |
                           ((aveG & 0xFF0000)>> 8) |
                           ((aveB & 0xFF0000)>>16));

        //if interpolation occurred in Linear RGB space, convert the
        //gradients back to SRGB using the lookup table
        if (colorSpace == LinearGradientPaint.LINEAR_RGB) {
            if (dataModel.getColorSpace() ==
                ColorSpace.getInstance(ColorSpace.CS_sRGB)) {
                for (int j = 0; j < gradients.length; j++) {
                    for (int i = 0; i < gradients[j].length; i++) {
                        gradients[j][i] =
                            convertEntireColorLinearRGBtoSRGB(gradients[j][i]);
                    }
                }
                gradientAverage =
                    convertEntireColorLinearRGBtoSRGB(gradientAverage);
            }
        } else {
            if (dataModel.getColorSpace() ==
                ColorSpace.getInstance(ColorSpace.CS_LINEAR_RGB)) {
                for (int j = 0; j < gradients.length; j++) {
                    for (int i = 0; i < gradients[j].length; i++) {
                        gradients[j][i] =
                            convertEntireColorSRGBtoLinearRGB(gradients[j][i]);
                    }
                }
                gradientAverage =
                    convertEntireColorSRGBtoLinearRGB(gradientAverage);
            }
        }
    }

    
Yet another helper function.  This one linearly interpolates between
2 colors, filling up the output array.
Params:
rgb1 – the start color
rgb2 – the end color
output – the output array of colors... assuming this is not null or length 0./** Yet another helper function.  This one linearly interpolates between
     * 2 colors, filling up the output array.
     *
     * @param rgb1 the start color
     * @param rgb2 the end color
     * @param output the output array of colors... assuming this is not null or length 0.
     */
    private void interpolate(int rgb1, int rgb2, int[] output) {

        int nSteps = output.length;

        //step between interpolated values.
        float stepSize = 1/(float)nSteps;

        //extract color components from packed integer
        int a1 = (rgb1 >> 24) & 0xff;
        int r1 = (rgb1 >> 16) & 0xff;
        int g1 = (rgb1 >>  8) & 0xff;
        int b1 = (rgb1      ) & 0xff;
        // calculate the total change in alpha, red, green, blue
        // the deltas can be negative !
        int da = ((rgb2 >> 24) & 0xff) - a1;
        int dr = ((rgb2 >> 16) & 0xff) - r1;
        int dg = ((rgb2 >>  8) & 0xff) - g1;
        int db = ((rgb2      ) & 0xff) - b1;

        // this method is a hotspot so we try to save some cycles
        // pre-compute some intermediate values.
        // the multiplication by 2 is used to help with rounding.
        float tempA = 2.0f * da * stepSize;
        float tempR = 2.0f * dr * stepSize;
        float tempG = 2.0f * dg * stepSize;
        float tempB = 2.0f * db * stepSize;

        //for each step in the interval calculate the in-between color by
        //multiplying the normalized current position by the total color change
        //(.5 is added to prevent truncation round-off error)

        // the previous implementation used a simple +0.5d to do some rounding.
        // but that is just rounding towards +inifitity. This results in
        // slightly different values (thus gradients) when you interpolate from
        //    color1 -> color2
        // versus
        //    color1 <- color2
        //
        // this implementation uses an implied multiplication by 2 ( in tempX )
        // and then a signed right-shift to do signed rounding.
        // this also spares a float-add per color-band.
        // we could also save the shift when we use a different and-mask and a different left-shift,
        // but that would obfuscate too much...
        //
        output[ 0 ] = rgb1;             // the start-color is fixed
        nSteps--;                       // upto, but not including the last slot
        output[ nSteps ] = rgb2;        // the last color is also fixed
        for (int i = 1; i < nSteps; i++) {
            float fI = i;
            output[i] =
                (( a1 + ((((int) ( fI * tempA )) +1) >> 1 ) & 0xff ) << 24) |
                (( r1 + ((((int) ( fI * tempR )) +1) >> 1 ) & 0xff ) << 16) |
                (( g1 + ((((int) ( fI * tempG )) +1) >> 1 ) & 0xff ) <<  8) |
                (( b1 + ((((int) ( fI * tempB )) +1) >> 1 ) & 0xff )      );
        }

    }


    
Yet another helper function.  This one extracts the color components
of an integer RGB triple, converts them from LinearRGB to SRGB, then
recompacts them into an int.
/** Yet another helper function.  This one extracts the color components
     * of an integer RGB triple, converts them from LinearRGB to SRGB, then
     * recompacts them into an int.
     */
    private static int convertEntireColorLinearRGBtoSRGB(int rgb) {

        //extract red, green, blue components
        int a1 = (rgb >> 24) & 0xff;
        int r1 = (rgb >> 16) & 0xff;
        int g1 = (rgb >>  8) & 0xff;
        int b1 =  rgb        & 0xff;

        //use the lookup table
        int[] workTbl = LinearRGBtoSRGB; // local is cheaper
        r1 =  workTbl[r1];
        g1 =  workTbl[g1];
        b1 =  workTbl[b1];

        //re-compact the components
        return ((a1 << 24) |
                (r1 << 16) |
                (g1 << 8) |
                b1);
    }

    
Yet another helper function.  This one extracts the color components
of an integer RGB triple, converts them from LinearRGB to SRGB, then
recompacts them into an int.
/** Yet another helper function.  This one extracts the color components
     * of an integer RGB triple, converts them from LinearRGB to SRGB, then
     * recompacts them into an int.
     */
    private static int convertEntireColorSRGBtoLinearRGB(int rgb) {

        //extract red, green, blue components
        int a1 = (rgb >> 24) & 0xff;
        int r1 = (rgb >> 16) & 0xff;
        int g1 = (rgb >>  8) & 0xff;
        int b1 =  rgb        & 0xff;

        //use the lookup table
        int[] workTbl = SRGBtoLinearRGB; // local is cheaper
        r1 =  workTbl[r1];
        g1 =  workTbl[g1];
        b1 =  workTbl[b1];

        //re-compact the components
        return ((a1 << 24) |
                (r1 << 16) |
                (g1 << 8) |
                b1);
    }


    
Helper function to index into the gradients array.  This is necessary
because each interval has an array of colors with uniform size 255.
However, the color intervals are not necessarily of uniform length, so
a conversion is required.
Params:
position – the unmanipulated position.  want to map this into the
range 0 to 1
Returns:
integer color to display/** Helper function to index into the gradients array.  This is necessary
     * because each interval has an array of colors with uniform size 255.
     * However, the color intervals are not necessarily of uniform length, so
     * a conversion is required.
     *
     * @param position the unmanipulated position.  want to map this into the
     * range 0 to 1
     *
     * @return integer color to display
     *
     */
    protected final int indexIntoGradientsArrays(float position) {

        //first, manipulate position value depending on the cycle method.

        if (cycleMethod == MultipleGradientPaint.NO_CYCLE) {

            if (position >= 1) { //upper bound is 1
                return gradientOverflow;
            }

            else if (position <= 0) { //lower bound is 0
                return gradientUnderflow;
            }
        }

        else if (cycleMethod == MultipleGradientPaint.REPEAT) {
            //get the fractional part
            //(modulo behavior discards integer component)
            position = position - (int)position;

            //position now be between -1 and 1

            if (position < 0) {
                position = position + 1; //force it to be in the range 0-1
            }

            int w=0, c1=0, c2=0;
            if (isSimpleLookup) {
              position *= gradient.length;
              int idx1 = (int)(position);
              if (idx1+1 < gradient.length)
                return gradient[idx1];

              w = (int)((position-idx1)*(1<<16));
              c1 = gradient[idx1];
              c2 = gradient[0];
            } else {
              //for all the gradient interval arrays
              for (int i = 0; i < gradientsLength; i++) {

                if (position < fractions[i+1]) { //this is the array we want

                  float delta = position - fractions[i];

                  delta = ((delta / normalizedIntervals[i]) * GRADIENT_SIZE);
                  //this is the interval we want.
                  int index = (int)delta;
                  if ((index+1<gradients[i].length) ||
                      (i+1 < gradientsLength))
                    return gradients[i][index];

                  w  = (int)((delta-index)*(1<<16));
                  c1 = gradients[i][index];
                  c2 = gradients[0][0];
                  break;
                }
              }
            }

            return
              ((((  (  (c1>>  8)           &0xFF0000)+
                    ((((c2>>>24)     )-((c1>>>24)     ))*w))&0xFF0000)<< 8) |

               (((  (  (c1     )           &0xFF0000)+
                    ((((c2>> 16)&0xFF)-((c1>> 16)&0xFF))*w))&0xFF0000)    ) |

               (((  (  (c1<<  8)           &0xFF0000)+
                    ((((c2>>  8)&0xFF)-((c1>>  8)&0xFF))*w))&0xFF0000)>> 8) |

               (((  (  (c1<< 16)           &0xFF0000)+
                    ((((c2     )&0xFF)-((c1     )&0xFF))*w))&0xFF0000)>>16));

            // return c1 +
            //   ((( ((((c2>>>24)     )-((c1>>>24)     ))*w)&0xFF0000)<< 8) |
            //    (( ((((c2>> 16)&0xFF)-((c1>> 16)&0xFF))*w)&0xFF0000)    ) |
            //    (( ((((c2>>  8)&0xFF)-((c1>>  8)&0xFF))*w)&0xFF0000)>> 8) |
            //    (( ((((c2     )&0xFF)-((c1     )&0xFF))*w)&0xFF0000)>>16));
        }

        else {  //cycleMethod == MultipleGradientPaint.REFLECT

            if (position < 0) {
                position = -position; //take absolute value
            }

            int part = (int)position; //take the integer part

            position = position - part; //get the fractional part

            if ((part & 0x00000001) == 1) { //if integer part is odd
                position = 1 - position; //want the reflected color instead
            }
        }

        //now, get the color based on this 0-1 position:

        if (isSimpleLookup) { //easy to compute: just scale index by array size
            return gradient[(int)(position * fastGradientArraySize)];
        }

        else { //more complicated computation, to save space

            //for all the gradient interval arrays
            for (int i = 0; i < gradientsLength; i++) {

                if (position < fractions[i+1]) { //this is the array we want

                    float delta = position - fractions[i];

                    //this is the interval we want.
                    int index = (int)((delta / normalizedIntervals[i])
                                      * (GRADIENT_SIZE_INDEX));

                    return gradients[i][index];
                }
            }

        }

        return gradientOverflow;
    }


    
Helper function to index into the gradients array.  This is necessary
because each interval has an array of colors with uniform size 255.
However, the color intervals are not necessarily of uniform length, so
a conversion is required.  This version also does anti-aliasing by
averaging the gradient over position+/-(sz/2).
Params:
position – the unmanipulated position.  want to map this into the
range 0 to 1
sz – the size in gradient space to average.
Returns:
ARGB integer color to display/** Helper function to index into the gradients array.  This is necessary
     * because each interval has an array of colors with uniform size 255.
     * However, the color intervals are not necessarily of uniform length, so
     * a conversion is required.  This version also does anti-aliasing by
     * averaging the gradient over position+/-(sz/2).
     *
     * @param position the unmanipulated position.  want to map this into the
     * range 0 to 1
     * @param sz the size in gradient space to average.
     *
     * @return ARGB integer color to display
     */
    protected final int indexGradientAntiAlias(float position, float sz) {
        //first, manipulate position value depending on the cycle method.
        if (cycleMethod == MultipleGradientPaint.NO_CYCLE) {
            if (DEBUG) System.out.println("NO_CYCLE");
            float p1 = position-(sz/2);
            float p2 = position+(sz/2);

            if (p1 >= 1)
                return gradientOverflow;

            if (p2 <= 0)
                return gradientUnderflow;

            int interior;
            float top_weight=0, bottom_weight=0, frac;
            if (p2 >= 1) {
                top_weight = (p2-1)/sz;
                if (p1 <= 0) {
                    bottom_weight = -p1/sz;
                    frac=1;
                    interior = gradientAverage;
                } else {
                    frac=1-p1;
                    interior = getAntiAlias(p1, true, 1, false, 1-p1, 1);
                }
            } else if (p1 <= 0) {
                bottom_weight = -p1/sz;
                frac = p2;
                interior = getAntiAlias(0, true, p2, false, p2, 1);
            } else
                return getAntiAlias(p1, true, p2, false, sz, 1);

            int norm = (int)((1<<16)*frac/sz);
            int pA = (((interior>>>20)&0xFF0)*norm)>>16;
            int pR = (((interior>> 12)&0xFF0)*norm)>>16;
            int pG = (((interior>>  4)&0xFF0)*norm)>>16;
            int pB = (((interior<<  4)&0xFF0)*norm)>>16;

            if (bottom_weight != 0) {
                int bPix = gradientUnderflow;
                // System.out.println("ave: " + gradientAverage);
                norm = (int)((1<<16)*bottom_weight);
                pA += (((bPix>>>20) & 0xFF0)*norm)>>16;
                pR += (((bPix>> 12) & 0xFF0)*norm)>>16;
                pG += (((bPix>>  4) & 0xFF0)*norm)>>16;
                pB += (((bPix<<  4) & 0xFF0)*norm)>>16;
            }

            if (top_weight != 0) {
                int tPix = gradientOverflow;

                norm = (int)((1<<16)*top_weight);
                pA += (((tPix>>>20) & 0xFF0)*norm)>>16;
                pR += (((tPix>> 12) & 0xFF0)*norm)>>16;
                pG += (((tPix>>  4) & 0xFF0)*norm)>>16;
                pB += (((tPix<<  4) & 0xFF0)*norm)>>16;
            }

            return (((pA&0xFF0)<<20)  |
                    ((pR&0xFF0)<<12)  |
                    ((pG&0xFF0)<< 4)  |
                    ((pB&0xFF0)>> 4));
        }

        // See how many times we are going to "wrap around" the gradient,
        // array.
        int intSz = (int)sz;

        float weight = 1.0f;
        if (intSz != 0) {
            // We need to make sure that sz is < 1.0 otherwise
            // p1 and p2 my pass each other which will cause no end of
            // trouble.
            sz -= intSz;
            weight = sz/(intSz+sz);
            if (weight < 0.1)
                // The part of the color from the location will be swamped
                // by the averaged part of the gradient so just use the
                // average color for the gradient.
                return gradientAverage;
        }

        // So close to full gradient just use the average value...
        if (sz > 0.99)
            return gradientAverage;

            // Go up and down from position by 1/2 sz.
        float p1 = position-(sz/2);
        float p2 = position+(sz/2);
        if (DEBUG) System.out.println("P1: " + p1 + " P2: " + p2);

        // These indicate the direction to go from p1 and p2 when
        // averaging...
        boolean p1_up=true;
        boolean p2_up=false;

        if (cycleMethod == MultipleGradientPaint.REPEAT) {
            if (DEBUG) System.out.println("REPEAT");

            // Get positions between -1 and 1
            p1=p1-(int)p1;
            p2=p2-(int)p2;

            // force to be in rage 0-1.
            if (p1 <0) p1 += 1;
            if (p2 <0) p2 += 1;
        }

        else {  //cycleMethod == MultipleGradientPaint.REFLECT
            if (DEBUG) System.out.println("REFLECT");

            //take absolute values
            // Note when we reflect we change sense of p1/2_up.
            if (p2 < 0) {
                p1 = -p1; p1_up = !p1_up;
                p2 = -p2; p2_up = !p2_up;
            } else if (p1 < 0) {
                p1 = -p1; p1_up = !p1_up;
            }

            int part1, part2;
            part1 = (int)p1;   // take the integer part
            p1   = p1 - part1; // get the fractional part

            part2 = (int)p2;   // take the integer part
            p2   = p2 - part2; // get the fractional part

            // if integer part is odd we want the reflected color instead.
            // Note when we reflect we change sense of p1/2_up.
            if ((part1 & 0x01) == 1) {
                p1 = 1-p1;
                p1_up = !p1_up;
            }

            if ((part2 & 0x01) == 1) {
                p2 = 1-p2;
                p2_up = !p2_up;
            }

            // Check if in the end they just got switched around.
            // this commonly happens if they both end up negative.
            if ((p1 > p2) && !p1_up && p2_up) {
                float t = p1;
                p1 = p2;
                p2 = t;
                p1_up = true;
                p2_up = false;
            }
        }

        return getAntiAlias(p1, p1_up, p2, p2_up, sz, weight);
    }


    private final int getAntiAlias(float p1, boolean p1_up,
                                   float p2, boolean p2_up,
                                   float sz, float weight) {

        // Until the last set of ops these are 28.4 fixed point values.
        int ach=0, rch=0, gch=0, bch=0;
        if (isSimpleLookup) {
            p1 *= fastGradientArraySize;
            p2 *= fastGradientArraySize;

            int idx1 = (int)p1;
            int idx2 = (int)p2;

            int i, pix;

            if (p1_up && !p2_up && (idx1 <= idx2)) {

                if (idx1 == idx2)
                    return gradient[idx1];

                // Sum between idx1 and idx2.
                for (i=idx1+1; i<idx2; i++) {
                    pix  = gradient[i];
                    ach += ((pix>>>20)&0xFF0);
                    rch += ((pix>>>12)&0xFF0);
                    gch += ((pix>>> 4)&0xFF0);
                    bch += ((pix<<  4)&0xFF0);
                }
            } else {
                // Do the bulk of the work, all the whole gradient entries
                // for idx1 and idx2.
                int iStart;
                int iEnd;
                if (p1_up) {
                    iStart = idx1+1;
                    iEnd = fastGradientArraySize;
                } else {
                    iStart = 0;
                    iEnd = idx1;
                }
                for ( i = iStart; i < iEnd; i++) {
                        pix  = gradient[i];
                        ach += ((pix>>>20)&0xFF0);
                        rch += ((pix>>>12)&0xFF0);
                        gch += ((pix>>> 4)&0xFF0);
                        bch += ((pix<<  4)&0xFF0);
                }

                if (p2_up) {
                    iStart = idx2 + 1;
                    iEnd = fastGradientArraySize;
                } else {
                    iStart = 0;
                    iEnd = idx2;
                }
                for (i= iStart; i < iEnd; i++) {
                        pix  = gradient[i];
                        ach += ((pix>>>20)&0xFF0);
                        rch += ((pix>>>12)&0xFF0);
                        gch += ((pix>>> 4)&0xFF0);
                        bch += ((pix<<  4)&0xFF0);
                }
            }

            int norm, isz;

            // Normalize the summation so far...
            isz = (int)((1<<16)/(sz*fastGradientArraySize));
            ach = (ach*isz)>>16;
            rch = (rch*isz)>>16;
            gch = (gch*isz)>>16;
            bch = (bch*isz)>>16;

            // Clean up with the partial buckets at each end.
            if (p1_up) norm = (int)((1-(p1-idx1))*isz);
            else       norm = (int)(   (p1-idx1) *isz);
            pix = gradient[idx1];
            ach += (((pix>>>20)&0xFF0) *norm)>>16;
            rch += (((pix>>>12)&0xFF0) *norm)>>16;
            gch += (((pix>>> 4)&0xFF0) *norm)>>16;
            bch += (((pix<<  4)&0xFF0) *norm)>>16;

            if (p2_up) norm = (int)((1-(p2-idx2))*isz);
            else       norm = (int)(   (p2-idx2) *isz);
            pix = gradient[idx2];
            ach += (((pix>>>20)&0xFF0) *norm)>>16;
            rch += (((pix>>>12)&0xFF0) *norm)>>16;
            gch += (((pix>>> 4)&0xFF0) *norm)>>16;
            bch += (((pix<<  4)&0xFF0) *norm)>>16;

            // Round and drop the 4bits frac.
            ach = (ach+0x08)>>4;
            rch = (rch+0x08)>>4;
            gch = (gch+0x08)>>4;
            bch = (bch+0x08)>>4;

        } else {
            int idx1=0, idx2=0;
            int i1=-1, i2=-1;
            float f1=0, f2=0;
            // Find which gradient interval our points fall into.
            for (int i = 0; i < gradientsLength; i++) {
                if ((p1 < fractions[i+1]) && (i1 == -1)) {
                    //this is the array we want
                    i1 = i;
                    f1 = p1 - fractions[i];

                    f1 = ((f1/normalizedIntervals[i])
                             *GRADIENT_SIZE_INDEX);
                    //this is the  interval we want.
                    idx1 = (int)f1;
                    if (i2 != -1) break;
                }
                if ((p2 < fractions[i+1]) && (i2 == -1)) {
                    //this is the array we want
                    i2 = i;
                    f2 = p2 - fractions[i];

                    f2 = ((f2/normalizedIntervals[i])
                             *GRADIENT_SIZE_INDEX);
                    //this is the interval we want.
                    idx2 = (int)f2;
                    if (i1 != -1) break;
                }
            }

            if (i1 == -1) {
                i1 = gradients.length - 1;
                f1 = idx1 = GRADIENT_SIZE_INDEX;
            }

            if (i2 == -1) {
                i2 = gradients.length - 1;
                f2 = idx2 = GRADIENT_SIZE_INDEX;
            }

            if (DEBUG) System.out.println("I1: " + i1 + " Idx1: " + idx1 +
                                          " I2: " + i2 + " Idx2: " + idx2);

            // Simple case within one gradient array (so the average
            // of the two idx gives us the true average of colors).
            if ((i1 == i2) && (idx1 <= idx2) && p1_up && !p2_up)
                return gradients[i1][(idx1+idx2+1)>>1];

            // i1 != i2

            int pix, norm;
            int base = (int)((1<<16)/sz);
            if ((i1 < i2) && p1_up && !p2_up) {
                norm = (int)((base
                              *normalizedIntervals[i1]
                              *(GRADIENT_SIZE_INDEX-f1))
                             /GRADIENT_SIZE_INDEX);
                pix  = gradients[i1][(idx1+GRADIENT_SIZE)>>1];
                ach += (((pix>>>20)&0xFF0) *norm)>>16;
                rch += (((pix>>>12)&0xFF0) *norm)>>16;
                gch += (((pix>>> 4)&0xFF0) *norm)>>16;
                bch += (((pix<<  4)&0xFF0) *norm)>>16;

                for (int i=i1+1; i<i2; i++) {
                    norm = (int)(base*normalizedIntervals[i]);
                    pix  = gradients[i][GRADIENT_SIZE>>1];

                    ach += (((pix>>>20)&0xFF0) *norm)>>16;
                    rch += (((pix>>>12)&0xFF0) *norm)>>16;
                    gch += (((pix>>> 4)&0xFF0) *norm)>>16;
                    bch += (((pix<<  4)&0xFF0) *norm)>>16;
                }

                norm = (int)((base*normalizedIntervals[i2]*f2)
                             /GRADIENT_SIZE_INDEX);
                pix  = gradients[i2][(idx2+1)>>1];
                ach += (((pix>>>20)&0xFF0) *norm)>>16;
                rch += (((pix>>>12)&0xFF0) *norm)>>16;
                gch += (((pix>>> 4)&0xFF0) *norm)>>16;
                bch += (((pix<<  4)&0xFF0) *norm)>>16;
            } else {
                if (p1_up) {
                    norm = (int)((base
                                  *normalizedIntervals[i1]
                                  *(GRADIENT_SIZE_INDEX-f1))
                                 /GRADIENT_SIZE_INDEX);
                    pix  = gradients[i1][(idx1+GRADIENT_SIZE)>>1];
                } else {
                    norm = (int)((base*normalizedIntervals[i1]*f1)
                                 /GRADIENT_SIZE_INDEX);
                    pix  = gradients[i1][(idx1+1)>>1];
                }
                ach += (((pix>>>20)&0xFF0) *norm)>>16;
                rch += (((pix>>>12)&0xFF0) *norm)>>16;
                gch += (((pix>>> 4)&0xFF0) *norm)>>16;
                bch += (((pix<<  4)&0xFF0) *norm)>>16;

                if (p2_up) {
                    norm = (int)((base
                                  *normalizedIntervals[i2]
                                  *(GRADIENT_SIZE_INDEX-f2))
                                 /GRADIENT_SIZE_INDEX);
                    pix  =  gradients[i2][(idx2+GRADIENT_SIZE)>>1];
                } else {
                    norm = (int)((base*normalizedIntervals[i2]*f2)
                                 /GRADIENT_SIZE_INDEX);
                    pix  = gradients[i2][(idx2+1)>>1];
                }
                ach += (((pix>>>20)&0xFF0) *norm)>>16;
                rch += (((pix>>>12)&0xFF0) *norm)>>16;
                gch += (((pix>>> 4)&0xFF0) *norm)>>16;
                bch += (((pix<<  4)&0xFF0) *norm)>>16;

                // p1_up and p2_up are just used to set the loop-boundarys,
                // then we loop from iStart to iEnd
                int iStart;
                int iEnd;

                if (p1_up) {
                    iStart = i1+1;
                    iEnd = gradientsLength;
                } else {
                    iStart = 0;
                    iEnd = i1;
                }
                for (int i=iStart; i < iEnd ; i++) {
                    norm = (int)(base*normalizedIntervals[i]);
                    pix  = gradients[i][GRADIENT_SIZE>>1];

                    ach += (((pix>>>20)&0xFF0) *norm)>>16;
                    rch += (((pix>>>12)&0xFF0) *norm)>>16;
                    gch += (((pix>>> 4)&0xFF0) *norm)>>16;
                    bch += (((pix<<  4)&0xFF0) *norm)>>16;
                }

                if (p2_up) {
                    iStart = i2+1;
                    iEnd = gradientsLength;
                } else {
                    iStart = 0;
                    iEnd = i2;
                }
                for (int i=iStart; i < iEnd ; i++) {
                    norm = (int)(base*normalizedIntervals[i]);
                    pix  = gradients[i][GRADIENT_SIZE>>1];

                    ach += (((pix>>>20)&0xFF0) *norm)>>16;
                    rch += (((pix>>>12)&0xFF0) *norm)>>16;
                    gch += (((pix>>> 4)&0xFF0) *norm)>>16;
                    bch += (((pix<<  4)&0xFF0) *norm)>>16;
                }


            }
            ach = (ach+0x08)>>4;
            rch = (rch+0x08)>>4;
            gch = (gch+0x08)>>4;
            bch = (bch+0x08)>>4;
            if (DEBUG) System.out.println("Pix: [" + ach + ", " + rch +
                                          ", " + gch + ", " + bch + ']' );
        }

        if (weight != 1) {
            // System.out.println("ave: " + gradientAverage);
            int aveW = (int)((1<<16)*(1-weight));
            int aveA = ((gradientAverage>>>24) & 0xFF)*aveW;
            int aveR = ((gradientAverage>> 16) & 0xFF)*aveW;
            int aveG = ((gradientAverage>>  8) & 0xFF)*aveW;
            int aveB = ((gradientAverage     ) & 0xFF)*aveW;

            int iw = (int)(weight*(1<<16));
            ach = ((ach*iw)+aveA)>>16;
            rch = ((rch*iw)+aveR)>>16;
            gch = ((gch*iw)+aveG)>>16;
            bch = ((bch*iw)+aveB)>>16;
        }

        return ((ach<<24) | (rch<<16) | (gch<<8) | bch);
    }


    
Helper function to convert a color component in sRGB space to linear
RGB space.  Used to build a static lookup table.
/**
     * Helper function to convert a color component in sRGB space to linear
     * RGB space.  Used to build a static lookup table.
     */
    private static int convertSRGBtoLinearRGB(int color) {

        // use of float and double arithmetic gives exactly same results
        float output;

        float input = color/255.0f;
        if (input <= 0.04045f) {
            output = input/12.92f;
        } else {
            output = (float) Math.pow((input + 0.055) / 1.055, 2.4);
        }
        int o = Math.round(output * 255.0f);

        return o;
    }

     
Helper function to convert a color component in linear RGB space to
 SRGB space. Used to build a static lookup table.
/** Helper function to convert a color component in linear RGB space to
      *  SRGB space. Used to build a static lookup table.
      */
    private static int convertLinearRGBtoSRGB(int color) {

         // use of float and double arithmetic gives exactly same results
        float output;

        float input = color/255.0f;

        if (input <= 0.0031308f) {
            output = input * 12.92f;
        } else {
            output = (1.055f * ((float) Math.pow(input, (1.0 / 2.4)))) - 0.055f;
        }

        int o = Math.round(output * 255.0f);

        return o;
    }


    
Superclass getRaster... /** Superclass getRaster... */
    public final Raster getRaster(int x, int y, int w, int h) {
        if (w == 0 || h == 0) {
            return null;
        }

        //
        // If working raster is big enough, reuse it. Otherwise,
        // build a large enough new one.
        //
        WritableRaster raster = saved;
        if (raster == null || raster.getWidth() < w || raster.getHeight() < h)
            {
                raster = getCachedRaster(dataModel, w, h);
                saved = raster;
                // NOTE:We would like to use 'x' & 'y' here instead of
                // '0', '0' but this will fail on MacOSX.  Since it
                // doesn't have an effect on other JVMs.
                raster = raster.createWritableChild
                    (raster.getMinX(), raster.getMinY(), w, h, 0, 0, null);
            }

        // Access raster internal int array. Because we use a DirectColorModel,
        // we know the DataBuffer is of type DataBufferInt and the SampleModel
        // is SinglePixelPackedSampleModel.
        // Adjust for initial offset in DataBuffer and also for the scanline
        // stride.
        //
        DataBufferInt rasterDB = (DataBufferInt)raster.getDataBuffer();
        int[] pixels = rasterDB.getBankData()[0];
        int off = rasterDB.getOffset();
        int scanlineStride = ((SinglePixelPackedSampleModel)
                              raster.getSampleModel()).getScanlineStride();
        int adjust = scanlineStride - w;

        fillRaster(pixels, off, adjust, x, y, w, h); //delegate to subclass.

        GraphicsUtil.coerceData(raster, dataModel,
                                model.isAlphaPremultiplied());


        return raster;
    }

    
Subclasses should implement this. /** Subclasses should implement this. */
    protected abstract void fillRaster(int[] pixels, int off, int adjust,
                                       int x, int y, int w, int h);


    
Took this cacheRaster code from GradientPaint. It appears to recycle
rasters for use by any other instance, as long as they are sufficiently
large.
/** Took this cacheRaster code from GradientPaint. It appears to recycle
     * rasters for use by any other instance, as long as they are sufficiently
     * large.
     */
    protected static final
    synchronized WritableRaster getCachedRaster
        (ColorModel cm, int w, int h) {
        if (cm == cachedModel) {
            if (cached != null) {
                WritableRaster ras = (WritableRaster) cached.get();
                if (ras != null &&
                    ras.getWidth() >= w &&
                    ras.getHeight() >= h)
                    {
                        cached = null;
                        return ras;
                    }
            }
        }
        // Don't create rediculously small rasters...
        if (w<32) w=32;
        if (h<32) h=32;
        return cm.createCompatibleWritableRaster(w, h);
    }

    
Took this cacheRaster code from GradientPaint. It appears to recycle
rasters for use by any other instance, as long as they are sufficiently
large.
/** Took this cacheRaster code from GradientPaint. It appears to recycle
     * rasters for use by any other instance, as long as they are sufficiently
     * large.
     */
    protected static final
    synchronized void putCachedRaster(ColorModel cm,
                                             WritableRaster ras) {
        if (cached != null) {
            WritableRaster cras = (WritableRaster) cached.get();
            if (cras != null) {
                int cw = cras.getWidth();
                int ch = cras.getHeight();
                int iw = ras.getWidth();
                int ih = ras.getHeight();
                if (cw >= iw && ch >= ih) {
                    return;
                }
                if (cw * ch >= iw * ih) {
                    return;
                }
            }
        }
        cachedModel = cm;
        cached = new WeakReference(ras);
    }

    
Release the resources allocated for the operation.
/**
     * Release the resources allocated for the operation.
     */
    public final void dispose() {
        if (saved != null) {
            putCachedRaster(model, saved);
            saved = null;
        }
    }

    
Return the ColorModel of the output.
/**
     * Return the ColorModel of the output.
     */
    public final ColorModel getColorModel() {
        return model;
    }
}