OLD | NEW |
(Empty) | |
| 1 // Copyright 2012 Google Inc. All Rights Reserved. |
| 2 // |
| 3 // This code is licensed under the same terms as WebM: |
| 4 // Software License Agreement: http://www.webmproject.org/license/software/ |
| 5 // Additional IP Rights Grant: http://www.webmproject.org/license/additional/ |
| 6 // ----------------------------------------------------------------------------- |
| 7 // |
| 8 // Image transforms and color space conversion methods for lossless decoder. |
| 9 // |
| 10 // Authors: Vikas Arora (vikaas.arora@gmail.com) |
| 11 // Jyrki Alakuijala (jyrki@google.com) |
| 12 // Urvang Joshi (urvang@google.com) |
| 13 |
| 14 #if defined(__cplusplus) || defined(c_plusplus) |
| 15 extern "C" { |
| 16 #endif |
| 17 |
| 18 #include <math.h> |
| 19 #include <stdlib.h> |
| 20 #include "./lossless.h" |
| 21 #include "../dec/vp8li.h" |
| 22 #include "../dsp/yuv.h" |
| 23 #include "../dsp/dsp.h" |
| 24 #include "../enc/histogram.h" |
| 25 |
| 26 #define MAX_DIFF_COST (1e30f) |
| 27 |
| 28 // lookup table for small values of log2(int) |
| 29 #define APPROX_LOG_MAX 4096 |
| 30 #define LOG_2_RECIPROCAL 1.44269504088896338700465094007086 |
| 31 #define LOG_LOOKUP_IDX_MAX 256 |
| 32 static const float kLog2Table[LOG_LOOKUP_IDX_MAX] = { |
| 33 0.0000000000000000f, 0.0000000000000000f, |
| 34 1.0000000000000000f, 1.5849625007211560f, |
| 35 2.0000000000000000f, 2.3219280948873621f, |
| 36 2.5849625007211560f, 2.8073549220576041f, |
| 37 3.0000000000000000f, 3.1699250014423121f, |
| 38 3.3219280948873621f, 3.4594316186372973f, |
| 39 3.5849625007211560f, 3.7004397181410921f, |
| 40 3.8073549220576041f, 3.9068905956085187f, |
| 41 4.0000000000000000f, 4.0874628412503390f, |
| 42 4.1699250014423121f, 4.2479275134435852f, |
| 43 4.3219280948873626f, 4.3923174227787606f, |
| 44 4.4594316186372973f, 4.5235619560570130f, |
| 45 4.5849625007211560f, 4.6438561897747243f, |
| 46 4.7004397181410917f, 4.7548875021634682f, |
| 47 4.8073549220576037f, 4.8579809951275718f, |
| 48 4.9068905956085187f, 4.9541963103868749f, |
| 49 5.0000000000000000f, 5.0443941193584533f, |
| 50 5.0874628412503390f, 5.1292830169449663f, |
| 51 5.1699250014423121f, 5.2094533656289501f, |
| 52 5.2479275134435852f, 5.2854022188622487f, |
| 53 5.3219280948873626f, 5.3575520046180837f, |
| 54 5.3923174227787606f, 5.4262647547020979f, |
| 55 5.4594316186372973f, 5.4918530963296747f, |
| 56 5.5235619560570130f, 5.5545888516776376f, |
| 57 5.5849625007211560f, 5.6147098441152083f, |
| 58 5.6438561897747243f, 5.6724253419714951f, |
| 59 5.7004397181410917f, 5.7279204545631987f, |
| 60 5.7548875021634682f, 5.7813597135246599f, |
| 61 5.8073549220576037f, 5.8328900141647412f, |
| 62 5.8579809951275718f, 5.8826430493618415f, |
| 63 5.9068905956085187f, 5.9307373375628866f, |
| 64 5.9541963103868749f, 5.9772799234999167f, |
| 65 6.0000000000000000f, 6.0223678130284543f, |
| 66 6.0443941193584533f, 6.0660891904577720f, |
| 67 6.0874628412503390f, 6.1085244567781691f, |
| 68 6.1292830169449663f, 6.1497471195046822f, |
| 69 6.1699250014423121f, 6.1898245588800175f, |
| 70 6.2094533656289501f, 6.2288186904958804f, |
| 71 6.2479275134435852f, 6.2667865406949010f, |
| 72 6.2854022188622487f, 6.3037807481771030f, |
| 73 6.3219280948873626f, 6.3398500028846243f, |
| 74 6.3575520046180837f, 6.3750394313469245f, |
| 75 6.3923174227787606f, 6.4093909361377017f, |
| 76 6.4262647547020979f, 6.4429434958487279f, |
| 77 6.4594316186372973f, 6.4757334309663976f, |
| 78 6.4918530963296747f, 6.5077946401986963f, |
| 79 6.5235619560570130f, 6.5391588111080309f, |
| 80 6.5545888516776376f, 6.5698556083309478f, |
| 81 6.5849625007211560f, 6.5999128421871278f, |
| 82 6.6147098441152083f, 6.6293566200796094f, |
| 83 6.6438561897747243f, 6.6582114827517946f, |
| 84 6.6724253419714951f, 6.6865005271832185f, |
| 85 6.7004397181410917f, 6.7142455176661224f, |
| 86 6.7279204545631987f, 6.7414669864011464f, |
| 87 6.7548875021634682f, 6.7681843247769259f, |
| 88 6.7813597135246599f, 6.7944158663501061f, |
| 89 6.8073549220576037f, 6.8201789624151878f, |
| 90 6.8328900141647412f, 6.8454900509443747f, |
| 91 6.8579809951275718f, 6.8703647195834047f, |
| 92 6.8826430493618415f, 6.8948177633079437f, |
| 93 6.9068905956085187f, 6.9188632372745946f, |
| 94 6.9307373375628866f, 6.9425145053392398f, |
| 95 6.9541963103868749f, 6.9657842846620869f, |
| 96 6.9772799234999167f, 6.9886846867721654f, |
| 97 7.0000000000000000f, 7.0112272554232539f, |
| 98 7.0223678130284543f, 7.0334230015374501f, |
| 99 7.0443941193584533f, 7.0552824355011898f, |
| 100 7.0660891904577720f, 7.0768155970508308f, |
| 101 7.0874628412503390f, 7.0980320829605263f, |
| 102 7.1085244567781691f, 7.1189410727235076f, |
| 103 7.1292830169449663f, 7.1395513523987936f, |
| 104 7.1497471195046822f, 7.1598713367783890f, |
| 105 7.1699250014423121f, 7.1799090900149344f, |
| 106 7.1898245588800175f, 7.1996723448363644f, |
| 107 7.2094533656289501f, 7.2191685204621611f, |
| 108 7.2288186904958804f, 7.2384047393250785f, |
| 109 7.2479275134435852f, 7.2573878426926521f, |
| 110 7.2667865406949010f, 7.2761244052742375f, |
| 111 7.2854022188622487f, 7.2946207488916270f, |
| 112 7.3037807481771030f, 7.3128829552843557f, |
| 113 7.3219280948873626f, 7.3309168781146167f, |
| 114 7.3398500028846243f, 7.3487281542310771f, |
| 115 7.3575520046180837f, 7.3663222142458160f, |
| 116 7.3750394313469245f, 7.3837042924740519f, |
| 117 7.3923174227787606f, 7.4008794362821843f, |
| 118 7.4093909361377017f, 7.4178525148858982f, |
| 119 7.4262647547020979f, 7.4346282276367245f, |
| 120 7.4429434958487279f, 7.4512111118323289f, |
| 121 7.4594316186372973f, 7.4676055500829976f, |
| 122 7.4757334309663976f, 7.4838157772642563f, |
| 123 7.4918530963296747f, 7.4998458870832056f, |
| 124 7.5077946401986963f, 7.5156998382840427f, |
| 125 7.5235619560570130f, 7.5313814605163118f, |
| 126 7.5391588111080309f, 7.5468944598876364f, |
| 127 7.5545888516776376f, 7.5622424242210728f, |
| 128 7.5698556083309478f, 7.5774288280357486f, |
| 129 7.5849625007211560f, 7.5924570372680806f, |
| 130 7.5999128421871278f, 7.6073303137496104f, |
| 131 7.6147098441152083f, 7.6220518194563764f, |
| 132 7.6293566200796094f, 7.6366246205436487f, |
| 133 7.6438561897747243f, 7.6510516911789281f, |
| 134 7.6582114827517946f, 7.6653359171851764f, |
| 135 7.6724253419714951f, 7.6794800995054464f, |
| 136 7.6865005271832185f, 7.6934869574993252f, |
| 137 7.7004397181410917f, 7.7073591320808825f, |
| 138 7.7142455176661224f, 7.7210991887071855f, |
| 139 7.7279204545631987f, 7.7347096202258383f, |
| 140 7.7414669864011464f, 7.7481928495894605f, |
| 141 7.7548875021634682f, 7.7615512324444795f, |
| 142 7.7681843247769259f, 7.7747870596011736f, |
| 143 7.7813597135246599f, 7.7879025593914317f, |
| 144 7.7944158663501061f, 7.8008998999203047f, |
| 145 7.8073549220576037f, 7.8137811912170374f, |
| 146 7.8201789624151878f, 7.8265484872909150f, |
| 147 7.8328900141647412f, 7.8392037880969436f, |
| 148 7.8454900509443747f, 7.8517490414160571f, |
| 149 7.8579809951275718f, 7.8641861446542797f, |
| 150 7.8703647195834047f, 7.8765169465649993f, |
| 151 7.8826430493618415f, 7.8887432488982591f, |
| 152 7.8948177633079437f, 7.9008668079807486f, |
| 153 7.9068905956085187f, 7.9128893362299619f, |
| 154 7.9188632372745946f, 7.9248125036057812f, |
| 155 7.9307373375628866f, 7.9366379390025709f, |
| 156 7.9425145053392398f, 7.9483672315846778f, |
| 157 7.9541963103868749f, 7.9600019320680805f, |
| 158 7.9657842846620869f, 7.9715435539507719f, |
| 159 7.9772799234999167f, 7.9829935746943103f, |
| 160 7.9886846867721654f, 7.9943534368588577f |
| 161 }; |
| 162 |
| 163 float VP8LFastLog2(int v) { |
| 164 if (v < LOG_LOOKUP_IDX_MAX) { |
| 165 return kLog2Table[v]; |
| 166 } else if (v < APPROX_LOG_MAX) { |
| 167 int log_cnt = 0; |
| 168 while (v >= LOG_LOOKUP_IDX_MAX) { |
| 169 ++log_cnt; |
| 170 v = v >> 1; |
| 171 } |
| 172 return kLog2Table[v] + (float)log_cnt; |
| 173 } else { |
| 174 return (float)(LOG_2_RECIPROCAL * log((double)v)); |
| 175 } |
| 176 } |
| 177 |
| 178 //------------------------------------------------------------------------------ |
| 179 // Image transforms. |
| 180 |
| 181 // In-place sum of each component with mod 256. |
| 182 static WEBP_INLINE void AddPixelsEq(uint32_t* a, uint32_t b) { |
| 183 const uint32_t alpha_and_green = (*a & 0xff00ff00u) + (b & 0xff00ff00u); |
| 184 const uint32_t red_and_blue = (*a & 0x00ff00ffu) + (b & 0x00ff00ffu); |
| 185 *a = (alpha_and_green & 0xff00ff00u) | (red_and_blue & 0x00ff00ffu); |
| 186 } |
| 187 |
| 188 static WEBP_INLINE uint32_t Average2(uint32_t a0, uint32_t a1) { |
| 189 return (((a0 ^ a1) & 0xfefefefeL) >> 1) + (a0 & a1); |
| 190 } |
| 191 |
| 192 static WEBP_INLINE uint32_t Average3(uint32_t a0, uint32_t a1, uint32_t a2) { |
| 193 return Average2(Average2(a0, a2), a1); |
| 194 } |
| 195 |
| 196 static WEBP_INLINE uint32_t Average4(uint32_t a0, uint32_t a1, |
| 197 uint32_t a2, uint32_t a3) { |
| 198 return Average2(Average2(a0, a1), Average2(a2, a3)); |
| 199 } |
| 200 |
| 201 static WEBP_INLINE uint32_t Clip255(uint32_t a) { |
| 202 if (a < 256) { |
| 203 return a; |
| 204 } |
| 205 // return 0, when a is a negative integer. |
| 206 // return 255, when a is positive. |
| 207 return ~a >> 24; |
| 208 } |
| 209 |
| 210 static WEBP_INLINE int AddSubtractComponentFull(int a, int b, int c) { |
| 211 return Clip255(a + b - c); |
| 212 } |
| 213 |
| 214 static WEBP_INLINE uint32_t ClampedAddSubtractFull(uint32_t c0, uint32_t c1, |
| 215 uint32_t c2) { |
| 216 const int a = AddSubtractComponentFull(c0 >> 24, c1 >> 24, c2 >> 24); |
| 217 const int r = AddSubtractComponentFull((c0 >> 16) & 0xff, |
| 218 (c1 >> 16) & 0xff, |
| 219 (c2 >> 16) & 0xff); |
| 220 const int g = AddSubtractComponentFull((c0 >> 8) & 0xff, |
| 221 (c1 >> 8) & 0xff, |
| 222 (c2 >> 8) & 0xff); |
| 223 const int b = AddSubtractComponentFull(c0 & 0xff, c1 & 0xff, c2 & 0xff); |
| 224 return (a << 24) | (r << 16) | (g << 8) | b; |
| 225 } |
| 226 |
| 227 static WEBP_INLINE int AddSubtractComponentHalf(int a, int b) { |
| 228 return Clip255(a + (a - b) / 2); |
| 229 } |
| 230 |
| 231 static WEBP_INLINE uint32_t ClampedAddSubtractHalf(uint32_t c0, uint32_t c1, |
| 232 uint32_t c2) { |
| 233 const uint32_t ave = Average2(c0, c1); |
| 234 const int a = AddSubtractComponentHalf(ave >> 24, c2 >> 24); |
| 235 const int r = AddSubtractComponentHalf((ave >> 16) & 0xff, (c2 >> 16) & 0xff); |
| 236 const int g = AddSubtractComponentHalf((ave >> 8) & 0xff, (c2 >> 8) & 0xff); |
| 237 const int b = AddSubtractComponentHalf((ave >> 0) & 0xff, (c2 >> 0) & 0xff); |
| 238 return (a << 24) | (r << 16) | (g << 8) | b; |
| 239 } |
| 240 |
| 241 static WEBP_INLINE int Sub3(int a, int b, int c) { |
| 242 const int pa = b - c; |
| 243 const int pb = a - c; |
| 244 return abs(pa) - abs(pb); |
| 245 } |
| 246 |
| 247 static WEBP_INLINE uint32_t Select(uint32_t a, uint32_t b, uint32_t c) { |
| 248 const int pa_minus_pb = |
| 249 Sub3((a >> 24) , (b >> 24) , (c >> 24) ) + |
| 250 Sub3((a >> 16) & 0xff, (b >> 16) & 0xff, (c >> 16) & 0xff) + |
| 251 Sub3((a >> 8) & 0xff, (b >> 8) & 0xff, (c >> 8) & 0xff) + |
| 252 Sub3((a ) & 0xff, (b ) & 0xff, (c ) & 0xff); |
| 253 |
| 254 return (pa_minus_pb <= 0) ? a : b; |
| 255 } |
| 256 |
| 257 //------------------------------------------------------------------------------ |
| 258 // Predictors |
| 259 |
| 260 static uint32_t Predictor0(uint32_t left, const uint32_t* const top) { |
| 261 (void)top; |
| 262 (void)left; |
| 263 return ARGB_BLACK; |
| 264 } |
| 265 static uint32_t Predictor1(uint32_t left, const uint32_t* const top) { |
| 266 (void)top; |
| 267 return left; |
| 268 } |
| 269 static uint32_t Predictor2(uint32_t left, const uint32_t* const top) { |
| 270 (void)left; |
| 271 return top[0]; |
| 272 } |
| 273 static uint32_t Predictor3(uint32_t left, const uint32_t* const top) { |
| 274 (void)left; |
| 275 return top[1]; |
| 276 } |
| 277 static uint32_t Predictor4(uint32_t left, const uint32_t* const top) { |
| 278 (void)left; |
| 279 return top[-1]; |
| 280 } |
| 281 static uint32_t Predictor5(uint32_t left, const uint32_t* const top) { |
| 282 const uint32_t pred = Average3(left, top[0], top[1]); |
| 283 return pred; |
| 284 } |
| 285 static uint32_t Predictor6(uint32_t left, const uint32_t* const top) { |
| 286 const uint32_t pred = Average2(left, top[-1]); |
| 287 return pred; |
| 288 } |
| 289 static uint32_t Predictor7(uint32_t left, const uint32_t* const top) { |
| 290 const uint32_t pred = Average2(left, top[0]); |
| 291 return pred; |
| 292 } |
| 293 static uint32_t Predictor8(uint32_t left, const uint32_t* const top) { |
| 294 const uint32_t pred = Average2(top[-1], top[0]); |
| 295 (void)left; |
| 296 return pred; |
| 297 } |
| 298 static uint32_t Predictor9(uint32_t left, const uint32_t* const top) { |
| 299 const uint32_t pred = Average2(top[0], top[1]); |
| 300 (void)left; |
| 301 return pred; |
| 302 } |
| 303 static uint32_t Predictor10(uint32_t left, const uint32_t* const top) { |
| 304 const uint32_t pred = Average4(left, top[-1], top[0], top[1]); |
| 305 return pred; |
| 306 } |
| 307 static uint32_t Predictor11(uint32_t left, const uint32_t* const top) { |
| 308 const uint32_t pred = Select(top[0], left, top[-1]); |
| 309 return pred; |
| 310 } |
| 311 static uint32_t Predictor12(uint32_t left, const uint32_t* const top) { |
| 312 const uint32_t pred = ClampedAddSubtractFull(left, top[0], top[-1]); |
| 313 return pred; |
| 314 } |
| 315 static uint32_t Predictor13(uint32_t left, const uint32_t* const top) { |
| 316 const uint32_t pred = ClampedAddSubtractHalf(left, top[0], top[-1]); |
| 317 return pred; |
| 318 } |
| 319 |
| 320 typedef uint32_t (*PredictorFunc)(uint32_t left, const uint32_t* const top); |
| 321 static const PredictorFunc kPredictors[16] = { |
| 322 Predictor0, Predictor1, Predictor2, Predictor3, |
| 323 Predictor4, Predictor5, Predictor6, Predictor7, |
| 324 Predictor8, Predictor9, Predictor10, Predictor11, |
| 325 Predictor12, Predictor13, |
| 326 Predictor0, Predictor0 // <- padding security sentinels |
| 327 }; |
| 328 |
| 329 // TODO(vikasa): Replace 256 etc with defines. |
| 330 static float PredictionCostSpatial(const int* counts, |
| 331 int weight_0, double exp_val) { |
| 332 const int significant_symbols = 16; |
| 333 const double exp_decay_factor = 0.6; |
| 334 double bits = weight_0 * counts[0]; |
| 335 int i; |
| 336 for (i = 1; i < significant_symbols; ++i) { |
| 337 bits += exp_val * (counts[i] + counts[256 - i]); |
| 338 exp_val *= exp_decay_factor; |
| 339 } |
| 340 return (float)(-0.1 * bits); |
| 341 } |
| 342 |
| 343 // Compute the Shanon's entropy: Sum(p*log2(p)) |
| 344 static float ShannonEntropy(const int* const array, int n) { |
| 345 int i; |
| 346 float retval = 0.f; |
| 347 int sum = 0; |
| 348 for (i = 0; i < n; ++i) { |
| 349 if (array[i] != 0) { |
| 350 sum += array[i]; |
| 351 retval -= VP8LFastSLog2(array[i]); |
| 352 } |
| 353 } |
| 354 retval += VP8LFastSLog2(sum); |
| 355 return retval; |
| 356 } |
| 357 |
| 358 static float PredictionCostSpatialHistogram(int accumulated[4][256], |
| 359 int tile[4][256]) { |
| 360 int i; |
| 361 int k; |
| 362 int combo[256]; |
| 363 double retval = 0; |
| 364 for (i = 0; i < 4; ++i) { |
| 365 const double exp_val = 0.94; |
| 366 retval += PredictionCostSpatial(&tile[i][0], 1, exp_val); |
| 367 retval += ShannonEntropy(&tile[i][0], 256); |
| 368 for (k = 0; k < 256; ++k) { |
| 369 combo[k] = accumulated[i][k] + tile[i][k]; |
| 370 } |
| 371 retval += ShannonEntropy(&combo[0], 256); |
| 372 } |
| 373 return (float)retval; |
| 374 } |
| 375 |
| 376 static int GetBestPredictorForTile(int width, int height, |
| 377 int tile_x, int tile_y, int bits, |
| 378 int accumulated[4][256], |
| 379 const uint32_t* const argb_scratch) { |
| 380 const int kNumPredModes = 14; |
| 381 const int col_start = tile_x << bits; |
| 382 const int row_start = tile_y << bits; |
| 383 const int tile_size = 1 << bits; |
| 384 const int ymax = (tile_size <= height - row_start) ? |
| 385 tile_size : height - row_start; |
| 386 const int xmax = (tile_size <= width - col_start) ? |
| 387 tile_size : width - col_start; |
| 388 int histo[4][256]; |
| 389 float best_diff = MAX_DIFF_COST; |
| 390 int best_mode = 0; |
| 391 |
| 392 int mode; |
| 393 for (mode = 0; mode < kNumPredModes; ++mode) { |
| 394 const uint32_t* current_row = argb_scratch; |
| 395 const PredictorFunc pred_func = kPredictors[mode]; |
| 396 float cur_diff; |
| 397 int y; |
| 398 memset(&histo[0][0], 0, sizeof(histo)); |
| 399 for (y = 0; y < ymax; ++y) { |
| 400 int x; |
| 401 const int row = row_start + y; |
| 402 const uint32_t* const upper_row = current_row; |
| 403 current_row = upper_row + width; |
| 404 for (x = 0; x < xmax; ++x) { |
| 405 const int col = col_start + x; |
| 406 uint32_t predict; |
| 407 uint32_t predict_diff; |
| 408 if (row == 0) { |
| 409 predict = (col == 0) ? ARGB_BLACK : current_row[col - 1]; // Left. |
| 410 } else if (col == 0) { |
| 411 predict = upper_row[col]; // Top. |
| 412 } else { |
| 413 predict = pred_func(current_row[col - 1], upper_row + col); |
| 414 } |
| 415 predict_diff = VP8LSubPixels(current_row[col], predict); |
| 416 ++histo[0][predict_diff >> 24]; |
| 417 ++histo[1][((predict_diff >> 16) & 0xff)]; |
| 418 ++histo[2][((predict_diff >> 8) & 0xff)]; |
| 419 ++histo[3][(predict_diff & 0xff)]; |
| 420 } |
| 421 } |
| 422 cur_diff = PredictionCostSpatialHistogram(accumulated, histo); |
| 423 if (cur_diff < best_diff) { |
| 424 best_diff = cur_diff; |
| 425 best_mode = mode; |
| 426 } |
| 427 } |
| 428 |
| 429 return best_mode; |
| 430 } |
| 431 |
| 432 static void CopyTileWithPrediction(int width, int height, |
| 433 int tile_x, int tile_y, int bits, int mode, |
| 434 const uint32_t* const argb_scratch, |
| 435 uint32_t* const argb) { |
| 436 const int col_start = tile_x << bits; |
| 437 const int row_start = tile_y << bits; |
| 438 const int tile_size = 1 << bits; |
| 439 const int ymax = (tile_size <= height - row_start) ? |
| 440 tile_size : height - row_start; |
| 441 const int xmax = (tile_size <= width - col_start) ? |
| 442 tile_size : width - col_start; |
| 443 const PredictorFunc pred_func = kPredictors[mode]; |
| 444 const uint32_t* current_row = argb_scratch; |
| 445 |
| 446 int y; |
| 447 for (y = 0; y < ymax; ++y) { |
| 448 int x; |
| 449 const int row = row_start + y; |
| 450 const uint32_t* const upper_row = current_row; |
| 451 current_row = upper_row + width; |
| 452 for (x = 0; x < xmax; ++x) { |
| 453 const int col = col_start + x; |
| 454 const int pix = row * width + col; |
| 455 uint32_t predict; |
| 456 if (row == 0) { |
| 457 predict = (col == 0) ? ARGB_BLACK : current_row[col - 1]; // Left. |
| 458 } else if (col == 0) { |
| 459 predict = upper_row[col]; // Top. |
| 460 } else { |
| 461 predict = pred_func(current_row[col - 1], upper_row + col); |
| 462 } |
| 463 argb[pix] = VP8LSubPixels(current_row[col], predict); |
| 464 } |
| 465 } |
| 466 } |
| 467 |
| 468 void VP8LResidualImage(int width, int height, int bits, |
| 469 uint32_t* const argb, uint32_t* const argb_scratch, |
| 470 uint32_t* const image) { |
| 471 const int max_tile_size = 1 << bits; |
| 472 const int tiles_per_row = VP8LSubSampleSize(width, bits); |
| 473 const int tiles_per_col = VP8LSubSampleSize(height, bits); |
| 474 uint32_t* const upper_row = argb_scratch; |
| 475 uint32_t* const current_tile_rows = argb_scratch + width; |
| 476 int tile_y; |
| 477 int histo[4][256]; |
| 478 memset(histo, 0, sizeof(histo)); |
| 479 for (tile_y = 0; tile_y < tiles_per_col; ++tile_y) { |
| 480 const int tile_y_offset = tile_y * max_tile_size; |
| 481 const int this_tile_height = |
| 482 (tile_y < tiles_per_col - 1) ? max_tile_size : height - tile_y_offset; |
| 483 int tile_x; |
| 484 if (tile_y > 0) { |
| 485 memcpy(upper_row, current_tile_rows + (max_tile_size - 1) * width, |
| 486 width * sizeof(*upper_row)); |
| 487 } |
| 488 memcpy(current_tile_rows, &argb[tile_y_offset * width], |
| 489 this_tile_height * width * sizeof(*current_tile_rows)); |
| 490 for (tile_x = 0; tile_x < tiles_per_row; ++tile_x) { |
| 491 int pred; |
| 492 int y; |
| 493 const int tile_x_offset = tile_x * max_tile_size; |
| 494 int all_x_max = tile_x_offset + max_tile_size; |
| 495 if (all_x_max > width) { |
| 496 all_x_max = width; |
| 497 } |
| 498 pred = GetBestPredictorForTile(width, height, tile_x, tile_y, bits, histo, |
| 499 argb_scratch); |
| 500 image[tile_y * tiles_per_row + tile_x] = 0xff000000u | (pred << 8); |
| 501 CopyTileWithPrediction(width, height, tile_x, tile_y, bits, pred, |
| 502 argb_scratch, argb); |
| 503 for (y = 0; y < max_tile_size; ++y) { |
| 504 int ix; |
| 505 int all_x; |
| 506 int all_y = tile_y_offset + y; |
| 507 if (all_y >= height) { |
| 508 break; |
| 509 } |
| 510 ix = all_y * width + tile_x_offset; |
| 511 for (all_x = tile_x_offset; all_x < all_x_max; ++all_x, ++ix) { |
| 512 const uint32_t a = argb[ix]; |
| 513 ++histo[0][a >> 24]; |
| 514 ++histo[1][((a >> 16) & 0xff)]; |
| 515 ++histo[2][((a >> 8) & 0xff)]; |
| 516 ++histo[3][(a & 0xff)]; |
| 517 } |
| 518 } |
| 519 } |
| 520 } |
| 521 } |
| 522 |
| 523 // Inverse prediction. |
| 524 static void PredictorInverseTransform(const VP8LTransform* const transform, |
| 525 int y_start, int y_end, uint32_t* data) { |
| 526 const int width = transform->xsize_; |
| 527 if (y_start == 0) { // First Row follows the L (mode=1) mode. |
| 528 int x; |
| 529 const uint32_t pred0 = Predictor0(data[-1], NULL); |
| 530 AddPixelsEq(data, pred0); |
| 531 for (x = 1; x < width; ++x) { |
| 532 const uint32_t pred1 = Predictor1(data[x - 1], NULL); |
| 533 AddPixelsEq(data + x, pred1); |
| 534 } |
| 535 data += width; |
| 536 ++y_start; |
| 537 } |
| 538 |
| 539 { |
| 540 int y = y_start; |
| 541 const int mask = (1 << transform->bits_) - 1; |
| 542 const int tiles_per_row = VP8LSubSampleSize(width, transform->bits_); |
| 543 const uint32_t* pred_mode_base = |
| 544 transform->data_ + (y >> transform->bits_) * tiles_per_row; |
| 545 |
| 546 while (y < y_end) { |
| 547 int x; |
| 548 const uint32_t pred2 = Predictor2(data[-1], data - width); |
| 549 const uint32_t* pred_mode_src = pred_mode_base; |
| 550 PredictorFunc pred_func; |
| 551 |
| 552 // First pixel follows the T (mode=2) mode. |
| 553 AddPixelsEq(data, pred2); |
| 554 |
| 555 // .. the rest: |
| 556 pred_func = kPredictors[((*pred_mode_src++) >> 8) & 0xf]; |
| 557 for (x = 1; x < width; ++x) { |
| 558 uint32_t pred; |
| 559 if ((x & mask) == 0) { // start of tile. Read predictor function. |
| 560 pred_func = kPredictors[((*pred_mode_src++) >> 8) & 0xf]; |
| 561 } |
| 562 pred = pred_func(data[x - 1], data + x - width); |
| 563 AddPixelsEq(data + x, pred); |
| 564 } |
| 565 data += width; |
| 566 ++y; |
| 567 if ((y & mask) == 0) { // Use the same mask, since tiles are squares. |
| 568 pred_mode_base += tiles_per_row; |
| 569 } |
| 570 } |
| 571 } |
| 572 } |
| 573 |
| 574 void VP8LSubtractGreenFromBlueAndRed(uint32_t* argb_data, int num_pixs) { |
| 575 int i; |
| 576 for (i = 0; i < num_pixs; ++i) { |
| 577 const uint32_t argb = argb_data[i]; |
| 578 const uint32_t green = (argb >> 8) & 0xff; |
| 579 const uint32_t new_r = (((argb >> 16) & 0xff) - green) & 0xff; |
| 580 const uint32_t new_b = ((argb & 0xff) - green) & 0xff; |
| 581 argb_data[i] = (argb & 0xff00ff00) | (new_r << 16) | new_b; |
| 582 } |
| 583 } |
| 584 |
| 585 // Add green to blue and red channels (i.e. perform the inverse transform of |
| 586 // 'subtract green'). |
| 587 static void AddGreenToBlueAndRed(const VP8LTransform* const transform, |
| 588 int y_start, int y_end, uint32_t* data) { |
| 589 const int width = transform->xsize_; |
| 590 const uint32_t* const data_end = data + (y_end - y_start) * width; |
| 591 while (data < data_end) { |
| 592 const uint32_t argb = *data; |
| 593 // "* 0001001u" is equivalent to "(green << 16) + green)" |
| 594 const uint32_t green = ((argb >> 8) & 0xff); |
| 595 uint32_t red_blue = (argb & 0x00ff00ffu); |
| 596 red_blue += (green << 16) | green; |
| 597 red_blue &= 0x00ff00ffu; |
| 598 *data++ = (argb & 0xff00ff00u) | red_blue; |
| 599 } |
| 600 } |
| 601 |
| 602 typedef struct { |
| 603 // Note: the members are uint8_t, so that any negative values are |
| 604 // automatically converted to "mod 256" values. |
| 605 uint8_t green_to_red_; |
| 606 uint8_t green_to_blue_; |
| 607 uint8_t red_to_blue_; |
| 608 } Multipliers; |
| 609 |
| 610 static WEBP_INLINE void MultipliersClear(Multipliers* m) { |
| 611 m->green_to_red_ = 0; |
| 612 m->green_to_blue_ = 0; |
| 613 m->red_to_blue_ = 0; |
| 614 } |
| 615 |
| 616 static WEBP_INLINE uint32_t ColorTransformDelta(int8_t color_pred, |
| 617 int8_t color) { |
| 618 return (uint32_t)((int)(color_pred) * color) >> 5; |
| 619 } |
| 620 |
| 621 static WEBP_INLINE void ColorCodeToMultipliers(uint32_t color_code, |
| 622 Multipliers* const m) { |
| 623 m->green_to_red_ = (color_code >> 0) & 0xff; |
| 624 m->green_to_blue_ = (color_code >> 8) & 0xff; |
| 625 m->red_to_blue_ = (color_code >> 16) & 0xff; |
| 626 } |
| 627 |
| 628 static WEBP_INLINE uint32_t MultipliersToColorCode(Multipliers* const m) { |
| 629 return 0xff000000u | |
| 630 ((uint32_t)(m->red_to_blue_) << 16) | |
| 631 ((uint32_t)(m->green_to_blue_) << 8) | |
| 632 m->green_to_red_; |
| 633 } |
| 634 |
| 635 static WEBP_INLINE uint32_t TransformColor(const Multipliers* const m, |
| 636 uint32_t argb, int inverse) { |
| 637 const uint32_t green = argb >> 8; |
| 638 const uint32_t red = argb >> 16; |
| 639 uint32_t new_red = red; |
| 640 uint32_t new_blue = argb; |
| 641 |
| 642 if (inverse) { |
| 643 new_red += ColorTransformDelta(m->green_to_red_, green); |
| 644 new_red &= 0xff; |
| 645 new_blue += ColorTransformDelta(m->green_to_blue_, green); |
| 646 new_blue += ColorTransformDelta(m->red_to_blue_, new_red); |
| 647 new_blue &= 0xff; |
| 648 } else { |
| 649 new_red -= ColorTransformDelta(m->green_to_red_, green); |
| 650 new_red &= 0xff; |
| 651 new_blue -= ColorTransformDelta(m->green_to_blue_, green); |
| 652 new_blue -= ColorTransformDelta(m->red_to_blue_, red); |
| 653 new_blue &= 0xff; |
| 654 } |
| 655 return (argb & 0xff00ff00u) | (new_red << 16) | (new_blue); |
| 656 } |
| 657 |
| 658 static WEBP_INLINE int SkipRepeatedPixels(const uint32_t* const argb, |
| 659 int ix, int xsize) { |
| 660 const uint32_t v = argb[ix]; |
| 661 if (ix >= xsize + 3) { |
| 662 if (v == argb[ix - xsize] && |
| 663 argb[ix - 1] == argb[ix - xsize - 1] && |
| 664 argb[ix - 2] == argb[ix - xsize - 2] && |
| 665 argb[ix - 3] == argb[ix - xsize - 3]) { |
| 666 return 1; |
| 667 } |
| 668 return v == argb[ix - 3] && v == argb[ix - 2] && v == argb[ix - 1]; |
| 669 } else if (ix >= 3) { |
| 670 return v == argb[ix - 3] && v == argb[ix - 2] && v == argb[ix - 1]; |
| 671 } |
| 672 return 0; |
| 673 } |
| 674 |
| 675 static float PredictionCostCrossColor(const int accumulated[256], |
| 676 const int counts[256]) { |
| 677 // Favor low entropy, locally and globally. |
| 678 int i; |
| 679 int combo[256]; |
| 680 for (i = 0; i < 256; ++i) { |
| 681 combo[i] = accumulated[i] + counts[i]; |
| 682 } |
| 683 return ShannonEntropy(combo, 256) + |
| 684 ShannonEntropy(counts, 256) + |
| 685 PredictionCostSpatial(counts, 3, 2.4); // Favor small absolute values. |
| 686 } |
| 687 |
| 688 static Multipliers GetBestColorTransformForTile( |
| 689 int tile_x, int tile_y, int bits, |
| 690 Multipliers prevX, |
| 691 Multipliers prevY, |
| 692 int step, int xsize, int ysize, |
| 693 int* accumulated_red_histo, |
| 694 int* accumulated_blue_histo, |
| 695 const uint32_t* const argb) { |
| 696 float best_diff = MAX_DIFF_COST; |
| 697 float cur_diff; |
| 698 const int halfstep = step / 2; |
| 699 const int max_tile_size = 1 << bits; |
| 700 const int tile_y_offset = tile_y * max_tile_size; |
| 701 const int tile_x_offset = tile_x * max_tile_size; |
| 702 int green_to_red; |
| 703 int green_to_blue; |
| 704 int red_to_blue; |
| 705 int all_x_max = tile_x_offset + max_tile_size; |
| 706 int all_y_max = tile_y_offset + max_tile_size; |
| 707 Multipliers best_tx; |
| 708 MultipliersClear(&best_tx); |
| 709 if (all_x_max > xsize) { |
| 710 all_x_max = xsize; |
| 711 } |
| 712 if (all_y_max > ysize) { |
| 713 all_y_max = ysize; |
| 714 } |
| 715 for (green_to_red = -64; green_to_red <= 64; green_to_red += halfstep) { |
| 716 int histo[256] = { 0 }; |
| 717 int all_y; |
| 718 Multipliers tx; |
| 719 MultipliersClear(&tx); |
| 720 tx.green_to_red_ = green_to_red & 0xff; |
| 721 |
| 722 for (all_y = tile_y_offset; all_y < all_y_max; ++all_y) { |
| 723 uint32_t predict; |
| 724 int ix = all_y * xsize + tile_x_offset; |
| 725 int all_x; |
| 726 for (all_x = tile_x_offset; all_x < all_x_max; ++all_x, ++ix) { |
| 727 if (SkipRepeatedPixels(argb, ix, xsize)) { |
| 728 continue; |
| 729 } |
| 730 predict = TransformColor(&tx, argb[ix], 0); |
| 731 ++histo[(predict >> 16) & 0xff]; // red. |
| 732 } |
| 733 } |
| 734 cur_diff = PredictionCostCrossColor(&accumulated_red_histo[0], &histo[0]); |
| 735 if (tx.green_to_red_ == prevX.green_to_red_) { |
| 736 cur_diff -= 3; // favor keeping the areas locally similar |
| 737 } |
| 738 if (tx.green_to_red_ == prevY.green_to_red_) { |
| 739 cur_diff -= 3; // favor keeping the areas locally similar |
| 740 } |
| 741 if (tx.green_to_red_ == 0) { |
| 742 cur_diff -= 3; |
| 743 } |
| 744 if (cur_diff < best_diff) { |
| 745 best_diff = cur_diff; |
| 746 best_tx = tx; |
| 747 } |
| 748 } |
| 749 best_diff = MAX_DIFF_COST; |
| 750 green_to_red = best_tx.green_to_red_; |
| 751 for (green_to_blue = -32; green_to_blue <= 32; green_to_blue += step) { |
| 752 for (red_to_blue = -32; red_to_blue <= 32; red_to_blue += step) { |
| 753 int all_y; |
| 754 int histo[256] = { 0 }; |
| 755 Multipliers tx; |
| 756 tx.green_to_red_ = green_to_red; |
| 757 tx.green_to_blue_ = green_to_blue; |
| 758 tx.red_to_blue_ = red_to_blue; |
| 759 for (all_y = tile_y_offset; all_y < all_y_max; ++all_y) { |
| 760 uint32_t predict; |
| 761 int all_x; |
| 762 int ix = all_y * xsize + tile_x_offset; |
| 763 for (all_x = tile_x_offset; all_x < all_x_max; ++all_x, ++ix) { |
| 764 if (SkipRepeatedPixels(argb, ix, xsize)) { |
| 765 continue; |
| 766 } |
| 767 predict = TransformColor(&tx, argb[ix], 0); |
| 768 ++histo[predict & 0xff]; // blue. |
| 769 } |
| 770 } |
| 771 cur_diff = |
| 772 PredictionCostCrossColor(&accumulated_blue_histo[0], &histo[0]); |
| 773 if (tx.green_to_blue_ == prevX.green_to_blue_) { |
| 774 cur_diff -= 3; // favor keeping the areas locally similar |
| 775 } |
| 776 if (tx.green_to_blue_ == prevY.green_to_blue_) { |
| 777 cur_diff -= 3; // favor keeping the areas locally similar |
| 778 } |
| 779 if (tx.red_to_blue_ == prevX.red_to_blue_) { |
| 780 cur_diff -= 3; // favor keeping the areas locally similar |
| 781 } |
| 782 if (tx.red_to_blue_ == prevY.red_to_blue_) { |
| 783 cur_diff -= 3; // favor keeping the areas locally similar |
| 784 } |
| 785 if (tx.green_to_blue_ == 0) { |
| 786 cur_diff -= 3; |
| 787 } |
| 788 if (tx.red_to_blue_ == 0) { |
| 789 cur_diff -= 3; |
| 790 } |
| 791 if (cur_diff < best_diff) { |
| 792 best_diff = cur_diff; |
| 793 best_tx = tx; |
| 794 } |
| 795 } |
| 796 } |
| 797 return best_tx; |
| 798 } |
| 799 |
| 800 static void CopyTileWithColorTransform(int xsize, int ysize, |
| 801 int tile_x, int tile_y, int bits, |
| 802 Multipliers color_transform, |
| 803 uint32_t* const argb) { |
| 804 int y; |
| 805 int xscan = 1 << bits; |
| 806 int yscan = 1 << bits; |
| 807 tile_x <<= bits; |
| 808 tile_y <<= bits; |
| 809 if (xscan > xsize - tile_x) { |
| 810 xscan = xsize - tile_x; |
| 811 } |
| 812 if (yscan > ysize - tile_y) { |
| 813 yscan = ysize - tile_y; |
| 814 } |
| 815 yscan += tile_y; |
| 816 for (y = tile_y; y < yscan; ++y) { |
| 817 int ix = y * xsize + tile_x; |
| 818 const int end_ix = ix + xscan; |
| 819 for (; ix < end_ix; ++ix) { |
| 820 argb[ix] = TransformColor(&color_transform, argb[ix], 0); |
| 821 } |
| 822 } |
| 823 } |
| 824 |
| 825 void VP8LColorSpaceTransform(int width, int height, int bits, int step, |
| 826 uint32_t* const argb, uint32_t* image) { |
| 827 const int max_tile_size = 1 << bits; |
| 828 int tile_xsize = VP8LSubSampleSize(width, bits); |
| 829 int tile_ysize = VP8LSubSampleSize(height, bits); |
| 830 int accumulated_red_histo[256] = { 0 }; |
| 831 int accumulated_blue_histo[256] = { 0 }; |
| 832 int tile_y; |
| 833 int tile_x; |
| 834 Multipliers prevX; |
| 835 Multipliers prevY; |
| 836 MultipliersClear(&prevY); |
| 837 MultipliersClear(&prevX); |
| 838 for (tile_y = 0; tile_y < tile_ysize; ++tile_y) { |
| 839 for (tile_x = 0; tile_x < tile_xsize; ++tile_x) { |
| 840 Multipliers color_transform; |
| 841 int all_x_max; |
| 842 int y; |
| 843 const int tile_y_offset = tile_y * max_tile_size; |
| 844 const int tile_x_offset = tile_x * max_tile_size; |
| 845 if (tile_y != 0) { |
| 846 ColorCodeToMultipliers(image[tile_y * tile_xsize + tile_x - 1], &prevX); |
| 847 ColorCodeToMultipliers(image[(tile_y - 1) * tile_xsize + tile_x], |
| 848 &prevY); |
| 849 } else if (tile_x != 0) { |
| 850 ColorCodeToMultipliers(image[tile_y * tile_xsize + tile_x - 1], &prevX); |
| 851 } |
| 852 color_transform = |
| 853 GetBestColorTransformForTile(tile_x, tile_y, bits, |
| 854 prevX, prevY, |
| 855 step, width, height, |
| 856 &accumulated_red_histo[0], |
| 857 &accumulated_blue_histo[0], |
| 858 argb); |
| 859 image[tile_y * tile_xsize + tile_x] = |
| 860 MultipliersToColorCode(&color_transform); |
| 861 CopyTileWithColorTransform(width, height, tile_x, tile_y, bits, |
| 862 color_transform, argb); |
| 863 |
| 864 // Gather accumulated histogram data. |
| 865 all_x_max = tile_x_offset + max_tile_size; |
| 866 if (all_x_max > width) { |
| 867 all_x_max = width; |
| 868 } |
| 869 for (y = 0; y < max_tile_size; ++y) { |
| 870 int ix; |
| 871 int all_x; |
| 872 int all_y = tile_y_offset + y; |
| 873 if (all_y >= height) { |
| 874 break; |
| 875 } |
| 876 ix = all_y * width + tile_x_offset; |
| 877 for (all_x = tile_x_offset; all_x < all_x_max; ++all_x, ++ix) { |
| 878 if (ix >= 2 && |
| 879 argb[ix] == argb[ix - 2] && |
| 880 argb[ix] == argb[ix - 1]) { |
| 881 continue; // repeated pixels are handled by backward references |
| 882 } |
| 883 if (ix >= width + 2 && |
| 884 argb[ix - 2] == argb[ix - width - 2] && |
| 885 argb[ix - 1] == argb[ix - width - 1] && |
| 886 argb[ix] == argb[ix - width]) { |
| 887 continue; // repeated pixels are handled by backward references |
| 888 } |
| 889 ++accumulated_red_histo[(argb[ix] >> 16) & 0xff]; |
| 890 ++accumulated_blue_histo[argb[ix] & 0xff]; |
| 891 } |
| 892 } |
| 893 } |
| 894 } |
| 895 } |
| 896 |
| 897 // Color space inverse transform. |
| 898 static void ColorSpaceInverseTransform(const VP8LTransform* const transform, |
| 899 int y_start, int y_end, uint32_t* data) { |
| 900 const int width = transform->xsize_; |
| 901 const int mask = (1 << transform->bits_) - 1; |
| 902 const int tiles_per_row = VP8LSubSampleSize(width, transform->bits_); |
| 903 int y = y_start; |
| 904 const uint32_t* pred_row = |
| 905 transform->data_ + (y >> transform->bits_) * tiles_per_row; |
| 906 |
| 907 while (y < y_end) { |
| 908 const uint32_t* pred = pred_row; |
| 909 Multipliers m = { 0, 0, 0 }; |
| 910 int x; |
| 911 |
| 912 for (x = 0; x < width; ++x) { |
| 913 if ((x & mask) == 0) ColorCodeToMultipliers(*pred++, &m); |
| 914 data[x] = TransformColor(&m, data[x], 1); |
| 915 } |
| 916 data += width; |
| 917 ++y; |
| 918 if ((y & mask) == 0) pred_row += tiles_per_row;; |
| 919 } |
| 920 } |
| 921 |
| 922 // Separate out pixels packed together using pixel-bundling. |
| 923 static void ColorIndexInverseTransform( |
| 924 const VP8LTransform* const transform, |
| 925 int y_start, int y_end, const uint32_t* src, uint32_t* dst) { |
| 926 int y; |
| 927 const int bits_per_pixel = 8 >> transform->bits_; |
| 928 const int width = transform->xsize_; |
| 929 const uint32_t* const color_map = transform->data_; |
| 930 if (bits_per_pixel < 8) { |
| 931 const int pixels_per_byte = 1 << transform->bits_; |
| 932 const int count_mask = pixels_per_byte - 1; |
| 933 const uint32_t bit_mask = (1 << bits_per_pixel) - 1; |
| 934 for (y = y_start; y < y_end; ++y) { |
| 935 uint32_t packed_pixels = 0; |
| 936 int x; |
| 937 for (x = 0; x < width; ++x) { |
| 938 // We need to load fresh 'packed_pixels' once every 'bytes_per_pixels' |
| 939 // increments of x. Fortunately, pixels_per_byte is a power of 2, so |
| 940 // can just use a mask for that, instead of decrementing a counter. |
| 941 if ((x & count_mask) == 0) packed_pixels = ((*src++) >> 8) & 0xff; |
| 942 *dst++ = color_map[packed_pixels & bit_mask]; |
| 943 packed_pixels >>= bits_per_pixel; |
| 944 } |
| 945 } |
| 946 } else { |
| 947 for (y = y_start; y < y_end; ++y) { |
| 948 int x; |
| 949 for (x = 0; x < width; ++x) { |
| 950 *dst++ = color_map[((*src++) >> 8) & 0xff]; |
| 951 } |
| 952 } |
| 953 } |
| 954 } |
| 955 |
| 956 void VP8LInverseTransform(const VP8LTransform* const transform, |
| 957 int row_start, int row_end, |
| 958 const uint32_t* const in, uint32_t* const out) { |
| 959 assert(row_start < row_end); |
| 960 assert(row_end <= transform->ysize_); |
| 961 switch (transform->type_) { |
| 962 case SUBTRACT_GREEN: |
| 963 AddGreenToBlueAndRed(transform, row_start, row_end, out); |
| 964 break; |
| 965 case PREDICTOR_TRANSFORM: |
| 966 PredictorInverseTransform(transform, row_start, row_end, out); |
| 967 if (row_end != transform->ysize_) { |
| 968 // The last predicted row in this iteration will be the top-pred row |
| 969 // for the first row in next iteration. |
| 970 const int width = transform->xsize_; |
| 971 memcpy(out - width, out + (row_end - row_start - 1) * width, |
| 972 width * sizeof(*out)); |
| 973 } |
| 974 break; |
| 975 case CROSS_COLOR_TRANSFORM: |
| 976 ColorSpaceInverseTransform(transform, row_start, row_end, out); |
| 977 break; |
| 978 case COLOR_INDEXING_TRANSFORM: |
| 979 ColorIndexInverseTransform(transform, row_start, row_end, in, out); |
| 980 break; |
| 981 } |
| 982 } |
| 983 |
| 984 //------------------------------------------------------------------------------ |
| 985 // Color space conversion. |
| 986 |
| 987 static int is_big_endian(void) { |
| 988 static const union { |
| 989 uint16_t w; |
| 990 uint8_t b[2]; |
| 991 } tmp = { 1 }; |
| 992 return (tmp.b[0] != 1); |
| 993 } |
| 994 |
| 995 static void ConvertBGRAToRGB(const uint32_t* src, |
| 996 int num_pixels, uint8_t* dst) { |
| 997 const uint32_t* const src_end = src + num_pixels; |
| 998 while (src < src_end) { |
| 999 const uint32_t argb = *src++; |
| 1000 *dst++ = (argb >> 16) & 0xff; |
| 1001 *dst++ = (argb >> 8) & 0xff; |
| 1002 *dst++ = (argb >> 0) & 0xff; |
| 1003 } |
| 1004 } |
| 1005 |
| 1006 static void ConvertBGRAToRGBA(const uint32_t* src, |
| 1007 int num_pixels, uint8_t* dst) { |
| 1008 const uint32_t* const src_end = src + num_pixels; |
| 1009 while (src < src_end) { |
| 1010 const uint32_t argb = *src++; |
| 1011 *dst++ = (argb >> 16) & 0xff; |
| 1012 *dst++ = (argb >> 8) & 0xff; |
| 1013 *dst++ = (argb >> 0) & 0xff; |
| 1014 *dst++ = (argb >> 24) & 0xff; |
| 1015 } |
| 1016 } |
| 1017 |
| 1018 static void ConvertBGRAToRGBA4444(const uint32_t* src, |
| 1019 int num_pixels, uint8_t* dst) { |
| 1020 const uint32_t* const src_end = src + num_pixels; |
| 1021 while (src < src_end) { |
| 1022 const uint32_t argb = *src++; |
| 1023 *dst++ = ((argb >> 16) & 0xf0) | ((argb >> 12) & 0xf); |
| 1024 *dst++ = ((argb >> 0) & 0xf0) | ((argb >> 28) & 0xf); |
| 1025 } |
| 1026 } |
| 1027 |
| 1028 static void ConvertBGRAToRGB565(const uint32_t* src, |
| 1029 int num_pixels, uint8_t* dst) { |
| 1030 const uint32_t* const src_end = src + num_pixels; |
| 1031 while (src < src_end) { |
| 1032 const uint32_t argb = *src++; |
| 1033 *dst++ = ((argb >> 16) & 0xf8) | ((argb >> 13) & 0x7); |
| 1034 *dst++ = ((argb >> 5) & 0xe0) | ((argb >> 3) & 0x1f); |
| 1035 } |
| 1036 } |
| 1037 |
| 1038 static void ConvertBGRAToBGR(const uint32_t* src, |
| 1039 int num_pixels, uint8_t* dst) { |
| 1040 const uint32_t* const src_end = src + num_pixels; |
| 1041 while (src < src_end) { |
| 1042 const uint32_t argb = *src++; |
| 1043 *dst++ = (argb >> 0) & 0xff; |
| 1044 *dst++ = (argb >> 8) & 0xff; |
| 1045 *dst++ = (argb >> 16) & 0xff; |
| 1046 } |
| 1047 } |
| 1048 |
| 1049 static void CopyOrSwap(const uint32_t* src, int num_pixels, uint8_t* dst, |
| 1050 int swap_on_big_endian) { |
| 1051 if (is_big_endian() == swap_on_big_endian) { |
| 1052 const uint32_t* const src_end = src + num_pixels; |
| 1053 while (src < src_end) { |
| 1054 uint32_t argb = *src++; |
| 1055 #if !defined(__BIG_ENDIAN__) && (defined(__i386__) || defined(__x86_64__)) |
| 1056 __asm__ volatile("bswap %0" : "=r"(argb) : "0"(argb)); |
| 1057 *(uint32_t*)dst = argb; |
| 1058 dst += sizeof(argb); |
| 1059 #elif !defined(__BIG_ENDIAN__) && defined(_MSC_VER) |
| 1060 argb = _byteswap_ulong(argb); |
| 1061 *(uint32_t*)dst = argb; |
| 1062 dst += sizeof(argb); |
| 1063 #else |
| 1064 *dst++ = (argb >> 24) & 0xff; |
| 1065 *dst++ = (argb >> 16) & 0xff; |
| 1066 *dst++ = (argb >> 8) & 0xff; |
| 1067 *dst++ = (argb >> 0) & 0xff; |
| 1068 #endif |
| 1069 } |
| 1070 } else { |
| 1071 memcpy(dst, src, num_pixels * sizeof(*src)); |
| 1072 } |
| 1073 } |
| 1074 |
| 1075 void VP8LConvertFromBGRA(const uint32_t* const in_data, int num_pixels, |
| 1076 WEBP_CSP_MODE out_colorspace, uint8_t* const rgba) { |
| 1077 switch (out_colorspace) { |
| 1078 case MODE_RGB: |
| 1079 ConvertBGRAToRGB(in_data, num_pixels, rgba); |
| 1080 break; |
| 1081 case MODE_RGBA: |
| 1082 ConvertBGRAToRGBA(in_data, num_pixels, rgba); |
| 1083 break; |
| 1084 case MODE_rgbA: |
| 1085 ConvertBGRAToRGBA(in_data, num_pixels, rgba); |
| 1086 WebPApplyAlphaMultiply(rgba, 0, num_pixels, 1, 0); |
| 1087 break; |
| 1088 case MODE_BGR: |
| 1089 ConvertBGRAToBGR(in_data, num_pixels, rgba); |
| 1090 break; |
| 1091 case MODE_BGRA: |
| 1092 CopyOrSwap(in_data, num_pixels, rgba, 1); |
| 1093 break; |
| 1094 case MODE_bgrA: |
| 1095 CopyOrSwap(in_data, num_pixels, rgba, 1); |
| 1096 WebPApplyAlphaMultiply(rgba, 0, num_pixels, 1, 0); |
| 1097 break; |
| 1098 case MODE_ARGB: |
| 1099 CopyOrSwap(in_data, num_pixels, rgba, 0); |
| 1100 break; |
| 1101 case MODE_Argb: |
| 1102 CopyOrSwap(in_data, num_pixels, rgba, 0); |
| 1103 WebPApplyAlphaMultiply(rgba, 1, num_pixels, 1, 0); |
| 1104 break; |
| 1105 case MODE_RGBA_4444: |
| 1106 ConvertBGRAToRGBA4444(in_data, num_pixels, rgba); |
| 1107 break; |
| 1108 case MODE_rgbA_4444: |
| 1109 ConvertBGRAToRGBA4444(in_data, num_pixels, rgba); |
| 1110 WebPApplyAlphaMultiply4444(rgba, num_pixels, 1, 0); |
| 1111 break; |
| 1112 case MODE_RGB_565: |
| 1113 ConvertBGRAToRGB565(in_data, num_pixels, rgba); |
| 1114 break; |
| 1115 default: |
| 1116 assert(0); // Code flow should not reach here. |
| 1117 } |
| 1118 } |
| 1119 |
| 1120 //------------------------------------------------------------------------------ |
| 1121 |
| 1122 #if defined(__cplusplus) || defined(c_plusplus) |
| 1123 } // extern "C" |
| 1124 #endif |
OLD | NEW |