|  | /* | 
|  | * Copyright 2012 Google Inc. | 
|  | * | 
|  | * Use of this source code is governed by a BSD-style license that can be | 
|  | * found in the LICENSE file. | 
|  | */ | 
|  | #include "SkLineParameters.h" | 
|  | #include "SkPathOpsCubic.h" | 
|  | #include "SkPathOpsLine.h" | 
|  | #include "SkPathOpsQuad.h" | 
|  | #include "SkPathOpsRect.h" | 
|  | #include "SkTSort.h" | 
|  |  | 
|  | const int SkDCubic::gPrecisionUnit = 256;  // FIXME: test different values in test framework | 
|  |  | 
|  | // give up when changing t no longer moves point | 
|  | // also, copy point rather than recompute it when it does change | 
|  | double SkDCubic::binarySearch(double min, double max, double axisIntercept, | 
|  | SearchAxis xAxis) const { | 
|  | double t = (min + max) / 2; | 
|  | double step = (t - min) / 2; | 
|  | SkDPoint cubicAtT = ptAtT(t); | 
|  | double calcPos = (&cubicAtT.fX)[xAxis]; | 
|  | double calcDist = calcPos - axisIntercept; | 
|  | do { | 
|  | double priorT = t - step; | 
|  | SkASSERT(priorT >= min); | 
|  | SkDPoint lessPt = ptAtT(priorT); | 
|  | if (approximately_equal(lessPt.fX, cubicAtT.fX) | 
|  | && approximately_equal(lessPt.fY, cubicAtT.fY)) { | 
|  | return -1;  // binary search found no point at this axis intercept | 
|  | } | 
|  | double lessDist = (&lessPt.fX)[xAxis] - axisIntercept; | 
|  | #if DEBUG_CUBIC_BINARY_SEARCH | 
|  | SkDebugf("t=%1.9g calc=%1.9g dist=%1.9g step=%1.9g less=%1.9g\n", t, calcPos, calcDist, | 
|  | step, lessDist); | 
|  | #endif | 
|  | double lastStep = step; | 
|  | step /= 2; | 
|  | if (calcDist > 0 ? calcDist > lessDist : calcDist < lessDist) { | 
|  | t = priorT; | 
|  | } else { | 
|  | double nextT = t + lastStep; | 
|  | SkASSERT(nextT <= max); | 
|  | SkDPoint morePt = ptAtT(nextT); | 
|  | if (approximately_equal(morePt.fX, cubicAtT.fX) | 
|  | && approximately_equal(morePt.fY, cubicAtT.fY)) { | 
|  | return -1;  // binary search found no point at this axis intercept | 
|  | } | 
|  | double moreDist = (&morePt.fX)[xAxis] - axisIntercept; | 
|  | if (calcDist > 0 ? calcDist <= moreDist : calcDist >= moreDist) { | 
|  | continue; | 
|  | } | 
|  | t = nextT; | 
|  | } | 
|  | SkDPoint testAtT = ptAtT(t); | 
|  | cubicAtT = testAtT; | 
|  | calcPos = (&cubicAtT.fX)[xAxis]; | 
|  | calcDist = calcPos - axisIntercept; | 
|  | } while (!approximately_equal(calcPos, axisIntercept)); | 
|  | return t; | 
|  | } | 
|  |  | 
|  | // FIXME: cache keep the bounds and/or precision with the caller? | 
|  | double SkDCubic::calcPrecision() const { | 
|  | SkDRect dRect; | 
|  | dRect.setBounds(*this);  // OPTIMIZATION: just use setRawBounds ? | 
|  | double width = dRect.fRight - dRect.fLeft; | 
|  | double height = dRect.fBottom - dRect.fTop; | 
|  | return (width > height ? width : height) / gPrecisionUnit; | 
|  | } | 
|  |  | 
|  | bool SkDCubic::clockwise() const { | 
|  | double sum = (fPts[0].fX - fPts[3].fX) * (fPts[0].fY + fPts[3].fY); | 
|  | for (int idx = 0; idx < 3; ++idx) { | 
|  | sum += (fPts[idx + 1].fX - fPts[idx].fX) * (fPts[idx + 1].fY + fPts[idx].fY); | 
|  | } | 
|  | return sum <= 0; | 
|  | } | 
|  |  | 
|  | void SkDCubic::Coefficients(const double* src, double* A, double* B, double* C, double* D) { | 
|  | *A = src[6];  // d | 
|  | *B = src[4] * 3;  // 3*c | 
|  | *C = src[2] * 3;  // 3*b | 
|  | *D = src[0];  // a | 
|  | *A -= *D - *C + *B;     // A =   -a + 3*b - 3*c + d | 
|  | *B += 3 * *D - 2 * *C;  // B =  3*a - 6*b + 3*c | 
|  | *C -= 3 * *D;           // C = -3*a + 3*b | 
|  | } | 
|  |  | 
|  | bool SkDCubic::controlsContainedByEnds() const { | 
|  | SkDVector startTan = fPts[1] - fPts[0]; | 
|  | if (startTan.fX == 0 && startTan.fY == 0) { | 
|  | startTan = fPts[2] - fPts[0]; | 
|  | } | 
|  | SkDVector endTan = fPts[2] - fPts[3]; | 
|  | if (endTan.fX == 0 && endTan.fY == 0) { | 
|  | endTan = fPts[1] - fPts[3]; | 
|  | } | 
|  | if (startTan.dot(endTan) >= 0) { | 
|  | return false; | 
|  | } | 
|  | SkDLine startEdge = {{fPts[0], fPts[0]}}; | 
|  | startEdge[1].fX -= startTan.fY; | 
|  | startEdge[1].fY += startTan.fX; | 
|  | SkDLine endEdge = {{fPts[3], fPts[3]}}; | 
|  | endEdge[1].fX -= endTan.fY; | 
|  | endEdge[1].fY += endTan.fX; | 
|  | double leftStart1 = startEdge.isLeft(fPts[1]); | 
|  | if (leftStart1 * startEdge.isLeft(fPts[2]) < 0) { | 
|  | return false; | 
|  | } | 
|  | double leftEnd1 = endEdge.isLeft(fPts[1]); | 
|  | if (leftEnd1 * endEdge.isLeft(fPts[2]) < 0) { | 
|  | return false; | 
|  | } | 
|  | return leftStart1 * leftEnd1 >= 0; | 
|  | } | 
|  |  | 
|  | bool SkDCubic::endsAreExtremaInXOrY() const { | 
|  | return (between(fPts[0].fX, fPts[1].fX, fPts[3].fX) | 
|  | && between(fPts[0].fX, fPts[2].fX, fPts[3].fX)) | 
|  | || (between(fPts[0].fY, fPts[1].fY, fPts[3].fY) | 
|  | && between(fPts[0].fY, fPts[2].fY, fPts[3].fY)); | 
|  | } | 
|  |  | 
|  | bool SkDCubic::isLinear(int startIndex, int endIndex) const { | 
|  | SkLineParameters lineParameters; | 
|  | lineParameters.cubicEndPoints(*this, startIndex, endIndex); | 
|  | // FIXME: maybe it's possible to avoid this and compare non-normalized | 
|  | lineParameters.normalize(); | 
|  | double distance = lineParameters.controlPtDistance(*this, 1); | 
|  | if (!approximately_zero(distance)) { | 
|  | return false; | 
|  | } | 
|  | distance = lineParameters.controlPtDistance(*this, 2); | 
|  | return approximately_zero(distance); | 
|  | } | 
|  |  | 
|  | bool SkDCubic::monotonicInY() const { | 
|  | return between(fPts[0].fY, fPts[1].fY, fPts[3].fY) | 
|  | && between(fPts[0].fY, fPts[2].fY, fPts[3].fY); | 
|  | } | 
|  |  | 
|  | int SkDCubic::searchRoots(double extremeTs[6], int extrema, double axisIntercept, | 
|  | SearchAxis xAxis, double* validRoots) const { | 
|  | extrema += findInflections(&extremeTs[extrema]); | 
|  | extremeTs[extrema++] = 0; | 
|  | extremeTs[extrema] = 1; | 
|  | SkTQSort(extremeTs, extremeTs + extrema); | 
|  | int validCount = 0; | 
|  | for (int index = 0; index < extrema; ) { | 
|  | double min = extremeTs[index]; | 
|  | double max = extremeTs[++index]; | 
|  | if (min == max) { | 
|  | continue; | 
|  | } | 
|  | double newT = binarySearch(min, max, axisIntercept, xAxis); | 
|  | if (newT >= 0) { | 
|  | validRoots[validCount++] = newT; | 
|  | } | 
|  | } | 
|  | return validCount; | 
|  | } | 
|  |  | 
|  | bool SkDCubic::serpentine() const { | 
|  | #if 0  // FIXME: enabling this fixes cubicOp114 but breaks cubicOp58d and cubicOp53d | 
|  | double tValues[2]; | 
|  | // OPTIMIZATION : another case where caching the present of cubic inflections would be useful | 
|  | return findInflections(tValues) > 1; | 
|  | #endif | 
|  | if (!controlsContainedByEnds()) { | 
|  | return false; | 
|  | } | 
|  | double wiggle = (fPts[0].fX - fPts[2].fX) * (fPts[0].fY + fPts[2].fY); | 
|  | for (int idx = 0; idx < 2; ++idx) { | 
|  | wiggle += (fPts[idx + 1].fX - fPts[idx].fX) * (fPts[idx + 1].fY + fPts[idx].fY); | 
|  | } | 
|  | double waggle = (fPts[1].fX - fPts[3].fX) * (fPts[1].fY + fPts[3].fY); | 
|  | for (int idx = 1; idx < 3; ++idx) { | 
|  | waggle += (fPts[idx + 1].fX - fPts[idx].fX) * (fPts[idx + 1].fY + fPts[idx].fY); | 
|  | } | 
|  | return wiggle * waggle < 0; | 
|  | } | 
|  |  | 
|  | // cubic roots | 
|  |  | 
|  | static const double PI = 3.141592653589793; | 
|  |  | 
|  | // from SkGeometry.cpp (and Numeric Solutions, 5.6) | 
|  | int SkDCubic::RootsValidT(double A, double B, double C, double D, double t[3]) { | 
|  | double s[3]; | 
|  | int realRoots = RootsReal(A, B, C, D, s); | 
|  | int foundRoots = SkDQuad::AddValidTs(s, realRoots, t); | 
|  | return foundRoots; | 
|  | } | 
|  |  | 
|  | int SkDCubic::RootsReal(double A, double B, double C, double D, double s[3]) { | 
|  | #ifdef SK_DEBUG | 
|  | // create a string mathematica understands | 
|  | // GDB set print repe 15 # if repeated digits is a bother | 
|  | //     set print elements 400 # if line doesn't fit | 
|  | char str[1024]; | 
|  | sk_bzero(str, sizeof(str)); | 
|  | SK_SNPRINTF(str, sizeof(str), "Solve[%1.19g x^3 + %1.19g x^2 + %1.19g x + %1.19g == 0, x]", | 
|  | A, B, C, D); | 
|  | SkPathOpsDebug::MathematicaIze(str, sizeof(str)); | 
|  | #if ONE_OFF_DEBUG && ONE_OFF_DEBUG_MATHEMATICA | 
|  | SkDebugf("%s\n", str); | 
|  | #endif | 
|  | #endif | 
|  | if (approximately_zero(A) | 
|  | && approximately_zero_when_compared_to(A, B) | 
|  | && approximately_zero_when_compared_to(A, C) | 
|  | && approximately_zero_when_compared_to(A, D)) {  // we're just a quadratic | 
|  | return SkDQuad::RootsReal(B, C, D, s); | 
|  | } | 
|  | if (approximately_zero_when_compared_to(D, A) | 
|  | && approximately_zero_when_compared_to(D, B) | 
|  | && approximately_zero_when_compared_to(D, C)) {  // 0 is one root | 
|  | int num = SkDQuad::RootsReal(A, B, C, s); | 
|  | for (int i = 0; i < num; ++i) { | 
|  | if (approximately_zero(s[i])) { | 
|  | return num; | 
|  | } | 
|  | } | 
|  | s[num++] = 0; | 
|  | return num; | 
|  | } | 
|  | if (approximately_zero(A + B + C + D)) {  // 1 is one root | 
|  | int num = SkDQuad::RootsReal(A, A + B, -D, s); | 
|  | for (int i = 0; i < num; ++i) { | 
|  | if (AlmostDequalUlps(s[i], 1)) { | 
|  | return num; | 
|  | } | 
|  | } | 
|  | s[num++] = 1; | 
|  | return num; | 
|  | } | 
|  | double a, b, c; | 
|  | { | 
|  | double invA = 1 / A; | 
|  | a = B * invA; | 
|  | b = C * invA; | 
|  | c = D * invA; | 
|  | } | 
|  | double a2 = a * a; | 
|  | double Q = (a2 - b * 3) / 9; | 
|  | double R = (2 * a2 * a - 9 * a * b + 27 * c) / 54; | 
|  | double R2 = R * R; | 
|  | double Q3 = Q * Q * Q; | 
|  | double R2MinusQ3 = R2 - Q3; | 
|  | double adiv3 = a / 3; | 
|  | double r; | 
|  | double* roots = s; | 
|  | if (R2MinusQ3 < 0) {   // we have 3 real roots | 
|  | double theta = acos(R / sqrt(Q3)); | 
|  | double neg2RootQ = -2 * sqrt(Q); | 
|  |  | 
|  | r = neg2RootQ * cos(theta / 3) - adiv3; | 
|  | *roots++ = r; | 
|  |  | 
|  | r = neg2RootQ * cos((theta + 2 * PI) / 3) - adiv3; | 
|  | if (!AlmostDequalUlps(s[0], r)) { | 
|  | *roots++ = r; | 
|  | } | 
|  | r = neg2RootQ * cos((theta - 2 * PI) / 3) - adiv3; | 
|  | if (!AlmostDequalUlps(s[0], r) && (roots - s == 1 || !AlmostDequalUlps(s[1], r))) { | 
|  | *roots++ = r; | 
|  | } | 
|  | } else {  // we have 1 real root | 
|  | double sqrtR2MinusQ3 = sqrt(R2MinusQ3); | 
|  | double A = fabs(R) + sqrtR2MinusQ3; | 
|  | A = SkDCubeRoot(A); | 
|  | if (R > 0) { | 
|  | A = -A; | 
|  | } | 
|  | if (A != 0) { | 
|  | A += Q / A; | 
|  | } | 
|  | r = A - adiv3; | 
|  | *roots++ = r; | 
|  | if (AlmostDequalUlps((double) R2, (double) Q3)) { | 
|  | r = -A / 2 - adiv3; | 
|  | if (!AlmostDequalUlps(s[0], r)) { | 
|  | *roots++ = r; | 
|  | } | 
|  | } | 
|  | } | 
|  | return static_cast<int>(roots - s); | 
|  | } | 
|  |  | 
|  | // from http://www.cs.sunysb.edu/~qin/courses/geometry/4.pdf | 
|  | // c(t)  = a(1-t)^3 + 3bt(1-t)^2 + 3c(1-t)t^2 + dt^3 | 
|  | // c'(t) = -3a(1-t)^2 + 3b((1-t)^2 - 2t(1-t)) + 3c(2t(1-t) - t^2) + 3dt^2 | 
|  | //       = 3(b-a)(1-t)^2 + 6(c-b)t(1-t) + 3(d-c)t^2 | 
|  | static double derivative_at_t(const double* src, double t) { | 
|  | double one_t = 1 - t; | 
|  | double a = src[0]; | 
|  | double b = src[2]; | 
|  | double c = src[4]; | 
|  | double d = src[6]; | 
|  | return 3 * ((b - a) * one_t * one_t + 2 * (c - b) * t * one_t + (d - c) * t * t); | 
|  | } | 
|  |  | 
|  | // OPTIMIZE? compute t^2, t(1-t), and (1-t)^2 and pass them to another version of derivative at t? | 
|  | SkDVector SkDCubic::dxdyAtT(double t) const { | 
|  | SkDVector result = { derivative_at_t(&fPts[0].fX, t), derivative_at_t(&fPts[0].fY, t) }; | 
|  | return result; | 
|  | } | 
|  |  | 
|  | // OPTIMIZE? share code with formulate_F1DotF2 | 
|  | int SkDCubic::findInflections(double tValues[]) const { | 
|  | double Ax = fPts[1].fX - fPts[0].fX; | 
|  | double Ay = fPts[1].fY - fPts[0].fY; | 
|  | double Bx = fPts[2].fX - 2 * fPts[1].fX + fPts[0].fX; | 
|  | double By = fPts[2].fY - 2 * fPts[1].fY + fPts[0].fY; | 
|  | double Cx = fPts[3].fX + 3 * (fPts[1].fX - fPts[2].fX) - fPts[0].fX; | 
|  | double Cy = fPts[3].fY + 3 * (fPts[1].fY - fPts[2].fY) - fPts[0].fY; | 
|  | return SkDQuad::RootsValidT(Bx * Cy - By * Cx, Ax * Cy - Ay * Cx, Ax * By - Ay * Bx, tValues); | 
|  | } | 
|  |  | 
|  | static void formulate_F1DotF2(const double src[], double coeff[4]) { | 
|  | double a = src[2] - src[0]; | 
|  | double b = src[4] - 2 * src[2] + src[0]; | 
|  | double c = src[6] + 3 * (src[2] - src[4]) - src[0]; | 
|  | coeff[0] = c * c; | 
|  | coeff[1] = 3 * b * c; | 
|  | coeff[2] = 2 * b * b + c * a; | 
|  | coeff[3] = a * b; | 
|  | } | 
|  |  | 
|  | /** SkDCubic'(t) = At^2 + Bt + C, where | 
|  | A = 3(-a + 3(b - c) + d) | 
|  | B = 6(a - 2b + c) | 
|  | C = 3(b - a) | 
|  | Solve for t, keeping only those that fit between 0 < t < 1 | 
|  | */ | 
|  | int SkDCubic::FindExtrema(double a, double b, double c, double d, double tValues[2]) { | 
|  | // we divide A,B,C by 3 to simplify | 
|  | double A = d - a + 3*(b - c); | 
|  | double B = 2*(a - b - b + c); | 
|  | double C = b - a; | 
|  |  | 
|  | return SkDQuad::RootsValidT(A, B, C, tValues); | 
|  | } | 
|  |  | 
|  | /*  from SkGeometry.cpp | 
|  | Looking for F' dot F'' == 0 | 
|  |  | 
|  | A = b - a | 
|  | B = c - 2b + a | 
|  | C = d - 3c + 3b - a | 
|  |  | 
|  | F' = 3Ct^2 + 6Bt + 3A | 
|  | F'' = 6Ct + 6B | 
|  |  | 
|  | F' dot F'' -> CCt^3 + 3BCt^2 + (2BB + CA)t + AB | 
|  | */ | 
|  | int SkDCubic::findMaxCurvature(double tValues[]) const { | 
|  | double coeffX[4], coeffY[4]; | 
|  | int i; | 
|  | formulate_F1DotF2(&fPts[0].fX, coeffX); | 
|  | formulate_F1DotF2(&fPts[0].fY, coeffY); | 
|  | for (i = 0; i < 4; i++) { | 
|  | coeffX[i] = coeffX[i] + coeffY[i]; | 
|  | } | 
|  | return RootsValidT(coeffX[0], coeffX[1], coeffX[2], coeffX[3], tValues); | 
|  | } | 
|  |  | 
|  | SkDPoint SkDCubic::top(double startT, double endT) const { | 
|  | SkDCubic sub = subDivide(startT, endT); | 
|  | SkDPoint topPt = sub[0]; | 
|  | if (topPt.fY > sub[3].fY || (topPt.fY == sub[3].fY && topPt.fX > sub[3].fX)) { | 
|  | topPt = sub[3]; | 
|  | } | 
|  | double extremeTs[2]; | 
|  | if (!sub.monotonicInY()) { | 
|  | int roots = FindExtrema(sub[0].fY, sub[1].fY, sub[2].fY, sub[3].fY, extremeTs); | 
|  | for (int index = 0; index < roots; ++index) { | 
|  | double t = startT + (endT - startT) * extremeTs[index]; | 
|  | SkDPoint mid = ptAtT(t); | 
|  | if (topPt.fY > mid.fY || (topPt.fY == mid.fY && topPt.fX > mid.fX)) { | 
|  | topPt = mid; | 
|  | } | 
|  | } | 
|  | } | 
|  | return topPt; | 
|  | } | 
|  |  | 
|  | SkDPoint SkDCubic::ptAtT(double t) const { | 
|  | if (0 == t) { | 
|  | return fPts[0]; | 
|  | } | 
|  | if (1 == t) { | 
|  | return fPts[3]; | 
|  | } | 
|  | double one_t = 1 - t; | 
|  | double one_t2 = one_t * one_t; | 
|  | double a = one_t2 * one_t; | 
|  | double b = 3 * one_t2 * t; | 
|  | double t2 = t * t; | 
|  | double c = 3 * one_t * t2; | 
|  | double d = t2 * t; | 
|  | SkDPoint result = {a * fPts[0].fX + b * fPts[1].fX + c * fPts[2].fX + d * fPts[3].fX, | 
|  | a * fPts[0].fY + b * fPts[1].fY + c * fPts[2].fY + d * fPts[3].fY}; | 
|  | return result; | 
|  | } | 
|  |  | 
|  | /* | 
|  | Given a cubic c, t1, and t2, find a small cubic segment. | 
|  |  | 
|  | The new cubic is defined as points A, B, C, and D, where | 
|  | s1 = 1 - t1 | 
|  | s2 = 1 - t2 | 
|  | A = c[0]*s1*s1*s1 + 3*c[1]*s1*s1*t1 + 3*c[2]*s1*t1*t1 + c[3]*t1*t1*t1 | 
|  | D = c[0]*s2*s2*s2 + 3*c[1]*s2*s2*t2 + 3*c[2]*s2*t2*t2 + c[3]*t2*t2*t2 | 
|  |  | 
|  | We don't have B or C. So We define two equations to isolate them. | 
|  | First, compute two reference T values 1/3 and 2/3 from t1 to t2: | 
|  |  | 
|  | c(at (2*t1 + t2)/3) == E | 
|  | c(at (t1 + 2*t2)/3) == F | 
|  |  | 
|  | Next, compute where those values must be if we know the values of B and C: | 
|  |  | 
|  | _12   =  A*2/3 + B*1/3 | 
|  | 12_   =  A*1/3 + B*2/3 | 
|  | _23   =  B*2/3 + C*1/3 | 
|  | 23_   =  B*1/3 + C*2/3 | 
|  | _34   =  C*2/3 + D*1/3 | 
|  | 34_   =  C*1/3 + D*2/3 | 
|  | _123  = (A*2/3 + B*1/3)*2/3 + (B*2/3 + C*1/3)*1/3 = A*4/9 + B*4/9 + C*1/9 | 
|  | 123_  = (A*1/3 + B*2/3)*1/3 + (B*1/3 + C*2/3)*2/3 = A*1/9 + B*4/9 + C*4/9 | 
|  | _234  = (B*2/3 + C*1/3)*2/3 + (C*2/3 + D*1/3)*1/3 = B*4/9 + C*4/9 + D*1/9 | 
|  | 234_  = (B*1/3 + C*2/3)*1/3 + (C*1/3 + D*2/3)*2/3 = B*1/9 + C*4/9 + D*4/9 | 
|  | _1234 = (A*4/9 + B*4/9 + C*1/9)*2/3 + (B*4/9 + C*4/9 + D*1/9)*1/3 | 
|  | =  A*8/27 + B*12/27 + C*6/27 + D*1/27 | 
|  | =  E | 
|  | 1234_ = (A*1/9 + B*4/9 + C*4/9)*1/3 + (B*1/9 + C*4/9 + D*4/9)*2/3 | 
|  | =  A*1/27 + B*6/27 + C*12/27 + D*8/27 | 
|  | =  F | 
|  | E*27  =  A*8    + B*12   + C*6     + D | 
|  | F*27  =  A      + B*6    + C*12    + D*8 | 
|  |  | 
|  | Group the known values on one side: | 
|  |  | 
|  | M       = E*27 - A*8 - D     = B*12 + C* 6 | 
|  | N       = F*27 - A   - D*8   = B* 6 + C*12 | 
|  | M*2 - N = B*18 | 
|  | N*2 - M = C*18 | 
|  | B       = (M*2 - N)/18 | 
|  | C       = (N*2 - M)/18 | 
|  | */ | 
|  |  | 
|  | static double interp_cubic_coords(const double* src, double t) { | 
|  | double ab = SkDInterp(src[0], src[2], t); | 
|  | double bc = SkDInterp(src[2], src[4], t); | 
|  | double cd = SkDInterp(src[4], src[6], t); | 
|  | double abc = SkDInterp(ab, bc, t); | 
|  | double bcd = SkDInterp(bc, cd, t); | 
|  | double abcd = SkDInterp(abc, bcd, t); | 
|  | return abcd; | 
|  | } | 
|  |  | 
|  | SkDCubic SkDCubic::subDivide(double t1, double t2) const { | 
|  | if (t1 == 0 || t2 == 1) { | 
|  | if (t1 == 0 && t2 == 1) { | 
|  | return *this; | 
|  | } | 
|  | SkDCubicPair pair = chopAt(t1 == 0 ? t2 : t1); | 
|  | SkDCubic dst = t1 == 0 ? pair.first() : pair.second(); | 
|  | return dst; | 
|  | } | 
|  | SkDCubic dst; | 
|  | double ax = dst[0].fX = interp_cubic_coords(&fPts[0].fX, t1); | 
|  | double ay = dst[0].fY = interp_cubic_coords(&fPts[0].fY, t1); | 
|  | double ex = interp_cubic_coords(&fPts[0].fX, (t1*2+t2)/3); | 
|  | double ey = interp_cubic_coords(&fPts[0].fY, (t1*2+t2)/3); | 
|  | double fx = interp_cubic_coords(&fPts[0].fX, (t1+t2*2)/3); | 
|  | double fy = interp_cubic_coords(&fPts[0].fY, (t1+t2*2)/3); | 
|  | double dx = dst[3].fX = interp_cubic_coords(&fPts[0].fX, t2); | 
|  | double dy = dst[3].fY = interp_cubic_coords(&fPts[0].fY, t2); | 
|  | double mx = ex * 27 - ax * 8 - dx; | 
|  | double my = ey * 27 - ay * 8 - dy; | 
|  | double nx = fx * 27 - ax - dx * 8; | 
|  | double ny = fy * 27 - ay - dy * 8; | 
|  | /* bx = */ dst[1].fX = (mx * 2 - nx) / 18; | 
|  | /* by = */ dst[1].fY = (my * 2 - ny) / 18; | 
|  | /* cx = */ dst[2].fX = (nx * 2 - mx) / 18; | 
|  | /* cy = */ dst[2].fY = (ny * 2 - my) / 18; | 
|  | // FIXME: call align() ? | 
|  | return dst; | 
|  | } | 
|  |  | 
|  | void SkDCubic::align(int endIndex, int ctrlIndex, SkDPoint* dstPt) const { | 
|  | if (fPts[endIndex].fX == fPts[ctrlIndex].fX) { | 
|  | dstPt->fX = fPts[endIndex].fX; | 
|  | } | 
|  | if (fPts[endIndex].fY == fPts[ctrlIndex].fY) { | 
|  | dstPt->fY = fPts[endIndex].fY; | 
|  | } | 
|  | } | 
|  |  | 
|  | void SkDCubic::subDivide(const SkDPoint& a, const SkDPoint& d, | 
|  | double t1, double t2, SkDPoint dst[2]) const { | 
|  | SkASSERT(t1 != t2); | 
|  | #if 0 | 
|  | double ex = interp_cubic_coords(&fPts[0].fX, (t1 * 2 + t2) / 3); | 
|  | double ey = interp_cubic_coords(&fPts[0].fY, (t1 * 2 + t2) / 3); | 
|  | double fx = interp_cubic_coords(&fPts[0].fX, (t1 + t2 * 2) / 3); | 
|  | double fy = interp_cubic_coords(&fPts[0].fY, (t1 + t2 * 2) / 3); | 
|  | double mx = ex * 27 - a.fX * 8 - d.fX; | 
|  | double my = ey * 27 - a.fY * 8 - d.fY; | 
|  | double nx = fx * 27 - a.fX - d.fX * 8; | 
|  | double ny = fy * 27 - a.fY - d.fY * 8; | 
|  | /* bx = */ dst[0].fX = (mx * 2 - nx) / 18; | 
|  | /* by = */ dst[0].fY = (my * 2 - ny) / 18; | 
|  | /* cx = */ dst[1].fX = (nx * 2 - mx) / 18; | 
|  | /* cy = */ dst[1].fY = (ny * 2 - my) / 18; | 
|  | #else | 
|  | // this approach assumes that the control points computed directly are accurate enough | 
|  | SkDCubic sub = subDivide(t1, t2); | 
|  | dst[0] = sub[1] + (a - sub[0]); | 
|  | dst[1] = sub[2] + (d - sub[3]); | 
|  | #endif | 
|  | if (t1 == 0 || t2 == 0) { | 
|  | align(0, 1, t1 == 0 ? &dst[0] : &dst[1]); | 
|  | } | 
|  | if (t1 == 1 || t2 == 1) { | 
|  | align(3, 2, t1 == 1 ? &dst[0] : &dst[1]); | 
|  | } | 
|  | if (AlmostBequalUlps(dst[0].fX, a.fX)) { | 
|  | dst[0].fX = a.fX; | 
|  | } | 
|  | if (AlmostBequalUlps(dst[0].fY, a.fY)) { | 
|  | dst[0].fY = a.fY; | 
|  | } | 
|  | if (AlmostBequalUlps(dst[1].fX, d.fX)) { | 
|  | dst[1].fX = d.fX; | 
|  | } | 
|  | if (AlmostBequalUlps(dst[1].fY, d.fY)) { | 
|  | dst[1].fY = d.fY; | 
|  | } | 
|  | } | 
|  |  | 
|  | /* classic one t subdivision */ | 
|  | static void interp_cubic_coords(const double* src, double* dst, double t) { | 
|  | double ab = SkDInterp(src[0], src[2], t); | 
|  | double bc = SkDInterp(src[2], src[4], t); | 
|  | double cd = SkDInterp(src[4], src[6], t); | 
|  | double abc = SkDInterp(ab, bc, t); | 
|  | double bcd = SkDInterp(bc, cd, t); | 
|  | double abcd = SkDInterp(abc, bcd, t); | 
|  |  | 
|  | dst[0] = src[0]; | 
|  | dst[2] = ab; | 
|  | dst[4] = abc; | 
|  | dst[6] = abcd; | 
|  | dst[8] = bcd; | 
|  | dst[10] = cd; | 
|  | dst[12] = src[6]; | 
|  | } | 
|  |  | 
|  | SkDCubicPair SkDCubic::chopAt(double t) const { | 
|  | SkDCubicPair dst; | 
|  | if (t == 0.5) { | 
|  | dst.pts[0] = fPts[0]; | 
|  | dst.pts[1].fX = (fPts[0].fX + fPts[1].fX) / 2; | 
|  | dst.pts[1].fY = (fPts[0].fY + fPts[1].fY) / 2; | 
|  | dst.pts[2].fX = (fPts[0].fX + 2 * fPts[1].fX + fPts[2].fX) / 4; | 
|  | dst.pts[2].fY = (fPts[0].fY + 2 * fPts[1].fY + fPts[2].fY) / 4; | 
|  | dst.pts[3].fX = (fPts[0].fX + 3 * (fPts[1].fX + fPts[2].fX) + fPts[3].fX) / 8; | 
|  | dst.pts[3].fY = (fPts[0].fY + 3 * (fPts[1].fY + fPts[2].fY) + fPts[3].fY) / 8; | 
|  | dst.pts[4].fX = (fPts[1].fX + 2 * fPts[2].fX + fPts[3].fX) / 4; | 
|  | dst.pts[4].fY = (fPts[1].fY + 2 * fPts[2].fY + fPts[3].fY) / 4; | 
|  | dst.pts[5].fX = (fPts[2].fX + fPts[3].fX) / 2; | 
|  | dst.pts[5].fY = (fPts[2].fY + fPts[3].fY) / 2; | 
|  | dst.pts[6] = fPts[3]; | 
|  | return dst; | 
|  | } | 
|  | interp_cubic_coords(&fPts[0].fX, &dst.pts[0].fX, t); | 
|  | interp_cubic_coords(&fPts[0].fY, &dst.pts[0].fY, t); | 
|  | return dst; | 
|  | } |