Home | History | Annotate | Line # | Download | only in pixman
      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 <pixman-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
     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
     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 void
     70 radial_write_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 		    int                            Bpp,
     79 		    pixman_gradient_walker_write_t write_pixel,
     80 		    uint32_t                      *buffer)
     81 {
     82     /*
     83      * In this function error propagation can lead to bad results:
     84      *  - discr can have an unbound error (if b*b-a*c is very small),
     85      *    potentially making it the opposite sign of what it should have been
     86      *    (thus clearing a pixel that would have been colored or vice-versa)
     87      *    or propagating the error to sqrtdiscr;
     88      *    if discr has the wrong sign or b is very small, this can lead to bad
     89      *    results
     90      *
     91      *  - the algorithm used to compute the solutions of the quadratic
     92      *    equation is not numerically stable (but saves one division compared
     93      *    to the numerically stable one);
     94      *    this can be a problem if a*c is much smaller than b*b
     95      *
     96      *  - the above problems are worse if a is small (as inva becomes bigger)
     97      */
     98     double discr;
     99 
    100     if (a == 0)
    101     {
    102 	double t;
    103 
    104 	if (b == 0)
    105 	{
    106 	    memset (buffer, 0, Bpp);
    107 	    return;
    108 	}
    109 
    110 	t = pixman_fixed_1 / 2 * c / b;
    111 	if (repeat == PIXMAN_REPEAT_NONE)
    112 	{
    113 	    if (0 <= t && t <= pixman_fixed_1)
    114 	    {
    115 		write_pixel (walker, t, buffer);
    116 		return;
    117 	    }
    118 	}
    119 	else
    120 	{
    121 	    if (t * dr >= mindr)
    122 	    {
    123 		write_pixel (walker, t, buffer);
    124 		return;
    125 	    }
    126 	}
    127 
    128 	memset (buffer, 0, Bpp);
    129 	return;
    130     }
    131 
    132     discr = fdot (b, a, 0, b, -c, 0);
    133     if (discr >= 0)
    134     {
    135 	double sqrtdiscr, t0, t1;
    136 
    137 	sqrtdiscr = sqrt (discr);
    138 	t0 = (b + sqrtdiscr) * inva;
    139 	t1 = (b - sqrtdiscr) * inva;
    140 
    141 	/*
    142 	 * The root that must be used is the biggest one that belongs
    143 	 * to the valid range ([0,1] for PIXMAN_REPEAT_NONE, any
    144 	 * solution that results in a positive radius otherwise).
    145 	 *
    146 	 * If a > 0, t0 is the biggest solution, so if it is valid, it
    147 	 * is the correct result.
    148 	 *
    149 	 * If a < 0, only one of the solutions can be valid, so the
    150 	 * order in which they are tested is not important.
    151 	 */
    152 	if (repeat == PIXMAN_REPEAT_NONE)
    153 	{
    154 	    if (0 <= t0 && t0 <= pixman_fixed_1)
    155 	    {
    156 		write_pixel (walker, t0, buffer);
    157 		return;
    158 	    }
    159 	    else if (0 <= t1 && t1 <= pixman_fixed_1)
    160 	    {
    161 		write_pixel (walker, t1, buffer);
    162 		return;
    163            }
    164 	}
    165 	else
    166 	{
    167 	    if (t0 * dr >= mindr)
    168 	    {
    169 		write_pixel (walker, t0, buffer);
    170 		return;
    171 	    }
    172 	    else if (t1 * dr >= mindr)
    173 	    {
    174 		write_pixel (walker, t1, buffer);
    175 		return;
    176 	    }
    177 	}
    178     }
    179 
    180     memset (buffer, 0, Bpp);
    181     return;
    182 }
    183 
    184 static uint32_t *
    185 radial_get_scanline (pixman_iter_t                 *iter,
    186 		     const uint32_t                *mask,
    187 		     int                            Bpp,
    188 		     pixman_gradient_walker_write_t write_pixel)
    189 {
    190     /*
    191      * Implementation of radial gradients following the PDF specification.
    192      * See section 8.7.4.5.4 Type 3 (Radial) Shadings of the PDF Reference
    193      * Manual (PDF 32000-1:2008 at the time of this writing).
    194      *
    195      * In the radial gradient problem we are given two circles (c,r) and
    196      * (c,r) that define the gradient itself.
    197      *
    198      * Mathematically the gradient can be defined as the family of circles
    199      *
    200      *     ((1-t)c + t(c), (1-t)r + tr)
    201      *
    202      * excluding those circles whose radius would be < 0. When a point
    203      * belongs to more than one circle, the one with a bigger t is the only
    204      * one that contributes to its color. When a point does not belong
    205      * to any of the circles, it is transparent black, i.e. RGBA (0, 0, 0, 0).
    206      * Further limitations on the range of values for t are imposed when
    207      * the gradient is not repeated, namely t must belong to [0,1].
    208      *
    209      * The graphical result is the same as drawing the valid (radius > 0)
    210      * circles with increasing t in [-inf, +inf] (or in [0,1] if the gradient
    211      * is not repeated) using SOURCE operator composition.
    212      *
    213      * It looks like a cone pointing towards the viewer if the ending circle
    214      * is smaller than the starting one, a cone pointing inside the page if
    215      * the starting circle is the smaller one and like a cylinder if they
    216      * have the same radius.
    217      *
    218      * What we actually do is, given the point whose color we are interested
    219      * in, compute the t values for that point, solving for t in:
    220      *
    221      *     length((1-t)c + t(c) - p) = (1-t)r + tr
    222      *
    223      * Let's rewrite it in a simpler way, by defining some auxiliary
    224      * variables:
    225      *
    226      *     cd = c - c
    227      *     pd = p - c
    228      *     dr = r - r
    229      *     length(tcd - pd) = r + tdr
    230      *
    231      * which actually means
    232      *
    233      *     hypot(tcdx - pdx, tcdy - pdy) = r + tdr
    234      *
    235      * or
    236      *
    237      *     ((tcdx - pdx) + (tcdy - pdy)) = r + tdr.
    238      *
    239      * If we impose (as stated earlier) that r + tdr >= 0, it becomes:
    240      *
    241      *     (tcdx - pdx) + (tcdy - pdy) = (r + tdr)
    242      *
    243      * where we can actually expand the squares and solve for t:
    244      *
    245      *     tcdx - 2tcdxpdx + pdx + tcdy - 2tcdypdy + pdy =
    246      *       = r + 2rtdr + tdr
    247      *
    248      *     (cdx + cdy - dr)t - 2(cdxpdx + cdypdy + rdr)t +
    249      *         (pdx + pdy - r) = 0
    250      *
    251      *     A = cdx + cdy - dr
    252      *     B = pdxcdx + pdycdy + rdr
    253      *     C = pdx + pdy - r
    254      *     At - 2Bt + C = 0
    255      *
    256      * The solutions (unless the equation degenerates because of A = 0) are:
    257      *
    258      *     t = (B  (B - AC)) / A
    259      *
    260      * The solution we are going to prefer is the bigger one, unless the
    261      * radius associated to it is negative (or it falls outside the valid t
    262      * range).
    263      *
    264      * Additional observations (useful for optimizations):
    265      * A does not depend on p
    266      *
    267      * A < 0 <=> one of the two circles completely contains the other one
    268      *   <=> for every p, the radiuses associated with the two t solutions
    269      *       have opposite sign
    270      */
    271     pixman_image_t *image = iter->image;
    272     int x = iter->x;
    273     int y = iter->y;
    274     int width = iter->width;
    275     uint32_t *buffer = iter->buffer;
    276 
    277     gradient_t *gradient = (gradient_t *)image;
    278     radial_gradient_t *radial = (radial_gradient_t *)image;
    279     uint32_t *end = buffer + width * (Bpp / 4);
    280     pixman_gradient_walker_t walker;
    281     pixman_vector_t v, unit;
    282 
    283     /* reference point is the center of the pixel */
    284     v.vector[0] = pixman_int_to_fixed (x) + pixman_fixed_1 / 2;
    285     v.vector[1] = pixman_int_to_fixed (y) + pixman_fixed_1 / 2;
    286     v.vector[2] = pixman_fixed_1;
    287 
    288     _pixman_gradient_walker_init (&walker, gradient, image->common.repeat);
    289 
    290     if (image->common.transform)
    291     {
    292 	if (!pixman_transform_point_3d (image->common.transform, &v))
    293 	    return iter->buffer;
    294 
    295 	unit.vector[0] = image->common.transform->matrix[0][0];
    296 	unit.vector[1] = image->common.transform->matrix[1][0];
    297 	unit.vector[2] = image->common.transform->matrix[2][0];
    298     }
    299     else
    300     {
    301 	unit.vector[0] = pixman_fixed_1;
    302 	unit.vector[1] = 0;
    303 	unit.vector[2] = 0;
    304     }
    305 
    306     if (unit.vector[2] == 0 && v.vector[2] == pixman_fixed_1)
    307     {
    308 	/*
    309 	 * Given:
    310 	 *
    311 	 * t = (B  (B - AC)) / A
    312 	 *
    313 	 * where
    314 	 *
    315 	 * A = cdx + cdy - dr
    316 	 * B = pdxcdx + pdycdy + rdr
    317 	 * C = pdx + pdy - r
    318 	 * det = B - AC
    319 	 *
    320 	 * Since we have an affine transformation, we know that (pdx, pdy)
    321 	 * increase linearly with each pixel,
    322 	 *
    323 	 * pdx = pdx + nux,
    324 	 * pdy = pdy + nuy,
    325 	 *
    326 	 * we can then express B, C and det through multiple differentiation.
    327 	 */
    328 	pixman_fixed_32_32_t b, db, c, dc, ddc;
    329 
    330 	/* warning: this computation may overflow */
    331 	v.vector[0] -= radial->c1.x;
    332 	v.vector[1] -= radial->c1.y;
    333 
    334 	/*
    335 	 * B and C are computed and updated exactly.
    336 	 * If fdot was used instead of dot, in the worst case it would
    337 	 * lose 11 bits of precision in each of the multiplication and
    338 	 * summing up would zero out all the bit that were preserved,
    339 	 * thus making the result 0 instead of the correct one.
    340 	 * This would mean a worst case of unbound relative error or
    341 	 * about 2^10 absolute error
    342 	 */
    343 	b = dot (v.vector[0], v.vector[1], radial->c1.radius,
    344 		 radial->delta.x, radial->delta.y, radial->delta.radius);
    345 	db = dot (unit.vector[0], unit.vector[1], 0,
    346 		  radial->delta.x, radial->delta.y, 0);
    347 
    348 	c = dot (v.vector[0], v.vector[1],
    349 		 -((pixman_fixed_48_16_t) radial->c1.radius),
    350 		 v.vector[0], v.vector[1], radial->c1.radius);
    351 	dc = dot (2 * (pixman_fixed_48_16_t) v.vector[0] + unit.vector[0],
    352 		  2 * (pixman_fixed_48_16_t) v.vector[1] + unit.vector[1],
    353 		  0,
    354 		  unit.vector[0], unit.vector[1], 0);
    355 	ddc = 2 * dot (unit.vector[0], unit.vector[1], 0,
    356 		       unit.vector[0], unit.vector[1], 0);
    357 
    358 	while (buffer < end)
    359 	{
    360 	    if (!mask || *mask++)
    361 	    {
    362 		radial_write_color (radial->a, b, c,
    363 				    radial->inva,
    364 				    radial->delta.radius,
    365 				    radial->mindr,
    366 				    &walker,
    367 				    image->common.repeat,
    368 				    Bpp,
    369 				    write_pixel,
    370 				    buffer);
    371 	    }
    372 
    373 	    b += db;
    374 	    c += dc;
    375 	    dc += ddc;
    376 	    buffer += (Bpp / 4);
    377 	}
    378     }
    379     else
    380     {
    381 	/* projective */
    382 	/* Warning:
    383 	 * error propagation guarantees are much looser than in the affine case
    384 	 */
    385 	while (buffer < end)
    386 	{
    387 	    if (!mask || *mask++)
    388 	    {
    389 		if (v.vector[2] != 0)
    390 		{
    391 		    double pdx, pdy, invv2, b, c;
    392 
    393 		    invv2 = 1. * pixman_fixed_1 / v.vector[2];
    394 
    395 		    pdx = v.vector[0] * invv2 - radial->c1.x;
    396 		    /*    / pixman_fixed_1 */
    397 
    398 		    pdy = v.vector[1] * invv2 - radial->c1.y;
    399 		    /*    / pixman_fixed_1 */
    400 
    401 		    b = fdot (pdx, pdy, radial->c1.radius,
    402 			      radial->delta.x, radial->delta.y,
    403 			      radial->delta.radius);
    404 		    /*  / pixman_fixed_1 / pixman_fixed_1 */
    405 
    406 		    c = fdot (pdx, pdy, -radial->c1.radius,
    407 			      pdx, pdy, radial->c1.radius);
    408 		    /*  / pixman_fixed_1 / pixman_fixed_1 */
    409 
    410 		    radial_write_color (radial->a, b, c,
    411 					radial->inva,
    412 					radial->delta.radius,
    413 					radial->mindr,
    414 					&walker,
    415 					image->common.repeat,
    416 					Bpp,
    417 					write_pixel,
    418 					buffer);
    419 		}
    420 		else
    421 		{
    422 		    memset (buffer, 0, Bpp);
    423 		}
    424 	    }
    425 
    426 	    buffer += (Bpp / 4);
    427 
    428 	    v.vector[0] += unit.vector[0];
    429 	    v.vector[1] += unit.vector[1];
    430 	    v.vector[2] += unit.vector[2];
    431 	}
    432     }
    433 
    434     iter->y++;
    435     return iter->buffer;
    436 }
    437 
    438 static uint32_t *
    439 radial_get_scanline_narrow (pixman_iter_t *iter, const uint32_t *mask)
    440 {
    441     return radial_get_scanline (iter, mask, 4,
    442 				_pixman_gradient_walker_write_narrow);
    443 }
    444 
    445 static uint32_t *
    446 radial_get_scanline_wide (pixman_iter_t *iter, const uint32_t *mask)
    447 {
    448     return radial_get_scanline (iter, NULL, 16,
    449 				_pixman_gradient_walker_write_wide);
    450 }
    451 
    452 void
    453 _pixman_radial_gradient_iter_init (pixman_image_t *image, pixman_iter_t *iter)
    454 {
    455     if (iter->iter_flags & ITER_NARROW)
    456 	iter->get_scanline = radial_get_scanline_narrow;
    457     else
    458 	iter->get_scanline = radial_get_scanline_wide;
    459 }
    460 
    461 PIXMAN_EXPORT pixman_image_t *
    462 pixman_image_create_radial_gradient (const pixman_point_fixed_t *  inner,
    463 				     const pixman_point_fixed_t *  outer,
    464 				     pixman_fixed_t                inner_radius,
    465 				     pixman_fixed_t                outer_radius,
    466 				     const pixman_gradient_stop_t *stops,
    467 				     int                           n_stops)
    468 {
    469     pixman_image_t *image;
    470     radial_gradient_t *radial;
    471 
    472     image = _pixman_image_allocate ();
    473 
    474     if (!image)
    475 	return NULL;
    476 
    477     radial = &image->radial;
    478 
    479     if (!_pixman_init_gradient (&radial->common, stops, n_stops))
    480     {
    481 	free (image);
    482 	return NULL;
    483     }
    484 
    485     image->type = RADIAL;
    486 
    487     radial->c1.x = inner->x;
    488     radial->c1.y = inner->y;
    489     radial->c1.radius = inner_radius;
    490     radial->c2.x = outer->x;
    491     radial->c2.y = outer->y;
    492     radial->c2.radius = outer_radius;
    493 
    494     /* warning: this computations may overflow */
    495     radial->delta.x = radial->c2.x - radial->c1.x;
    496     radial->delta.y = radial->c2.y - radial->c1.y;
    497     radial->delta.radius = radial->c2.radius - radial->c1.radius;
    498 
    499     /* computed exactly, then cast to double -> every bit of the double
    500        representation is correct (53 bits) */
    501     radial->a = dot (radial->delta.x, radial->delta.y, -radial->delta.radius,
    502 		     radial->delta.x, radial->delta.y, radial->delta.radius);
    503     if (radial->a != 0)
    504 	radial->inva = 1. * pixman_fixed_1 / radial->a;
    505 
    506     radial->mindr = -1. * pixman_fixed_1 * radial->c1.radius;
    507 
    508     return image;
    509 }
    510