| OLD | NEW |
| 1 // Copyright 2014 Google Inc. All Rights Reserved. | 1 // Copyright 2014 Google Inc. All Rights Reserved. |
| 2 // | 2 // |
| 3 // Use of this source code is governed by a BSD-style license | 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 | 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 | 5 // tree. An additional intellectual property rights grant can be found |
| 6 // in the file PATENTS. All contributing project authors may | 6 // in the file PATENTS. All contributing project authors may |
| 7 // be found in the AUTHORS file in the root of the source tree. | 7 // be found in the AUTHORS file in the root of the source tree. |
| 8 // ----------------------------------------------------------------------------- | 8 // ----------------------------------------------------------------------------- |
| 9 // | 9 // |
| 10 // WebPPicture utils for colorspace conversion | 10 // WebPPicture utils for colorspace conversion |
| 11 // | 11 // |
| 12 // Author: Skal (pascal.massimino@gmail.com) | 12 // Author: Skal (pascal.massimino@gmail.com) |
| 13 | 13 |
| 14 #include <assert.h> | 14 #include <assert.h> |
| 15 #include <stdlib.h> | 15 #include <stdlib.h> |
| 16 #include <math.h> | 16 #include <math.h> |
| 17 | 17 |
| 18 #include "./vp8enci.h" | 18 #include "./vp8i_enc.h" |
| 19 #include "../utils/random.h" | 19 #include "../utils/random_utils.h" |
| 20 #include "../utils/utils.h" | 20 #include "../utils/utils.h" |
| 21 #include "../dsp/yuv.h" | 21 #include "../dsp/yuv.h" |
| 22 | 22 |
| 23 // Uncomment to disable gamma-compression during RGB->U/V averaging | 23 // Uncomment to disable gamma-compression during RGB->U/V averaging |
| 24 #define USE_GAMMA_COMPRESSION | 24 #define USE_GAMMA_COMPRESSION |
| 25 | 25 |
| 26 // If defined, use table to compute x / alpha. | 26 // If defined, use table to compute x / alpha. |
| 27 #define USE_INVERSE_ALPHA_TABLE | 27 #define USE_INVERSE_ALPHA_TABLE |
| 28 | 28 |
| 29 static const union { | 29 static const union { |
| (...skipping 116 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 146 return (rg == NULL) ? VP8RGBToU(r, g, b, YUV_HALF << 2) | 146 return (rg == NULL) ? VP8RGBToU(r, g, b, YUV_HALF << 2) |
| 147 : VP8RGBToU(r, g, b, VP8RandomBits(rg, YUV_FIX + 2)); | 147 : VP8RGBToU(r, g, b, VP8RandomBits(rg, YUV_FIX + 2)); |
| 148 } | 148 } |
| 149 | 149 |
| 150 static int RGBToV(int r, int g, int b, VP8Random* const rg) { | 150 static int RGBToV(int r, int g, int b, VP8Random* const rg) { |
| 151 return (rg == NULL) ? VP8RGBToV(r, g, b, YUV_HALF << 2) | 151 return (rg == NULL) ? VP8RGBToV(r, g, b, YUV_HALF << 2) |
| 152 : VP8RGBToV(r, g, b, VP8RandomBits(rg, YUV_FIX + 2)); | 152 : VP8RGBToV(r, g, b, VP8RandomBits(rg, YUV_FIX + 2)); |
| 153 } | 153 } |
| 154 | 154 |
| 155 //------------------------------------------------------------------------------ | 155 //------------------------------------------------------------------------------ |
| 156 // Smart RGB->YUV conversion | 156 // Sharp RGB->YUV conversion |
| 157 | 157 |
| 158 static const int kNumIterations = 6; | 158 static const int kNumIterations = 4; |
| 159 static const int kMinDimensionIterativeConversion = 4; | 159 static const int kMinDimensionIterativeConversion = 4; |
| 160 | 160 |
| 161 // We could use SFIX=0 and only uint8_t for fixed_y_t, but it produces some | 161 // We could use SFIX=0 and only uint8_t for fixed_y_t, but it produces some |
| 162 // banding sometimes. Better use extra precision. | 162 // banding sometimes. Better use extra precision. |
| 163 #define SFIX 2 // fixed-point precision of RGB and Y/W | 163 #define SFIX 2 // fixed-point precision of RGB and Y/W |
| 164 typedef int16_t fixed_t; // signed type with extra SFIX precision for UV | 164 typedef int16_t fixed_t; // signed type with extra SFIX precision for UV |
| 165 typedef uint16_t fixed_y_t; // unsigned type with extra SFIX precision for W | 165 typedef uint16_t fixed_y_t; // unsigned type with extra SFIX precision for W |
| 166 | 166 |
| 167 #define SHALF (1 << SFIX >> 1) | 167 #define SHALF (1 << SFIX >> 1) |
| 168 #define MAX_Y_T ((256 << SFIX) - 1) | 168 #define MAX_Y_T ((256 << SFIX) - 1) |
| 169 #define SROUNDER (1 << (YUV_FIX + SFIX - 1)) | 169 #define SROUNDER (1 << (YUV_FIX + SFIX - 1)) |
| 170 | 170 |
| 171 #if defined(USE_GAMMA_COMPRESSION) | 171 #if defined(USE_GAMMA_COMPRESSION) |
| 172 | 172 |
| 173 // float variant of gamma-correction | 173 // float variant of gamma-correction |
| 174 // We use tables of different size and precision, along with a 'real-world' | 174 // We use tables of different size and precision for the Rec709 |
| 175 // Gamma value close to ~2. | 175 // transfer function. |
| 176 #define kGammaF 2.2 | 176 #define kGammaF (1./0.45) |
| 177 static float kGammaToLinearTabF[MAX_Y_T + 1]; // size scales with Y_FIX | 177 static float kGammaToLinearTabF[MAX_Y_T + 1]; // size scales with Y_FIX |
| 178 static float kLinearToGammaTabF[kGammaTabSize + 2]; | 178 static float kLinearToGammaTabF[kGammaTabSize + 2]; |
| 179 static volatile int kGammaTablesFOk = 0; | 179 static volatile int kGammaTablesFOk = 0; |
| 180 | 180 |
| 181 static WEBP_TSAN_IGNORE_FUNCTION void InitGammaTablesF(void) { | 181 static WEBP_TSAN_IGNORE_FUNCTION void InitGammaTablesF(void) { |
| 182 if (!kGammaTablesFOk) { | 182 if (!kGammaTablesFOk) { |
| 183 int v; | 183 int v; |
| 184 const double norm = 1. / MAX_Y_T; | 184 const double norm = 1. / MAX_Y_T; |
| 185 const double scale = 1. / kGammaTabSize; | 185 const double scale = 1. / kGammaTabSize; |
| 186 const double a = 0.099; |
| 187 const double thresh = 0.018; |
| 186 for (v = 0; v <= MAX_Y_T; ++v) { | 188 for (v = 0; v <= MAX_Y_T; ++v) { |
| 187 kGammaToLinearTabF[v] = (float)pow(norm * v, kGammaF); | 189 const double g = norm * v; |
| 190 if (g <= thresh * 4.5) { |
| 191 kGammaToLinearTabF[v] = (float)(g / 4.5); |
| 192 } else { |
| 193 const double a_rec = 1. / (1. + a); |
| 194 kGammaToLinearTabF[v] = (float)pow(a_rec * (g + a), kGammaF); |
| 195 } |
| 188 } | 196 } |
| 189 for (v = 0; v <= kGammaTabSize; ++v) { | 197 for (v = 0; v <= kGammaTabSize; ++v) { |
| 190 kLinearToGammaTabF[v] = (float)(MAX_Y_T * pow(scale * v, 1. / kGammaF)); | 198 const double g = scale * v; |
| 199 double value; |
| 200 if (g <= thresh) { |
| 201 value = 4.5 * g; |
| 202 } else { |
| 203 value = (1. + a) * pow(g, 1. / kGammaF) - a; |
| 204 } |
| 205 kLinearToGammaTabF[v] = (float)(MAX_Y_T * value); |
| 191 } | 206 } |
| 192 // to prevent small rounding errors to cause read-overflow: | 207 // to prevent small rounding errors to cause read-overflow: |
| 193 kLinearToGammaTabF[kGammaTabSize + 1] = kLinearToGammaTabF[kGammaTabSize]; | 208 kLinearToGammaTabF[kGammaTabSize + 1] = kLinearToGammaTabF[kGammaTabSize]; |
| 194 kGammaTablesFOk = 1; | 209 kGammaTablesFOk = 1; |
| 195 } | 210 } |
| 196 } | 211 } |
| 197 | 212 |
| 198 static WEBP_INLINE float GammaToLinearF(int v) { | 213 static WEBP_INLINE float GammaToLinearF(int v) { |
| 199 return kGammaToLinearTabF[v]; | 214 return kGammaToLinearTabF[v]; |
| 200 } | 215 } |
| (...skipping 27 matching lines...) Expand all Loading... |
| 228 return (!(v & ~0xff)) ? (uint8_t)v : (v < 0) ? 0u : 255u; | 243 return (!(v & ~0xff)) ? (uint8_t)v : (v < 0) ? 0u : 255u; |
| 229 } | 244 } |
| 230 | 245 |
| 231 static fixed_y_t clip_y(int y) { | 246 static fixed_y_t clip_y(int y) { |
| 232 return (!(y & ~MAX_Y_T)) ? (fixed_y_t)y : (y < 0) ? 0 : MAX_Y_T; | 247 return (!(y & ~MAX_Y_T)) ? (fixed_y_t)y : (y < 0) ? 0 : MAX_Y_T; |
| 233 } | 248 } |
| 234 | 249 |
| 235 //------------------------------------------------------------------------------ | 250 //------------------------------------------------------------------------------ |
| 236 | 251 |
| 237 static int RGBToGray(int r, int g, int b) { | 252 static int RGBToGray(int r, int g, int b) { |
| 238 const int luma = 19595 * r + 38470 * g + 7471 * b + YUV_HALF; | 253 const int luma = 13933 * r + 46871 * g + 4732 * b + YUV_HALF; |
| 239 return (luma >> YUV_FIX); | 254 return (luma >> YUV_FIX); |
| 240 } | 255 } |
| 241 | 256 |
| 242 static float RGBToGrayF(float r, float g, float b) { | 257 static float RGBToGrayF(float r, float g, float b) { |
| 243 return 0.299f * r + 0.587f * g + 0.114f * b; | 258 return (float)(0.2126 * r + 0.7152 * g + 0.0722 * b); |
| 244 } | 259 } |
| 245 | 260 |
| 246 static int ScaleDown(int a, int b, int c, int d) { | 261 static int ScaleDown(int a, int b, int c, int d) { |
| 247 const float A = GammaToLinearF(a); | 262 const float A = GammaToLinearF(a); |
| 248 const float B = GammaToLinearF(b); | 263 const float B = GammaToLinearF(b); |
| 249 const float C = GammaToLinearF(c); | 264 const float C = GammaToLinearF(c); |
| 250 const float D = GammaToLinearF(d); | 265 const float D = GammaToLinearF(d); |
| 251 return LinearToGammaF(0.25f * (A + B + C + D)); | 266 return LinearToGammaF(0.25f * (A + B + C + D)); |
| 252 } | 267 } |
| 253 | 268 |
| 254 static WEBP_INLINE void UpdateW(const fixed_y_t* src, fixed_y_t* dst, int len) { | 269 static WEBP_INLINE void UpdateW(const fixed_y_t* src, fixed_y_t* dst, int w) { |
| 255 while (len-- > 0) { | 270 int i; |
| 256 const float R = GammaToLinearF(src[0]); | 271 for (i = 0; i < w; ++i) { |
| 257 const float G = GammaToLinearF(src[1]); | 272 const float R = GammaToLinearF(src[0 * w + i]); |
| 258 const float B = GammaToLinearF(src[2]); | 273 const float G = GammaToLinearF(src[1 * w + i]); |
| 274 const float B = GammaToLinearF(src[2 * w + i]); |
| 259 const float Y = RGBToGrayF(R, G, B); | 275 const float Y = RGBToGrayF(R, G, B); |
| 260 *dst++ = (fixed_y_t)LinearToGammaF(Y); | 276 dst[i] = (fixed_y_t)LinearToGammaF(Y); |
| 261 src += 3; | |
| 262 } | 277 } |
| 263 } | 278 } |
| 264 | 279 |
| 265 static int UpdateChroma(const fixed_y_t* src1, | 280 static void UpdateChroma(const fixed_y_t* src1, const fixed_y_t* src2, |
| 266 const fixed_y_t* src2, | 281 fixed_t* dst, int uv_w) { |
| 267 fixed_t* dst, fixed_y_t* tmp, int len) { | 282 int i; |
| 268 int diff = 0; | 283 for (i = 0; i < uv_w; ++i) { |
| 269 while (len--> 0) { | 284 const int r = ScaleDown(src1[0 * uv_w + 0], src1[0 * uv_w + 1], |
| 270 const int r = ScaleDown(src1[0], src1[3], src2[0], src2[3]); | 285 src2[0 * uv_w + 0], src2[0 * uv_w + 1]); |
| 271 const int g = ScaleDown(src1[1], src1[4], src2[1], src2[4]); | 286 const int g = ScaleDown(src1[2 * uv_w + 0], src1[2 * uv_w + 1], |
| 272 const int b = ScaleDown(src1[2], src1[5], src2[2], src2[5]); | 287 src2[2 * uv_w + 0], src2[2 * uv_w + 1]); |
| 288 const int b = ScaleDown(src1[4 * uv_w + 0], src1[4 * uv_w + 1], |
| 289 src2[4 * uv_w + 0], src2[4 * uv_w + 1]); |
| 273 const int W = RGBToGray(r, g, b); | 290 const int W = RGBToGray(r, g, b); |
| 274 const int r_avg = (src1[0] + src1[3] + src2[0] + src2[3] + 2) >> 2; | 291 dst[0 * uv_w] = (fixed_t)(r - W); |
| 275 const int g_avg = (src1[1] + src1[4] + src2[1] + src2[4] + 2) >> 2; | 292 dst[1 * uv_w] = (fixed_t)(g - W); |
| 276 const int b_avg = (src1[2] + src1[5] + src2[2] + src2[5] + 2) >> 2; | 293 dst[2 * uv_w] = (fixed_t)(b - W); |
| 277 dst[0] = (fixed_t)(r - W); | 294 dst += 1; |
| 278 dst[1] = (fixed_t)(g - W); | 295 src1 += 2; |
| 279 dst[2] = (fixed_t)(b - W); | 296 src2 += 2; |
| 280 dst += 3; | |
| 281 src1 += 6; | |
| 282 src2 += 6; | |
| 283 if (tmp != NULL) { | |
| 284 tmp[0] = tmp[1] = clip_y(W); | |
| 285 tmp += 2; | |
| 286 } | |
| 287 diff += abs(RGBToGray(r_avg, g_avg, b_avg) - W); | |
| 288 } | 297 } |
| 289 return diff; | 298 } |
| 299 |
| 300 static void StoreGray(const fixed_y_t* rgb, fixed_y_t* y, int w) { |
| 301 int i; |
| 302 for (i = 0; i < w; ++i) { |
| 303 y[i] = RGBToGray(rgb[0 * w + i], rgb[1 * w + i], rgb[2 * w + i]); |
| 304 } |
| 290 } | 305 } |
| 291 | 306 |
| 292 //------------------------------------------------------------------------------ | 307 //------------------------------------------------------------------------------ |
| 293 | 308 |
| 294 static WEBP_INLINE int Filter(const fixed_t* const A, const fixed_t* const B, | 309 static WEBP_INLINE fixed_y_t Filter2(int A, int B, int W0) { |
| 295 int rightwise) { | 310 const int v0 = (A * 3 + B + 2) >> 2; |
| 296 int v; | 311 return clip_y(v0 + W0); |
| 297 if (!rightwise) { | |
| 298 v = (A[0] * 9 + A[-3] * 3 + B[0] * 3 + B[-3]); | |
| 299 } else { | |
| 300 v = (A[0] * 9 + A[+3] * 3 + B[0] * 3 + B[+3]); | |
| 301 } | |
| 302 return (v + 8) >> 4; | |
| 303 } | 312 } |
| 304 | 313 |
| 305 static WEBP_INLINE int Filter2(int A, int B) { return (A * 3 + B + 2) >> 2; } | |
| 306 | |
| 307 //------------------------------------------------------------------------------ | 314 //------------------------------------------------------------------------------ |
| 308 | 315 |
| 309 static WEBP_INLINE fixed_y_t UpLift(uint8_t a) { // 8bit -> SFIX | 316 static WEBP_INLINE fixed_y_t UpLift(uint8_t a) { // 8bit -> SFIX |
| 310 return ((fixed_y_t)a << SFIX) | SHALF; | 317 return ((fixed_y_t)a << SFIX) | SHALF; |
| 311 } | 318 } |
| 312 | 319 |
| 313 static void ImportOneRow(const uint8_t* const r_ptr, | 320 static void ImportOneRow(const uint8_t* const r_ptr, |
| 314 const uint8_t* const g_ptr, | 321 const uint8_t* const g_ptr, |
| 315 const uint8_t* const b_ptr, | 322 const uint8_t* const b_ptr, |
| 316 int step, | 323 int step, |
| 317 int pic_width, | 324 int pic_width, |
| 318 fixed_y_t* const dst) { | 325 fixed_y_t* const dst) { |
| 319 int i; | 326 int i; |
| 327 const int w = (pic_width + 1) & ~1; |
| 320 for (i = 0; i < pic_width; ++i) { | 328 for (i = 0; i < pic_width; ++i) { |
| 321 const int off = i * step; | 329 const int off = i * step; |
| 322 dst[3 * i + 0] = UpLift(r_ptr[off]); | 330 dst[i + 0 * w] = UpLift(r_ptr[off]); |
| 323 dst[3 * i + 1] = UpLift(g_ptr[off]); | 331 dst[i + 1 * w] = UpLift(g_ptr[off]); |
| 324 dst[3 * i + 2] = UpLift(b_ptr[off]); | 332 dst[i + 2 * w] = UpLift(b_ptr[off]); |
| 325 } | 333 } |
| 326 if (pic_width & 1) { // replicate rightmost pixel | 334 if (pic_width & 1) { // replicate rightmost pixel |
| 327 memcpy(dst + 3 * pic_width, dst + 3 * (pic_width - 1), 3 * sizeof(*dst)); | 335 dst[pic_width + 0 * w] = dst[pic_width + 0 * w - 1]; |
| 336 dst[pic_width + 1 * w] = dst[pic_width + 1 * w - 1]; |
| 337 dst[pic_width + 2 * w] = dst[pic_width + 2 * w - 1]; |
| 328 } | 338 } |
| 329 } | 339 } |
| 330 | 340 |
| 331 static void InterpolateTwoRows(const fixed_y_t* const best_y, | 341 static void InterpolateTwoRows(const fixed_y_t* const best_y, |
| 332 const fixed_t* const prev_uv, | 342 const fixed_t* prev_uv, |
| 333 const fixed_t* const cur_uv, | 343 const fixed_t* cur_uv, |
| 334 const fixed_t* const next_uv, | 344 const fixed_t* next_uv, |
| 335 int w, | 345 int w, |
| 336 fixed_y_t* const out1, | 346 fixed_y_t* out1, |
| 337 fixed_y_t* const out2) { | 347 fixed_y_t* out2) { |
| 338 int i, k; | 348 const int uv_w = w >> 1; |
| 339 { // special boundary case for i==0 | 349 const int len = (w - 1) >> 1; // length to filter |
| 340 const int W0 = best_y[0]; | 350 int k = 3; |
| 341 const int W1 = best_y[w]; | 351 while (k-- > 0) { // process each R/G/B segments in turn |
| 342 for (k = 0; k <= 2; ++k) { | 352 // special boundary case for i==0 |
| 343 out1[k] = clip_y(Filter2(cur_uv[k], prev_uv[k]) + W0); | 353 out1[0] = Filter2(cur_uv[0], prev_uv[0], best_y[0]); |
| 344 out2[k] = clip_y(Filter2(cur_uv[k], next_uv[k]) + W1); | 354 out2[0] = Filter2(cur_uv[0], next_uv[0], best_y[w]); |
| 355 |
| 356 WebPSharpYUVFilterRow(cur_uv, prev_uv, len, best_y + 0 + 1, out1 + 1); |
| 357 WebPSharpYUVFilterRow(cur_uv, next_uv, len, best_y + w + 1, out2 + 1); |
| 358 |
| 359 // special boundary case for i == w - 1 when w is even |
| 360 if (!(w & 1)) { |
| 361 out1[w - 1] = Filter2(cur_uv[uv_w - 1], prev_uv[uv_w - 1], |
| 362 best_y[w - 1 + 0]); |
| 363 out2[w - 1] = Filter2(cur_uv[uv_w - 1], next_uv[uv_w - 1], |
| 364 best_y[w - 1 + w]); |
| 345 } | 365 } |
| 346 } | 366 out1 += w; |
| 347 for (i = 1; i < w - 1; ++i) { | 367 out2 += w; |
| 348 const int W0 = best_y[i + 0]; | 368 prev_uv += uv_w; |
| 349 const int W1 = best_y[i + w]; | 369 cur_uv += uv_w; |
| 350 const int off = 3 * (i >> 1); | 370 next_uv += uv_w; |
| 351 for (k = 0; k <= 2; ++k) { | |
| 352 const int tmp0 = Filter(cur_uv + off + k, prev_uv + off + k, i & 1); | |
| 353 const int tmp1 = Filter(cur_uv + off + k, next_uv + off + k, i & 1); | |
| 354 out1[3 * i + k] = clip_y(tmp0 + W0); | |
| 355 out2[3 * i + k] = clip_y(tmp1 + W1); | |
| 356 } | |
| 357 } | |
| 358 { // special boundary case for i == w - 1 | |
| 359 const int W0 = best_y[i + 0]; | |
| 360 const int W1 = best_y[i + w]; | |
| 361 const int off = 3 * (i >> 1); | |
| 362 for (k = 0; k <= 2; ++k) { | |
| 363 out1[3 * i + k] = clip_y(Filter2(cur_uv[off + k], prev_uv[off + k]) + W0); | |
| 364 out2[3 * i + k] = clip_y(Filter2(cur_uv[off + k], next_uv[off + k]) + W1); | |
| 365 } | |
| 366 } | 371 } |
| 367 } | 372 } |
| 368 | 373 |
| 369 static WEBP_INLINE uint8_t ConvertRGBToY(int r, int g, int b) { | 374 static WEBP_INLINE uint8_t ConvertRGBToY(int r, int g, int b) { |
| 370 const int luma = 16839 * r + 33059 * g + 6420 * b + SROUNDER; | 375 const int luma = 16839 * r + 33059 * g + 6420 * b + SROUNDER; |
| 371 return clip_8b(16 + (luma >> (YUV_FIX + SFIX))); | 376 return clip_8b(16 + (luma >> (YUV_FIX + SFIX))); |
| 372 } | 377 } |
| 373 | 378 |
| 374 static WEBP_INLINE uint8_t ConvertRGBToU(int r, int g, int b) { | 379 static WEBP_INLINE uint8_t ConvertRGBToU(int r, int g, int b) { |
| 375 const int u = -9719 * r - 19081 * g + 28800 * b + SROUNDER; | 380 const int u = -9719 * r - 19081 * g + 28800 * b + SROUNDER; |
| (...skipping 11 matching lines...) Expand all Loading... |
| 387 uint8_t* dst_y = picture->y; | 392 uint8_t* dst_y = picture->y; |
| 388 uint8_t* dst_u = picture->u; | 393 uint8_t* dst_u = picture->u; |
| 389 uint8_t* dst_v = picture->v; | 394 uint8_t* dst_v = picture->v; |
| 390 const fixed_t* const best_uv_base = best_uv; | 395 const fixed_t* const best_uv_base = best_uv; |
| 391 const int w = (picture->width + 1) & ~1; | 396 const int w = (picture->width + 1) & ~1; |
| 392 const int h = (picture->height + 1) & ~1; | 397 const int h = (picture->height + 1) & ~1; |
| 393 const int uv_w = w >> 1; | 398 const int uv_w = w >> 1; |
| 394 const int uv_h = h >> 1; | 399 const int uv_h = h >> 1; |
| 395 for (best_uv = best_uv_base, j = 0; j < picture->height; ++j) { | 400 for (best_uv = best_uv_base, j = 0; j < picture->height; ++j) { |
| 396 for (i = 0; i < picture->width; ++i) { | 401 for (i = 0; i < picture->width; ++i) { |
| 397 const int off = 3 * (i >> 1); | 402 const int off = (i >> 1); |
| 398 const int W = best_y[i]; | 403 const int W = best_y[i]; |
| 399 const int r = best_uv[off + 0] + W; | 404 const int r = best_uv[off + 0 * uv_w] + W; |
| 400 const int g = best_uv[off + 1] + W; | 405 const int g = best_uv[off + 1 * uv_w] + W; |
| 401 const int b = best_uv[off + 2] + W; | 406 const int b = best_uv[off + 2 * uv_w] + W; |
| 402 dst_y[i] = ConvertRGBToY(r, g, b); | 407 dst_y[i] = ConvertRGBToY(r, g, b); |
| 403 } | 408 } |
| 404 best_y += w; | 409 best_y += w; |
| 405 best_uv += (j & 1) * 3 * uv_w; | 410 best_uv += (j & 1) * 3 * uv_w; |
| 406 dst_y += picture->y_stride; | 411 dst_y += picture->y_stride; |
| 407 } | 412 } |
| 408 for (best_uv = best_uv_base, j = 0; j < uv_h; ++j) { | 413 for (best_uv = best_uv_base, j = 0; j < uv_h; ++j) { |
| 409 for (i = 0; i < uv_w; ++i) { | 414 for (i = 0; i < uv_w; ++i) { |
| 410 const int off = 3 * i; | 415 const int off = i; |
| 411 const int r = best_uv[off + 0]; | 416 const int r = best_uv[off + 0 * uv_w]; |
| 412 const int g = best_uv[off + 1]; | 417 const int g = best_uv[off + 1 * uv_w]; |
| 413 const int b = best_uv[off + 2]; | 418 const int b = best_uv[off + 2 * uv_w]; |
| 414 dst_u[i] = ConvertRGBToU(r, g, b); | 419 dst_u[i] = ConvertRGBToU(r, g, b); |
| 415 dst_v[i] = ConvertRGBToV(r, g, b); | 420 dst_v[i] = ConvertRGBToV(r, g, b); |
| 416 } | 421 } |
| 417 best_uv += 3 * uv_w; | 422 best_uv += 3 * uv_w; |
| 418 dst_u += picture->uv_stride; | 423 dst_u += picture->uv_stride; |
| 419 dst_v += picture->uv_stride; | 424 dst_v += picture->uv_stride; |
| 420 } | 425 } |
| 421 return 1; | 426 return 1; |
| 422 } | 427 } |
| 423 | 428 |
| 424 //------------------------------------------------------------------------------ | 429 //------------------------------------------------------------------------------ |
| 425 // Main function | 430 // Main function |
| 426 | 431 |
| 427 #define SAFE_ALLOC(W, H, T) ((T*)WebPSafeMalloc((W) * (H), sizeof(T))) | 432 #define SAFE_ALLOC(W, H, T) ((T*)WebPSafeMalloc((W) * (H), sizeof(T))) |
| 428 | 433 |
| 429 static int PreprocessARGB(const uint8_t* r_ptr, | 434 static int PreprocessARGB(const uint8_t* r_ptr, |
| 430 const uint8_t* g_ptr, | 435 const uint8_t* g_ptr, |
| 431 const uint8_t* b_ptr, | 436 const uint8_t* b_ptr, |
| 432 int step, int rgb_stride, | 437 int step, int rgb_stride, |
| 433 WebPPicture* const picture) { | 438 WebPPicture* const picture) { |
| 434 // we expand the right/bottom border if needed | 439 // we expand the right/bottom border if needed |
| 435 const int w = (picture->width + 1) & ~1; | 440 const int w = (picture->width + 1) & ~1; |
| 436 const int h = (picture->height + 1) & ~1; | 441 const int h = (picture->height + 1) & ~1; |
| 437 const int uv_w = w >> 1; | 442 const int uv_w = w >> 1; |
| 438 const int uv_h = h >> 1; | 443 const int uv_h = h >> 1; |
| 439 int i, j, iter; | 444 uint64_t prev_diff_y_sum = ~0; |
| 445 int j, iter; |
| 440 | 446 |
| 441 // TODO(skal): allocate one big memory chunk. But for now, it's easier | 447 // TODO(skal): allocate one big memory chunk. But for now, it's easier |
| 442 // for valgrind debugging to have several chunks. | 448 // for valgrind debugging to have several chunks. |
| 443 fixed_y_t* const tmp_buffer = SAFE_ALLOC(w * 3, 2, fixed_y_t); // scratch | 449 fixed_y_t* const tmp_buffer = SAFE_ALLOC(w * 3, 2, fixed_y_t); // scratch |
| 444 fixed_y_t* const best_y_base = SAFE_ALLOC(w, h, fixed_y_t); | 450 fixed_y_t* const best_y_base = SAFE_ALLOC(w, h, fixed_y_t); |
| 445 fixed_y_t* const target_y_base = SAFE_ALLOC(w, h, fixed_y_t); | 451 fixed_y_t* const target_y_base = SAFE_ALLOC(w, h, fixed_y_t); |
| 446 fixed_y_t* const best_rgb_y = SAFE_ALLOC(w, 2, fixed_y_t); | 452 fixed_y_t* const best_rgb_y = SAFE_ALLOC(w, 2, fixed_y_t); |
| 447 fixed_t* const best_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t); | 453 fixed_t* const best_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t); |
| 448 fixed_t* const target_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t); | 454 fixed_t* const target_uv_base = SAFE_ALLOC(uv_w * 3, uv_h, fixed_t); |
| 449 fixed_t* const best_rgb_uv = SAFE_ALLOC(uv_w * 3, 1, fixed_t); | 455 fixed_t* const best_rgb_uv = SAFE_ALLOC(uv_w * 3, 1, fixed_t); |
| 450 fixed_y_t* best_y = best_y_base; | 456 fixed_y_t* best_y = best_y_base; |
| 451 fixed_y_t* target_y = target_y_base; | 457 fixed_y_t* target_y = target_y_base; |
| 452 fixed_t* best_uv = best_uv_base; | 458 fixed_t* best_uv = best_uv_base; |
| 453 fixed_t* target_uv = target_uv_base; | 459 fixed_t* target_uv = target_uv_base; |
| 460 const uint64_t diff_y_threshold = (uint64_t)(3.0 * w * h); |
| 454 int ok; | 461 int ok; |
| 455 int diff_sum = 0; | |
| 456 const int first_diff_threshold = (int)(2.5 * w * h); | |
| 457 const int min_improvement = 5; // stop if improvement is below this % | |
| 458 const int min_first_improvement = 80; | |
| 459 | 462 |
| 460 if (best_y_base == NULL || best_uv_base == NULL || | 463 if (best_y_base == NULL || best_uv_base == NULL || |
| 461 target_y_base == NULL || target_uv_base == NULL || | 464 target_y_base == NULL || target_uv_base == NULL || |
| 462 best_rgb_y == NULL || best_rgb_uv == NULL || | 465 best_rgb_y == NULL || best_rgb_uv == NULL || |
| 463 tmp_buffer == NULL) { | 466 tmp_buffer == NULL) { |
| 464 ok = WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY); | 467 ok = WebPEncodingSetError(picture, VP8_ENC_ERROR_OUT_OF_MEMORY); |
| 465 goto End; | 468 goto End; |
| 466 } | 469 } |
| 467 assert(picture->width >= kMinDimensionIterativeConversion); | 470 assert(picture->width >= kMinDimensionIterativeConversion); |
| 468 assert(picture->height >= kMinDimensionIterativeConversion); | 471 assert(picture->height >= kMinDimensionIterativeConversion); |
| 469 | 472 |
| 473 WebPInitConvertARGBToYUV(); |
| 474 |
| 470 // Import RGB samples to W/RGB representation. | 475 // Import RGB samples to W/RGB representation. |
| 471 for (j = 0; j < picture->height; j += 2) { | 476 for (j = 0; j < picture->height; j += 2) { |
| 472 const int is_last_row = (j == picture->height - 1); | 477 const int is_last_row = (j == picture->height - 1); |
| 473 fixed_y_t* const src1 = tmp_buffer; | 478 fixed_y_t* const src1 = tmp_buffer + 0 * w; |
| 474 fixed_y_t* const src2 = tmp_buffer + 3 * w; | 479 fixed_y_t* const src2 = tmp_buffer + 3 * w; |
| 475 | 480 |
| 476 // prepare two rows of input | 481 // prepare two rows of input |
| 477 ImportOneRow(r_ptr, g_ptr, b_ptr, step, picture->width, src1); | 482 ImportOneRow(r_ptr, g_ptr, b_ptr, step, picture->width, src1); |
| 478 if (!is_last_row) { | 483 if (!is_last_row) { |
| 479 ImportOneRow(r_ptr + rgb_stride, g_ptr + rgb_stride, b_ptr + rgb_stride, | 484 ImportOneRow(r_ptr + rgb_stride, g_ptr + rgb_stride, b_ptr + rgb_stride, |
| 480 step, picture->width, src2); | 485 step, picture->width, src2); |
| 481 } else { | 486 } else { |
| 482 memcpy(src2, src1, 3 * w * sizeof(*src2)); | 487 memcpy(src2, src1, 3 * w * sizeof(*src2)); |
| 483 } | 488 } |
| 489 StoreGray(src1, best_y + 0, w); |
| 490 StoreGray(src2, best_y + w, w); |
| 491 |
| 484 UpdateW(src1, target_y, w); | 492 UpdateW(src1, target_y, w); |
| 485 UpdateW(src2, target_y + w, w); | 493 UpdateW(src2, target_y + w, w); |
| 486 diff_sum += UpdateChroma(src1, src2, target_uv, best_y, uv_w); | 494 UpdateChroma(src1, src2, target_uv, uv_w); |
| 487 memcpy(best_uv, target_uv, 3 * uv_w * sizeof(*best_uv)); | 495 memcpy(best_uv, target_uv, 3 * uv_w * sizeof(*best_uv)); |
| 488 memcpy(best_y + w, best_y, w * sizeof(*best_y)); | |
| 489 best_y += 2 * w; | 496 best_y += 2 * w; |
| 490 best_uv += 3 * uv_w; | 497 best_uv += 3 * uv_w; |
| 491 target_y += 2 * w; | 498 target_y += 2 * w; |
| 492 target_uv += 3 * uv_w; | 499 target_uv += 3 * uv_w; |
| 493 r_ptr += 2 * rgb_stride; | 500 r_ptr += 2 * rgb_stride; |
| 494 g_ptr += 2 * rgb_stride; | 501 g_ptr += 2 * rgb_stride; |
| 495 b_ptr += 2 * rgb_stride; | 502 b_ptr += 2 * rgb_stride; |
| 496 } | 503 } |
| 497 | 504 |
| 498 // Iterate and resolve clipping conflicts. | 505 // Iterate and resolve clipping conflicts. |
| 499 for (iter = 0; iter < kNumIterations; ++iter) { | 506 for (iter = 0; iter < kNumIterations; ++iter) { |
| 500 int k; | |
| 501 const fixed_t* cur_uv = best_uv_base; | 507 const fixed_t* cur_uv = best_uv_base; |
| 502 const fixed_t* prev_uv = best_uv_base; | 508 const fixed_t* prev_uv = best_uv_base; |
| 503 const int old_diff_sum = diff_sum; | 509 uint64_t diff_y_sum = 0; |
| 504 diff_sum = 0; | |
| 505 | 510 |
| 506 best_y = best_y_base; | 511 best_y = best_y_base; |
| 507 best_uv = best_uv_base; | 512 best_uv = best_uv_base; |
| 508 target_y = target_y_base; | 513 target_y = target_y_base; |
| 509 target_uv = target_uv_base; | 514 target_uv = target_uv_base; |
| 510 for (j = 0; j < h; j += 2) { | 515 for (j = 0; j < h; j += 2) { |
| 511 fixed_y_t* const src1 = tmp_buffer; | 516 fixed_y_t* const src1 = tmp_buffer + 0 * w; |
| 512 fixed_y_t* const src2 = tmp_buffer + 3 * w; | 517 fixed_y_t* const src2 = tmp_buffer + 3 * w; |
| 513 { | 518 { |
| 514 const fixed_t* const next_uv = cur_uv + ((j < h - 2) ? 3 * uv_w : 0); | 519 const fixed_t* const next_uv = cur_uv + ((j < h - 2) ? 3 * uv_w : 0); |
| 515 InterpolateTwoRows(best_y, prev_uv, cur_uv, next_uv, w, src1, src2); | 520 InterpolateTwoRows(best_y, prev_uv, cur_uv, next_uv, w, src1, src2); |
| 516 prev_uv = cur_uv; | 521 prev_uv = cur_uv; |
| 517 cur_uv = next_uv; | 522 cur_uv = next_uv; |
| 518 } | 523 } |
| 519 | 524 |
| 520 UpdateW(src1, best_rgb_y + 0 * w, w); | 525 UpdateW(src1, best_rgb_y + 0 * w, w); |
| 521 UpdateW(src2, best_rgb_y + 1 * w, w); | 526 UpdateW(src2, best_rgb_y + 1 * w, w); |
| 522 diff_sum += UpdateChroma(src1, src2, best_rgb_uv, NULL, uv_w); | 527 UpdateChroma(src1, src2, best_rgb_uv, uv_w); |
| 523 | 528 |
| 524 // update two rows of Y and one row of RGB | 529 // update two rows of Y and one row of RGB |
| 525 for (i = 0; i < 2 * w; ++i) { | 530 diff_y_sum += WebPSharpYUVUpdateY(target_y, best_rgb_y, best_y, 2 * w); |
| 526 const int diff_y = target_y[i] - best_rgb_y[i]; | 531 WebPSharpYUVUpdateRGB(target_uv, best_rgb_uv, best_uv, 3 * uv_w); |
| 527 const int new_y = (int)best_y[i] + diff_y; | 532 |
| 528 best_y[i] = clip_y(new_y); | |
| 529 } | |
| 530 for (i = 0; i < uv_w; ++i) { | |
| 531 const int off = 3 * i; | |
| 532 int W; | |
| 533 for (k = 0; k <= 2; ++k) { | |
| 534 const int diff_uv = (int)target_uv[off + k] - best_rgb_uv[off + k]; | |
| 535 best_uv[off + k] += diff_uv; | |
| 536 } | |
| 537 W = RGBToGray(best_uv[off + 0], best_uv[off + 1], best_uv[off + 2]); | |
| 538 for (k = 0; k <= 2; ++k) { | |
| 539 best_uv[off + k] -= W; | |
| 540 } | |
| 541 } | |
| 542 best_y += 2 * w; | 533 best_y += 2 * w; |
| 543 best_uv += 3 * uv_w; | 534 best_uv += 3 * uv_w; |
| 544 target_y += 2 * w; | 535 target_y += 2 * w; |
| 545 target_uv += 3 * uv_w; | 536 target_uv += 3 * uv_w; |
| 546 } | 537 } |
| 547 // test exit condition | 538 // test exit condition |
| 548 if (diff_sum > 0) { | 539 if (iter > 0) { |
| 549 const int improvement = 100 * abs(diff_sum - old_diff_sum) / diff_sum; | 540 if (diff_y_sum < diff_y_threshold) break; |
| 550 // Check if first iteration gave good result already, without a large | 541 if (diff_y_sum > prev_diff_y_sum) break; |
| 551 // jump of improvement (otherwise it means we need to try few extra | |
| 552 // iterations, just to be sure). | |
| 553 if (iter == 0 && diff_sum < first_diff_threshold && | |
| 554 improvement < min_first_improvement) { | |
| 555 break; | |
| 556 } | |
| 557 // then, check if improvement is stalling. | |
| 558 if (improvement < min_improvement) { | |
| 559 break; | |
| 560 } | |
| 561 } else { | |
| 562 break; | |
| 563 } | 542 } |
| 543 prev_diff_y_sum = diff_y_sum; |
| 564 } | 544 } |
| 565 | |
| 566 // final reconstruction | 545 // final reconstruction |
| 567 ok = ConvertWRGBToYUV(best_y_base, best_uv_base, picture); | 546 ok = ConvertWRGBToYUV(best_y_base, best_uv_base, picture); |
| 568 | 547 |
| 569 End: | 548 End: |
| 570 WebPSafeFree(best_y_base); | 549 WebPSafeFree(best_y_base); |
| 571 WebPSafeFree(best_uv_base); | 550 WebPSafeFree(best_uv_base); |
| 572 WebPSafeFree(target_y_base); | 551 WebPSafeFree(target_y_base); |
| 573 WebPSafeFree(target_uv_base); | 552 WebPSafeFree(target_uv_base); |
| 574 WebPSafeFree(best_rgb_y); | 553 WebPSafeFree(best_rgb_y); |
| 575 WebPSafeFree(best_rgb_uv); | 554 WebPSafeFree(best_rgb_uv); |
| (...skipping 449 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 1025 | 1004 |
| 1026 int WebPPictureARGBToYUVADithered(WebPPicture* picture, WebPEncCSP colorspace, | 1005 int WebPPictureARGBToYUVADithered(WebPPicture* picture, WebPEncCSP colorspace, |
| 1027 float dithering) { | 1006 float dithering) { |
| 1028 return PictureARGBToYUVA(picture, colorspace, dithering, 0); | 1007 return PictureARGBToYUVA(picture, colorspace, dithering, 0); |
| 1029 } | 1008 } |
| 1030 | 1009 |
| 1031 int WebPPictureARGBToYUVA(WebPPicture* picture, WebPEncCSP colorspace) { | 1010 int WebPPictureARGBToYUVA(WebPPicture* picture, WebPEncCSP colorspace) { |
| 1032 return PictureARGBToYUVA(picture, colorspace, 0.f, 0); | 1011 return PictureARGBToYUVA(picture, colorspace, 0.f, 0); |
| 1033 } | 1012 } |
| 1034 | 1013 |
| 1014 int WebPPictureSharpARGBToYUVA(WebPPicture* picture) { |
| 1015 return PictureARGBToYUVA(picture, WEBP_YUV420, 0.f, 1); |
| 1016 } |
| 1017 // for backward compatibility |
| 1035 int WebPPictureSmartARGBToYUVA(WebPPicture* picture) { | 1018 int WebPPictureSmartARGBToYUVA(WebPPicture* picture) { |
| 1036 return PictureARGBToYUVA(picture, WEBP_YUV420, 0.f, 1); | 1019 return WebPPictureSharpARGBToYUVA(picture); |
| 1037 } | 1020 } |
| 1038 | 1021 |
| 1039 //------------------------------------------------------------------------------ | 1022 //------------------------------------------------------------------------------ |
| 1040 // call for YUVA -> ARGB conversion | 1023 // call for YUVA -> ARGB conversion |
| 1041 | 1024 |
| 1042 int WebPPictureYUVAToARGB(WebPPicture* picture) { | 1025 int WebPPictureYUVAToARGB(WebPPicture* picture) { |
| 1043 if (picture == NULL) return 0; | 1026 if (picture == NULL) return 0; |
| 1044 if (picture->y == NULL || picture->u == NULL || picture->v == NULL) { | 1027 if (picture->y == NULL || picture->u == NULL || picture->v == NULL) { |
| 1045 return WebPEncodingSetError(picture, VP8_ENC_ERROR_NULL_PARAMETER); | 1028 return WebPEncodingSetError(picture, VP8_ENC_ERROR_NULL_PARAMETER); |
| 1046 } | 1029 } |
| (...skipping 136 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 1183 } | 1166 } |
| 1184 | 1167 |
| 1185 int WebPPictureImportBGRX(WebPPicture* picture, | 1168 int WebPPictureImportBGRX(WebPPicture* picture, |
| 1186 const uint8_t* rgba, int rgba_stride) { | 1169 const uint8_t* rgba, int rgba_stride) { |
| 1187 return (picture != NULL && rgba != NULL) | 1170 return (picture != NULL && rgba != NULL) |
| 1188 ? Import(picture, rgba, rgba_stride, 4, 1, 0) | 1171 ? Import(picture, rgba, rgba_stride, 4, 1, 0) |
| 1189 : 0; | 1172 : 0; |
| 1190 } | 1173 } |
| 1191 | 1174 |
| 1192 //------------------------------------------------------------------------------ | 1175 //------------------------------------------------------------------------------ |
| OLD | NEW |