1/*
2 * Texture block compression
3 * Copyright (C) 2015 Vittorio Giovara <vittorio.giovara@gmail.com>
4 * Based on public domain code by Fabian Giesen, Sean Barrett and Yann Collet.
5 *
6 * This file is part of FFmpeg
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a
9 * copy of this software and associated documentation files (the "Software"),
10 * to deal in the Software without restriction, including without limitation
11 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
12 * and/or sell copies of the Software, and to permit persons to whom the
13 * Software is furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice shall be included
15 * in all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
20 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
23 * IN THE SOFTWARE.
24 */
25
26#include <stddef.h>
27#include <stdint.h>
28
29#include "libavutil/attributes.h"
30#include "libavutil/common.h"
31#include "libavutil/intreadwrite.h"
32
33#include "texturedsp.h"
34
35static const uint8_t expand5[32] = {
36      0,   8,  16,  24,  33,  41,  49,  57,  66,  74,  82,  90,
37     99, 107, 115, 123, 132, 140, 148, 156, 165, 173, 181, 189,
38    198, 206, 214, 222, 231, 239, 247, 255,
39};
40
41static const uint8_t expand6[64] = {
42      0,   4,   8,  12,  16,  20,  24,  28,  32,  36,  40,  44,
43     48,  52,  56,  60,  65,  69,  73,  77,  81,  85,  89,  93,
44     97, 101, 105, 109, 113, 117, 121, 125, 130, 134, 138, 142,
45    146, 150, 154, 158, 162, 166, 170, 174, 178, 182, 186, 190,
46    195, 199, 203, 207, 211, 215, 219, 223, 227, 231, 235, 239,
47    243, 247, 251, 255,
48};
49
50static const uint8_t match5[256][2] = {
51    {  0,  0 }, {  0,  0 }, {  0,  1 }, {  0,  1 }, {  1,  0 }, {  1,  0 },
52    {  1,  0 }, {  1,  1 }, {  1,  1 }, {  2,  0 }, {  2,  0 }, {  0,  4 },
53    {  2,  1 }, {  2,  1 }, {  2,  1 }, {  3,  0 }, {  3,  0 }, {  3,  0 },
54    {  3,  1 }, {  1,  5 }, {  3,  2 }, {  3,  2 }, {  4,  0 }, {  4,  0 },
55    {  4,  1 }, {  4,  1 }, {  4,  2 }, {  4,  2 }, {  4,  2 }, {  3,  5 },
56    {  5,  1 }, {  5,  1 }, {  5,  2 }, {  4,  4 }, {  5,  3 }, {  5,  3 },
57    {  5,  3 }, {  6,  2 }, {  6,  2 }, {  6,  2 }, {  6,  3 }, {  5,  5 },
58    {  6,  4 }, {  6,  4 }, {  4,  8 }, {  7,  3 }, {  7,  3 }, {  7,  3 },
59    {  7,  4 }, {  7,  4 }, {  7,  4 }, {  7,  5 }, {  5,  9 }, {  7,  6 },
60    {  7,  6 }, {  8,  4 }, {  8,  4 }, {  8,  5 }, {  8,  5 }, {  8,  6 },
61    {  8,  6 }, {  8,  6 }, {  7,  9 }, {  9,  5 }, {  9,  5 }, {  9,  6 },
62    {  8,  8 }, {  9,  7 }, {  9,  7 }, {  9,  7 }, { 10,  6 }, { 10,  6 },
63    { 10,  6 }, { 10,  7 }, {  9,  9 }, { 10,  8 }, { 10,  8 }, {  8, 12 },
64    { 11,  7 }, { 11,  7 }, { 11,  7 }, { 11,  8 }, { 11,  8 }, { 11,  8 },
65    { 11,  9 }, {  9, 13 }, { 11, 10 }, { 11, 10 }, { 12,  8 }, { 12,  8 },
66    { 12,  9 }, { 12,  9 }, { 12, 10 }, { 12, 10 }, { 12, 10 }, { 11, 13 },
67    { 13,  9 }, { 13,  9 }, { 13, 10 }, { 12, 12 }, { 13, 11 }, { 13, 11 },
68    { 13, 11 }, { 14, 10 }, { 14, 10 }, { 14, 10 }, { 14, 11 }, { 13, 13 },
69    { 14, 12 }, { 14, 12 }, { 12, 16 }, { 15, 11 }, { 15, 11 }, { 15, 11 },
70    { 15, 12 }, { 15, 12 }, { 15, 12 }, { 15, 13 }, { 13, 17 }, { 15, 14 },
71    { 15, 14 }, { 16, 12 }, { 16, 12 }, { 16, 13 }, { 16, 13 }, { 16, 14 },
72    { 16, 14 }, { 16, 14 }, { 15, 17 }, { 17, 13 }, { 17, 13 }, { 17, 14 },
73    { 16, 16 }, { 17, 15 }, { 17, 15 }, { 17, 15 }, { 18, 14 }, { 18, 14 },
74    { 18, 14 }, { 18, 15 }, { 17, 17 }, { 18, 16 }, { 18, 16 }, { 16, 20 },
75    { 19, 15 }, { 19, 15 }, { 19, 15 }, { 19, 16 }, { 19, 16 }, { 19, 16 },
76    { 19, 17 }, { 17, 21 }, { 19, 18 }, { 19, 18 }, { 20, 16 }, { 20, 16 },
77    { 20, 17 }, { 20, 17 }, { 20, 18 }, { 20, 18 }, { 20, 18 }, { 19, 21 },
78    { 21, 17 }, { 21, 17 }, { 21, 18 }, { 20, 20 }, { 21, 19 }, { 21, 19 },
79    { 21, 19 }, { 22, 18 }, { 22, 18 }, { 22, 18 }, { 22, 19 }, { 21, 21 },
80    { 22, 20 }, { 22, 20 }, { 20, 24 }, { 23, 19 }, { 23, 19 }, { 23, 19 },
81    { 23, 20 }, { 23, 20 }, { 23, 20 }, { 23, 21 }, { 21, 25 }, { 23, 22 },
82    { 23, 22 }, { 24, 20 }, { 24, 20 }, { 24, 21 }, { 24, 21 }, { 24, 22 },
83    { 24, 22 }, { 24, 22 }, { 23, 25 }, { 25, 21 }, { 25, 21 }, { 25, 22 },
84    { 24, 24 }, { 25, 23 }, { 25, 23 }, { 25, 23 }, { 26, 22 }, { 26, 22 },
85    { 26, 22 }, { 26, 23 }, { 25, 25 }, { 26, 24 }, { 26, 24 }, { 24, 28 },
86    { 27, 23 }, { 27, 23 }, { 27, 23 }, { 27, 24 }, { 27, 24 }, { 27, 24 },
87    { 27, 25 }, { 25, 29 }, { 27, 26 }, { 27, 26 }, { 28, 24 }, { 28, 24 },
88    { 28, 25 }, { 28, 25 }, { 28, 26 }, { 28, 26 }, { 28, 26 }, { 27, 29 },
89    { 29, 25 }, { 29, 25 }, { 29, 26 }, { 28, 28 }, { 29, 27 }, { 29, 27 },
90    { 29, 27 }, { 30, 26 }, { 30, 26 }, { 30, 26 }, { 30, 27 }, { 29, 29 },
91    { 30, 28 }, { 30, 28 }, { 30, 28 }, { 31, 27 }, { 31, 27 }, { 31, 27 },
92    { 31, 28 }, { 31, 28 }, { 31, 28 }, { 31, 29 }, { 31, 29 }, { 31, 30 },
93    { 31, 30 }, { 31, 30 }, { 31, 31 }, { 31, 31 },
94};
95
96static const uint8_t match6[256][2] = {
97    {  0,  0 }, {  0,  1 }, {  1,  0 }, {  1,  0 }, {  1,  1 }, {  2,  0 },
98    {  2,  1 }, {  3,  0 }, {  3,  0 }, {  3,  1 }, {  4,  0 }, {  4,  0 },
99    {  4,  1 }, {  5,  0 }, {  5,  1 }, {  6,  0 }, {  6,  0 }, {  6,  1 },
100    {  7,  0 }, {  7,  0 }, {  7,  1 }, {  8,  0 }, {  8,  1 }, {  8,  1 },
101    {  8,  2 }, {  9,  1 }, {  9,  2 }, {  9,  2 }, {  9,  3 }, { 10,  2 },
102    { 10,  3 }, { 10,  3 }, { 10,  4 }, { 11,  3 }, { 11,  4 }, { 11,  4 },
103    { 11,  5 }, { 12,  4 }, { 12,  5 }, { 12,  5 }, { 12,  6 }, { 13,  5 },
104    { 13,  6 }, {  8, 16 }, { 13,  7 }, { 14,  6 }, { 14,  7 }, {  9, 17 },
105    { 14,  8 }, { 15,  7 }, { 15,  8 }, { 11, 16 }, { 15,  9 }, { 15, 10 },
106    { 16,  8 }, { 16,  9 }, { 16, 10 }, { 15, 13 }, { 17,  9 }, { 17, 10 },
107    { 17, 11 }, { 15, 16 }, { 18, 10 }, { 18, 11 }, { 18, 12 }, { 16, 16 },
108    { 19, 11 }, { 19, 12 }, { 19, 13 }, { 17, 17 }, { 20, 12 }, { 20, 13 },
109    { 20, 14 }, { 19, 16 }, { 21, 13 }, { 21, 14 }, { 21, 15 }, { 20, 17 },
110    { 22, 14 }, { 22, 15 }, { 25, 10 }, { 22, 16 }, { 23, 15 }, { 23, 16 },
111    { 26, 11 }, { 23, 17 }, { 24, 16 }, { 24, 17 }, { 27, 12 }, { 24, 18 },
112    { 25, 17 }, { 25, 18 }, { 28, 13 }, { 25, 19 }, { 26, 18 }, { 26, 19 },
113    { 29, 14 }, { 26, 20 }, { 27, 19 }, { 27, 20 }, { 30, 15 }, { 27, 21 },
114    { 28, 20 }, { 28, 21 }, { 28, 21 }, { 28, 22 }, { 29, 21 }, { 29, 22 },
115    { 24, 32 }, { 29, 23 }, { 30, 22 }, { 30, 23 }, { 25, 33 }, { 30, 24 },
116    { 31, 23 }, { 31, 24 }, { 27, 32 }, { 31, 25 }, { 31, 26 }, { 32, 24 },
117    { 32, 25 }, { 32, 26 }, { 31, 29 }, { 33, 25 }, { 33, 26 }, { 33, 27 },
118    { 31, 32 }, { 34, 26 }, { 34, 27 }, { 34, 28 }, { 32, 32 }, { 35, 27 },
119    { 35, 28 }, { 35, 29 }, { 33, 33 }, { 36, 28 }, { 36, 29 }, { 36, 30 },
120    { 35, 32 }, { 37, 29 }, { 37, 30 }, { 37, 31 }, { 36, 33 }, { 38, 30 },
121    { 38, 31 }, { 41, 26 }, { 38, 32 }, { 39, 31 }, { 39, 32 }, { 42, 27 },
122    { 39, 33 }, { 40, 32 }, { 40, 33 }, { 43, 28 }, { 40, 34 }, { 41, 33 },
123    { 41, 34 }, { 44, 29 }, { 41, 35 }, { 42, 34 }, { 42, 35 }, { 45, 30 },
124    { 42, 36 }, { 43, 35 }, { 43, 36 }, { 46, 31 }, { 43, 37 }, { 44, 36 },
125    { 44, 37 }, { 44, 37 }, { 44, 38 }, { 45, 37 }, { 45, 38 }, { 40, 48 },
126    { 45, 39 }, { 46, 38 }, { 46, 39 }, { 41, 49 }, { 46, 40 }, { 47, 39 },
127    { 47, 40 }, { 43, 48 }, { 47, 41 }, { 47, 42 }, { 48, 40 }, { 48, 41 },
128    { 48, 42 }, { 47, 45 }, { 49, 41 }, { 49, 42 }, { 49, 43 }, { 47, 48 },
129    { 50, 42 }, { 50, 43 }, { 50, 44 }, { 48, 48 }, { 51, 43 }, { 51, 44 },
130    { 51, 45 }, { 49, 49 }, { 52, 44 }, { 52, 45 }, { 52, 46 }, { 51, 48 },
131    { 53, 45 }, { 53, 46 }, { 53, 47 }, { 52, 49 }, { 54, 46 }, { 54, 47 },
132    { 57, 42 }, { 54, 48 }, { 55, 47 }, { 55, 48 }, { 58, 43 }, { 55, 49 },
133    { 56, 48 }, { 56, 49 }, { 59, 44 }, { 56, 50 }, { 57, 49 }, { 57, 50 },
134    { 60, 45 }, { 57, 51 }, { 58, 50 }, { 58, 51 }, { 61, 46 }, { 58, 52 },
135    { 59, 51 }, { 59, 52 }, { 62, 47 }, { 59, 53 }, { 60, 52 }, { 60, 53 },
136    { 60, 53 }, { 60, 54 }, { 61, 53 }, { 61, 54 }, { 61, 54 }, { 61, 55 },
137    { 62, 54 }, { 62, 55 }, { 62, 55 }, { 62, 56 }, { 63, 55 }, { 63, 56 },
138    { 63, 56 }, { 63, 57 }, { 63, 58 }, { 63, 59 }, { 63, 59 }, { 63, 60 },
139    { 63, 61 }, { 63, 62 }, { 63, 62 }, { 63, 63 },
140};
141
142/* Multiplication over 8 bit emulation */
143#define mul8(a, b) (((a) * (b) + 128 + (((a) * (b) + 128) >> 8)) >> 8)
144
145/* Conversion from rgb24 to rgb565 */
146#define rgb2rgb565(r, g, b) \
147    ((mul8(r, 31) << 11) | (mul8(g, 63) << 5) | (mul8(b, 31) << 0))
148
149/* Linear interpolation at 1/3 point between a and b */
150#define lerp13(a, b) ((2 * (a) + (b)) / 3)
151
152/* Linear interpolation on an RGB pixel */
153static inline void lerp13rgb(uint8_t *out, uint8_t *p1, uint8_t *p2)
154{
155    out[0] = lerp13(p1[0], p2[0]);
156    out[1] = lerp13(p1[1], p2[1]);
157    out[2] = lerp13(p1[2], p2[2]);
158}
159
160/* Conversion from rgb565 to rgb24 */
161static inline void rgb5652rgb(uint8_t *out, uint16_t v)
162{
163    int rv = (v & 0xf800) >> 11;
164    int gv = (v & 0x07e0) >> 5;
165    int bv = (v & 0x001f) >> 0;
166
167    out[0] = expand5[rv];
168    out[1] = expand6[gv];
169    out[2] = expand5[bv];
170    out[3] = 0;
171}
172
173/* Color matching function */
174static unsigned int match_colors(const uint8_t *block, ptrdiff_t stride,
175                                 uint16_t c0, uint16_t c1)
176{
177    uint32_t mask = 0;
178    int dirr, dirg, dirb;
179    int dots[16];
180    int stops[4];
181    int x, y, k = 0;
182    int c0_point, half_point, c3_point;
183    uint8_t color[16];
184    static const uint32_t indexMap[8] = {
185        0U << 30, 2U << 30, 0U << 30, 2U << 30,
186        3U << 30, 3U << 30, 1U << 30, 1U << 30,
187    };
188
189    /* Fill color and compute direction for each component */
190    rgb5652rgb(color + 0, c0);
191    rgb5652rgb(color + 4, c1);
192    lerp13rgb(color + 8, color + 0, color + 4);
193    lerp13rgb(color + 12, color + 4, color + 0);
194
195    dirr = color[0 * 4 + 0] - color[1 * 4 + 0];
196    dirg = color[0 * 4 + 1] - color[1 * 4 + 1];
197    dirb = color[0 * 4 + 2] - color[1 * 4 + 2];
198
199    for (y = 0; y < 4; y++) {
200        for (x = 0; x < 4; x++)
201            dots[k++] = block[0 + x * 4 + y * stride] * dirr +
202                        block[1 + x * 4 + y * stride] * dirg +
203                        block[2 + x * 4 + y * stride] * dirb;
204
205        stops[y] = color[0 + y * 4] * dirr +
206                   color[1 + y * 4] * dirg +
207                   color[2 + y * 4] * dirb;
208    }
209
210    /* Think of the colors as arranged on a line; project point onto that line,
211     * then choose next color out of available ones. we compute the crossover
212     * points for 'best color in top half'/'best in bottom half' and then
213     * the same inside that subinterval.
214     *
215     * Relying on this 1d approximation isn't always optimal in terms of
216     * Euclidean distance, but it's very close and a lot faster.
217     *
218     * http://cbloomrants.blogspot.com/2008/12/12-08-08-dxtc-summary.html */
219    c0_point   = (stops[1] + stops[3]) >> 1;
220    half_point = (stops[3] + stops[2]) >> 1;
221    c3_point   = (stops[2] + stops[0]) >> 1;
222
223    for (x = 0; x < 16; x++) {
224        int dot  = dots[x];
225        int bits = (dot < half_point ? 4 : 0) |
226                   (dot < c0_point   ? 2 : 0) |
227                   (dot < c3_point   ? 1 : 0);
228
229        mask >>= 2;
230        mask  |= indexMap[bits];
231    }
232
233    return mask;
234}
235
236/* Color optimization function */
237static void optimize_colors(const uint8_t *block, ptrdiff_t stride,
238                            uint16_t *pmax16, uint16_t *pmin16)
239{
240    const uint8_t *minp;
241    const uint8_t *maxp;
242    const int iter_power = 4;
243    double magn;
244    int v_r, v_g, v_b;
245    float covf[6], vfr, vfg, vfb;
246    int mind, maxd;
247    int cov[6] = { 0 };
248    int mu[3], min[3], max[3];
249    int ch, iter, x, y;
250
251    /* Determine color distribution */
252    for (ch = 0; ch < 3; ch++) {
253        const uint8_t *bp = &block[ch];
254        int muv, minv, maxv;
255
256        muv = minv = maxv = bp[0];
257        for (y = 0; y < 4; y++) {
258            for (x = 0; x < 4; x++) {
259                muv += bp[x * 4 + y * stride];
260                if (bp[x * 4 + y * stride] < minv)
261                    minv = bp[x * 4 + y * stride];
262                else if (bp[x * 4 + y * stride] > maxv)
263                    maxv = bp[x * 4 + y * stride];
264            }
265        }
266
267        mu[ch]  = (muv + 8) >> 4;
268        min[ch] = minv;
269        max[ch] = maxv;
270    }
271
272    /* Determine covariance matrix */
273    for (y = 0; y < 4; y++) {
274        for (x = 0; x < 4; x++) {
275            int r = block[x * 4 + stride * y + 0] - mu[0];
276            int g = block[x * 4 + stride * y + 1] - mu[1];
277            int b = block[x * 4 + stride * y + 2] - mu[2];
278
279            cov[0] += r * r;
280            cov[1] += r * g;
281            cov[2] += r * b;
282            cov[3] += g * g;
283            cov[4] += g * b;
284            cov[5] += b * b;
285        }
286    }
287
288    /* Convert covariance matrix to float, find principal axis via power iter */
289    for (x = 0; x < 6; x++)
290        covf[x] = cov[x] / 255.0f;
291
292    vfr = (float) (max[0] - min[0]);
293    vfg = (float) (max[1] - min[1]);
294    vfb = (float) (max[2] - min[2]);
295
296    for (iter = 0; iter < iter_power; iter++) {
297        float r = vfr * covf[0] + vfg * covf[1] + vfb * covf[2];
298        float g = vfr * covf[1] + vfg * covf[3] + vfb * covf[4];
299        float b = vfr * covf[2] + vfg * covf[4] + vfb * covf[5];
300
301        vfr = r;
302        vfg = g;
303        vfb = b;
304    }
305
306    magn = fabs(vfr);
307    if (fabs(vfg) > magn)
308        magn = fabs(vfg);
309    if (fabs(vfb) > magn)
310        magn = fabs(vfb);
311
312    /* if magnitude is too small, default to luminance */
313    if (magn < 4.0f) {
314        /* JPEG YCbCr luma coefs, scaled by 1000 */
315        v_r = 299;
316        v_g = 587;
317        v_b = 114;
318    } else {
319        magn = 512.0 / magn;
320        v_r  = (int) (vfr * magn);
321        v_g  = (int) (vfg * magn);
322        v_b  = (int) (vfb * magn);
323    }
324
325    /* Pick colors at extreme points */
326    mind = maxd = block[0] * v_r + block[1] * v_g + block[2] * v_b;
327    minp = maxp = block;
328    for (y = 0; y < 4; y++) {
329        for (x = 0; x < 4; x++) {
330            int dot = block[x * 4 + y * stride + 0] * v_r +
331                      block[x * 4 + y * stride + 1] * v_g +
332                      block[x * 4 + y * stride + 2] * v_b;
333
334            if (dot < mind) {
335                mind = dot;
336                minp = block + x * 4 + y * stride;
337            } else if (dot > maxd) {
338                maxd = dot;
339                maxp = block + x * 4 + y * stride;
340            }
341        }
342    }
343
344    *pmax16 = rgb2rgb565(maxp[0], maxp[1], maxp[2]);
345    *pmin16 = rgb2rgb565(minp[0], minp[1], minp[2]);
346}
347
348/* Try to optimize colors to suit block contents better, by solving
349 * a least squares system via normal equations + Cramer's rule. */
350static int refine_colors(const uint8_t *block, ptrdiff_t stride,
351                         uint16_t *pmax16, uint16_t *pmin16, uint32_t mask)
352{
353    uint32_t cm = mask;
354    uint16_t oldMin = *pmin16;
355    uint16_t oldMax = *pmax16;
356    uint16_t min16, max16;
357    int x, y;
358
359    /* Additional magic to save a lot of multiplies in the accumulating loop.
360     * The tables contain precomputed products of weights for least squares
361     * system, accumulated inside one 32-bit register */
362    static const int w1tab[4] = { 3, 0, 2, 1 };
363    static const int prods[4] = { 0x090000, 0x000900, 0x040102, 0x010402 };
364
365    /* Check if all pixels have the same index */
366    if ((mask ^ (mask << 2)) < 4) {
367        /* If so, linear system would be singular; solve using optimal
368         * single-color match on average color. */
369        int r = 8, g = 8, b = 8;
370        for (y = 0; y < 4; y++) {
371            for (x = 0; x < 4; x++) {
372                r += block[0 + x * 4 + y * stride];
373                g += block[1 + x * 4 + y * stride];
374                b += block[2 + x * 4 + y * stride];
375            }
376        }
377
378        r >>= 4;
379        g >>= 4;
380        b >>= 4;
381
382        max16 = (match5[r][0] << 11) | (match6[g][0] << 5) | match5[b][0];
383        min16 = (match5[r][1] << 11) | (match6[g][1] << 5) | match5[b][1];
384    } else {
385        float fr, fg, fb;
386        int at1_r = 0, at1_g = 0, at1_b = 0;
387        int at2_r = 0, at2_g = 0, at2_b = 0;
388        int akku = 0;
389        int xx, xy, yy;
390
391        for (y = 0; y < 4; y++) {
392            for (x = 0; x < 4; x++) {
393                int step = cm & 3;
394                int w1 = w1tab[step];
395                int r = block[0 + x * 4 + y * stride];
396                int g = block[1 + x * 4 + y * stride];
397                int b = block[2 + x * 4 + y * stride];
398
399                akku  += prods[step];
400                at1_r += w1 * r;
401                at1_g += w1 * g;
402                at1_b += w1 * b;
403                at2_r += r;
404                at2_g += g;
405                at2_b += b;
406
407                cm >>= 2;
408            }
409        }
410
411        at2_r = 3 * at2_r - at1_r;
412        at2_g = 3 * at2_g - at1_g;
413        at2_b = 3 * at2_b - at1_b;
414
415        /* Extract solutions and decide solvability */
416        xx =  akku >> 16;
417        yy = (akku >>  8) & 0xFF;
418        xy = (akku >>  0) & 0xFF;
419
420        fr = 3.0f * 31.0f / 255.0f / (xx * yy - xy * xy);
421        fg = fr * 63.0f / 31.0f;
422        fb = fr;
423
424        /* Solve */
425        max16  = av_clip_uintp2((at1_r * yy - at2_r * xy) * fr + 0.5f, 5) << 11;
426        max16 |= av_clip_uintp2((at1_g * yy - at2_g * xy) * fg + 0.5f, 6) <<  5;
427        max16 |= av_clip_uintp2((at1_b * yy - at2_b * xy) * fb + 0.5f, 5) <<  0;
428
429        min16  = av_clip_uintp2((at2_r * xx - at1_r * xy) * fr + 0.5f, 5) << 11;
430        min16 |= av_clip_uintp2((at2_g * xx - at1_g * xy) * fg + 0.5f, 6) <<  5;
431        min16 |= av_clip_uintp2((at2_b * xx - at1_b * xy) * fb + 0.5f, 5) <<  0;
432    }
433
434    *pmin16 = min16;
435    *pmax16 = max16;
436    return oldMin != min16 || oldMax != max16;
437}
438
439/* Check if input block is a constant color */
440static int constant_color(const uint8_t *block, ptrdiff_t stride)
441{
442    int x, y;
443    uint32_t first = AV_RL32(block);
444
445    for (y = 0; y < 4; y++)
446        for (x = 0; x < 4; x++)
447            if (first != AV_RL32(block + x * 4 + y * stride))
448                return 0;
449    return 1;
450}
451
452/* Main color compression function */
453static void compress_color(uint8_t *dst, ptrdiff_t stride, const uint8_t *block)
454{
455    uint32_t mask;
456    uint16_t max16, min16;
457    int constant = constant_color(block, stride);
458
459    /* Constant color will load values from tables */
460    if (constant) {
461        int r = block[0];
462        int g = block[1];
463        int b = block[2];
464        mask  = 0xAAAAAAAA;
465        max16 = (match5[r][0] << 11) | (match6[g][0] << 5) | match5[b][0];
466        min16 = (match5[r][1] << 11) | (match6[g][1] << 5) | match5[b][1];
467    } else {
468        int refine;
469
470        /* Otherwise find pca and map along principal axis */
471        optimize_colors(block, stride, &max16, &min16);
472        if (max16 != min16)
473            mask = match_colors(block, stride, max16, min16);
474        else
475            mask = 0;
476
477        /* One pass refinement */
478        refine  = refine_colors(block, stride, &max16, &min16, mask);
479        if (refine) {
480            if (max16 != min16)
481                mask = match_colors(block, stride, max16, min16);
482            else
483                mask = 0;
484        }
485    }
486
487    /* Finally write the color block */
488    if (max16 < min16) {
489        FFSWAP(uint16_t, min16, max16);
490        mask ^= 0x55555555;
491    }
492
493    AV_WL16(dst + 0, max16);
494    AV_WL16(dst + 2, min16);
495    AV_WL32(dst + 4, mask);
496}
497
498/* Alpha compression function */
499static void compress_alpha(uint8_t *dst, ptrdiff_t stride, const uint8_t *block)
500{
501    int x, y;
502    int dist, bias, dist4, dist2;
503    int mn, mx;
504    int bits = 0;
505    int mask = 0;
506
507    memset(dst, 0, 8);
508
509    /* Find min/max color */
510    mn = mx = block[3];
511    for (y = 0; y < 4; y++) {
512        for (x = 0; x < 4; x++) {
513            int val = block[3 + x * 4 + y * stride];
514            if (val < mn)
515                mn = val;
516            else if (val > mx)
517                mx = val;
518        }
519    }
520
521    /* Encode them */
522    dst[0] = (uint8_t) mx;
523    dst[1] = (uint8_t) mn;
524    dst += 2;
525
526    /* Mono-alpha shortcut */
527    if (mn == mx)
528        return;
529
530    /* Determine bias and emit color indices.
531     * Given the choice of mx/mn, these indices are optimal:
532     * fgiesen.wordpress.com/2009/12/15/dxt5-alpha-block-index-determination */
533    dist = mx - mn;
534
535    dist4 = dist * 4;
536    dist2 = dist * 2;
537    if (dist < 8)
538        bias = dist - 1 - mn * 7;
539    else
540        bias = dist / 2 + 2 - mn * 7;
541
542    for (y = 0; y < 4; y++) {
543        for (x = 0; x < 4; x++) {
544            int alp = block[3 + x * 4 + y * stride] * 7 + bias;
545            int ind, tmp;
546
547            /* This is a "linear scale" lerp factor between 0 (val=min)
548             * and 7 (val=max) to select index. */
549            tmp  = (alp >= dist4) ? -1 : 0;
550            ind  = tmp & 4;
551            alp -= dist4 & tmp;
552            tmp  = (alp >= dist2) ? -1 : 0;
553            ind += tmp & 2;
554            alp -= dist2 & tmp;
555            ind += (alp >= dist);
556
557            /* Turn linear scale into DXT index (0/1 are extreme points) */
558            ind  = -ind & 7;
559            ind ^= (2 > ind);
560
561            /* Write index */
562            mask |= ind << bits;
563            bits += 3;
564            if (bits >= 8) {
565                *dst++ = mask;
566                mask >>= 8;
567                bits  -= 8;
568            }
569        }
570    }
571}
572
573/**
574 * Convert a RGBA buffer to unscaled YCoCg.
575 * Scale is usually introduced to avoid banding over a certain range of colors,
576 * but this version of the algorithm does not introduce it as much as other
577 * implementations, allowing for a simpler and faster conversion.
578 */
579static void rgba2ycocg(uint8_t *dst, const uint8_t *pixel)
580{
581    int r =  pixel[0];
582    int g = (pixel[1] + 1) >> 1;
583    int b =  pixel[2];
584    int t = (2 + r + b) >> 2;
585
586    dst[0] = av_clip_uint8(128 + ((r - b + 1) >> 1));   /* Co */
587    dst[1] = av_clip_uint8(128 + g - t);                /* Cg */
588    dst[2] = 0;
589    dst[3] = av_clip_uint8(g + t);                      /* Y */
590}
591
592/**
593 * Compress one block of RGBA pixels in a DXT1 texture and store the
594 * resulting bytes in 'dst'. Alpha is not preserved.
595 *
596 * @param dst    output buffer.
597 * @param stride scanline in bytes.
598 * @param block  block to compress.
599 * @return how much texture data has been written.
600 */
601static int dxt1_block(uint8_t *dst, ptrdiff_t stride, const uint8_t *block)
602{
603    compress_color(dst, stride, block);
604
605    return 8;
606}
607
608/**
609 * Compress one block of RGBA pixels in a DXT5 texture and store the
610 * resulting bytes in 'dst'. Alpha is preserved.
611 *
612 * @param dst    output buffer.
613 * @param stride scanline in bytes.
614 * @param block  block to compress.
615 * @return how much texture data has been written.
616 */
617static int dxt5_block(uint8_t *dst, ptrdiff_t stride, const uint8_t *block)
618{
619    compress_alpha(dst, stride, block);
620    compress_color(dst + 8, stride, block);
621
622    return 16;
623}
624
625/**
626 * Compress one block of RGBA pixels in a DXT5-YCoCg texture and store the
627 * resulting bytes in 'dst'. Alpha is not preserved.
628 *
629 * @param dst    output buffer.
630 * @param stride scanline in bytes.
631 * @param block  block to compress.
632 * @return how much texture data has been written.
633 */
634static int dxt5ys_block(uint8_t *dst, ptrdiff_t stride, const uint8_t *block)
635{
636    int x, y;
637    uint8_t reorder[64];
638
639    /* Reorder the components and then run a normal DXT5 compression. */
640    for (y = 0; y < 4; y++)
641        for (x = 0; x < 4; x++)
642            rgba2ycocg(reorder + x * 4 + y * 16, block + x * 4 + y * stride);
643
644    compress_alpha(dst + 0, 16, reorder);
645    compress_color(dst + 8, 16, reorder);
646
647    return 16;
648}
649
650/**
651 * Compress one block of RGBA pixels in a RGTC1U texture and store the
652 * resulting bytes in 'dst'. Use the alpha channel of the input image.
653 *
654 * @param dst    output buffer.
655 * @param stride scanline in bytes.
656 * @param block  block to compress.
657 * @return how much texture data has been written.
658 */
659static int rgtc1u_alpha_block(uint8_t *dst, ptrdiff_t stride, const uint8_t *block)
660{
661    compress_alpha(dst, stride, block);
662
663    return 8;
664}
665
666av_cold void ff_texturedspenc_init(TextureDSPContext *c)
667{
668    c->dxt1_block         = dxt1_block;
669    c->dxt5_block         = dxt5_block;
670    c->dxt5ys_block       = dxt5ys_block;
671    c->rgtc1u_alpha_block = rgtc1u_alpha_block;
672}
673
674#define TEXTUREDSP_FUNC_NAME ff_texturedsp_compress_thread
675#define TEXTUREDSP_TEX_FUNC(a, b, c) tex_funct(c, b, a)
676#include "texturedsp_template.c"
677