1 /* -*- Mode: c; c-basic-offset: 4; tab-width: 8; indent-tabs-mode: t; -*- */
2 /*
3 *
4 * Copyright © 2000 Keith Packard, member of The XFree86 Project, Inc.
5 * Copyright © 2000 SuSE, Inc.
6 * 2005 Lars Knoll & Zack Rusin, Trolltech
7 * Copyright © 2007 Red Hat, Inc.
8 *
9 *
10 * Permission to use, copy, modify, distribute, and sell this software and its
11 * documentation for any purpose is hereby granted without fee, provided that
12 * the above copyright notice appear in all copies and that both that
13 * copyright notice and this permission notice appear in supporting
14 * documentation, and that the name of Keith Packard not be used in
15 * advertising or publicity pertaining to distribution of the software without
16 * specific, written prior permission. Keith Packard makes no
17 * representations about the suitability of this software for any purpose. It
18 * is provided "as is" without express or implied warranty.
19 *
20 * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
21 * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
22 * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
23 * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
24 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
25 * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
26 * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
27 * SOFTWARE.
28 */
29
30 #ifdef HAVE_CONFIG_H
31 #include <config.h>
32 #endif
33 #include <stdlib.h>
34 #include <math.h>
35 #include "pixman-private.h"
36
37 static inline pixman_fixed_32_32_t
dot(pixman_fixed_48_16_t x1,pixman_fixed_48_16_t y1,pixman_fixed_48_16_t z1,pixman_fixed_48_16_t x2,pixman_fixed_48_16_t y2,pixman_fixed_48_16_t z2)38 dot (pixman_fixed_48_16_t x1,
39 pixman_fixed_48_16_t y1,
40 pixman_fixed_48_16_t z1,
41 pixman_fixed_48_16_t x2,
42 pixman_fixed_48_16_t y2,
43 pixman_fixed_48_16_t z2)
44 {
45 /*
46 * Exact computation, assuming that the input values can
47 * be represented as pixman_fixed_16_16_t
48 */
49 return x1 * x2 + y1 * y2 + z1 * z2;
50 }
51
52 static inline double
fdot(double x1,double y1,double z1,double x2,double y2,double z2)53 fdot (double x1,
54 double y1,
55 double z1,
56 double x2,
57 double y2,
58 double z2)
59 {
60 /*
61 * Error can be unbound in some special cases.
62 * Using clever dot product algorithms (for example compensated
63 * dot product) would improve this but make the code much less
64 * obvious
65 */
66 return x1 * x2 + y1 * y2 + z1 * z2;
67 }
68
69 static uint32_t
radial_compute_color(double a,double b,double c,double inva,double dr,double mindr,pixman_gradient_walker_t * walker,pixman_repeat_t repeat)70 radial_compute_color (double a,
71 double b,
72 double c,
73 double inva,
74 double dr,
75 double mindr,
76 pixman_gradient_walker_t *walker,
77 pixman_repeat_t repeat)
78 {
79 /*
80 * In this function error propagation can lead to bad results:
81 * - discr can have an unbound error (if b*b-a*c is very small),
82 * potentially making it the opposite sign of what it should have been
83 * (thus clearing a pixel that would have been colored or vice-versa)
84 * or propagating the error to sqrtdiscr;
85 * if discr has the wrong sign or b is very small, this can lead to bad
86 * results
87 *
88 * - the algorithm used to compute the solutions of the quadratic
89 * equation is not numerically stable (but saves one division compared
90 * to the numerically stable one);
91 * this can be a problem if a*c is much smaller than b*b
92 *
93 * - the above problems are worse if a is small (as inva becomes bigger)
94 */
95 double discr;
96
97 if (a == 0)
98 {
99 double t;
100
101 if (b == 0)
102 return 0;
103
104 t = pixman_fixed_1 / 2 * c / b;
105 if (repeat == PIXMAN_REPEAT_NONE)
106 {
107 if (0 <= t && t <= pixman_fixed_1)
108 return _pixman_gradient_walker_pixel (walker, t);
109 }
110 else
111 {
112 if (t * dr >= mindr)
113 return _pixman_gradient_walker_pixel (walker, t);
114 }
115
116 return 0;
117 }
118
119 discr = fdot (b, a, 0, b, -c, 0);
120 if (discr >= 0)
121 {
122 double sqrtdiscr, t0, t1;
123
124 sqrtdiscr = sqrt (discr);
125 t0 = (b + sqrtdiscr) * inva;
126 t1 = (b - sqrtdiscr) * inva;
127
128 /*
129 * The root that must be used is the biggest one that belongs
130 * to the valid range ([0,1] for PIXMAN_REPEAT_NONE, any
131 * solution that results in a positive radius otherwise).
132 *
133 * If a > 0, t0 is the biggest solution, so if it is valid, it
134 * is the correct result.
135 *
136 * If a < 0, only one of the solutions can be valid, so the
137 * order in which they are tested is not important.
138 */
139 if (repeat == PIXMAN_REPEAT_NONE)
140 {
141 if (0 <= t0 && t0 <= pixman_fixed_1)
142 return _pixman_gradient_walker_pixel (walker, t0);
143 else if (0 <= t1 && t1 <= pixman_fixed_1)
144 return _pixman_gradient_walker_pixel (walker, t1);
145 }
146 else
147 {
148 if (t0 * dr >= mindr)
149 return _pixman_gradient_walker_pixel (walker, t0);
150 else if (t1 * dr >= mindr)
151 return _pixman_gradient_walker_pixel (walker, t1);
152 }
153 }
154
155 return 0;
156 }
157
158 static uint32_t *
radial_get_scanline_narrow(pixman_iter_t * iter,const uint32_t * mask)159 radial_get_scanline_narrow (pixman_iter_t *iter, const uint32_t *mask)
160 {
161 /*
162 * Implementation of radial gradients following the PDF specification.
163 * See section 8.7.4.5.4 Type 3 (Radial) Shadings of the PDF Reference
164 * Manual (PDF 32000-1:2008 at the time of this writing).
165 *
166 * In the radial gradient problem we are given two circles (c₁,r₁) and
167 * (c₂,r₂) that define the gradient itself.
168 *
169 * Mathematically the gradient can be defined as the family of circles
170 *
171 * ((1-t)·c₁ + t·(c₂), (1-t)·r₁ + t·r₂)
172 *
173 * excluding those circles whose radius would be < 0. When a point
174 * belongs to more than one circle, the one with a bigger t is the only
175 * one that contributes to its color. When a point does not belong
176 * to any of the circles, it is transparent black, i.e. RGBA (0, 0, 0, 0).
177 * Further limitations on the range of values for t are imposed when
178 * the gradient is not repeated, namely t must belong to [0,1].
179 *
180 * The graphical result is the same as drawing the valid (radius > 0)
181 * circles with increasing t in [-inf, +inf] (or in [0,1] if the gradient
182 * is not repeated) using SOURCE operator composition.
183 *
184 * It looks like a cone pointing towards the viewer if the ending circle
185 * is smaller than the starting one, a cone pointing inside the page if
186 * the starting circle is the smaller one and like a cylinder if they
187 * have the same radius.
188 *
189 * What we actually do is, given the point whose color we are interested
190 * in, compute the t values for that point, solving for t in:
191 *
192 * length((1-t)·c₁ + t·(c₂) - p) = (1-t)·r₁ + t·r₂
193 *
194 * Let's rewrite it in a simpler way, by defining some auxiliary
195 * variables:
196 *
197 * cd = c₂ - c₁
198 * pd = p - c₁
199 * dr = r₂ - r₁
200 * length(t·cd - pd) = r₁ + t·dr
201 *
202 * which actually means
203 *
204 * hypot(t·cdx - pdx, t·cdy - pdy) = r₁ + t·dr
205 *
206 * or
207 *
208 * ⎷((t·cdx - pdx)² + (t·cdy - pdy)²) = r₁ + t·dr.
209 *
210 * If we impose (as stated earlier) that r₁ + t·dr >= 0, it becomes:
211 *
212 * (t·cdx - pdx)² + (t·cdy - pdy)² = (r₁ + t·dr)²
213 *
214 * where we can actually expand the squares and solve for t:
215 *
216 * t²cdx² - 2t·cdx·pdx + pdx² + t²cdy² - 2t·cdy·pdy + pdy² =
217 * = r₁² + 2·r₁·t·dr + t²·dr²
218 *
219 * (cdx² + cdy² - dr²)t² - 2(cdx·pdx + cdy·pdy + r₁·dr)t +
220 * (pdx² + pdy² - r₁²) = 0
221 *
222 * A = cdx² + cdy² - dr²
223 * B = pdx·cdx + pdy·cdy + r₁·dr
224 * C = pdx² + pdy² - r₁²
225 * At² - 2Bt + C = 0
226 *
227 * The solutions (unless the equation degenerates because of A = 0) are:
228 *
229 * t = (B ± ⎷(B² - A·C)) / A
230 *
231 * The solution we are going to prefer is the bigger one, unless the
232 * radius associated to it is negative (or it falls outside the valid t
233 * range).
234 *
235 * Additional observations (useful for optimizations):
236 * A does not depend on p
237 *
238 * A < 0 <=> one of the two circles completely contains the other one
239 * <=> for every p, the radiuses associated with the two t solutions
240 * have opposite sign
241 */
242 pixman_image_t *image = iter->image;
243 int x = iter->x;
244 int y = iter->y;
245 int width = iter->width;
246 uint32_t *buffer = iter->buffer;
247
248 gradient_t *gradient = (gradient_t *)image;
249 radial_gradient_t *radial = (radial_gradient_t *)image;
250 uint32_t *end = buffer + width;
251 pixman_gradient_walker_t walker;
252 pixman_vector_t v, unit;
253
254 /* reference point is the center of the pixel */
255 v.vector[0] = pixman_int_to_fixed (x) + pixman_fixed_1 / 2;
256 v.vector[1] = pixman_int_to_fixed (y) + pixman_fixed_1 / 2;
257 v.vector[2] = pixman_fixed_1;
258
259 _pixman_gradient_walker_init (&walker, gradient, image->common.repeat);
260
261 if (image->common.transform)
262 {
263 if (!pixman_transform_point_3d (image->common.transform, &v))
264 return iter->buffer;
265
266 unit.vector[0] = image->common.transform->matrix[0][0];
267 unit.vector[1] = image->common.transform->matrix[1][0];
268 unit.vector[2] = image->common.transform->matrix[2][0];
269 }
270 else
271 {
272 unit.vector[0] = pixman_fixed_1;
273 unit.vector[1] = 0;
274 unit.vector[2] = 0;
275 }
276
277 if (unit.vector[2] == 0 && v.vector[2] == pixman_fixed_1)
278 {
279 /*
280 * Given:
281 *
282 * t = (B ± ⎷(B² - A·C)) / A
283 *
284 * where
285 *
286 * A = cdx² + cdy² - dr²
287 * B = pdx·cdx + pdy·cdy + r₁·dr
288 * C = pdx² + pdy² - r₁²
289 * det = B² - A·C
290 *
291 * Since we have an affine transformation, we know that (pdx, pdy)
292 * increase linearly with each pixel,
293 *
294 * pdx = pdx₀ + n·ux,
295 * pdy = pdy₀ + n·uy,
296 *
297 * we can then express B, C and det through multiple differentiation.
298 */
299 pixman_fixed_32_32_t b, db, c, dc, ddc;
300
301 /* warning: this computation may overflow */
302 v.vector[0] -= radial->c1.x;
303 v.vector[1] -= radial->c1.y;
304
305 /*
306 * B and C are computed and updated exactly.
307 * If fdot was used instead of dot, in the worst case it would
308 * lose 11 bits of precision in each of the multiplication and
309 * summing up would zero out all the bit that were preserved,
310 * thus making the result 0 instead of the correct one.
311 * This would mean a worst case of unbound relative error or
312 * about 2^10 absolute error
313 */
314 b = dot (v.vector[0], v.vector[1], radial->c1.radius,
315 radial->delta.x, radial->delta.y, radial->delta.radius);
316 db = dot (unit.vector[0], unit.vector[1], 0,
317 radial->delta.x, radial->delta.y, 0);
318
319 c = dot (v.vector[0], v.vector[1],
320 -((pixman_fixed_48_16_t) radial->c1.radius),
321 v.vector[0], v.vector[1], radial->c1.radius);
322 dc = dot (2 * (pixman_fixed_48_16_t) v.vector[0] + unit.vector[0],
323 2 * (pixman_fixed_48_16_t) v.vector[1] + unit.vector[1],
324 0,
325 unit.vector[0], unit.vector[1], 0);
326 ddc = 2 * dot (unit.vector[0], unit.vector[1], 0,
327 unit.vector[0], unit.vector[1], 0);
328
329 while (buffer < end)
330 {
331 if (!mask || *mask++)
332 {
333 *buffer = radial_compute_color (radial->a, b, c,
334 radial->inva,
335 radial->delta.radius,
336 radial->mindr,
337 &walker,
338 image->common.repeat);
339 }
340
341 b += db;
342 c += dc;
343 dc += ddc;
344 ++buffer;
345 }
346 }
347 else
348 {
349 /* projective */
350 /* Warning:
351 * error propagation guarantees are much looser than in the affine case
352 */
353 while (buffer < end)
354 {
355 if (!mask || *mask++)
356 {
357 if (v.vector[2] != 0)
358 {
359 double pdx, pdy, invv2, b, c;
360
361 invv2 = 1. * pixman_fixed_1 / v.vector[2];
362
363 pdx = v.vector[0] * invv2 - radial->c1.x;
364 /* / pixman_fixed_1 */
365
366 pdy = v.vector[1] * invv2 - radial->c1.y;
367 /* / pixman_fixed_1 */
368
369 b = fdot (pdx, pdy, radial->c1.radius,
370 radial->delta.x, radial->delta.y,
371 radial->delta.radius);
372 /* / pixman_fixed_1 / pixman_fixed_1 */
373
374 c = fdot (pdx, pdy, -radial->c1.radius,
375 pdx, pdy, radial->c1.radius);
376 /* / pixman_fixed_1 / pixman_fixed_1 */
377
378 *buffer = radial_compute_color (radial->a, b, c,
379 radial->inva,
380 radial->delta.radius,
381 radial->mindr,
382 &walker,
383 image->common.repeat);
384 }
385 else
386 {
387 *buffer = 0;
388 }
389 }
390
391 ++buffer;
392
393 v.vector[0] += unit.vector[0];
394 v.vector[1] += unit.vector[1];
395 v.vector[2] += unit.vector[2];
396 }
397 }
398
399 iter->y++;
400 return iter->buffer;
401 }
402
403 static uint32_t *
radial_get_scanline_wide(pixman_iter_t * iter,const uint32_t * mask)404 radial_get_scanline_wide (pixman_iter_t *iter, const uint32_t *mask)
405 {
406 uint32_t *buffer = radial_get_scanline_narrow (iter, NULL);
407
408 pixman_expand_to_float (
409 (argb_t *)buffer, buffer, PIXMAN_a8r8g8b8, iter->width);
410
411 return buffer;
412 }
413
414 void
_pixman_radial_gradient_iter_init(pixman_image_t * image,pixman_iter_t * iter)415 _pixman_radial_gradient_iter_init (pixman_image_t *image, pixman_iter_t *iter)
416 {
417 if (iter->iter_flags & ITER_NARROW)
418 iter->get_scanline = radial_get_scanline_narrow;
419 else
420 iter->get_scanline = radial_get_scanline_wide;
421 }
422
423 PIXMAN_EXPORT pixman_image_t *
pixman_image_create_radial_gradient(const pixman_point_fixed_t * inner,const pixman_point_fixed_t * outer,pixman_fixed_t inner_radius,pixman_fixed_t outer_radius,const pixman_gradient_stop_t * stops,int n_stops)424 pixman_image_create_radial_gradient (const pixman_point_fixed_t * inner,
425 const pixman_point_fixed_t * outer,
426 pixman_fixed_t inner_radius,
427 pixman_fixed_t outer_radius,
428 const pixman_gradient_stop_t *stops,
429 int n_stops)
430 {
431 pixman_image_t *image;
432 radial_gradient_t *radial;
433
434 image = _pixman_image_allocate ();
435
436 if (!image)
437 return NULL;
438
439 radial = &image->radial;
440
441 if (!_pixman_init_gradient (&radial->common, stops, n_stops))
442 {
443 free (image);
444 return NULL;
445 }
446
447 image->type = RADIAL;
448
449 radial->c1.x = inner->x;
450 radial->c1.y = inner->y;
451 radial->c1.radius = inner_radius;
452 radial->c2.x = outer->x;
453 radial->c2.y = outer->y;
454 radial->c2.radius = outer_radius;
455
456 /* warning: this computations may overflow */
457 radial->delta.x = radial->c2.x - radial->c1.x;
458 radial->delta.y = radial->c2.y - radial->c1.y;
459 radial->delta.radius = radial->c2.radius - radial->c1.radius;
460
461 /* computed exactly, then cast to double -> every bit of the double
462 representation is correct (53 bits) */
463 radial->a = dot (radial->delta.x, radial->delta.y, -radial->delta.radius,
464 radial->delta.x, radial->delta.y, radial->delta.radius);
465 if (radial->a != 0)
466 radial->inva = 1. * pixman_fixed_1 / radial->a;
467
468 radial->mindr = -1. * pixman_fixed_1 * radial->c1.radius;
469
470 return image;
471 }
472