1 | /* |
2 | * Copyright 2011 Google Inc. |
3 | * |
4 | * Use of this source code is governed by a BSD-style license that can be |
5 | * found in the LICENSE file. |
6 | */ |
7 | |
8 | #ifndef GrPathUtils_DEFINED |
9 | #define GrPathUtils_DEFINED |
10 | |
11 | #include "include/core/SkRect.h" |
12 | #include "include/private/SkTArray.h" |
13 | #include "src/core/SkGeometry.h" |
14 | #include "src/core/SkPathPriv.h" |
15 | |
16 | class SkMatrix; |
17 | |
18 | /** |
19 | * Utilities for evaluating paths. |
20 | */ |
21 | namespace GrPathUtils { |
22 | // Very small tolerances will be increased to a minimum threshold value, to avoid division |
23 | // problems in subsequent math. |
24 | SkScalar scaleToleranceToSrc(SkScalar devTol, |
25 | const SkMatrix& viewM, |
26 | const SkRect& pathBounds); |
27 | |
28 | int worstCasePointCount(const SkPath&, |
29 | int* subpaths, |
30 | SkScalar tol); |
31 | |
32 | uint32_t quadraticPointCount(const SkPoint points[], SkScalar tol); |
33 | |
34 | uint32_t generateQuadraticPoints(const SkPoint& p0, |
35 | const SkPoint& p1, |
36 | const SkPoint& p2, |
37 | SkScalar tolSqd, |
38 | SkPoint** points, |
39 | uint32_t pointsLeft); |
40 | |
41 | uint32_t cubicPointCount(const SkPoint points[], SkScalar tol); |
42 | |
43 | uint32_t generateCubicPoints(const SkPoint& p0, |
44 | const SkPoint& p1, |
45 | const SkPoint& p2, |
46 | const SkPoint& p3, |
47 | SkScalar tolSqd, |
48 | SkPoint** points, |
49 | uint32_t pointsLeft); |
50 | |
51 | // A 2x3 matrix that goes from the 2d space coordinates to UV space where |
52 | // u^2-v = 0 specifies the quad. The matrix is determined by the control |
53 | // points of the quadratic. |
54 | class QuadUVMatrix { |
55 | public: |
56 | QuadUVMatrix() {} |
57 | // Initialize the matrix from the control pts |
58 | QuadUVMatrix(const SkPoint controlPts[3]) { this->set(controlPts); } |
59 | void set(const SkPoint controlPts[3]); |
60 | |
61 | /** |
62 | * Applies the matrix to vertex positions to compute UV coords. |
63 | * |
64 | * vertices is a pointer to the first vertex. |
65 | * vertexCount is the number of vertices. |
66 | * stride is the size of each vertex. |
67 | * uvOffset is the offset of the UV values within each vertex. |
68 | */ |
69 | void apply(void* vertices, int vertexCount, size_t stride, size_t uvOffset) const { |
70 | intptr_t xyPtr = reinterpret_cast<intptr_t>(vertices); |
71 | intptr_t uvPtr = reinterpret_cast<intptr_t>(vertices) + uvOffset; |
72 | float sx = fM[0]; |
73 | float kx = fM[1]; |
74 | float tx = fM[2]; |
75 | float ky = fM[3]; |
76 | float sy = fM[4]; |
77 | float ty = fM[5]; |
78 | for (int i = 0; i < vertexCount; ++i) { |
79 | const SkPoint* xy = reinterpret_cast<const SkPoint*>(xyPtr); |
80 | SkPoint* uv = reinterpret_cast<SkPoint*>(uvPtr); |
81 | uv->fX = sx * xy->fX + kx * xy->fY + tx; |
82 | uv->fY = ky * xy->fX + sy * xy->fY + ty; |
83 | xyPtr += stride; |
84 | uvPtr += stride; |
85 | } |
86 | } |
87 | private: |
88 | float fM[6]; |
89 | }; |
90 | |
91 | // Input is 3 control points and a weight for a bezier conic. Calculates the |
92 | // three linear functionals (K,L,M) that represent the implicit equation of the |
93 | // conic, k^2 - lm. |
94 | // |
95 | // Output: klm holds the linear functionals K,L,M as row vectors: |
96 | // |
97 | // | ..K.. | | x | | k | |
98 | // | ..L.. | * | y | == | l | |
99 | // | ..M.. | | 1 | | m | |
100 | // |
101 | void getConicKLM(const SkPoint p[3], const SkScalar weight, SkMatrix* klm); |
102 | |
103 | // Converts a cubic into a sequence of quads. If working in device space |
104 | // use tolScale = 1, otherwise set based on stretchiness of the matrix. The |
105 | // result is sets of 3 points in quads. This will preserve the starting and |
106 | // ending tangent vectors (modulo FP precision). |
107 | void convertCubicToQuads(const SkPoint p[4], |
108 | SkScalar tolScale, |
109 | SkTArray<SkPoint, true>* quads); |
110 | |
111 | // When we approximate a cubic {a,b,c,d} with a quadratic we may have to |
112 | // ensure that the new control point lies between the lines ab and cd. The |
113 | // convex path renderer requires this. It starts with a path where all the |
114 | // control points taken together form a convex polygon. It relies on this |
115 | // property and the quadratic approximation of cubics step cannot alter it. |
116 | // This variation enforces this constraint. The cubic must be simple and dir |
117 | // must specify the orientation of the contour containing the cubic. |
118 | void convertCubicToQuadsConstrainToTangents(const SkPoint p[4], |
119 | SkScalar tolScale, |
120 | SkPathPriv::FirstDirection dir, |
121 | SkTArray<SkPoint, true>* quads); |
122 | |
123 | enum class ExcludedTerm { |
124 | kNonInvertible, |
125 | kQuadraticTerm, |
126 | kLinearTerm |
127 | }; |
128 | |
129 | // Computes the inverse-transpose of the cubic's power basis matrix, after removing a specific |
130 | // row of coefficients. |
131 | // |
132 | // E.g. if the cubic is defined in power basis form as follows: |
133 | // |
134 | // | x3 y3 0 | |
135 | // C(t,s) = [t^3 t^2*s t*s^2 s^3] * | x2 y2 0 | |
136 | // | x1 y1 0 | |
137 | // | x0 y0 1 | |
138 | // |
139 | // And the excluded term is "kQuadraticTerm", then the resulting inverse-transpose will be: |
140 | // |
141 | // | x3 y3 0 | -1 T |
142 | // | x1 y1 0 | |
143 | // | x0 y0 1 | |
144 | // |
145 | // (The term to exclude is chosen based on maximizing the resulting matrix determinant.) |
146 | // |
147 | // This can be used to find the KLM linear functionals: |
148 | // |
149 | // | ..K.. | | ..kcoeffs.. | |
150 | // | ..L.. | = | ..lcoeffs.. | * inverse_transpose_power_basis_matrix |
151 | // | ..M.. | | ..mcoeffs.. | |
152 | // |
153 | // NOTE: the same term that was excluded here must also be removed from the corresponding column |
154 | // of the klmcoeffs matrix. |
155 | // |
156 | // Returns which row of coefficients was removed, or kNonInvertible if the cubic was degenerate. |
157 | ExcludedTerm calcCubicInverseTransposePowerBasisMatrix(const SkPoint p[4], SkMatrix* out); |
158 | |
159 | // Computes the KLM linear functionals for the cubic implicit form. The "right" side of the |
160 | // curve (when facing in the direction of increasing parameter values) will be the area that |
161 | // satisfies: |
162 | // |
163 | // k^3 < l*m |
164 | // |
165 | // Output: |
166 | // |
167 | // klm: Holds the linear functionals K,L,M as row vectors: |
168 | // |
169 | // | ..K.. | | x | | k | |
170 | // | ..L.. | * | y | == | l | |
171 | // | ..M.. | | 1 | | m | |
172 | // |
173 | // NOTE: the KLM lines are calculated in the same space as the input control points. If you |
174 | // transform the points the lines will also need to be transformed. This can be done by mapping |
175 | // the lines with the inverse-transpose of the matrix used to map the points. |
176 | // |
177 | // t[],s[]: These are set to the two homogeneous parameter values at which points the lines L&M |
178 | // intersect with K (See SkClassifyCubic). |
179 | // |
180 | // Returns the cubic's classification. |
181 | SkCubicType getCubicKLM(const SkPoint src[4], SkMatrix* klm, double t[2], double s[2]); |
182 | |
183 | // Chops the cubic bezier passed in by src, at the double point (intersection point) |
184 | // if the curve is a cubic loop. If it is a loop, there will be two parametric values for |
185 | // the double point: t1 and t2. We chop the cubic at these values if they are between 0 and 1. |
186 | // Return value: |
187 | // Value of 3: t1 and t2 are both between (0,1), and dst will contain the three cubics, |
188 | // dst[0..3], dst[3..6], and dst[6..9] if dst is not nullptr |
189 | // Value of 2: Only one of t1 and t2 are between (0,1), and dst will contain the two cubics, |
190 | // dst[0..3] and dst[3..6] if dst is not nullptr |
191 | // Value of 1: Neither t1 nor t2 are between (0,1), and dst will contain the one original cubic, |
192 | // src[0..3] |
193 | // |
194 | // Output: |
195 | // |
196 | // klm: Holds the linear functionals K,L,M as row vectors. (See getCubicKLM().) |
197 | // |
198 | // loopIndex: This value will tell the caller which of the chopped sections (if any) are the |
199 | // actual loop. A value of -1 means there is no loop section. The caller can then use |
200 | // this value to decide how/if they want to flip the orientation of this section. |
201 | // The flip should be done by negating the k and l values as follows: |
202 | // |
203 | // KLM.postScale(-1, -1) |
204 | int chopCubicAtLoopIntersection(const SkPoint src[4], SkPoint dst[10], SkMatrix* klm, |
205 | int* loopIndex); |
206 | |
207 | // When tessellating curved paths into linear segments, this defines the maximum distance |
208 | // in screen space which a segment may deviate from the mathmatically correct value. |
209 | // Above this value, the segment will be subdivided. |
210 | // This value was chosen to approximate the supersampling accuracy of the raster path (16 |
211 | // samples, or one quarter pixel). |
212 | static const SkScalar kDefaultTolerance = SkDoubleToScalar(0.25); |
213 | |
214 | // We guarantee that no quad or cubic will ever produce more than this many points |
215 | static const int kMaxPointsPerCurve = 1 << 10; |
216 | }; |
217 | #endif |
218 | |