1// Copyright 2016 Google Inc. All Rights Reserved.
2//
3// Use of this source code is governed by a BSD-style license
4// that can be found in the COPYING file in the root of the source
5// tree. An additional intellectual property rights grant can be found
6// in the file PATENTS. All contributing project authors may
7// be found in the AUTHORS file in the root of the source tree.
8// -----------------------------------------------------------------------------
9//
10// Image transform methods for lossless encoder.
11//
12// Authors: Vikas Arora (vikaas.arora@gmail.com)
13// Jyrki Alakuijala (jyrki@google.com)
14// Urvang Joshi (urvang@google.com)
15// Vincent Rabaud (vrabaud@google.com)
16
17#include "src/dsp/lossless.h"
18#include "src/dsp/lossless_common.h"
19#include "src/enc/vp8li_enc.h"
20
21#define MAX_DIFF_COST (1e30f)
22
23static const float kSpatialPredictorBias = 15.f;
24static const int kPredLowEffort = 11;
25static const uint32_t kMaskAlpha = 0xff000000;
26
27// Mostly used to reduce code size + readability
28static WEBP_INLINE int GetMin(int a, int b) { return (a > b) ? b : a; }
29
30//------------------------------------------------------------------------------
31// Methods to calculate Entropy (Shannon).
32
33static float PredictionCostSpatial(const int counts[256], int weight_0,
34 double exp_val) {
35 const int significant_symbols = 256 >> 4;
36 const double exp_decay_factor = 0.6;
37 double bits = weight_0 * counts[0];
38 int i;
39 for (i = 1; i < significant_symbols; ++i) {
40 bits += exp_val * (counts[i] + counts[256 - i]);
41 exp_val *= exp_decay_factor;
42 }
43 return (float)(-0.1 * bits);
44}
45
46static float PredictionCostSpatialHistogram(const int accumulated[4][256],
47 const int tile[4][256]) {
48 int i;
49 double retval = 0;
50 for (i = 0; i < 4; ++i) {
51 const double kExpValue = 0.94;
52 retval += PredictionCostSpatial(tile[i], 1, kExpValue);
53 retval += VP8LCombinedShannonEntropy(tile[i], accumulated[i]);
54 }
55 return (float)retval;
56}
57
58static WEBP_INLINE void UpdateHisto(int histo_argb[4][256], uint32_t argb) {
59 ++histo_argb[0][argb >> 24];
60 ++histo_argb[1][(argb >> 16) & 0xff];
61 ++histo_argb[2][(argb >> 8) & 0xff];
62 ++histo_argb[3][argb & 0xff];
63}
64
65//------------------------------------------------------------------------------
66// Spatial transform functions.
67
68static WEBP_INLINE void PredictBatch(int mode, int x_start, int y,
69 int num_pixels, const uint32_t* current,
70 const uint32_t* upper, uint32_t* out) {
71 if (x_start == 0) {
72 if (y == 0) {
73 // ARGB_BLACK.
74 VP8LPredictorsSub[0](current, NULL, 1, out);
75 } else {
76 // Top one.
77 VP8LPredictorsSub[2](current, upper, 1, out);
78 }
79 ++x_start;
80 ++out;
81 --num_pixels;
82 }
83 if (y == 0) {
84 // Left one.
85 VP8LPredictorsSub[1](current + x_start, NULL, num_pixels, out);
86 } else {
87 VP8LPredictorsSub[mode](current + x_start, upper + x_start, num_pixels,
88 out);
89 }
90}
91
92#if (WEBP_NEAR_LOSSLESS == 1)
93static WEBP_INLINE int GetMax(int a, int b) { return (a < b) ? b : a; }
94
95static int MaxDiffBetweenPixels(uint32_t p1, uint32_t p2) {
96 const int diff_a = abs((int)(p1 >> 24) - (int)(p2 >> 24));
97 const int diff_r = abs((int)((p1 >> 16) & 0xff) - (int)((p2 >> 16) & 0xff));
98 const int diff_g = abs((int)((p1 >> 8) & 0xff) - (int)((p2 >> 8) & 0xff));
99 const int diff_b = abs((int)(p1 & 0xff) - (int)(p2 & 0xff));
100 return GetMax(GetMax(diff_a, diff_r), GetMax(diff_g, diff_b));
101}
102
103static int MaxDiffAroundPixel(uint32_t current, uint32_t up, uint32_t down,
104 uint32_t left, uint32_t right) {
105 const int diff_up = MaxDiffBetweenPixels(current, up);
106 const int diff_down = MaxDiffBetweenPixels(current, down);
107 const int diff_left = MaxDiffBetweenPixels(current, left);
108 const int diff_right = MaxDiffBetweenPixels(current, right);
109 return GetMax(GetMax(diff_up, diff_down), GetMax(diff_left, diff_right));
110}
111
112static uint32_t AddGreenToBlueAndRed(uint32_t argb) {
113 const uint32_t green = (argb >> 8) & 0xff;
114 uint32_t red_blue = argb & 0x00ff00ffu;
115 red_blue += (green << 16) | green;
116 red_blue &= 0x00ff00ffu;
117 return (argb & 0xff00ff00u) | red_blue;
118}
119
120static void MaxDiffsForRow(int width, int stride, const uint32_t* const argb,
121 uint8_t* const max_diffs, int used_subtract_green) {
122 uint32_t current, up, down, left, right;
123 int x;
124 if (width <= 2) return;
125 current = argb[0];
126 right = argb[1];
127 if (used_subtract_green) {
128 current = AddGreenToBlueAndRed(current);
129 right = AddGreenToBlueAndRed(right);
130 }
131 // max_diffs[0] and max_diffs[width - 1] are never used.
132 for (x = 1; x < width - 1; ++x) {
133 up = argb[-stride + x];
134 down = argb[stride + x];
135 left = current;
136 current = right;
137 right = argb[x + 1];
138 if (used_subtract_green) {
139 up = AddGreenToBlueAndRed(up);
140 down = AddGreenToBlueAndRed(down);
141 right = AddGreenToBlueAndRed(right);
142 }
143 max_diffs[x] = MaxDiffAroundPixel(current, up, down, left, right);
144 }
145}
146
147// Quantize the difference between the actual component value and its prediction
148// to a multiple of quantization, working modulo 256, taking care not to cross
149// a boundary (inclusive upper limit).
150static uint8_t NearLosslessComponent(uint8_t value, uint8_t predict,
151 uint8_t boundary, int quantization) {
152 const int residual = (value - predict) & 0xff;
153 const int boundary_residual = (boundary - predict) & 0xff;
154 const int lower = residual & ~(quantization - 1);
155 const int upper = lower + quantization;
156 // Resolve ties towards a value closer to the prediction (i.e. towards lower
157 // if value comes after prediction and towards upper otherwise).
158 const int bias = ((boundary - value) & 0xff) < boundary_residual;
159 if (residual - lower < upper - residual + bias) {
160 // lower is closer to residual than upper.
161 if (residual > boundary_residual && lower <= boundary_residual) {
162 // Halve quantization step to avoid crossing boundary. This midpoint is
163 // on the same side of boundary as residual because midpoint >= residual
164 // (since lower is closer than upper) and residual is above the boundary.
165 return lower + (quantization >> 1);
166 }
167 return lower;
168 } else {
169 // upper is closer to residual than lower.
170 if (residual <= boundary_residual && upper > boundary_residual) {
171 // Halve quantization step to avoid crossing boundary. This midpoint is
172 // on the same side of boundary as residual because midpoint <= residual
173 // (since upper is closer than lower) and residual is below the boundary.
174 return lower + (quantization >> 1);
175 }
176 return upper & 0xff;
177 }
178}
179
180static WEBP_INLINE uint8_t NearLosslessDiff(uint8_t a, uint8_t b) {
181 return (uint8_t)((((int)(a) - (int)(b))) & 0xff);
182}
183
184// Quantize every component of the difference between the actual pixel value and
185// its prediction to a multiple of a quantization (a power of 2, not larger than
186// max_quantization which is a power of 2, smaller than max_diff). Take care if
187// value and predict have undergone subtract green, which means that red and
188// blue are represented as offsets from green.
189static uint32_t NearLossless(uint32_t value, uint32_t predict,
190 int max_quantization, int max_diff,
191 int used_subtract_green) {
192 int quantization;
193 uint8_t new_green = 0;
194 uint8_t green_diff = 0;
195 uint8_t a, r, g, b;
196 if (max_diff <= 2) {
197 return VP8LSubPixels(value, predict);
198 }
199 quantization = max_quantization;
200 while (quantization >= max_diff) {
201 quantization >>= 1;
202 }
203 if ((value >> 24) == 0 || (value >> 24) == 0xff) {
204 // Preserve transparency of fully transparent or fully opaque pixels.
205 a = NearLosslessDiff((value >> 24) & 0xff, (predict >> 24) & 0xff);
206 } else {
207 a = NearLosslessComponent(value >> 24, predict >> 24, 0xff, quantization);
208 }
209 g = NearLosslessComponent((value >> 8) & 0xff, (predict >> 8) & 0xff, 0xff,
210 quantization);
211 if (used_subtract_green) {
212 // The green offset will be added to red and blue components during decoding
213 // to obtain the actual red and blue values.
214 new_green = ((predict >> 8) + g) & 0xff;
215 // The amount by which green has been adjusted during quantization. It is
216 // subtracted from red and blue for compensation, to avoid accumulating two
217 // quantization errors in them.
218 green_diff = NearLosslessDiff(new_green, (value >> 8) & 0xff);
219 }
220 r = NearLosslessComponent(NearLosslessDiff((value >> 16) & 0xff, green_diff),
221 (predict >> 16) & 0xff, 0xff - new_green,
222 quantization);
223 b = NearLosslessComponent(NearLosslessDiff(value & 0xff, green_diff),
224 predict & 0xff, 0xff - new_green, quantization);
225 return ((uint32_t)a << 24) | ((uint32_t)r << 16) | ((uint32_t)g << 8) | b;
226}
227#endif // (WEBP_NEAR_LOSSLESS == 1)
228
229// Stores the difference between the pixel and its prediction in "out".
230// In case of a lossy encoding, updates the source image to avoid propagating
231// the deviation further to pixels which depend on the current pixel for their
232// predictions.
233static WEBP_INLINE void GetResidual(
234 int width, int height, uint32_t* const upper_row,
235 uint32_t* const current_row, const uint8_t* const max_diffs, int mode,
236 int x_start, int x_end, int y, int max_quantization, int exact,
237 int used_subtract_green, uint32_t* const out) {
238 if (exact) {
239 PredictBatch(mode, x_start, y, x_end - x_start, current_row, upper_row,
240 out);
241 } else {
242 const VP8LPredictorFunc pred_func = VP8LPredictors[mode];
243 int x;
244 for (x = x_start; x < x_end; ++x) {
245 uint32_t predict;
246 uint32_t residual;
247 if (y == 0) {
248 predict = (x == 0) ? ARGB_BLACK : current_row[x - 1]; // Left.
249 } else if (x == 0) {
250 predict = upper_row[x]; // Top.
251 } else {
252 predict = pred_func(current_row[x - 1], upper_row + x);
253 }
254#if (WEBP_NEAR_LOSSLESS == 1)
255 if (max_quantization == 1 || mode == 0 || y == 0 || y == height - 1 ||
256 x == 0 || x == width - 1) {
257 residual = VP8LSubPixels(current_row[x], predict);
258 } else {
259 residual = NearLossless(current_row[x], predict, max_quantization,
260 max_diffs[x], used_subtract_green);
261 // Update the source image.
262 current_row[x] = VP8LAddPixels(predict, residual);
263 // x is never 0 here so we do not need to update upper_row like below.
264 }
265#else
266 (void)max_diffs;
267 (void)height;
268 (void)max_quantization;
269 (void)used_subtract_green;
270 residual = VP8LSubPixels(current_row[x], predict);
271#endif
272 if ((current_row[x] & kMaskAlpha) == 0) {
273 // If alpha is 0, cleanup RGB. We can choose the RGB values of the
274 // residual for best compression. The prediction of alpha itself can be
275 // non-zero and must be kept though. We choose RGB of the residual to be
276 // 0.
277 residual &= kMaskAlpha;
278 // Update the source image.
279 current_row[x] = predict & ~kMaskAlpha;
280 // The prediction for the rightmost pixel in a row uses the leftmost
281 // pixel
282 // in that row as its top-right context pixel. Hence if we change the
283 // leftmost pixel of current_row, the corresponding change must be
284 // applied
285 // to upper_row as well where top-right context is being read from.
286 if (x == 0 && y != 0) upper_row[width] = current_row[0];
287 }
288 out[x - x_start] = residual;
289 }
290 }
291}
292
293// Returns best predictor and updates the accumulated histogram.
294// If max_quantization > 1, assumes that near lossless processing will be
295// applied, quantizing residuals to multiples of quantization levels up to
296// max_quantization (the actual quantization level depends on smoothness near
297// the given pixel).
298static int GetBestPredictorForTile(int width, int height,
299 int tile_x, int tile_y, int bits,
300 int accumulated[4][256],
301 uint32_t* const argb_scratch,
302 const uint32_t* const argb,
303 int max_quantization,
304 int exact, int used_subtract_green,
305 const uint32_t* const modes) {
306 const int kNumPredModes = 14;
307 const int start_x = tile_x << bits;
308 const int start_y = tile_y << bits;
309 const int tile_size = 1 << bits;
310 const int max_y = GetMin(tile_size, height - start_y);
311 const int max_x = GetMin(tile_size, width - start_x);
312 // Whether there exist columns just outside the tile.
313 const int have_left = (start_x > 0);
314 // Position and size of the strip covering the tile and adjacent columns if
315 // they exist.
316 const int context_start_x = start_x - have_left;
317#if (WEBP_NEAR_LOSSLESS == 1)
318 const int context_width = max_x + have_left + (max_x < width - start_x);
319#endif
320 const int tiles_per_row = VP8LSubSampleSize(width, bits);
321 // Prediction modes of the left and above neighbor tiles.
322 const int left_mode = (tile_x > 0) ?
323 (modes[tile_y * tiles_per_row + tile_x - 1] >> 8) & 0xff : 0xff;
324 const int above_mode = (tile_y > 0) ?
325 (modes[(tile_y - 1) * tiles_per_row + tile_x] >> 8) & 0xff : 0xff;
326 // The width of upper_row and current_row is one pixel larger than image width
327 // to allow the top right pixel to point to the leftmost pixel of the next row
328 // when at the right edge.
329 uint32_t* upper_row = argb_scratch;
330 uint32_t* current_row = upper_row + width + 1;
331 uint8_t* const max_diffs = (uint8_t*)(current_row + width + 1);
332 float best_diff = MAX_DIFF_COST;
333 int best_mode = 0;
334 int mode;
335 int histo_stack_1[4][256];
336 int histo_stack_2[4][256];
337 // Need pointers to be able to swap arrays.
338 int (*histo_argb)[256] = histo_stack_1;
339 int (*best_histo)[256] = histo_stack_2;
340 int i, j;
341 uint32_t residuals[1 << MAX_TRANSFORM_BITS];
342 assert(bits <= MAX_TRANSFORM_BITS);
343 assert(max_x <= (1 << MAX_TRANSFORM_BITS));
344
345 for (mode = 0; mode < kNumPredModes; ++mode) {
346 float cur_diff;
347 int relative_y;
348 memset(histo_argb, 0, sizeof(histo_stack_1));
349 if (start_y > 0) {
350 // Read the row above the tile which will become the first upper_row.
351 // Include a pixel to the left if it exists; include a pixel to the right
352 // in all cases (wrapping to the leftmost pixel of the next row if it does
353 // not exist).
354 memcpy(current_row + context_start_x,
355 argb + (start_y - 1) * width + context_start_x,
356 sizeof(*argb) * (max_x + have_left + 1));
357 }
358 for (relative_y = 0; relative_y < max_y; ++relative_y) {
359 const int y = start_y + relative_y;
360 int relative_x;
361 uint32_t* tmp = upper_row;
362 upper_row = current_row;
363 current_row = tmp;
364 // Read current_row. Include a pixel to the left if it exists; include a
365 // pixel to the right in all cases except at the bottom right corner of
366 // the image (wrapping to the leftmost pixel of the next row if it does
367 // not exist in the current row).
368 memcpy(current_row + context_start_x,
369 argb + y * width + context_start_x,
370 sizeof(*argb) * (max_x + have_left + (y + 1 < height)));
371#if (WEBP_NEAR_LOSSLESS == 1)
372 if (max_quantization > 1 && y >= 1 && y + 1 < height) {
373 MaxDiffsForRow(context_width, width, argb + y * width + context_start_x,
374 max_diffs + context_start_x, used_subtract_green);
375 }
376#endif
377
378 GetResidual(width, height, upper_row, current_row, max_diffs, mode,
379 start_x, start_x + max_x, y, max_quantization, exact,
380 used_subtract_green, residuals);
381 for (relative_x = 0; relative_x < max_x; ++relative_x) {
382 UpdateHisto(histo_argb, residuals[relative_x]);
383 }
384 }
385 cur_diff = PredictionCostSpatialHistogram(
386 (const int (*)[256])accumulated, (const int (*)[256])histo_argb);
387 // Favor keeping the areas locally similar.
388 if (mode == left_mode) cur_diff -= kSpatialPredictorBias;
389 if (mode == above_mode) cur_diff -= kSpatialPredictorBias;
390
391 if (cur_diff < best_diff) {
392 int (*tmp)[256] = histo_argb;
393 histo_argb = best_histo;
394 best_histo = tmp;
395 best_diff = cur_diff;
396 best_mode = mode;
397 }
398 }
399
400 for (i = 0; i < 4; i++) {
401 for (j = 0; j < 256; j++) {
402 accumulated[i][j] += best_histo[i][j];
403 }
404 }
405
406 return best_mode;
407}
408
409// Converts pixels of the image to residuals with respect to predictions.
410// If max_quantization > 1, applies near lossless processing, quantizing
411// residuals to multiples of quantization levels up to max_quantization
412// (the actual quantization level depends on smoothness near the given pixel).
413static void CopyImageWithPrediction(int width, int height,
414 int bits, uint32_t* const modes,
415 uint32_t* const argb_scratch,
416 uint32_t* const argb,
417 int low_effort, int max_quantization,
418 int exact, int used_subtract_green) {
419 const int tiles_per_row = VP8LSubSampleSize(width, bits);
420 // The width of upper_row and current_row is one pixel larger than image width
421 // to allow the top right pixel to point to the leftmost pixel of the next row
422 // when at the right edge.
423 uint32_t* upper_row = argb_scratch;
424 uint32_t* current_row = upper_row + width + 1;
425 uint8_t* current_max_diffs = (uint8_t*)(current_row + width + 1);
426#if (WEBP_NEAR_LOSSLESS == 1)
427 uint8_t* lower_max_diffs = current_max_diffs + width;
428#endif
429 int y;
430
431 for (y = 0; y < height; ++y) {
432 int x;
433 uint32_t* const tmp32 = upper_row;
434 upper_row = current_row;
435 current_row = tmp32;
436 memcpy(current_row, argb + y * width,
437 sizeof(*argb) * (width + (y + 1 < height)));
438
439 if (low_effort) {
440 PredictBatch(kPredLowEffort, 0, y, width, current_row, upper_row,
441 argb + y * width);
442 } else {
443#if (WEBP_NEAR_LOSSLESS == 1)
444 if (max_quantization > 1) {
445 // Compute max_diffs for the lower row now, because that needs the
446 // contents of argb for the current row, which we will overwrite with
447 // residuals before proceeding with the next row.
448 uint8_t* const tmp8 = current_max_diffs;
449 current_max_diffs = lower_max_diffs;
450 lower_max_diffs = tmp8;
451 if (y + 2 < height) {
452 MaxDiffsForRow(width, width, argb + (y + 1) * width, lower_max_diffs,
453 used_subtract_green);
454 }
455 }
456#endif
457 for (x = 0; x < width;) {
458 const int mode =
459 (modes[(y >> bits) * tiles_per_row + (x >> bits)] >> 8) & 0xff;
460 int x_end = x + (1 << bits);
461 if (x_end > width) x_end = width;
462 GetResidual(width, height, upper_row, current_row, current_max_diffs,
463 mode, x, x_end, y, max_quantization, exact,
464 used_subtract_green, argb + y * width + x);
465 x = x_end;
466 }
467 }
468 }
469}
470
471// Finds the best predictor for each tile, and converts the image to residuals
472// with respect to predictions. If near_lossless_quality < 100, applies
473// near lossless processing, shaving off more bits of residuals for lower
474// qualities.
475void VP8LResidualImage(int width, int height, int bits, int low_effort,
476 uint32_t* const argb, uint32_t* const argb_scratch,
477 uint32_t* const image, int near_lossless_quality,
478 int exact, int used_subtract_green) {
479 const int tiles_per_row = VP8LSubSampleSize(width, bits);
480 const int tiles_per_col = VP8LSubSampleSize(height, bits);
481 int tile_y;
482 int histo[4][256];
483 const int max_quantization = 1 << VP8LNearLosslessBits(near_lossless_quality);
484 if (low_effort) {
485 int i;
486 for (i = 0; i < tiles_per_row * tiles_per_col; ++i) {
487 image[i] = ARGB_BLACK | (kPredLowEffort << 8);
488 }
489 } else {
490 memset(histo, 0, sizeof(histo));
491 for (tile_y = 0; tile_y < tiles_per_col; ++tile_y) {
492 int tile_x;
493 for (tile_x = 0; tile_x < tiles_per_row; ++tile_x) {
494 const int pred = GetBestPredictorForTile(width, height, tile_x, tile_y,
495 bits, histo, argb_scratch, argb, max_quantization, exact,
496 used_subtract_green, image);
497 image[tile_y * tiles_per_row + tile_x] = ARGB_BLACK | (pred << 8);
498 }
499 }
500 }
501
502 CopyImageWithPrediction(width, height, bits, image, argb_scratch, argb,
503 low_effort, max_quantization, exact,
504 used_subtract_green);
505}
506
507//------------------------------------------------------------------------------
508// Color transform functions.
509
510static WEBP_INLINE void MultipliersClear(VP8LMultipliers* const m) {
511 m->green_to_red_ = 0;
512 m->green_to_blue_ = 0;
513 m->red_to_blue_ = 0;
514}
515
516static WEBP_INLINE void ColorCodeToMultipliers(uint32_t color_code,
517 VP8LMultipliers* const m) {
518 m->green_to_red_ = (color_code >> 0) & 0xff;
519 m->green_to_blue_ = (color_code >> 8) & 0xff;
520 m->red_to_blue_ = (color_code >> 16) & 0xff;
521}
522
523static WEBP_INLINE uint32_t MultipliersToColorCode(
524 const VP8LMultipliers* const m) {
525 return 0xff000000u |
526 ((uint32_t)(m->red_to_blue_) << 16) |
527 ((uint32_t)(m->green_to_blue_) << 8) |
528 m->green_to_red_;
529}
530
531static float PredictionCostCrossColor(const int accumulated[256],
532 const int counts[256]) {
533 // Favor low entropy, locally and globally.
534 // Favor small absolute values for PredictionCostSpatial
535 static const double kExpValue = 2.4;
536 return VP8LCombinedShannonEntropy(counts, accumulated) +
537 PredictionCostSpatial(counts, 3, kExpValue);
538}
539
540static float GetPredictionCostCrossColorRed(
541 const uint32_t* argb, int stride, int tile_width, int tile_height,
542 VP8LMultipliers prev_x, VP8LMultipliers prev_y, int green_to_red,
543 const int accumulated_red_histo[256]) {
544 int histo[256] = { 0 };
545 float cur_diff;
546
547 VP8LCollectColorRedTransforms(argb, stride, tile_width, tile_height,
548 green_to_red, histo);
549
550 cur_diff = PredictionCostCrossColor(accumulated_red_histo, histo);
551 if ((uint8_t)green_to_red == prev_x.green_to_red_) {
552 cur_diff -= 3; // favor keeping the areas locally similar
553 }
554 if ((uint8_t)green_to_red == prev_y.green_to_red_) {
555 cur_diff -= 3; // favor keeping the areas locally similar
556 }
557 if (green_to_red == 0) {
558 cur_diff -= 3;
559 }
560 return cur_diff;
561}
562
563static void GetBestGreenToRed(
564 const uint32_t* argb, int stride, int tile_width, int tile_height,
565 VP8LMultipliers prev_x, VP8LMultipliers prev_y, int quality,
566 const int accumulated_red_histo[256], VP8LMultipliers* const best_tx) {
567 const int kMaxIters = 4 + ((7 * quality) >> 8); // in range [4..6]
568 int green_to_red_best = 0;
569 int iter, offset;
570 float best_diff = GetPredictionCostCrossColorRed(
571 argb, stride, tile_width, tile_height, prev_x, prev_y,
572 green_to_red_best, accumulated_red_histo);
573 for (iter = 0; iter < kMaxIters; ++iter) {
574 // ColorTransformDelta is a 3.5 bit fixed point, so 32 is equal to
575 // one in color computation. Having initial delta here as 1 is sufficient
576 // to explore the range of (-2, 2).
577 const int delta = 32 >> iter;
578 // Try a negative and a positive delta from the best known value.
579 for (offset = -delta; offset <= delta; offset += 2 * delta) {
580 const int green_to_red_cur = offset + green_to_red_best;
581 const float cur_diff = GetPredictionCostCrossColorRed(
582 argb, stride, tile_width, tile_height, prev_x, prev_y,
583 green_to_red_cur, accumulated_red_histo);
584 if (cur_diff < best_diff) {
585 best_diff = cur_diff;
586 green_to_red_best = green_to_red_cur;
587 }
588 }
589 }
590 best_tx->green_to_red_ = (green_to_red_best & 0xff);
591}
592
593static float GetPredictionCostCrossColorBlue(
594 const uint32_t* argb, int stride, int tile_width, int tile_height,
595 VP8LMultipliers prev_x, VP8LMultipliers prev_y,
596 int green_to_blue, int red_to_blue, const int accumulated_blue_histo[256]) {
597 int histo[256] = { 0 };
598 float cur_diff;
599
600 VP8LCollectColorBlueTransforms(argb, stride, tile_width, tile_height,
601 green_to_blue, red_to_blue, histo);
602
603 cur_diff = PredictionCostCrossColor(accumulated_blue_histo, histo);
604 if ((uint8_t)green_to_blue == prev_x.green_to_blue_) {
605 cur_diff -= 3; // favor keeping the areas locally similar
606 }
607 if ((uint8_t)green_to_blue == prev_y.green_to_blue_) {
608 cur_diff -= 3; // favor keeping the areas locally similar
609 }
610 if ((uint8_t)red_to_blue == prev_x.red_to_blue_) {
611 cur_diff -= 3; // favor keeping the areas locally similar
612 }
613 if ((uint8_t)red_to_blue == prev_y.red_to_blue_) {
614 cur_diff -= 3; // favor keeping the areas locally similar
615 }
616 if (green_to_blue == 0) {
617 cur_diff -= 3;
618 }
619 if (red_to_blue == 0) {
620 cur_diff -= 3;
621 }
622 return cur_diff;
623}
624
625#define kGreenRedToBlueNumAxis 8
626#define kGreenRedToBlueMaxIters 7
627static void GetBestGreenRedToBlue(
628 const uint32_t* argb, int stride, int tile_width, int tile_height,
629 VP8LMultipliers prev_x, VP8LMultipliers prev_y, int quality,
630 const int accumulated_blue_histo[256],
631 VP8LMultipliers* const best_tx) {
632 const int8_t offset[kGreenRedToBlueNumAxis][2] =
633 {{0, -1}, {0, 1}, {-1, 0}, {1, 0}, {-1, -1}, {-1, 1}, {1, -1}, {1, 1}};
634 const int8_t delta_lut[kGreenRedToBlueMaxIters] = { 16, 16, 8, 4, 2, 2, 2 };
635 const int iters =
636 (quality < 25) ? 1 : (quality > 50) ? kGreenRedToBlueMaxIters : 4;
637 int green_to_blue_best = 0;
638 int red_to_blue_best = 0;
639 int iter;
640 // Initial value at origin:
641 float best_diff = GetPredictionCostCrossColorBlue(
642 argb, stride, tile_width, tile_height, prev_x, prev_y,
643 green_to_blue_best, red_to_blue_best, accumulated_blue_histo);
644 for (iter = 0; iter < iters; ++iter) {
645 const int delta = delta_lut[iter];
646 int axis;
647 for (axis = 0; axis < kGreenRedToBlueNumAxis; ++axis) {
648 const int green_to_blue_cur =
649 offset[axis][0] * delta + green_to_blue_best;
650 const int red_to_blue_cur = offset[axis][1] * delta + red_to_blue_best;
651 const float cur_diff = GetPredictionCostCrossColorBlue(
652 argb, stride, tile_width, tile_height, prev_x, prev_y,
653 green_to_blue_cur, red_to_blue_cur, accumulated_blue_histo);
654 if (cur_diff < best_diff) {
655 best_diff = cur_diff;
656 green_to_blue_best = green_to_blue_cur;
657 red_to_blue_best = red_to_blue_cur;
658 }
659 if (quality < 25 && iter == 4) {
660 // Only axis aligned diffs for lower quality.
661 break; // next iter.
662 }
663 }
664 if (delta == 2 && green_to_blue_best == 0 && red_to_blue_best == 0) {
665 // Further iterations would not help.
666 break; // out of iter-loop.
667 }
668 }
669 best_tx->green_to_blue_ = green_to_blue_best & 0xff;
670 best_tx->red_to_blue_ = red_to_blue_best & 0xff;
671}
672#undef kGreenRedToBlueMaxIters
673#undef kGreenRedToBlueNumAxis
674
675static VP8LMultipliers GetBestColorTransformForTile(
676 int tile_x, int tile_y, int bits,
677 VP8LMultipliers prev_x,
678 VP8LMultipliers prev_y,
679 int quality, int xsize, int ysize,
680 const int accumulated_red_histo[256],
681 const int accumulated_blue_histo[256],
682 const uint32_t* const argb) {
683 const int max_tile_size = 1 << bits;
684 const int tile_y_offset = tile_y * max_tile_size;
685 const int tile_x_offset = tile_x * max_tile_size;
686 const int all_x_max = GetMin(tile_x_offset + max_tile_size, xsize);
687 const int all_y_max = GetMin(tile_y_offset + max_tile_size, ysize);
688 const int tile_width = all_x_max - tile_x_offset;
689 const int tile_height = all_y_max - tile_y_offset;
690 const uint32_t* const tile_argb = argb + tile_y_offset * xsize
691 + tile_x_offset;
692 VP8LMultipliers best_tx;
693 MultipliersClear(&best_tx);
694
695 GetBestGreenToRed(tile_argb, xsize, tile_width, tile_height,
696 prev_x, prev_y, quality, accumulated_red_histo, &best_tx);
697 GetBestGreenRedToBlue(tile_argb, xsize, tile_width, tile_height,
698 prev_x, prev_y, quality, accumulated_blue_histo,
699 &best_tx);
700 return best_tx;
701}
702
703static void CopyTileWithColorTransform(int xsize, int ysize,
704 int tile_x, int tile_y,
705 int max_tile_size,
706 VP8LMultipliers color_transform,
707 uint32_t* argb) {
708 const int xscan = GetMin(max_tile_size, xsize - tile_x);
709 int yscan = GetMin(max_tile_size, ysize - tile_y);
710 argb += tile_y * xsize + tile_x;
711 while (yscan-- > 0) {
712 VP8LTransformColor(&color_transform, argb, xscan);
713 argb += xsize;
714 }
715}
716
717void VP8LColorSpaceTransform(int width, int height, int bits, int quality,
718 uint32_t* const argb, uint32_t* image) {
719 const int max_tile_size = 1 << bits;
720 const int tile_xsize = VP8LSubSampleSize(width, bits);
721 const int tile_ysize = VP8LSubSampleSize(height, bits);
722 int accumulated_red_histo[256] = { 0 };
723 int accumulated_blue_histo[256] = { 0 };
724 int tile_x, tile_y;
725 VP8LMultipliers prev_x, prev_y;
726 MultipliersClear(&prev_y);
727 MultipliersClear(&prev_x);
728 for (tile_y = 0; tile_y < tile_ysize; ++tile_y) {
729 for (tile_x = 0; tile_x < tile_xsize; ++tile_x) {
730 int y;
731 const int tile_x_offset = tile_x * max_tile_size;
732 const int tile_y_offset = tile_y * max_tile_size;
733 const int all_x_max = GetMin(tile_x_offset + max_tile_size, width);
734 const int all_y_max = GetMin(tile_y_offset + max_tile_size, height);
735 const int offset = tile_y * tile_xsize + tile_x;
736 if (tile_y != 0) {
737 ColorCodeToMultipliers(image[offset - tile_xsize], &prev_y);
738 }
739 prev_x = GetBestColorTransformForTile(tile_x, tile_y, bits,
740 prev_x, prev_y,
741 quality, width, height,
742 accumulated_red_histo,
743 accumulated_blue_histo,
744 argb);
745 image[offset] = MultipliersToColorCode(&prev_x);
746 CopyTileWithColorTransform(width, height, tile_x_offset, tile_y_offset,
747 max_tile_size, prev_x, argb);
748
749 // Gather accumulated histogram data.
750 for (y = tile_y_offset; y < all_y_max; ++y) {
751 int ix = y * width + tile_x_offset;
752 const int ix_end = ix + all_x_max - tile_x_offset;
753 for (; ix < ix_end; ++ix) {
754 const uint32_t pix = argb[ix];
755 if (ix >= 2 &&
756 pix == argb[ix - 2] &&
757 pix == argb[ix - 1]) {
758 continue; // repeated pixels are handled by backward references
759 }
760 if (ix >= width + 2 &&
761 argb[ix - 2] == argb[ix - width - 2] &&
762 argb[ix - 1] == argb[ix - width - 1] &&
763 pix == argb[ix - width]) {
764 continue; // repeated pixels are handled by backward references
765 }
766 ++accumulated_red_histo[(pix >> 16) & 0xff];
767 ++accumulated_blue_histo[(pix >> 0) & 0xff];
768 }
769 }
770 }
771 }
772}
773