/*
* Copyright (c) 2007, 2013, Oracle and/or its affiliates. All rights reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation. Oracle designates this
* particular file as subject to the "Classpath" exception as provided
* by Oracle in the LICENSE file that accompanied this code.
*
* This code is distributed in the hope that it will be useful, but WITHOUT
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
* version 2 for more details (a copy is included in the LICENSE file that
* accompanied this code).
*
* You should have received a copy of the GNU General Public License version
* 2 along with this work; if not, write to the Free Software Foundation,
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
*
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
* or visit www.oracle.com if you need additional information or have any
* questions.
*/
package sun.java2d.pipe;
import java.awt.Color;
import java.awt.GradientPaint;
import java.awt.LinearGradientPaint;
import java.awt.MultipleGradientPaint;
import java.awt.MultipleGradientPaint.ColorSpaceType;
import java.awt.MultipleGradientPaint.CycleMethod;
import java.awt.Paint;
import java.awt.RadialGradientPaint;
import java.awt.TexturePaint;
import java.awt.geom.AffineTransform;
import java.awt.geom.Point2D;
import java.awt.geom.Rectangle2D;
import java.awt.image.AffineTransformOp;
import java.awt.image.BufferedImage;
import sun.awt.image.PixelConverter;
import sun.java2d.SunGraphics2D;
import sun.java2d.SurfaceData;
import sun.java2d.loops.CompositeType;
import sun.java2d.loops.SurfaceType;
import static sun.java2d.pipe.BufferedOpCodes.*;
import java.lang.annotation.Native;
public class BufferedPaints {
static void setPaint(RenderQueue rq, SunGraphics2D sg2d,
Paint paint, int ctxflags)
{
if (sg2d.paintState <= SunGraphics2D.PAINT_ALPHACOLOR) {
setColor(rq, sg2d.pixel);
} else {
boolean useMask = (ctxflags & BufferedContext.USE_MASK) != 0;
switch (sg2d.paintState) {
case SunGraphics2D.PAINT_GRADIENT:
setGradientPaint(rq, sg2d,
(GradientPaint)paint, useMask);
break;
case SunGraphics2D.PAINT_LIN_GRADIENT:
setLinearGradientPaint(rq, sg2d,
(LinearGradientPaint)paint, useMask);
break;
case SunGraphics2D.PAINT_RAD_GRADIENT:
setRadialGradientPaint(rq, sg2d,
(RadialGradientPaint)paint, useMask);
break;
case SunGraphics2D.PAINT_TEXTURE:
setTexturePaint(rq, sg2d,
(TexturePaint)paint, useMask);
break;
default:
break;
}
}
}
static void resetPaint(RenderQueue rq) {
// assert rq.lock.isHeldByCurrentThread();
rq.ensureCapacity(4);
RenderBuffer buf = rq.getBuffer();
buf.putInt(RESET_PAINT);
}
Color support /****************************** Color support *******************************/
private static void setColor(RenderQueue rq, int pixel) {
// assert rq.lock.isHeldByCurrentThread();
rq.ensureCapacity(8);
RenderBuffer buf = rq.getBuffer();
buf.putInt(SET_COLOR);
buf.putInt(pixel);
}
/************************* GradientPaint support ****************************/
Note: This code is factored out into a separate static method
so that it can be shared by both the Gradient and LinearGradient
implementations. LinearGradient uses this code (for the
two-color sRGB case only) because it can be much faster than the
equivalent implementation that uses fragment shaders.
We use OpenGL's texture coordinate generator to automatically
apply a smooth gradient (either cyclic or acyclic) to the geometry
being rendered. This technique is almost identical to the one
described in the comments for BufferedPaints.setTexturePaint(),
except the calculations take place in one dimension instead of two.
Instead of an anchor rectangle in the TexturePaint case, we use
the vector between the two GradientPaint end points in our
calculations. The generator uses a single plane equation that
takes the (x,y) location (in device space) of the fragment being
rendered to calculate a (u) texture coordinate for that fragment:
u = Ax + By + Cz + Dw
The gradient renderer uses a two-pixel 1D texture where the first
pixel contains the first GradientPaint color, and the second pixel
contains the second GradientPaint color. (Note that we use the
GL_CLAMP_TO_EDGE wrapping mode for acyclic gradients so that we
clamp the colors properly at the extremes.) The following diagram
attempts to show the layout of the texture containing the two
GradientPaint colors (C1 and C2):
+-----------------+
| C1 | C2 |
| | |
+-----------------+
u=0 .25 .5 .75 1
We calculate our plane equation constants (A,B,D) such that u=0.25
corresponds to the first GradientPaint end point in user space and
u=0.75 corresponds to the second end point. This is somewhat
non-obvious, but since the gradient colors are generated by
interpolating between C1 and C2, we want the pure color at the
end points, and we will get the pure color only when u correlates
to the center of a texel. The following chart shows the expected
color for some sample values of u (where C' is the color halfway
between C1 and C2):
u value acyclic (GL_CLAMP) cyclic (GL_REPEAT)
------- ------------------ ------------------
-0.25 C1 C2
0.0 C1 C'
0.25 C1 C1
0.5 C' C'
0.75 C2 C2
1.0 C2 C'
1.25 C2 C1
Original inspiration for this technique came from UMD's Agile2D
project (GradientManager.java).
/**
* Note: This code is factored out into a separate static method
* so that it can be shared by both the Gradient and LinearGradient
* implementations. LinearGradient uses this code (for the
* two-color sRGB case only) because it can be much faster than the
* equivalent implementation that uses fragment shaders.
*
* We use OpenGL's texture coordinate generator to automatically
* apply a smooth gradient (either cyclic or acyclic) to the geometry
* being rendered. This technique is almost identical to the one
* described in the comments for BufferedPaints.setTexturePaint(),
* except the calculations take place in one dimension instead of two.
* Instead of an anchor rectangle in the TexturePaint case, we use
* the vector between the two GradientPaint end points in our
* calculations. The generator uses a single plane equation that
* takes the (x,y) location (in device space) of the fragment being
* rendered to calculate a (u) texture coordinate for that fragment:
* u = Ax + By + Cz + Dw
*
* The gradient renderer uses a two-pixel 1D texture where the first
* pixel contains the first GradientPaint color, and the second pixel
* contains the second GradientPaint color. (Note that we use the
* GL_CLAMP_TO_EDGE wrapping mode for acyclic gradients so that we
* clamp the colors properly at the extremes.) The following diagram
* attempts to show the layout of the texture containing the two
* GradientPaint colors (C1 and C2):
*
* +-----------------+
* | C1 | C2 |
* | | |
* +-----------------+
* u=0 .25 .5 .75 1
*
* We calculate our plane equation constants (A,B,D) such that u=0.25
* corresponds to the first GradientPaint end point in user space and
* u=0.75 corresponds to the second end point. This is somewhat
* non-obvious, but since the gradient colors are generated by
* interpolating between C1 and C2, we want the pure color at the
* end points, and we will get the pure color only when u correlates
* to the center of a texel. The following chart shows the expected
* color for some sample values of u (where C' is the color halfway
* between C1 and C2):
*
* u value acyclic (GL_CLAMP) cyclic (GL_REPEAT)
* ------- ------------------ ------------------
* -0.25 C1 C2
* 0.0 C1 C'
* 0.25 C1 C1
* 0.5 C' C'
* 0.75 C2 C2
* 1.0 C2 C'
* 1.25 C2 C1
*
* Original inspiration for this technique came from UMD's Agile2D
* project (GradientManager.java).
*/
private static void setGradientPaint(RenderQueue rq, AffineTransform at,
Color c1, Color c2,
Point2D pt1, Point2D pt2,
boolean isCyclic, boolean useMask)
{
// convert gradient colors to IntArgbPre format
PixelConverter pc = PixelConverter.ArgbPre.instance;
int pixel1 = pc.rgbToPixel(c1.getRGB(), null);
int pixel2 = pc.rgbToPixel(c2.getRGB(), null);
// calculate plane equation constants
double x = pt1.getX();
double y = pt1.getY();
at.translate(x, y);
// now gradient point 1 is at the origin
x = pt2.getX() - x;
y = pt2.getY() - y;
double len = Math.sqrt(x * x + y * y);
at.rotate(x, y);
// now gradient point 2 is on the positive x-axis
at.scale(2*len, 1);
// now gradient point 2 is at (0.5, 0)
at.translate(-0.25, 0);
// now gradient point 1 is at (0.25, 0), point 2 is at (0.75, 0)
double p0, p1, p3;
try {
at.invert();
p0 = at.getScaleX();
p1 = at.getShearX();
p3 = at.getTranslateX();
} catch (java.awt.geom.NoninvertibleTransformException e) {
p0 = p1 = p3 = 0.0;
}
// assert rq.lock.isHeldByCurrentThread();
rq.ensureCapacityAndAlignment(44, 12);
RenderBuffer buf = rq.getBuffer();
buf.putInt(SET_GRADIENT_PAINT);
buf.putInt(useMask ? 1 : 0);
buf.putInt(isCyclic ? 1 : 0);
buf.putDouble(p0).putDouble(p1).putDouble(p3);
buf.putInt(pixel1).putInt(pixel2);
}
private static void setGradientPaint(RenderQueue rq,
SunGraphics2D sg2d,
GradientPaint paint,
boolean useMask)
{
setGradientPaint(rq, (AffineTransform)sg2d.transform.clone(),
paint.getColor1(), paint.getColor2(),
paint.getPoint1(), paint.getPoint2(),
paint.isCyclic(), useMask);
}
/************************** TexturePaint support ****************************/
We use OpenGL's texture coordinate generator to automatically
map the TexturePaint image to the geometry being rendered. The
generator uses two separate plane equations that take the (x,y)
location (in device space) of the fragment being rendered to
calculate (u,v) texture coordinates for that fragment:
u = Ax + By + Cz + Dw
v = Ex + Fy + Gz + Hw
Since we use a 2D orthographic projection, we can assume that z=0
and w=1 for any fragment. So we need to calculate appropriate
values for the plane equation constants (A,B,D) and (E,F,H) such
that {u,v}=0 for the top-left of the TexturePaint's anchor
rectangle and {u,v}=1 for the bottom-right of the anchor rectangle.
We can easily make the texture image repeat for {u,v} values
outside the range [0,1] by specifying the GL_REPEAT texture wrap
mode.
Calculating the plane equation constants is surprisingly simple.
We can think of it as an inverse matrix operation that takes
device space coordinates and transforms them into user space
coordinates that correspond to a location relative to the anchor
rectangle. First, we translate and scale the current user space
transform by applying the anchor rectangle bounds. We then take
the inverse of this affine transform. The rows of the resulting
inverse matrix correlate nicely to the plane equation constants
we were seeking.
/**
* We use OpenGL's texture coordinate generator to automatically
* map the TexturePaint image to the geometry being rendered. The
* generator uses two separate plane equations that take the (x,y)
* location (in device space) of the fragment being rendered to
* calculate (u,v) texture coordinates for that fragment:
* u = Ax + By + Cz + Dw
* v = Ex + Fy + Gz + Hw
*
* Since we use a 2D orthographic projection, we can assume that z=0
* and w=1 for any fragment. So we need to calculate appropriate
* values for the plane equation constants (A,B,D) and (E,F,H) such
* that {u,v}=0 for the top-left of the TexturePaint's anchor
* rectangle and {u,v}=1 for the bottom-right of the anchor rectangle.
* We can easily make the texture image repeat for {u,v} values
* outside the range [0,1] by specifying the GL_REPEAT texture wrap
* mode.
*
* Calculating the plane equation constants is surprisingly simple.
* We can think of it as an inverse matrix operation that takes
* device space coordinates and transforms them into user space
* coordinates that correspond to a location relative to the anchor
* rectangle. First, we translate and scale the current user space
* transform by applying the anchor rectangle bounds. We then take
* the inverse of this affine transform. The rows of the resulting
* inverse matrix correlate nicely to the plane equation constants
* we were seeking.
*/
private static void setTexturePaint(RenderQueue rq,
SunGraphics2D sg2d,
TexturePaint paint,
boolean useMask)
{
BufferedImage bi = paint.getImage();
SurfaceData dstData = sg2d.surfaceData;
SurfaceData srcData =
dstData.getSourceSurfaceData(bi, SunGraphics2D.TRANSFORM_ISIDENT,
CompositeType.SrcOver, null);
boolean filter =
(sg2d.interpolationType !=
AffineTransformOp.TYPE_NEAREST_NEIGHBOR);
// calculate plane equation constants
AffineTransform at = (AffineTransform)sg2d.transform.clone();
Rectangle2D anchor = paint.getAnchorRect();
at.translate(anchor.getX(), anchor.getY());
at.scale(anchor.getWidth(), anchor.getHeight());
double xp0, xp1, xp3, yp0, yp1, yp3;
try {
at.invert();
xp0 = at.getScaleX();
xp1 = at.getShearX();
xp3 = at.getTranslateX();
yp0 = at.getShearY();
yp1 = at.getScaleY();
yp3 = at.getTranslateY();
} catch (java.awt.geom.NoninvertibleTransformException e) {
xp0 = xp1 = xp3 = yp0 = yp1 = yp3 = 0.0;
}
// assert rq.lock.isHeldByCurrentThread();
rq.ensureCapacityAndAlignment(68, 12);
RenderBuffer buf = rq.getBuffer();
buf.putInt(SET_TEXTURE_PAINT);
buf.putInt(useMask ? 1 : 0);
buf.putInt(filter ? 1 : 0);
buf.putLong(srcData.getNativeOps());
buf.putDouble(xp0).putDouble(xp1).putDouble(xp3);
buf.putDouble(yp0).putDouble(yp1).putDouble(yp3);
}
/****************** Shared MultipleGradientPaint support ********************/
The maximum number of gradient "stops" supported by our native
fragment shader implementations.
This value has been empirically determined and capped to allow
our native shaders to run on all shader-level graphics hardware,
even on the older, more limited GPUs. Even the oldest Nvidia
hardware could handle 16, or even 32 fractions without any problem.
But the first-generation boards from ATI would fall back into
software mode (which is unusably slow) for values larger than 12;
it appears that those boards do not have enough native registers
to support the number of array accesses required by our gradient
shaders. So for now we will cap this value at 12, but we can
re-evaluate this in the future as hardware becomes more capable.
/**
* The maximum number of gradient "stops" supported by our native
* fragment shader implementations.
*
* This value has been empirically determined and capped to allow
* our native shaders to run on all shader-level graphics hardware,
* even on the older, more limited GPUs. Even the oldest Nvidia
* hardware could handle 16, or even 32 fractions without any problem.
* But the first-generation boards from ATI would fall back into
* software mode (which is unusably slow) for values larger than 12;
* it appears that those boards do not have enough native registers
* to support the number of array accesses required by our gradient
* shaders. So for now we will cap this value at 12, but we can
* re-evaluate this in the future as hardware becomes more capable.
*/
@Native public static final int MULTI_MAX_FRACTIONS = 12;
Helper function to convert a color component in sRGB space to
linear RGB space. Copied directly from the
MultipleGradientPaintContext class.
/**
* Helper function to convert a color component in sRGB space to
* linear RGB space. Copied directly from the
* MultipleGradientPaintContext class.
*/
public static int convertSRGBtoLinearRGB(int color) {
float input, output;
input = color / 255.0f;
if (input <= 0.04045f) {
output = input / 12.92f;
} else {
output = (float)Math.pow((input + 0.055) / 1.055, 2.4);
}
return Math.round(output * 255.0f);
}
Helper function to convert a (non-premultiplied) Color in sRGB
space to an IntArgbPre pixel value, optionally in linear RGB space.
Based on the PixelConverter.ArgbPre.rgbToPixel() method.
/**
* Helper function to convert a (non-premultiplied) Color in sRGB
* space to an IntArgbPre pixel value, optionally in linear RGB space.
* Based on the PixelConverter.ArgbPre.rgbToPixel() method.
*/
private static int colorToIntArgbPrePixel(Color c, boolean linear) {
int rgb = c.getRGB();
if (!linear && ((rgb >> 24) == -1)) {
return rgb;
}
int a = rgb >>> 24;
int r = (rgb >> 16) & 0xff;
int g = (rgb >> 8) & 0xff;
int b = (rgb ) & 0xff;
if (linear) {
r = convertSRGBtoLinearRGB(r);
g = convertSRGBtoLinearRGB(g);
b = convertSRGBtoLinearRGB(b);
}
int a2 = a + (a >> 7);
r = (r * a2) >> 8;
g = (g * a2) >> 8;
b = (b * a2) >> 8;
return ((a << 24) | (r << 16) | (g << 8) | (b));
}
Converts the given array of Color objects into an int array
containing IntArgbPre pixel values. If the linear parameter
is true, the Color values will be converted into a linear RGB
color space before being returned.
/**
* Converts the given array of Color objects into an int array
* containing IntArgbPre pixel values. If the linear parameter
* is true, the Color values will be converted into a linear RGB
* color space before being returned.
*/
private static int[] convertToIntArgbPrePixels(Color[] colors,
boolean linear)
{
int[] pixels = new int[colors.length];
for (int i = 0; i < colors.length; i++) {
pixels[i] = colorToIntArgbPrePixel(colors[i], linear);
}
return pixels;
}
/********************** LinearGradientPaint support *************************/
This method uses techniques that are nearly identical to those
employed in setGradientPaint() above. The primary difference
is that at the native level we use a fragment shader to manually
apply the plane equation constants to the current fragment position
to calculate the gradient position in the range [0,1] (the native
code for GradientPaint does the same, except that it uses OpenGL's
automatic texture coordinate generation facilities).
One other minor difference worth mentioning is that
setGradientPaint() calculates the plane equation constants
such that the gradient end points are positioned at 0.25 and 0.75
(for reasons discussed in the comments for that method). In
contrast, for LinearGradientPaint we setup the equation constants
such that the gradient end points fall at 0.0 and 1.0. The
reason for this difference is that in the fragment shader we
have more control over how the gradient values are interpreted
(depending on the paint's CycleMethod).
/**
* This method uses techniques that are nearly identical to those
* employed in setGradientPaint() above. The primary difference
* is that at the native level we use a fragment shader to manually
* apply the plane equation constants to the current fragment position
* to calculate the gradient position in the range [0,1] (the native
* code for GradientPaint does the same, except that it uses OpenGL's
* automatic texture coordinate generation facilities).
*
* One other minor difference worth mentioning is that
* setGradientPaint() calculates the plane equation constants
* such that the gradient end points are positioned at 0.25 and 0.75
* (for reasons discussed in the comments for that method). In
* contrast, for LinearGradientPaint we setup the equation constants
* such that the gradient end points fall at 0.0 and 1.0. The
* reason for this difference is that in the fragment shader we
* have more control over how the gradient values are interpreted
* (depending on the paint's CycleMethod).
*/
private static void setLinearGradientPaint(RenderQueue rq,
SunGraphics2D sg2d,
LinearGradientPaint paint,
boolean useMask)
{
boolean linear =
(paint.getColorSpace() == ColorSpaceType.LINEAR_RGB);
Color[] colors = paint.getColors();
int numStops = colors.length;
Point2D pt1 = paint.getStartPoint();
Point2D pt2 = paint.getEndPoint();
AffineTransform at = paint.getTransform();
at.preConcatenate(sg2d.transform);
if (!linear && numStops == 2 &&
paint.getCycleMethod() != CycleMethod.REPEAT)
{
// delegate to the optimized two-color gradient codepath
boolean isCyclic =
(paint.getCycleMethod() != CycleMethod.NO_CYCLE);
setGradientPaint(rq, at,
colors[0], colors[1],
pt1, pt2,
isCyclic, useMask);
return;
}
int cycleMethod = paint.getCycleMethod().ordinal();
float[] fractions = paint.getFractions();
int[] pixels = convertToIntArgbPrePixels(colors, linear);
// calculate plane equation constants
double x = pt1.getX();
double y = pt1.getY();
at.translate(x, y);
// now gradient point 1 is at the origin
x = pt2.getX() - x;
y = pt2.getY() - y;
double len = Math.sqrt(x * x + y * y);
at.rotate(x, y);
// now gradient point 2 is on the positive x-axis
at.scale(len, 1);
// now gradient point 1 is at (0.0, 0), point 2 is at (1.0, 0)
float p0, p1, p3;
try {
at.invert();
p0 = (float)at.getScaleX();
p1 = (float)at.getShearX();
p3 = (float)at.getTranslateX();
} catch (java.awt.geom.NoninvertibleTransformException e) {
p0 = p1 = p3 = 0.0f;
}
// assert rq.lock.isHeldByCurrentThread();
rq.ensureCapacity(20 + 12 + (numStops*4*2));
RenderBuffer buf = rq.getBuffer();
buf.putInt(SET_LINEAR_GRADIENT_PAINT);
buf.putInt(useMask ? 1 : 0);
buf.putInt(linear ? 1 : 0);
buf.putInt(cycleMethod);
buf.putInt(numStops);
buf.putFloat(p0);
buf.putFloat(p1);
buf.putFloat(p3);
buf.put(fractions);
buf.put(pixels);
}
/********************** RadialGradientPaint support *************************/
This method calculates six m** values and a focusX value that
are used by the native fragment shader. These techniques are
based on a whitepaper by Daniel Rice on radial gradient performance
(attached to the bug report for 6521533). One can refer to that
document for the complete set of formulas and calculations, but
the basic goal is to compose a transform that will convert an
(x,y) position in device space into a "u" value that represents
the relative distance to the gradient focus point. The resulting
value can be used to look up the appropriate color by linearly
interpolating between the two nearest colors in the gradient.
/**
* This method calculates six m** values and a focusX value that
* are used by the native fragment shader. These techniques are
* based on a whitepaper by Daniel Rice on radial gradient performance
* (attached to the bug report for 6521533). One can refer to that
* document for the complete set of formulas and calculations, but
* the basic goal is to compose a transform that will convert an
* (x,y) position in device space into a "u" value that represents
* the relative distance to the gradient focus point. The resulting
* value can be used to look up the appropriate color by linearly
* interpolating between the two nearest colors in the gradient.
*/
private static void setRadialGradientPaint(RenderQueue rq,
SunGraphics2D sg2d,
RadialGradientPaint paint,
boolean useMask)
{
boolean linear =
(paint.getColorSpace() == ColorSpaceType.LINEAR_RGB);
int cycleMethod = paint.getCycleMethod().ordinal();
float[] fractions = paint.getFractions();
Color[] colors = paint.getColors();
int numStops = colors.length;
int[] pixels = convertToIntArgbPrePixels(colors, linear);
Point2D center = paint.getCenterPoint();
Point2D focus = paint.getFocusPoint();
float radius = paint.getRadius();
// save original (untransformed) center and focus points
double cx = center.getX();
double cy = center.getY();
double fx = focus.getX();
double fy = focus.getY();
// transform from gradient coords to device coords
AffineTransform at = paint.getTransform();
at.preConcatenate(sg2d.transform);
focus = at.transform(focus, focus);
// transform unit circle to gradient coords; we start with the
// unit circle (center=(0,0), focus on positive x-axis, radius=1)
// and then transform into gradient space
at.translate(cx, cy);
at.rotate(fx - cx, fy - cy);
at.scale(radius, radius);
// invert to get mapping from device coords to unit circle
try {
at.invert();
} catch (Exception e) {
at.setToScale(0.0, 0.0);
}
focus = at.transform(focus, focus);
// clamp the focus point so that it does not rest on, or outside
// of, the circumference of the gradient circle
fx = Math.min(focus.getX(), 0.99);
// assert rq.lock.isHeldByCurrentThread();
rq.ensureCapacity(20 + 28 + (numStops*4*2));
RenderBuffer buf = rq.getBuffer();
buf.putInt(SET_RADIAL_GRADIENT_PAINT);
buf.putInt(useMask ? 1 : 0);
buf.putInt(linear ? 1 : 0);
buf.putInt(numStops);
buf.putInt(cycleMethod);
buf.putFloat((float)at.getScaleX());
buf.putFloat((float)at.getShearX());
buf.putFloat((float)at.getTranslateX());
buf.putFloat((float)at.getShearY());
buf.putFloat((float)at.getScaleY());
buf.putFloat((float)at.getTranslateY());
buf.putFloat((float)fx);
buf.put(fractions);
buf.put(pixels);
}
}