blob: e56f5be2a21ab9cdaf31a0d993d7cc40c3a80d2c [file] [log] [blame]
/*
* Copyright (c) 2005, 2018, 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.
*/
#include <math.h>
#include <assert.h>
#include <stdlib.h>
#include <string.h>
#include "jni.h"
#include "j2d_md.h"
#include "java_awt_geom_PathIterator.h"
#include "ProcessPath.h"
/*
* This framework performs filling and drawing of paths with sub-pixel
* precision. Also, it performs clipping by the specified view area.
*
* Drawing of the shapes is performed not pixel by pixel but segment by segment
* except several pixels near endpoints of the drawn line. This approach saves
* lot's of cpu cycles especially in case of large primitives (like ovals with
* sizes more than 50) and helps in achieving appropriate visual quality. Also,
* such method of drawing is useful for the accelerated pipelines where
* overhead of the per-pixel drawing could eliminate all benefits of the
* hardware acceleration.
*
* Filling of the path was taken from
*
* [Graphics Gems, edited by Andrew S Glassner. Academic Press 1990,
* ISBN 0-12-286165-5 (Concave polygon scan conversion), 87-91]
*
* and modified to work with sub-pixel precision and non-continuous paths.
* It's also speeded up by using hash table by rows of the filled objects.
*
* Here is high level scheme showing the rendering process:
*
* doDrawPath doFillPath
* \ /
* ProcessPath
* |
* CheckPathSegment
* |
* --------+------
* | |
* | |
* | |
* _->ProcessCurve |
* / / | |
* \___/ | |
* | |
* DrawCurve ProcessLine
* \ /
* \ /
* \ /
* \ /
* ------+------
* (filling) / \ (drawing)
* / \
* Clipping and Clipping
* clamping \
* | \
* StoreFixedLine ProcessFixedLine
* | / \
* | / \
* FillPolygon PROCESS_LINE PROCESS_POINT
*
*
*
* CheckPathSegment - rough checking and skipping path's segments in case of
* invalid or huge coordinates of the control points to
* avoid calculation problems with NaNs and values close
* to the FLT_MAX
*
* ProcessCurve - (ProcessQuad, ProcessCubic) Splitting the curve into
* monotonic parts having appropriate size (calculated as
* boundary box of the control points)
*
* DrawMonotonicCurve - (DrawMonotonicQuad, DrawMonotonicCubic) flattening
* monotonic curve using adaptive forward differencing
*
* StoreFixedLine - storing segment from the flattened path to the
* FillData structure. Performing clipping and clamping if
* necessary.
*
* PROCESS_LINE, PROCESS_POINT - Helpers for calling appropriate primitive from
* DrawHandler structure
*
* ProcessFixedLine - Drawing line segment with subpixel precision.
*
*/
#define PROCESS_LINE(hnd, fX0, fY0, fX1, fY1, checkBounds, pixelInfo) \
do { \
jint X0 = (fX0) >> MDP_PREC; \
jint Y0 = (fY0) >> MDP_PREC; \
jint X1 = (fX1) >> MDP_PREC; \
jint Y1 = (fY1) >> MDP_PREC; \
jint res; \
\
/* Checking bounds and clipping if necessary. \
* REMIND: It's temporary solution to avoid OOB in rendering code. \
* Current approach uses float equations which are unreliable for \
* clipping and makes assumptions about the line biases of the \
* rendering algorithm. Also, clipping code should be moved down \
* into only those output renderers that need it. \
*/ \
if (checkBounds) { \
jfloat xMinf = hnd->dhnd->xMinf + 0.5f; \
jfloat yMinf = hnd->dhnd->yMinf + 0.5f; \
jfloat xMaxf = hnd->dhnd->xMaxf + 0.5f; \
jfloat yMaxf = hnd->dhnd->yMaxf + 0.5f; \
TESTANDCLIP(yMinf, yMaxf, Y0, X0, Y1, X1, jint, res); \
if (res == CRES_INVISIBLE) break; \
TESTANDCLIP(yMinf, yMaxf, Y1, X1, Y0, X0, jint, res); \
if (res == CRES_INVISIBLE) break; \
TESTANDCLIP(xMinf, xMaxf, X0, Y0, X1, Y1, jint, res); \
if (res == CRES_INVISIBLE) break; \
TESTANDCLIP(xMinf, xMaxf, X1, Y1, X0, Y0, jint, res); \
if (res == CRES_INVISIBLE) break; \
} \
\
/* Handling lines having just one pixel */ \
if (((X0^X1) | (Y0^Y1)) == 0) { \
if (pixelInfo[0] == 0) { \
pixelInfo[0] = 1; \
pixelInfo[1] = X0; \
pixelInfo[2] = Y0; \
pixelInfo[3] = X0; \
pixelInfo[4] = Y0; \
hnd->dhnd->pDrawPixel(hnd->dhnd, X0, Y0); \
} else if ((X0 != pixelInfo[3] || Y0 != pixelInfo[4]) && \
(X0 != pixelInfo[1] || Y0 != pixelInfo[2])) { \
hnd->dhnd->pDrawPixel(hnd->dhnd, X0, Y0); \
pixelInfo[3] = X0; \
pixelInfo[4] = Y0; \
} \
break; \
} \
\
if (pixelInfo[0] && \
((pixelInfo[1] == X0 && pixelInfo[2] == Y0) || \
(pixelInfo[3] == X0 && pixelInfo[4] == Y0))) \
{ \
hnd->dhnd->pDrawPixel(hnd->dhnd, X0, Y0); \
} \
\
hnd->dhnd->pDrawLine(hnd->dhnd, X0, Y0, X1, Y1); \
\
if (pixelInfo[0] == 0) { \
pixelInfo[0] = 1; \
pixelInfo[1] = X0; \
pixelInfo[2] = Y0; \
pixelInfo[3] = X0; \
pixelInfo[4] = Y0; \
} \
\
/* Switch on last pixel of the line if it was already \
* drawn during rendering of the previous segments \
*/ \
if ((pixelInfo[1] == X1 && pixelInfo[2] == Y1) || \
(pixelInfo[3] == X1 && pixelInfo[4] == Y1)) \
{ \
hnd->dhnd->pDrawPixel(hnd->dhnd, X1, Y1); \
} \
pixelInfo[3] = X1; \
pixelInfo[4] = Y1; \
} while(0)
#define PROCESS_POINT(hnd, fX, fY, checkBounds, pixelInfo) \
do { \
jint X_ = (fX)>> MDP_PREC; \
jint Y_ = (fY)>> MDP_PREC; \
if (checkBounds && \
(hnd->dhnd->yMin > Y_ || \
hnd->dhnd->yMax <= Y_ || \
hnd->dhnd->xMin > X_ || \
hnd->dhnd->xMax <= X_)) break; \
/* \
* (X_,Y_) should be inside boundaries \
* \
* assert(hnd->dhnd->yMin <= Y_ && \
* hnd->dhnd->yMax > Y_ && \
* hnd->dhnd->xMin <= X_ && \
* hnd->dhnd->xMax > X_); \
* \
*/ \
if (pixelInfo[0] == 0) { \
pixelInfo[0] = 1; \
pixelInfo[1] = X_; \
pixelInfo[2] = Y_; \
pixelInfo[3] = X_; \
pixelInfo[4] = Y_; \
hnd->dhnd->pDrawPixel(hnd->dhnd, X_, Y_); \
} else if ((X_ != pixelInfo[3] || Y_ != pixelInfo[4]) && \
(X_ != pixelInfo[1] || Y_ != pixelInfo[2])) { \
hnd->dhnd->pDrawPixel(hnd->dhnd, X_, Y_); \
pixelInfo[3] = X_; \
pixelInfo[4] = Y_; \
} \
} while(0)
/*
* Constants for the forward differencing
* of the cubic and quad curves
*/
/* Maximum size of the cubic curve (calculated as the size of the bounding box
* of the control points) which could be rendered without splitting
*/
#define MAX_CUB_SIZE 256
/* Maximum size of the quad curve (calculated as the size of the bounding box
* of the control points) which could be rendered without splitting
*/
#define MAX_QUAD_SIZE 1024
/* Default power of 2 steps used in the forward differencing. Here DF prefix
* stands for DeFault. Constants below are used as initial values for the
* adaptive forward differencing algorithm.
*/
#define DF_CUB_STEPS 3
#define DF_QUAD_STEPS 2
/* Shift of the current point of the curve for preparing to the midpoint
* rounding
*/
#define DF_CUB_SHIFT (FWD_PREC + DF_CUB_STEPS*3 - MDP_PREC)
#define DF_QUAD_SHIFT (FWD_PREC + DF_QUAD_STEPS*2 - MDP_PREC)
/* Default amount of steps of the forward differencing */
#define DF_CUB_COUNT (1<<DF_CUB_STEPS)
#define DF_QUAD_COUNT (1<<DF_QUAD_STEPS)
/* Default boundary constants used to check the necessity of the restepping */
#define DF_CUB_DEC_BND (1<<(DF_CUB_STEPS*3 + FWD_PREC + 2))
#define DF_CUB_INC_BND (1<<(DF_CUB_STEPS*3 + FWD_PREC - 1))
#define DF_QUAD_DEC_BND (1<<(DF_QUAD_STEPS*2 + FWD_PREC + 2))
/* Multiplyers for the coefficients of the polynomial form of the cubic and
* quad curves representation
*/
#define CUB_A_SHIFT FWD_PREC
#define CUB_B_SHIFT (DF_CUB_STEPS + FWD_PREC + 1)
#define CUB_C_SHIFT (DF_CUB_STEPS*2 + FWD_PREC)
#define CUB_A_MDP_MULT (1<<CUB_A_SHIFT)
#define CUB_B_MDP_MULT (1<<CUB_B_SHIFT)
#define CUB_C_MDP_MULT (1<<CUB_C_SHIFT)
#define QUAD_A_SHIFT FWD_PREC
#define QUAD_B_SHIFT (DF_QUAD_STEPS + FWD_PREC)
#define QUAD_A_MDP_MULT (1<<QUAD_A_SHIFT)
#define QUAD_B_MDP_MULT (1<<QUAD_B_SHIFT)
#define CALC_MAX(MAX, X) ((MAX)=((X)>(MAX))?(X):(MAX))
#define CALC_MIN(MIN, X) ((MIN)=((X)<(MIN))?(X):(MIN))
#define MAX(MAX, X) (((X)>(MAX))?(X):(MAX))
#define MIN(MIN, X) (((X)<(MIN))?(X):(MIN))
#define ABS32(X) (((X)^((X)>>31))-((X)>>31))
#define SIGN32(X) ((X) >> 31) | ((juint)(-(X)) >> 31)
/* Boundaries used for clipping large path segments (those are inside
* [UPPER/LOWER]_BND boundaries)
*/
#define UPPER_OUT_BND (1 << (30 - MDP_PREC))
#define LOWER_OUT_BND (-UPPER_OUT_BND)
#define ADJUST(X, LBND, UBND) \
do { \
if ((X) < (LBND)) { \
(X) = (LBND); \
} else if ((X) > UBND) { \
(X) = (UBND); \
} \
} while(0)
/* Following constants are used for providing open boundaries of the intervals
*/
#define EPSFX 1
#define EPSF (((jfloat)EPSFX)/MDP_MULT)
/* Calculation boundary. It is used for switching to the more slow but allowing
* larger input values method of calculation of the initial values of the scan
* converted line segments inside the FillPolygon.
*/
#define CALC_BND (1 << (30 - MDP_PREC))
/* Clipping macros for drawing and filling algorithms */
#define CLIP(a1, b1, a2, b2, t) \
(b1 + ((jdouble)(t - a1)*(b2 - b1)) / (a2 - a1))
enum {
CRES_MIN_CLIPPED,
CRES_MAX_CLIPPED,
CRES_NOT_CLIPPED,
CRES_INVISIBLE
};
#define IS_CLIPPED(res) (res == CRES_MIN_CLIPPED || res == CRES_MAX_CLIPPED)
#define TESTANDCLIP(LINE_MIN, LINE_MAX, a1, b1, a2, b2, TYPE, res) \
do { \
jdouble t; \
res = CRES_NOT_CLIPPED; \
if (a1 < (LINE_MIN) || a1 > (LINE_MAX)) { \
if (a1 < (LINE_MIN)) { \
if (a2 < (LINE_MIN)) { \
res = CRES_INVISIBLE; \
break; \
}; \
res = CRES_MIN_CLIPPED; \
t = (LINE_MIN); \
} else { \
if (a2 > (LINE_MAX)) { \
res = CRES_INVISIBLE; \
break; \
}; \
res = CRES_MAX_CLIPPED; \
t = (LINE_MAX); \
} \
b1 = (TYPE)CLIP(a1, b1, a2, b2, t); \
a1 = (TYPE)t; \
} \
} while (0)
/* Following macro is used for clipping and clumping filled shapes.
* An example of this process is shown on the picture below:
* ----+ ----+
* |/ | |/ |
* + | + |
* /| | I |
* / | | I |
* | | | ===> I |
* \ | | I |
* \| | I |
* + | + |
* |\ | |\ |
* | ----+ | ----+
* boundary boundary
*
* We can only perform clipping in case of right side of the output area
* because all segments passed out the right boundary don't influence on the
* result of scan conversion algorithm (it correctly handles half open
* contours).
*
*/
#define CLIPCLAMP(LINE_MIN, LINE_MAX, a1, b1, a2, b2, a3, b3, TYPE, res) \
do { \
a3 = a1; \
b3 = b1; \
TESTANDCLIP(LINE_MIN, LINE_MAX, a1, b1, a2, b2, TYPE, res); \
if (res == CRES_MIN_CLIPPED) { \
a3 = a1; \
} else if (res == CRES_MAX_CLIPPED) { \
a3 = a1; \
res = CRES_MAX_CLIPPED; \
} else if (res == CRES_INVISIBLE) { \
if (a1 > LINE_MAX) { \
res = CRES_INVISIBLE; \
} else { \
a1 = (TYPE)LINE_MIN; \
a2 = (TYPE)LINE_MIN; \
res = CRES_NOT_CLIPPED; \
} \
} \
} while (0)
/* Following macro is used for solving quadratic equations:
* A*t^2 + B*t + C = 0
* in (0,1) range. That means we put to the RES the only roots which
* belongs to the (0,1) range. Note: 0 and 1 are not included.
* See solveQuadratic method in
* src/share/classes/java/awt/geom/QuadCurve2D.java
* for more info about calculations
*/
#define SOLVEQUADINRANGE(A,B,C,RES,RCNT) \
do { \
double param; \
if ((A) != 0) { \
/* Calculating roots of the following equation \
* A*t^2 + B*t + C = 0 \
*/ \
double d = (B)*(B) - 4*(A)*(C); \
double q; \
if (d < 0) { \
break; \
} \
d = sqrt(d); \
/* For accuracy, calculate one root using: \
* (-B +/- d) / 2*A \
* and the other using: \
* 2*C / (-B +/- d) \
* Choose the sign of the +/- so that B+D gets larger \
* in magnitude \
*/ \
if ((B) < 0) { \
d = -d; \
} \
q = ((B) + d) / -2.0; \
param = q/(A); \
if (param < 1.0 && param > 0.0) { \
(RES)[(RCNT)++] = param; \
} \
if (d == 0 || q == 0) { \
break; \
} \
param = (C)/q; \
if (param < 1.0 && param > 0.0) { \
(RES)[(RCNT)++] = param; \
} \
} else { \
/* Calculating root of the following equation \
* B*t + C = 0 \
*/ \
if ((B) == 0) { \
break; \
} \
param = -(C)/(B); \
if (param < 1.0 && param > 0.0) { \
(RES)[(RCNT)++] = param; \
} \
} \
} while(0)
/* Drawing line with subpixel endpoints
*
* (x1, y1), (x2, y2) - fixed point coordinates of the endpoints
* with MDP_PREC bits for the fractional part
*
* pixelInfo - structure which keeps drawing info for avoiding
* multiple drawing at the same position on the
* screen (required for the XOR mode of drawing)
*
* pixelInfo[0] - state of the drawing
* 0 - no pixel drawn between
* moveTo/close of the path
* 1 - there are drawn pixels
*
* pixelInfo[1,2] - first pixel of the path
* between moveTo/close of the
* path
*
* pixelInfo[3,4] - last drawn pixel between
* moveTo/close of the path
*
* checkBounds - flag showing necessity of checking the clip
*
*/
void ProcessFixedLine(ProcessHandler* hnd,jint x1,jint y1,jint x2,jint y2,
jint* pixelInfo,jboolean checkBounds,
jboolean endSubPath)
{
/* Checking if line is inside a (X,Y),(X+MDP_MULT,Y+MDP_MULT) box */
jint c = ((x1 ^ x2) | (y1 ^ y2));
jint rx1, ry1, rx2, ry2;
if ((c & MDP_W_MASK) == 0) {
/* Checking for the segments with integer coordinates having
* the same start and end points
*/
if (c == 0) {
PROCESS_POINT(hnd, x1 + MDP_HALF_MULT, y1 + MDP_HALF_MULT,
checkBounds, pixelInfo);
}
return;
}
if (x1 == x2 || y1 == y2) {
rx1 = x1 + MDP_HALF_MULT;
rx2 = x2 + MDP_HALF_MULT;
ry1 = y1 + MDP_HALF_MULT;
ry2 = y2 + MDP_HALF_MULT;
} else {
/* Neither dx nor dy can be zero because of the check above */
jint dx = x2 - x1;
jint dy = y2 - y1;
/* Floor of x1, y1, x2, y2 */
jint fx1 = x1 & MDP_W_MASK;
jint fy1 = y1 & MDP_W_MASK;
jint fx2 = x2 & MDP_W_MASK;
jint fy2 = y2 & MDP_W_MASK;
/* Processing first endpoint */
if (fx1 == x1 || fy1 == y1) {
/* Adding MDP_HALF_MULT to the [xy]1 if f[xy]1 == [xy]1 will not
* affect the result
*/
rx1 = x1 + MDP_HALF_MULT;
ry1 = y1 + MDP_HALF_MULT;
} else {
/* Boundary at the direction from (x1,y1) to (x2,y2) */
jint bx1 = (x1 < x2) ? fx1 + MDP_MULT : fx1;
jint by1 = (y1 < y2) ? fy1 + MDP_MULT : fy1;
/* intersection with column bx1 */
jint cross = y1 + ((bx1 - x1)*dy)/dx;
if (cross >= fy1 && cross <= fy1 + MDP_MULT) {
rx1 = bx1;
ry1 = cross + MDP_HALF_MULT;
} else {
/* intersection with row by1 */
cross = x1 + ((by1 - y1)*dx)/dy;
rx1 = cross + MDP_HALF_MULT;
ry1 = by1;
}
}
/* Processing second endpoint */
if (fx2 == x2 || fy2 == y2) {
/* Adding MDP_HALF_MULT to the [xy]2 if f[xy]2 == [xy]2 will not
* affect the result
*/
rx2 = x2 + MDP_HALF_MULT;
ry2 = y2 + MDP_HALF_MULT;
} else {
/* Boundary at the direction from (x2,y2) to (x1,y1) */
jint bx2 = (x1 > x2) ? fx2 + MDP_MULT : fx2;
jint by2 = (y1 > y2) ? fy2 + MDP_MULT : fy2;
/* intersection with column bx2 */
jint cross = y2 + ((bx2 - x2)*dy)/dx;
if (cross >= fy2 && cross <= fy2 + MDP_MULT) {
rx2 = bx2;
ry2 = cross + MDP_HALF_MULT;
} else {
/* intersection with row by2 */
cross = x2 + ((by2 - y2)*dx)/dy;
rx2 = cross + MDP_HALF_MULT;
ry2 = by2;
}
}
}
PROCESS_LINE(hnd, rx1, ry1, rx2, ry2, checkBounds, pixelInfo);
}
/* Performing drawing of the monotonic in X and Y quadratic curves with sizes
* less than MAX_QUAD_SIZE by using forward differencing method of calculation.
* See comments to the DrawMonotonicCubic.
*/
static void DrawMonotonicQuad(ProcessHandler* hnd,
jfloat *coords,
jboolean checkBounds,
jint* pixelInfo)
{
jint x0 = (jint)(coords[0]*MDP_MULT);
jint y0 = (jint)(coords[1]*MDP_MULT);
jint xe = (jint)(coords[4]*MDP_MULT);
jint ye = (jint)(coords[5]*MDP_MULT);
/* Extracting fractional part of coordinates of first control point */
jint px = (x0 & (~MDP_W_MASK)) << DF_QUAD_SHIFT;
jint py = (y0 & (~MDP_W_MASK)) << DF_QUAD_SHIFT;
/* Setting default amount of steps */
jint count = DF_QUAD_COUNT;
/* Setting default shift for preparing to the midpoint rounding */
jint shift = DF_QUAD_SHIFT;
jint ax = (jint)((coords[0] - 2*coords[2] +
coords[4])*QUAD_A_MDP_MULT);
jint ay = (jint)((coords[1] - 2*coords[3] +
coords[5])*QUAD_A_MDP_MULT);
jint bx = (jint)((-2*coords[0] + 2*coords[2])*QUAD_B_MDP_MULT);
jint by = (jint)((-2*coords[1] + 2*coords[3])*QUAD_B_MDP_MULT);
jint ddpx = 2*ax;
jint ddpy = 2*ay;
jint dpx = ax + bx;
jint dpy = ay + by;
jint x1, y1;
jint x2 = x0;
jint y2 = y0;
jint maxDD = MAX(ABS32(ddpx),ABS32(ddpy));
jint x0w = x0 & MDP_W_MASK;
jint y0w = y0 & MDP_W_MASK;
jint dx = xe - x0;
jint dy = ye - y0;
/* Perform decreasing step in 2 times if slope of the second forward
* difference changes too quickly (more than a pixel per step in X or Y
* direction). We can perform adjusting of the step size before the
* rendering loop because the curvature of the quad curve remains the same
* along all the curve
*/
while (maxDD > DF_QUAD_DEC_BND) {
dpx = (dpx<<1) - ax;
dpy = (dpy<<1) - ay;
count <<= 1;
maxDD >>= 2;
px <<=2;
py <<=2;
shift += 2;
}
while(count-- > 1) {
px += dpx;
py += dpy;
dpx += ddpx;
dpy += ddpy;
x1 = x2;
y1 = y2;
x2 = x0w + (px >> shift);
y2 = y0w + (py >> shift);
/* Checking that we are not running out of the endpoint and bounding
* violating coordinate. The check is pretty simple because the curve
* passed to the DrawMonotonicQuad already split into the monotonic
* in X and Y pieces
*/
/* Bounding x2 by xe */
if (((xe-x2)^dx) < 0) {
x2 = xe;
}
/* Bounding y2 by ye */
if (((ye-y2)^dy) < 0) {
y2 = ye;
}
hnd->pProcessFixedLine(hnd, x1, y1, x2, y2, pixelInfo, checkBounds,
JNI_FALSE);
}
/* We are performing one step less than necessary and use actual (xe,ye)
* curve's endpoint instead of calculated. This prevent us from accumulated
* errors at the last point.
*/
hnd->pProcessFixedLine(hnd, x2, y2, xe, ye, pixelInfo, checkBounds,
JNI_FALSE);
}
/*
* Checking size of the quad curves and split them if necessary.
* Calling DrawMonotonicQuad for the curves of the appropriate size.
* Note: coords array could be changed
*/
static void ProcessMonotonicQuad(ProcessHandler* hnd,
jfloat *coords,
jint* pixelInfo) {
jfloat coords1[6];
jfloat xMin, xMax;
jfloat yMin, yMax;
xMin = xMax = coords[0];
yMin = yMax = coords[1];
CALC_MIN(xMin, coords[2]);
CALC_MAX(xMax, coords[2]);
CALC_MIN(yMin, coords[3]);
CALC_MAX(yMax, coords[3]);
CALC_MIN(xMin, coords[4]);
CALC_MAX(xMax, coords[4]);
CALC_MIN(yMin, coords[5]);
CALC_MAX(yMax, coords[5]);
if (hnd->clipMode == PH_MODE_DRAW_CLIP) {
/* In case of drawing we could just skip curves which are completely
* out of bounds
*/
if (hnd->dhnd->xMaxf < xMin || hnd->dhnd->xMinf > xMax ||
hnd->dhnd->yMaxf < yMin || hnd->dhnd->yMinf > yMax) {
return;
}
} else {
/* In case of filling we could skip curves which are above,
* below and behind the right boundary of the visible area
*/
if (hnd->dhnd->yMaxf < yMin || hnd->dhnd->yMinf > yMax ||
hnd->dhnd->xMaxf < xMin)
{
return;
}
/* We could clamp x coordinates to the corresponding boundary
* if the curve is completely behind the left one
*/
if (hnd->dhnd->xMinf > xMax) {
coords[0] = coords[2] = coords[4] = hnd->dhnd->xMinf;
}
}
if (xMax - xMin > MAX_QUAD_SIZE || yMax - yMin > MAX_QUAD_SIZE) {
coords1[4] = coords[4];
coords1[5] = coords[5];
coords1[2] = (coords[2] + coords[4])/2.0f;
coords1[3] = (coords[3] + coords[5])/2.0f;
coords[2] = (coords[0] + coords[2])/2.0f;
coords[3] = (coords[1] + coords[3])/2.0f;
coords[4] = coords1[0] = (coords[2] + coords1[2])/2.0f;
coords[5] = coords1[1] = (coords[3] + coords1[3])/2.0f;
ProcessMonotonicQuad(hnd, coords, pixelInfo);
ProcessMonotonicQuad(hnd, coords1, pixelInfo);
} else {
DrawMonotonicQuad(hnd, coords,
/* Set checkBounds parameter if curve intersects
* boundary of the visible area. We know that the
* curve is visible, so the check is pretty simple
*/
hnd->dhnd->xMinf >= xMin || hnd->dhnd->xMaxf <= xMax ||
hnd->dhnd->yMinf >= yMin || hnd->dhnd->yMaxf <= yMax,
pixelInfo);
}
}
/*
* Bite the piece of the quadratic curve from start point till the point
* corresponding to the specified parameter then call ProcessQuad for the
* bitten part.
* Note: coords array will be changed
*/
static void ProcessFirstMonotonicPartOfQuad(ProcessHandler* hnd, jfloat* coords,
jint* pixelInfo, jfloat t)
{
jfloat coords1[6];
coords1[0] = coords[0];
coords1[1] = coords[1];
coords1[2] = coords[0] + t*(coords[2] - coords[0]);
coords1[3] = coords[1] + t*(coords[3] - coords[1]);
coords[2] = coords[2] + t*(coords[4] - coords[2]);
coords[3] = coords[3] + t*(coords[5] - coords[3]);
coords[0] = coords1[4] = coords1[2] + t*(coords[2] - coords1[2]);
coords[1] = coords1[5] = coords1[3] + t*(coords[3] - coords1[3]);
ProcessMonotonicQuad(hnd, coords1, pixelInfo);
}
/*
* Split quadratic curve into monotonic in X and Y parts. Calling
* ProcessMonotonicQuad for each monotonic piece of the curve.
* Note: coords array could be changed
*/
static void ProcessQuad(ProcessHandler* hnd, jfloat* coords, jint* pixelInfo) {
/* Temporary array for holding parameters corresponding to the extreme in X
* and Y points. The values are inside the (0,1) range (0 and 1 excluded)
* and in ascending order.
*/
double params[2];
jint cnt = 0;
double param;
/* Simple check for monotonicity in X before searching for the extreme
* points of the X(t) function. We first check if the curve is monotonic
* in X by seeing if all of the X coordinates are strongly ordered.
*/
if ((coords[0] > coords[2] || coords[2] > coords[4]) &&
(coords[0] < coords[2] || coords[2] < coords[4]))
{
/* Searching for extreme points of the X(t) function by solving
* dX(t)
* ---- = 0 equation
* dt
*/
double ax = coords[0] - 2*coords[2] + coords[4];
if (ax != 0) {
/* Calculating root of the following equation
* ax*t + bx = 0
*/
double bx = coords[0] - coords[2];
param = bx/ax;
if (param < 1.0 && param > 0.0) {
params[cnt++] = param;
}
}
}
/* Simple check for monotonicity in Y before searching for the extreme
* points of the Y(t) function. We first check if the curve is monotonic
* in Y by seeing if all of the Y coordinates are strongly ordered.
*/
if ((coords[1] > coords[3] || coords[3] > coords[5]) &&
(coords[1] < coords[3] || coords[3] < coords[5]))
{
/* Searching for extreme points of the Y(t) function by solving
* dY(t)
* ----- = 0 equation
* dt
*/
double ay = coords[1] - 2*coords[3] + coords[5];
if (ay != 0) {
/* Calculating root of the following equation
* ay*t + by = 0
*/
double by = coords[1] - coords[3];
param = by/ay;
if (param < 1.0 && param > 0.0) {
if (cnt > 0) {
/* Inserting parameter only if it differs from
* already stored
*/
if (params[0] > param) {
params[cnt++] = params[0];
params[0] = param;
} else if (params[0] < param) {
params[cnt++] = param;
}
} else {
params[cnt++] = param;
}
}
}
}
/* Processing obtained monotonic parts */
switch(cnt) {
case 0:
break;
case 1:
ProcessFirstMonotonicPartOfQuad(hnd, coords, pixelInfo,
(jfloat)params[0]);
break;
case 2:
ProcessFirstMonotonicPartOfQuad(hnd, coords, pixelInfo,
(jfloat)params[0]);
param = params[1] - params[0];
if (param > 0) {
ProcessFirstMonotonicPartOfQuad(hnd, coords, pixelInfo,
/* Scale parameter to match with rest of the curve */
(jfloat)(param/(1.0 - params[0])));
}
break;
}
ProcessMonotonicQuad(hnd,coords,pixelInfo);
}
/*
* Performing drawing of the monotonic in X and Y cubic curves with sizes less
* than MAX_CUB_SIZE by using forward differencing method of calculation.
*
* Here is some math used in the code below.
*
* If we express the parametric equation for the coordinates as
* simple polynomial:
*
* V(t) = a * t^3 + b * t^2 + c * t + d
*
* The equations for how we derive these polynomial coefficients
* from the Bezier control points can be found in the method comments
* for the CubicCurve.fillEqn Java method.
*
* From this polynomial, we can derive the forward differences to
* allow us to calculate V(t+K) from V(t) as follows:
*
* 1) V1(0)
* = V(K)-V(0)
* = aK^3 + bK^2 + cK + d - d
* = aK^3 + bK^2 + cK
*
* 2) V1(K)
* = V(2K)-V(K)
* = 8aK^3 + 4bK^2 + 2cK + d - aK^3 - bK^2 - cK - d
* = 7aK^3 + 3bK^2 + cK
*
* 3) V1(2K)
* = V(3K)-V(2K)
* = 27aK^3 + 9bK^2 + 3cK + d - 8aK^3 - 4bK^2 - 2cK - d
* = 19aK^3 + 5bK^2 + cK
*
* 4) V2(0)
* = V1(K) - V1(0)
* = 7aK^3 + 3bK^2 + cK - aK^3 - bK^2 - cK
* = 6aK^3 + 2bK^2
*
* 5) V2(K)
* = V1(2K) - V1(K)
* = 19aK^3 + 5bK^2 + cK - 7aK^3 - 3bK^2 - cK
* = 12aK^3 + 2bK^2
*
* 6) V3(0)
* = V2(K) - V2(0)
* = 12aK^3 + 2bK^2 - 6aK^3 - 2bK^2
* = 6aK^3
*
* Note that if we continue on to calculate V1(3K), V2(2K) and
* V3(K) we will see that V3(K) == V3(0) so we need at most
* 3 cascading forward differences to step through the cubic
* curve.
*
* Note, b coefficient calculating in the DrawCubic is actually twice the b
* coefficient seen above. It's been done for the better accuracy.
*
* In our case, initialy K is chosen as 1/(2^DF_CUB_STEPS) this value is taken
* with FWD_PREC bits precision. This means that we should do 2^DF_CUB_STEPS
* steps to pass through all the curve.
*
* On each step we examine how far we are stepping by examining our first(V1)
* and second (V2) order derivatives and verifying that they are met following
* conditions:
*
* abs(V2) <= DF_CUB_DEC_BND
* abs(V1) > DF_CUB_INC_BND
*
* So, ensures that we step through the curve more slowly when its curvature is
* high and faster when its curvature is lower. If the step size needs
* adjustment we adjust it so that we step either twice as fast, or twice as
* slow until our step size is within range. This modifies our stepping
* variables as follows:
*
* Decreasing step size
* (See Graphics Gems/by A.Glassner,(Tutorial on forward differencing),601-602)
*
* V3 = oV3/8
* V2 = oV2/4 - V3
* V1 = (oV1 - V2)/2
*
* Here V1-V3 stands for new values of the forward differencies and oV1 - oV3
* for the old ones
*
* Using the equations above it's easy to calculating stepping variables for
* the increasing step size:
*
* V1 = 2*oV1 + oV2
* V2 = 4*oV2 + 4*oV3
* V3 = 8*oV3
*
* And then for not to running out of 32 bit precision we are performing 3 bit
* shift of the forward differencing precision (keeping in shift variable) in
* left or right direction depending on what is happening (decreasing or
* increasing). So, all oV1 - oV3 variables should be thought as appropriately
* shifted in regard to the V1 - V3.
*
* Taking all of the above into account we will have following:
*
* Decreasing step size:
*
* shift = shift + 3
* V3 keeps the same
* V2 = 2*oV2 - V3
* V1 = 4*oV1 - V2/2
*
* Increasing step size:
*
* shift = shift - 3
* V1 = oV1/4 + oV2/8
* V2 = oV2/2 + oV3/2
* V3 keeps the same
*
*/
static void DrawMonotonicCubic(ProcessHandler* hnd,
jfloat *coords,
jboolean checkBounds,
jint* pixelInfo)
{
jint x0 = (jint)(coords[0]*MDP_MULT);
jint y0 = (jint)(coords[1]*MDP_MULT);
jint xe = (jint)(coords[6]*MDP_MULT);
jint ye = (jint)(coords[7]*MDP_MULT);
/* Extracting fractional part of coordinates of first control point */
jint px = (x0 & (~MDP_W_MASK)) << DF_CUB_SHIFT;
jint py = (y0 & (~MDP_W_MASK)) << DF_CUB_SHIFT;
/* Setting default boundary values for checking first and second forward
* difference for the necessity of the restepping. See comments to the
* boundary values in ProcessQuad for more info.
*/
jint incStepBnd1 = DF_CUB_INC_BND;
jint incStepBnd2 = DF_CUB_INC_BND << 1;
jint decStepBnd1 = DF_CUB_DEC_BND;
jint decStepBnd2 = DF_CUB_DEC_BND << 1;
/* Setting default amount of steps */
jint count = DF_CUB_COUNT;
/* Setting default shift for preparing to the midpoint rounding */
jint shift = DF_CUB_SHIFT;
jint ax = (jint)((-coords[0] + 3*coords[2] - 3*coords[4] +
coords[6])*CUB_A_MDP_MULT);
jint ay = (jint)((-coords[1] + 3*coords[3] - 3*coords[5] +
coords[7])*CUB_A_MDP_MULT);
jint bx = (jint)((3*coords[0] - 6*coords[2] +
3*coords[4])*CUB_B_MDP_MULT);
jint by = (jint)((3*coords[1] - 6*coords[3] +
3*coords[5])*CUB_B_MDP_MULT);
jint cx = (jint)((-3*coords[0] + 3*coords[2])*(CUB_C_MDP_MULT));
jint cy = (jint)((-3*coords[1] + 3*coords[3])*(CUB_C_MDP_MULT));
jint dddpx = 6*ax;
jint dddpy = 6*ay;
jint ddpx = dddpx + bx;
jint ddpy = dddpy + by;
jint dpx = ax + (bx>>1) + cx;
jint dpy = ay + (by>>1) + cy;
jint x1, y1;
jint x2 = x0;
jint y2 = y0;
/* Calculating whole part of the first point of the curve */
jint x0w = x0 & MDP_W_MASK;
jint y0w = y0 & MDP_W_MASK;
jint dx = xe - x0;
jint dy = ye - y0;
while (count > 0) {
/* Perform decreasing step in 2 times if necessary */
while (
/* The code below is an optimized version of the checks:
* abs(ddpx) > decStepBnd1 ||
* abs(ddpy) > decStepBnd1
*/
(juint)(ddpx + decStepBnd1) > (juint)decStepBnd2 ||
(juint)(ddpy + decStepBnd1) > (juint)decStepBnd2)
{
ddpx = (ddpx<<1) - dddpx;
ddpy = (ddpy<<1) - dddpy;
dpx = (dpx<<2) - (ddpx>>1);
dpy = (dpy<<2) - (ddpy>>1);
count <<=1;
decStepBnd1 <<=3;
decStepBnd2 <<=3;
incStepBnd1 <<=3;
incStepBnd2 <<=3;
px <<=3;
py <<=3;
shift += 3;
}
/* Perform increasing step in 2 times if necessary.
* Note: we could do it only in even steps
*/
while (((count & 1) ^ 1) && shift > DF_CUB_SHIFT &&
/* The code below is an optimized version of the check:
* abs(dpx) <= incStepBnd1 &&
* abs(dpy) <= incStepBnd1
*/
(juint)(dpx + incStepBnd1) <= (juint)incStepBnd2 &&
(juint)(dpy + incStepBnd1) <= (juint)incStepBnd2)
{
dpx = (dpx>>2) + (ddpx>>3);
dpy = (dpy>>2) + (ddpy>>3);
ddpx = (ddpx + dddpx)>>1;
ddpy = (ddpy + dddpy)>>1;
count >>=1;
decStepBnd1 >>=3;
decStepBnd2 >>=3;
incStepBnd1 >>=3;
incStepBnd2 >>=3;
px >>=3;
py >>=3;
shift -= 3;
}
count--;
/* We are performing one step less than necessary and use actual
* (xe,ye) endpoint of the curve instead of calculated. This prevent
* us from accumulated errors at the last point.
*/
if (count) {
px += dpx;
py += dpy;
dpx += ddpx;
dpy += ddpy;
ddpx += dddpx;
ddpy += dddpy;
x1 = x2;
y1 = y2;
x2 = x0w + (px >> shift);
y2 = y0w + (py >> shift);
/* Checking that we are not running out of the endpoint and
* bounding violating coordinate. The check is pretty simple
* because the curve passed to the DrawMonotonicCubic already
* split into the monotonic in X and Y pieces
*/
/* Bounding x2 by xe */
if (((xe-x2)^dx) < 0) {
x2 = xe;
}
/* Bounding y2 by ye */
if (((ye-y2)^dy) < 0) {
y2 = ye;
}
hnd->pProcessFixedLine(hnd, x1, y1, x2, y2, pixelInfo, checkBounds,
JNI_FALSE);
} else {
hnd->pProcessFixedLine(hnd, x2, y2, xe, ye, pixelInfo, checkBounds,
JNI_FALSE);
}
}
}
/*
* Checking size of the cubic curves and split them if necessary.
* Calling DrawMonotonicCubic for the curves of the appropriate size.
* Note: coords array could be changed
*/
static void ProcessMonotonicCubic(ProcessHandler* hnd,
jfloat *coords,
jint* pixelInfo) {
jfloat coords1[8];
jfloat tx, ty;
jfloat xMin, xMax;
jfloat yMin, yMax;
xMin = xMax = coords[0];
yMin = yMax = coords[1];
CALC_MIN(xMin, coords[2]);
CALC_MAX(xMax, coords[2]);
CALC_MIN(yMin, coords[3]);
CALC_MAX(yMax, coords[3]);
CALC_MIN(xMin, coords[4]);
CALC_MAX(xMax, coords[4]);
CALC_MIN(yMin, coords[5]);
CALC_MAX(yMax, coords[5]);
CALC_MIN(xMin, coords[6]);
CALC_MAX(xMax, coords[6]);
CALC_MIN(yMin, coords[7]);
CALC_MAX(yMax, coords[7]);
if (hnd->clipMode == PH_MODE_DRAW_CLIP) {
/* In case of drawing we could just skip curves which are completely
* out of bounds
*/
if (hnd->dhnd->xMaxf < xMin || hnd->dhnd->xMinf > xMax ||
hnd->dhnd->yMaxf < yMin || hnd->dhnd->yMinf > yMax) {
return;
}
} else {
/* In case of filling we could skip curves which are above,
* below and behind the right boundary of the visible area
*/
if (hnd->dhnd->yMaxf < yMin || hnd->dhnd->yMinf > yMax ||
hnd->dhnd->xMaxf < xMin)
{
return;
}
/* We could clamp x coordinates to the corresponding boundary
* if the curve is completely behind the left one
*/
if (hnd->dhnd->xMinf > xMax) {
coords[0] = coords[2] = coords[4] = coords[6] =
hnd->dhnd->xMinf;
}
}
if (xMax - xMin > MAX_CUB_SIZE || yMax - yMin > MAX_CUB_SIZE) {
coords1[6] = coords[6];
coords1[7] = coords[7];
coords1[4] = (coords[4] + coords[6])/2.0f;
coords1[5] = (coords[5] + coords[7])/2.0f;
tx = (coords[2] + coords[4])/2.0f;
ty = (coords[3] + coords[5])/2.0f;
coords1[2] = (tx + coords1[4])/2.0f;
coords1[3] = (ty + coords1[5])/2.0f;
coords[2] = (coords[0] + coords[2])/2.0f;
coords[3] = (coords[1] + coords[3])/2.0f;
coords[4] = (coords[2] + tx)/2.0f;
coords[5] = (coords[3] + ty)/2.0f;
coords[6]=coords1[0]=(coords[4] + coords1[2])/2.0f;
coords[7]=coords1[1]=(coords[5] + coords1[3])/2.0f;
ProcessMonotonicCubic(hnd, coords, pixelInfo);
ProcessMonotonicCubic(hnd, coords1, pixelInfo);
} else {
DrawMonotonicCubic(hnd, coords,
/* Set checkBounds parameter if curve intersects
* boundary of the visible area. We know that the
* curve is visible, so the check is pretty simple
*/
hnd->dhnd->xMinf > xMin || hnd->dhnd->xMaxf < xMax ||
hnd->dhnd->yMinf > yMin || hnd->dhnd->yMaxf < yMax,
pixelInfo);
}
}
/*
* Bite the piece of the cubic curve from start point till the point
* corresponding to the specified parameter then call ProcessMonotonicCubic for
* the bitten part.
* Note: coords array will be changed
*/
static void ProcessFirstMonotonicPartOfCubic(ProcessHandler* hnd,
jfloat* coords, jint* pixelInfo,
jfloat t)
{
jfloat coords1[8];
jfloat tx, ty;
coords1[0] = coords[0];
coords1[1] = coords[1];
tx = coords[2] + t*(coords[4] - coords[2]);
ty = coords[3] + t*(coords[5] - coords[3]);
coords1[2] = coords[0] + t*(coords[2] - coords[0]);
coords1[3] = coords[1] + t*(coords[3] - coords[1]);
coords1[4] = coords1[2] + t*(tx - coords1[2]);
coords1[5] = coords1[3] + t*(ty - coords1[3]);
coords[4] = coords[4] + t*(coords[6] - coords[4]);
coords[5] = coords[5] + t*(coords[7] - coords[5]);
coords[2] = tx + t*(coords[4] - tx);
coords[3] = ty + t*(coords[5] - ty);
coords[0]=coords1[6]=coords1[4] + t*(coords[2] - coords1[4]);
coords[1]=coords1[7]=coords1[5] + t*(coords[3] - coords1[5]);
ProcessMonotonicCubic(hnd, coords1, pixelInfo);
}
/*
* Split cubic curve into monotonic in X and Y parts. Calling ProcessCubic for
* each monotonic piece of the curve.
*
* Note: coords array could be changed
*/
static void ProcessCubic(ProcessHandler* hnd, jfloat* coords, jint* pixelInfo)
{
/* Temporary array for holding parameters corresponding to the extreme in X
* and Y points. The values are inside the (0,1) range (0 and 1 excluded)
* and in ascending order.
*/
double params[4];
jint cnt = 0, i;
/* Simple check for monotonicity in X before searching for the extreme
* points of the X(t) function. We first check if the curve is monotonic in
* X by seeing if all of the X coordinates are strongly ordered.
*/
if ((coords[0] > coords[2] || coords[2] > coords[4] ||
coords[4] > coords[6]) &&
(coords[0] < coords[2] || coords[2] < coords[4] ||
coords[4] < coords[6]))
{
/* Searching for extreme points of the X(t) function by solving
* dX(t)
* ---- = 0 equation
* dt
*/
double ax = -coords[0] + 3*coords[2] - 3*coords[4] + coords[6];
double bx = 2*(coords[0] - 2*coords[2] + coords[4]);
double cx = -coords[0] + coords[2];
SOLVEQUADINRANGE(ax,bx,cx,params,cnt);
}
/* Simple check for monotonicity in Y before searching for the extreme
* points of the Y(t) function. We first check if the curve is monotonic in
* Y by seeing if all of the Y coordinates are strongly ordered.
*/
if ((coords[1] > coords[3] || coords[3] > coords[5] ||
coords[5] > coords[7]) &&
(coords[1] < coords[3] || coords[3] < coords[5] ||
coords[5] < coords[7]))
{
/* Searching for extreme points of the Y(t) function by solving
* dY(t)
* ----- = 0 equation
* dt
*/
double ay = -coords[1] + 3*coords[3] - 3*coords[5] + coords[7];
double by = 2*(coords[1] - 2*coords[3] + coords[5]);
double cy = -coords[1] + coords[3];
SOLVEQUADINRANGE(ay,by,cy,params,cnt);
}
if (cnt > 0) {
/* Sorting parameter values corresponding to the extremum points of
* the curve. We are using insertion sort because of tiny size of the
* array.
*/
jint j;
for(i = 1; i < cnt; i++) {
double value = params[i];
for (j = i - 1; j >= 0 && params[j] > value; j--) {
params[j + 1] = params[j];
}
params[j + 1] = value;
}
/* Processing obtained monotonic parts */
ProcessFirstMonotonicPartOfCubic(hnd, coords, pixelInfo,
(jfloat)params[0]);
for (i = 1; i < cnt; i++) {
double param = params[i] - params[i-1];
if (param > 0) {
ProcessFirstMonotonicPartOfCubic(hnd, coords, pixelInfo,
/* Scale parameter to match with rest of the curve */
(float)(param/(1.0 - params[i - 1])));
}
}
}
ProcessMonotonicCubic(hnd,coords,pixelInfo);
}
static void ProcessLine(ProcessHandler* hnd,
jfloat *coord1, jfloat *coord2, jint* pixelInfo) {
jfloat xMin, yMin, xMax, yMax;
jint X1, Y1, X2, Y2, X3, Y3, res;
jboolean clipped = JNI_FALSE;
jfloat x1 = coord1[0];
jfloat y1 = coord1[1];
jfloat x2 = coord2[0];
jfloat y2 = coord2[1];
jfloat x3,y3;
jboolean lastClipped;
xMin = hnd->dhnd->xMinf;
yMin = hnd->dhnd->yMinf;
xMax = hnd->dhnd->xMaxf;
yMax = hnd->dhnd->yMaxf;
TESTANDCLIP(yMin, yMax, y1, x1, y2, x2, jfloat, res);
if (res == CRES_INVISIBLE) return;
clipped = IS_CLIPPED(res);
TESTANDCLIP(yMin, yMax, y2, x2, y1, x1, jfloat, res);
if (res == CRES_INVISIBLE) return;
lastClipped = IS_CLIPPED(res);
clipped = clipped || lastClipped;
if (hnd->clipMode == PH_MODE_DRAW_CLIP) {
TESTANDCLIP(xMin, xMax,
x1, y1, x2, y2, jfloat, res);
if (res == CRES_INVISIBLE) return;
clipped = clipped || IS_CLIPPED(res);
TESTANDCLIP(xMin, xMax,
x2, y2, x1, y1, jfloat, res);
if (res == CRES_INVISIBLE) return;
lastClipped = lastClipped || IS_CLIPPED(res);
clipped = clipped || lastClipped;
X1 = (jint)(x1*MDP_MULT);
Y1 = (jint)(y1*MDP_MULT);
X2 = (jint)(x2*MDP_MULT);
Y2 = (jint)(y2*MDP_MULT);
hnd->pProcessFixedLine(hnd, X1, Y1, X2, Y2, pixelInfo,
clipped, /* enable boundary checking in case
of clipping to avoid entering
out of bounds which could
happens during rounding
*/
lastClipped /* Notify pProcessFixedLine that
this is the end of the
subpath (because of exiting
out of boundaries)
*/
);
} else {
/* Clamping starting from first vertex of the processed segment
*/
CLIPCLAMP(xMin, xMax, x1, y1, x2, y2, x3, y3, jfloat, res);
X1 = (jint)(x1*MDP_MULT);
Y1 = (jint)(y1*MDP_MULT);
/* Clamping only by left boundary */
if (res == CRES_MIN_CLIPPED) {
X3 = (jint)(x3*MDP_MULT);
Y3 = (jint)(y3*MDP_MULT);
hnd->pProcessFixedLine(hnd, X3, Y3, X1, Y1, pixelInfo,
JNI_FALSE, lastClipped);
} else if (res == CRES_INVISIBLE) {
return;
}
/* Clamping starting from last vertex of the processed segment
*/
CLIPCLAMP(xMin, xMax, x2, y2, x1, y1, x3, y3, jfloat, res);
/* Checking if there was a clip by right boundary */
lastClipped = lastClipped || (res == CRES_MAX_CLIPPED);
X2 = (jint)(x2*MDP_MULT);
Y2 = (jint)(y2*MDP_MULT);
hnd->pProcessFixedLine(hnd, X1, Y1, X2, Y2, pixelInfo,
JNI_FALSE, lastClipped);
/* Clamping only by left boundary */
if (res == CRES_MIN_CLIPPED) {
X3 = (jint)(x3*MDP_MULT);
Y3 = (jint)(y3*MDP_MULT);
hnd->pProcessFixedLine(hnd, X2, Y2, X3, Y3, pixelInfo,
JNI_FALSE, lastClipped);
}
}
}
jboolean ProcessPath(ProcessHandler* hnd,
jfloat transXf, jfloat transYf,
jfloat* coords, jint maxCoords,
jbyte* types, jint numTypes)
{
jfloat tCoords[8];
jfloat closeCoord[2];
jint pixelInfo[5];
jboolean skip = JNI_FALSE;
jboolean subpathStarted = JNI_FALSE;
jfloat lastX, lastY;
int i, index = 0;
pixelInfo[0] = 0;
/* Adding support of the KEY_STROKE_CONTROL rendering hint.
* Now we are supporting two modes: "pixels at centers" and
* "pixels at corners".
* First one is disabled by default but could be enabled by setting
* VALUE_STROKE_PURE to the rendering hint. It means that pixel at the
* screen (x,y) has (x + 0.5, y + 0.5) float coordinates.
*
* Second one is enabled by default and means straightforward mapping
* (x,y) --> (x,y)
*
*/
if (hnd->stroke == PH_STROKE_PURE) {
closeCoord[0] = -0.5f;
closeCoord[1] = -0.5f;
transXf -= 0.5;
transYf -= 0.5;
} else {
closeCoord[0] = 0.0f;
closeCoord[1] = 0.0f;
}
/* Adjusting boundaries to the capabilities of the ProcessPath code */
ADJUST(hnd->dhnd->xMin, LOWER_OUT_BND, UPPER_OUT_BND);
ADJUST(hnd->dhnd->yMin, LOWER_OUT_BND, UPPER_OUT_BND);
ADJUST(hnd->dhnd->xMax, LOWER_OUT_BND, UPPER_OUT_BND);
ADJUST(hnd->dhnd->yMax, LOWER_OUT_BND, UPPER_OUT_BND);
/* Setting up fractional clipping box
*
* We are using following float -> int mapping:
*
* xi = floor(xf + 0.5)
*
* So, fractional values that hit the [xmin, xmax) integer interval will be
* situated inside the [xmin-0.5, xmax - 0.5) fractional interval. We are
* using EPSF constant to provide that upper boundary is not included.
*/
hnd->dhnd->xMinf = hnd->dhnd->xMin - 0.5f;
hnd->dhnd->yMinf = hnd->dhnd->yMin - 0.5f;
hnd->dhnd->xMaxf = hnd->dhnd->xMax - 0.5f - EPSF;
hnd->dhnd->yMaxf = hnd->dhnd->yMax - 0.5f - EPSF;
for (i = 0; i < numTypes; i++) {
switch (types[i]) {
case java_awt_geom_PathIterator_SEG_MOVETO:
if (index + 2 <= maxCoords) {
/* Performing closing of the unclosed segments */
if (subpathStarted & !skip) {
if (hnd->clipMode == PH_MODE_FILL_CLIP) {
if (tCoords[0] != closeCoord[0] ||
tCoords[1] != closeCoord[1])
{
ProcessLine(hnd, tCoords, closeCoord,
pixelInfo);
}
}
hnd->pProcessEndSubPath(hnd);
}
tCoords[0] = coords[index++] + transXf;
tCoords[1] = coords[index++] + transYf;
/* Checking SEG_MOVETO coordinates if they are out of the
* [LOWER_BND, UPPER_BND] range. This check also handles
* NaN and Infinity values. Skipping next path segment in
* case of invalid data.
*/
if (tCoords[0] < UPPER_BND &&
tCoords[0] > LOWER_BND &&
tCoords[1] < UPPER_BND &&
tCoords[1] > LOWER_BND)
{
subpathStarted = JNI_TRUE;
skip = JNI_FALSE;
closeCoord[0] = tCoords[0];
closeCoord[1] = tCoords[1];
} else {
skip = JNI_TRUE;
}
} else {
return JNI_FALSE;
}
break;
case java_awt_geom_PathIterator_SEG_LINETO:
if (index + 2 <= maxCoords) {
lastX = tCoords[2] = coords[index++] + transXf;
lastY = tCoords[3] = coords[index++] + transYf;
/* Checking SEG_LINETO coordinates if they are out of the
* [LOWER_BND, UPPER_BND] range. This check also handles
* NaN and Infinity values. Ignoring current path segment
* in case of invalid data. If segment is skipped its
* endpoint (if valid) is used to begin new subpath.
*/
if (lastX < UPPER_BND &&
lastX > LOWER_BND &&
lastY < UPPER_BND &&
lastY > LOWER_BND)
{
if (skip) {
tCoords[0] = closeCoord[0] = lastX;
tCoords[1] = closeCoord[1] = lastY;
subpathStarted = JNI_TRUE;
skip = JNI_FALSE;
} else {
ProcessLine(hnd, tCoords, tCoords + 2,
pixelInfo);
tCoords[0] = lastX;
tCoords[1] = lastY;
}
}
} else {
return JNI_FALSE;
}
break;
case java_awt_geom_PathIterator_SEG_QUADTO:
if (index + 4 <= maxCoords) {
tCoords[2] = coords[index++] + transXf;
tCoords[3] = coords[index++] + transYf;
lastX = tCoords[4] = coords[index++] + transXf;
lastY = tCoords[5] = coords[index++] + transYf;
/* Checking SEG_QUADTO coordinates if they are out of the
* [LOWER_BND, UPPER_BND] range. This check also handles
* NaN and Infinity values. Ignoring current path segment
* in case of invalid endpoints's data. Equivalent to
* the SEG_LINETO if endpoint coordinates are valid but
* there are invalid data among other coordinates
*/
if (lastX < UPPER_BND &&
lastX > LOWER_BND &&
lastY < UPPER_BND &&
lastY > LOWER_BND)
{
if (skip) {
tCoords[0] = closeCoord[0] = lastX;
tCoords[1] = closeCoord[1] = lastY;
subpathStarted = JNI_TRUE;
skip = JNI_FALSE;
} else {
if (tCoords[2] < UPPER_BND &&
tCoords[2] > LOWER_BND &&
tCoords[3] < UPPER_BND &&
tCoords[3] > LOWER_BND)
{
ProcessQuad(hnd, tCoords, pixelInfo);
} else {
ProcessLine(hnd, tCoords,
tCoords + 4, pixelInfo);
}
tCoords[0] = lastX;
tCoords[1] = lastY;
}
}
} else {
return JNI_FALSE;
}
break;
case java_awt_geom_PathIterator_SEG_CUBICTO:
if (index + 6 <= maxCoords) {
tCoords[2] = coords[index++] + transXf;
tCoords[3] = coords[index++] + transYf;
tCoords[4] = coords[index++] + transXf;
tCoords[5] = coords[index++] + transYf;
lastX = tCoords[6] = coords[index++] + transXf;
lastY = tCoords[7] = coords[index++] + transYf;
/* Checking SEG_CUBICTO coordinates if they are out of the
* [LOWER_BND, UPPER_BND] range. This check also handles
* NaN and Infinity values. Ignoring current path segment
* in case of invalid endpoints's data. Equivalent to
* the SEG_LINETO if endpoint coordinates are valid but
* there are invalid data among other coordinates
*/
if (lastX < UPPER_BND &&
lastX > LOWER_BND &&
lastY < UPPER_BND &&
lastY > LOWER_BND)
{
if (skip) {
tCoords[0] = closeCoord[0] = tCoords[6];
tCoords[1] = closeCoord[1] = tCoords[7];
subpathStarted = JNI_TRUE;
skip = JNI_FALSE;
} else {
if (tCoords[2] < UPPER_BND &&
tCoords[2] > LOWER_BND &&
tCoords[3] < UPPER_BND &&
tCoords[3] > LOWER_BND &&
tCoords[4] < UPPER_BND &&
tCoords[4] > LOWER_BND &&
tCoords[5] < UPPER_BND &&
tCoords[5] > LOWER_BND)
{
ProcessCubic(hnd, tCoords, pixelInfo);
} else {
ProcessLine(hnd, tCoords, tCoords + 6,
pixelInfo);
}
tCoords[0] = lastX;
tCoords[1] = lastY;
}
}
} else {
return JNI_FALSE;
}
break;
case java_awt_geom_PathIterator_SEG_CLOSE:
if (subpathStarted && !skip) {
skip = JNI_FALSE;
if (tCoords[0] != closeCoord[0] ||
tCoords[1] != closeCoord[1])
{
ProcessLine(hnd, tCoords, closeCoord, pixelInfo);
/* Storing last path's point for using in
* following segments without initial moveTo
*/
tCoords[0] = closeCoord[0];
tCoords[1] = closeCoord[1];
}
hnd->pProcessEndSubPath(hnd);
}
break;
}
}
/* Performing closing of the unclosed segments */
if (subpathStarted & !skip) {
if (hnd->clipMode == PH_MODE_FILL_CLIP) {
if (tCoords[0] != closeCoord[0] ||
tCoords[1] != closeCoord[1])
{
ProcessLine(hnd, tCoords, closeCoord,
pixelInfo);
}
}
hnd->pProcessEndSubPath(hnd);
}
return JNI_TRUE;
}
/* TODO Add checking of the result of the malloc/realloc functions to handle
* out of memory error and don't leak earlier allocated data
*/
#define ALLOC(ptr, type, n) \
ptr = (type *)malloc((n)*sizeof(type))
#define REALLOC(ptr, type, n) \
ptr = (type *)realloc(ptr, (n)*sizeof(type))
struct _Edge;
typedef struct _Point {
jint x;
jint y;
jboolean lastPoint;
struct _Point* prev;
struct _Point* next;
struct _Point* nextByY;
jboolean endSL;
struct _Edge* edge;
} Point;
typedef struct _Edge {
jint x;
jint dx;
Point* p;
jint dir;
struct _Edge* prev;
struct _Edge* next;
} Edge;
/* Size of the default buffer in the FillData structure. This buffer is
* replaced with heap allocated in case of large paths.
*/
#define DF_MAX_POINT 256
/* Following structure accumulates points of the non-continuous flattened
* path during iteration through the origin path's segments . The end
* of the each subpath is marked as lastPoint flag set at the last point
*/
typedef struct {
Point *plgPnts;
Point dfPlgPnts[DF_MAX_POINT];
jint plgSize;
jint plgMax;
jint plgYMin;
jint plgYMax;
} FillData;
#define FD_INIT(PTR) \
do { \
(PTR)->plgPnts = (PTR)->dfPlgPnts; \
(PTR)->plgSize = 0; \
(PTR)->plgMax = DF_MAX_POINT; \
} while(0)
#define FD_ADD_POINT(PTR, X, Y, LASTPT) \
do { \
Point* _pnts = (PTR)->plgPnts; \
jint _size = (PTR)->plgSize; \
if (_size >= (PTR)->plgMax) { \
jint newMax = (PTR)->plgMax*2; \
if ((PTR)->plgPnts == (PTR)->dfPlgPnts) { \
(PTR)->plgPnts = (Point*)malloc(newMax*sizeof(Point)); \
memcpy((PTR)->plgPnts, _pnts, _size*sizeof(Point)); \
} else { \
(PTR)->plgPnts = (Point*)realloc( \
_pnts, newMax*sizeof(Point)); \
} \
_pnts = (PTR)->plgPnts; \
(PTR)->plgMax = newMax; \
} \
_pnts += _size; \
_pnts->x = X; \
_pnts->y = Y; \
_pnts->lastPoint = LASTPT; \
if (_size) { \
if ((PTR)->plgYMin > Y) (PTR)->plgYMin = Y; \
if ((PTR)->plgYMax < Y) (PTR)->plgYMax = Y; \
} else { \
(PTR)->plgYMin = Y; \
(PTR)->plgYMax = Y; \
} \
(PTR)->plgSize = _size + 1; \
} while(0)
#define FD_FREE_POINTS(PTR) \
do { \
if ((PTR)->plgPnts != (PTR)->dfPlgPnts) { \
free((PTR)->plgPnts); \
} \
} while(0)
#define FD_IS_EMPTY(PTR) (!((PTR)->plgSize))
#define FD_IS_ENDED(PTR) ((PTR)->plgPnts[(PTR)->plgSize - 1].lastPoint)
#define FD_SET_ENDED(PTR) \
do { \
(PTR)->plgPnts[(PTR)->plgSize - 1].lastPoint = JNI_TRUE; \
} while(0)
#define PFD(HND) ((FillData*)(HND)->pData)
/* Bubble sorting in the ascending order of the linked list. This
* implementation stops processing the list if there were no changes during the
* previous pass.
*
* LIST - ptr to the ptr to the first element of the list
* ETYPE - type of the element in the list
* NEXT - accessor to the next field in the list element
* GET_LKEY - accessor to the key of the list element
*/
#define LBUBBLE_SORT(LIST, ETYPE, NEXT, GET_LKEY) \
do { \
ETYPE *p, *q, *r, *s = NULL, *temp ; \
jint wasSwap = 1; \
/* r precedes p and s points to the node up to which comparisons \
* are to be made */ \
while ( s != NEXT(*LIST) && wasSwap) { \
r = p = *LIST; \
q = NEXT(p); \
wasSwap = 0; \
while ( p != s ) { \
if (GET_LKEY(p) >= GET_LKEY(q)) { \
wasSwap = 1; \
if ( p == *LIST ) { \
temp = NEXT(q); \
NEXT(q) = p ; \
NEXT(p) = temp ; \
*LIST = q ; \
r = q ; \
} else { \
temp = NEXT(q); \
NEXT(q) = p ; \
NEXT(p) = temp ; \
NEXT(r) = q ; \
r = q ; \
} \
} else { \
r = p ; \
p = NEXT(p); \
} \
q = NEXT(p); \
if ( q == s ) s = p ; \
} \
} \
} while(0);
/* Accessors for the Edge structure to work with LBUBBLE_SORT */
#define GET_ACTIVE_KEY(a) (a->x)
#define GET_ACTIVE_NEXT(a) ((a)->next)
/* TODO: Implement stack/heap allocation technique for active edges
*/
#define DELETE_ACTIVE(head,pnt) \
do { \
Edge *prevp = pnt->prev; \
Edge *nextp = pnt->next; \
if (prevp) { \
prevp->next = nextp; \
} else { \
head = nextp; \
} \
if (nextp) { \
nextp->prev = prevp; \
} \
} while(0);
#define INSERT_ACTIVE(head,pnt,cy) \
do { \
Point *np = pnt->next; \
Edge *ne = active + nact; \
if (pnt->y == np->y) { \
/* Skipping horizontal segments */ \
break; \
} else { \
jint dX = np->x - pnt->x; \
jint dY = np->y - pnt->y; \
jint dy; \
if (pnt->y < np->y) { \
ne->dir = -1; \
ne->p = pnt; \
ne->x = pnt->x; \
dy = cy - pnt->y; \
} else { /* pnt->y > np->y */ \
ne->dir = 1; \
ne->p = np; \
ne->x = np->x; \
dy = cy - np->y; \
} \
\
/* We need to worry only about dX because dY is in */\
/* denominator and abs(dy) < MDP_MULT (cy is a first */\
/* scanline of the scan converted segment and we subtract */\
/* y coordinate of the nearest segment's end from it to */\
/* obtain dy) */\
if (ABS32(dX) > CALC_BND) { \
ne->dx = (jint)((((jdouble)dX)*MDP_MULT)/dY); \
ne->x += (jint)((((jdouble)dX)*dy)/dY); \
} else { \
ne->dx = ((dX)<<MDP_PREC)/dY; \
ne->x += (dX*dy)/dY; \
} \
} \
ne->next = head; \
ne->prev = NULL; \
if (head) { \
head->prev = ne; \
} \
head = active + nact; \
pnt->edge = head; \
nact++; \
} while(0);
void FillPolygon(ProcessHandler* hnd,
jint fillRule) {
jint k, y, xl, xr;
jint drawing;
Edge* activeList, *active;
Edge* curEdge, *prevEdge;
jint nact;
jint n;
Point* pt, *curpt, *ept;
Point** yHash;
Point** curHash;
jint rightBnd = hnd->dhnd->xMax - 1;
FillData* pfd = (FillData*)(hnd->pData);
jint yMin = pfd->plgYMin;
jint yMax = pfd->plgYMax;
jint hashSize = ((yMax - yMin)>>MDP_PREC) + 4;
/* Because of support of the KEY_STROKE_CONTROL hint we are performing
* shift of the coordinates at the higher level
*/
jint hashOffset = ((yMin - 1) & MDP_W_MASK);
// TODO creating lists using fake first element to avoid special casing of
// the first element in the list (which otherwise should be performed in each
// list operation)
/* Winding counter */
jint counter;
/* Calculating mask to be applied to the winding counter */
jint counterMask =
(fillRule == java_awt_geom_PathIterator_WIND_NON_ZERO)? -1:1;
pt = pfd->plgPnts;
n = pfd->plgSize;
if (n <=1) return;
ALLOC(yHash, Point*, hashSize);
for (k = 0; k < hashSize; k++) {
yHash[k] = NULL;
}
ALLOC(active, Edge, n);
/* Creating double linked list (prev, next links) describing path order and
* hash table with points which fall between scanlines. nextByY link is
* used for the points which are between same scanlines. Scanlines are
* passed through the centers of the pixels.
*/
curpt = pt;
curpt->prev = NULL;
ept = pt + n - 1;
for (curpt = pt; curpt != ept; curpt++) {
Point* nextpt = curpt + 1;
curHash = yHash + ((curpt->y - hashOffset - 1) >> MDP_PREC);
curpt->nextByY = *curHash;
*curHash = curpt;
curpt->next = nextpt;
nextpt->prev = curpt;
curpt->edge = NULL;
}
curHash = yHash + ((ept->y - hashOffset - 1) >> MDP_PREC);
ept->nextByY = *curHash;
*curHash = ept;
ept->next = NULL;
ept->edge = NULL;
nact = 0;
activeList = NULL;
for (y=hashOffset + MDP_MULT,k = 0;
y<=yMax && k < hashSize; y += MDP_MULT, k++)
{
for(pt = yHash[k];pt; pt=pt->nextByY) {
/* pt->y should be inside hashed interval
* assert(y-MDP_MULT <= pt->y && pt->y < y);
*/
if (pt->prev && !pt->prev->lastPoint) {
if (pt->prev->edge && pt->prev->y <= y) {
DELETE_ACTIVE(activeList, pt->prev->edge);
pt->prev->edge = NULL;
} else if (pt->prev->y > y) {
INSERT_ACTIVE(activeList, pt->prev, y);
}
}
if (!pt->lastPoint && pt->next) {
if (pt->edge && pt->next->y <= y) {
DELETE_ACTIVE(activeList, pt->edge);
pt->edge = NULL;
} else if (pt->next->y > y) {
INSERT_ACTIVE(activeList, pt, y);
}
}
}
if (!activeList) continue;
/* We could not use O(N) Radix sort here because in most cases list of
* edges almost sorted. So, bubble sort (O(N^2))is working much
* better. Note, in case of array of edges Shell sort is more
* efficient.
*/
LBUBBLE_SORT((&activeList), Edge, GET_ACTIVE_NEXT, GET_ACTIVE_KEY);
/* Correction of the back links in the double linked edge list */
curEdge=activeList;
prevEdge = NULL;
while (curEdge) {
curEdge->prev = prevEdge;
prevEdge = curEdge;
curEdge = curEdge->next;
}
xl = xr = hnd->dhnd->xMin;
curEdge = activeList;
counter = 0;
drawing = 0;
for(;curEdge; curEdge = curEdge->next) {
counter += curEdge->dir;
if ((counter & counterMask) && !drawing) {
xl = (curEdge->x + MDP_MULT - 1)>>MDP_PREC;
drawing = 1;
}
if (!(counter & counterMask) && drawing) {
xr = (curEdge->x - 1)>>MDP_PREC;
if (xl <= xr) {
hnd->dhnd->pDrawScanline(hnd->dhnd, xl, xr, y >> MDP_PREC);
}
drawing = 0;
}
curEdge->x += curEdge->dx;
}
/* Performing drawing till the right boundary (for correct rendering
* shapes clipped at the right side)
*/
if (drawing && xl <= rightBnd) {
hnd->dhnd->pDrawScanline(hnd->dhnd, xl, rightBnd, y >> MDP_PREC);
}
}
free(active);
free(yHash);
}
void StoreFixedLine(ProcessHandler* hnd,jint x1,jint y1,jint x2,jint y2,
jint* pixelInfo,jboolean checkBounds,
jboolean endSubPath) {
FillData* pfd;
jint outXMin, outXMax, outYMin, outYMax;
jint x3, y3, res;
/* There is no need to round line coordinates to the forward differencing
* precision anymore. Such a rounding was used for preventing the curve go
* out the endpoint (this sometimes does not help). The problem was fixed
* in the forward differencing loops.
*/
if (checkBounds) {
jboolean lastClipped = JNI_FALSE;
/* This function is used only for filling shapes, so there is no
* check for the type of clipping
*/
outXMin = (jint)(hnd->dhnd->xMinf * MDP_MULT);
outXMax = (jint)(hnd->dhnd->xMaxf * MDP_MULT);
outYMin = (jint)(hnd->dhnd->yMinf * MDP_MULT);
outYMax = (jint)(hnd->dhnd->yMaxf * MDP_MULT);
TESTANDCLIP(outYMin, outYMax, y1, x1, y2, x2, jint, res);
if (res == CRES_INVISIBLE) return;
TESTANDCLIP(outYMin, outYMax, y2, x2, y1, x1, jint, res);
if (res == CRES_INVISIBLE) return;
lastClipped = IS_CLIPPED(res);
/* Clamping starting from first vertex of the processed segment */
CLIPCLAMP(outXMin, outXMax, x1, y1, x2, y2, x3, y3, jint, res);
/* Clamping only by left boundary */
if (res == CRES_MIN_CLIPPED) {
StoreFixedLine(hnd, x3, y3, x1, y1, pixelInfo,
JNI_FALSE, lastClipped);
} else if (res == CRES_INVISIBLE) {
return;
}
/* Clamping starting from last vertex of the processed segment */
CLIPCLAMP(outXMin, outXMax, x2, y2, x1, y1, x3, y3, jint, res);
/* Checking if there was a clip by right boundary */
lastClipped = lastClipped || (res == CRES_MAX_CLIPPED);
StoreFixedLine(hnd, x1, y1, x2, y2, pixelInfo,
JNI_FALSE, lastClipped);
/* Clamping only by left boundary */
if (res == CRES_MIN_CLIPPED) {
StoreFixedLine(hnd, x2, y2, x3, y3, pixelInfo,
JNI_FALSE, lastClipped);
}
return;
}
pfd = (FillData*)(hnd->pData);
/* Adding first point of the line only in case of empty or just finished
* path
*/
if (FD_IS_EMPTY(pfd) || FD_IS_ENDED(pfd)) {
FD_ADD_POINT(pfd, x1, y1, JNI_FALSE);
}
FD_ADD_POINT(pfd, x2, y2, JNI_FALSE);
if (endSubPath) {
FD_SET_ENDED(pfd);
}
}
static void endSubPath(ProcessHandler* hnd) {
FillData* pfd = (FillData*)(hnd->pData);
if (!FD_IS_EMPTY(pfd)) {
FD_SET_ENDED(pfd);
}
}
static void stubEndSubPath(ProcessHandler* hnd) {
}
JNIEXPORT jboolean JNICALL
doFillPath(DrawHandler* dhnd,
jint transX, jint transY,
jfloat* coords, jint maxCoords,
jbyte* types, jint numTypes,
PHStroke stroke, jint fillRule)
{
jint res;
FillData fillData;
ProcessHandler hnd =
{
&StoreFixedLine,
&endSubPath,
NULL,
PH_STROKE_DEFAULT,
PH_MODE_FILL_CLIP,
NULL
};
/* Initialization of the following fields in the declaration of the hnd
* above causes warnings on sun studio compiler with -xc99=%none option
* applied (this option means compliance with C90 standard instead of C99)
*/
hnd.dhnd = dhnd;
hnd.pData = &fillData;
hnd.stroke = stroke;
FD_INIT(&fillData);
res = ProcessPath(&hnd, (jfloat)transX, (jfloat)transY,
coords, maxCoords, types, numTypes);
if (!res) {
FD_FREE_POINTS(&fillData);
return JNI_FALSE;
}
FillPolygon(&hnd, fillRule);
FD_FREE_POINTS(&fillData);
return JNI_TRUE;
}
JNIEXPORT jboolean JNICALL
doDrawPath(DrawHandler* dhnd,
void (*pProcessEndSubPath)(ProcessHandler*),
jint transX, jint transY,
jfloat* coords, jint maxCoords,
jbyte* types, jint numTypes, PHStroke stroke)
{
ProcessHandler hnd =
{
&ProcessFixedLine,
NULL,
NULL,
PH_STROKE_DEFAULT,
PH_MODE_DRAW_CLIP,
NULL
};
/* Initialization of the following fields in the declaration of the hnd
* above causes warnings on sun studio compiler with -xc99=%none option
* applied (this option means compliance with C90 standard instead of C99)
*/
hnd.dhnd = dhnd;
hnd.stroke = stroke;
hnd.pProcessEndSubPath = (pProcessEndSubPath == NULL)?
stubEndSubPath : pProcessEndSubPath;
return ProcessPath(&hnd, (jfloat)transX, (jfloat)transY, coords, maxCoords,
types, numTypes);
}