libswscale/yuv2rgb.c

   1 /*
   2  * software YUV to RGB converter
   3  *
   4  * Copyright (C) 2009 Konstantin Shishkov
   5  *
   6  * 1,4,8bpp support and context / deglobalize stuff
   7  * by Michael Niedermayer (michaelni@gmx.at)
   8  *
   9  * This file is part of FFmpeg.
  10  *
  11  * FFmpeg is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2.1 of the License, or (at your option) any later version.
  15  *
  16  * FFmpeg is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with FFmpeg; if not, write to the Free Software
  23  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  24  */
  25
  26 #include <stdio.h>
  27 #include <stdlib.h>
  28 #include <inttypes.h>
  29 #include <assert.h>
  30
  31 #include "config.h"
  32 #include "rgb2rgb.h"
  33 #include "swscale.h"
  34 #include "swscale_internal.h"
  35 #include "libavutil/x86_cpu.h"
  36
  37 extern const uint8_t dither_8x8_32[8][8];
  38 extern const uint8_t dither_8x8_73[8][8];
  39 extern const uint8_t dither_8x8_220[8][8];
  40
  41 const int32_t ff_yuv2rgb_coeffs[8][4] = {
  42     {117504, 138453, 13954, 34903}, /* no sequence_display_extension */
  43     {117504, 138453, 13954, 34903}, /* ITU-R Rec. 709 (1990) */
  44     {104597, 132201, 25675, 53279}, /* unspecified */
  45     {104597, 132201, 25675, 53279}, /* reserved */
  46     {104448, 132798, 24759, 53109}, /* FCC */
  47     {104597, 132201, 25675, 53279}, /* ITU-R Rec. 624-4 System B, G */
  48     {104597, 132201, 25675, 53279}, /* SMPTE 170M */
  49     {117579, 136230, 16907, 35559}  /* SMPTE 240M (1987) */
  50 };
  51
  52 #define LOADCHROMA(i)                               \
  53     U = pu[i];                                      \
  54     V = pv[i];                                      \
  55     r = (void *)c->table_rV[V];                     \
  56     g = (void *)(c->table_gU[U] + c->table_gV[V]);  \
  57     b = (void *)c->table_bU[U];
  58
  59 #define PUTRGB(dst,src,i,o)          \
  60     Y = src[2*i+o];                  \
  61     dst[2*i  ] = r[Y] + g[Y] + b[Y]; \
  62     Y = src[2*i+1-o];                \
  63     dst[2*i+1] = r[Y] + g[Y] + b[Y];
  64
  65 #define PUTRGB24(dst,src,i)                                  \
  66     Y = src[2*i];                                            \
  67     dst[6*i+0] = r[Y]; dst[6*i+1] = g[Y]; dst[6*i+2] = b[Y]; \
  68     Y = src[2*i+1];                                          \
  69     dst[6*i+3] = r[Y]; dst[6*i+4] = g[Y]; dst[6*i+5] = b[Y];
  70
  71 #define PUTBGR24(dst,src,i)                                  \
  72     Y = src[2*i];                                            \
  73     dst[6*i+0] = b[Y]; dst[6*i+1] = g[Y]; dst[6*i+2] = r[Y]; \
  74     Y = src[2*i+1];                                          \
  75     dst[6*i+3] = b[Y]; dst[6*i+4] = g[Y]; dst[6*i+5] = r[Y];
  76
  77 #define PUTRGBA(dst,ysrc,asrc,i,o,s)                    \
  78     Y = ysrc[2*i+o];                                    \
  79     dst[2*i  ] = r[Y] + g[Y] + b[Y] + (asrc[2*i  ]<<s); \
  80     Y = ysrc[2*i+1-o];                                  \
  81     dst[2*i+1] = r[Y] + g[Y] + b[Y] + (asrc[2*i+1]<<s);
  82
  83 #define YUV2RGBFUNC(func_name, dst_type, alpha) \
  84 static int func_name(SwsContext *c, uint8_t* src[], int srcStride[], int srcSliceY, \
  85                      int srcSliceH, uint8_t* dst[], int dstStride[]){\
  86     int y;\
  87 \
  88     if (!alpha && c->srcFormat == PIX_FMT_YUV422P) {\
  89         srcStride[1] *= 2;\
  90         srcStride[2] *= 2;\
  91     }\
  92     for (y=0; y<srcSliceH; y+=2) {\
  93         dst_type *dst_1 = (dst_type*)(dst[0] + (y+srcSliceY  )*dstStride[0]);\
  94         dst_type *dst_2 = (dst_type*)(dst[0] + (y+srcSliceY+1)*dstStride[0]);\
  95         dst_type av_unused *r, *b;\
  96         dst_type *g;\
  97         uint8_t *py_1 = src[0] + y*srcStride[0];\
  98         uint8_t *py_2 = py_1 + srcStride[0];\
  99         uint8_t *pu = src[1] + (y>>1)*srcStride[1];\
 100         uint8_t *pv = src[2] + (y>>1)*srcStride[2];\
 101         uint8_t av_unused *pa_1, *pa_2;\
 102         unsigned int h_size = c->dstW>>3;\
 103         if (alpha){\
 104             pa_1 = src[3] + y*srcStride[3];\
 105             pa_2 = pa_1 + srcStride[3];\
 106         }\
 107         while (h_size--) {\
 108             int av_unused U, V;\
 109             int Y;\
 110
 111 #define ENDYUV2RGBLINE(dst_delta)\
 112             pu += 4;\
 113             pv += 4;\
 114             py_1 += 8;\
 115             py_2 += 8;\
 116             dst_1 += dst_delta;\
 117             dst_2 += dst_delta;\
 118         }\
 119         if (c->dstW & 4) {\
 120             int av_unused Y, U, V;\
 121
 122 #define ENDYUV2RGBFUNC()\
 123         }\
 124     }\
 125     return srcSliceH;\
 126 }
 127
 128 #define CLOSEYUV2RGBFUNC(dst_delta)\
 129     ENDYUV2RGBLINE(dst_delta)\
 130     ENDYUV2RGBFUNC()
 131
 132 YUV2RGBFUNC(yuv2rgb_c_32, uint32_t, 0)
 133     LOADCHROMA(0);
 134     PUTRGB(dst_1,py_1,0,0);
 135     PUTRGB(dst_2,py_2,0,1);
 136
 137     LOADCHROMA(1);
 138     PUTRGB(dst_2,py_2,1,1);
 139     PUTRGB(dst_1,py_1,1,0);
 140     LOADCHROMA(1);
 141     PUTRGB(dst_2,py_2,1,1);
 142     PUTRGB(dst_1,py_1,1,0);
 143
 144     LOADCHROMA(2);
 145     PUTRGB(dst_1,py_1,2,0);
 146     PUTRGB(dst_2,py_2,2,1);
 147
 148     LOADCHROMA(3);
 149     PUTRGB(dst_2,py_2,3,1);
 150     PUTRGB(dst_1,py_1,3,0);
 151 ENDYUV2RGBLINE(8)
 152     LOADCHROMA(0);
 153     PUTRGB(dst_1,py_1,0,0);
 154     PUTRGB(dst_2,py_2,0,1);
 155
 156     LOADCHROMA(1);
 157     PUTRGB(dst_2,py_2,1,1);
 158     PUTRGB(dst_1,py_1,1,0);
 159 ENDYUV2RGBFUNC()
 160
 161 YUV2RGBFUNC(yuva2rgba_c, uint32_t, 1)
 162     LOADCHROMA(0);
 163     PUTRGBA(dst_1,py_1,pa_1,0,0,24);
 164     PUTRGBA(dst_2,py_2,pa_2,0,1,24);
 165
 166     LOADCHROMA(1);
 167     PUTRGBA(dst_2,py_2,pa_1,1,1,24);
 168     PUTRGBA(dst_1,py_1,pa_2,1,0,24);
 169     LOADCHROMA(1);
 170     PUTRGBA(dst_2,py_2,pa_1,1,1,24);
 171     PUTRGBA(dst_1,py_1,pa_2,1,0,24);
 172
 173     LOADCHROMA(2);
 174     PUTRGBA(dst_1,py_1,pa_1,2,0,24);
 175     PUTRGBA(dst_2,py_2,pa_2,2,1,24);
 176
 177     LOADCHROMA(3);
 178     PUTRGBA(dst_2,py_2,pa_1,3,1,24);
 179     PUTRGBA(dst_1,py_1,pa_2,3,0,24);
 180     pa_1 += 8;\
 181     pa_2 += 8;\
 182 ENDYUV2RGBLINE(8)
 183     LOADCHROMA(0);
 184     PUTRGBA(dst_1,py_1,pa_1,0,0,24);
 185     PUTRGBA(dst_2,py_2,pa_2,0,1,24);
 186
 187     LOADCHROMA(1);
 188     PUTRGBA(dst_2,py_2,pa_1,1,1,24);
 189     PUTRGBA(dst_1,py_1,pa_2,1,0,24);
 190 ENDYUV2RGBFUNC()
 191
 192 YUV2RGBFUNC(yuva2argb_c, uint32_t, 1)
 193     LOADCHROMA(0);
 194     PUTRGBA(dst_1,py_1,pa_1,0,0,0);
 195     PUTRGBA(dst_2,py_2,pa_2,0,1,0);
 196
 197     LOADCHROMA(1);
 198     PUTRGBA(dst_2,py_2,pa_2,1,1,0);
 199     PUTRGBA(dst_1,py_1,pa_1,1,0,0);
 200     LOADCHROMA(1);
 201     PUTRGBA(dst_2,py_2,pa_2,1,1,0);
 202     PUTRGBA(dst_1,py_1,pa_1,1,0,0);
 203
 204     LOADCHROMA(2);
 205     PUTRGBA(dst_1,py_1,pa_1,2,0,0);
 206     PUTRGBA(dst_2,py_2,pa_2,2,1,0);
 207
 208     LOADCHROMA(3);
 209     PUTRGBA(dst_2,py_2,pa_2,3,1,0);
 210     PUTRGBA(dst_1,py_1,pa_1,3,0,0);
 211     pa_1 += 8;\
 212     pa_2 += 8;\
 213 ENDYUV2RGBLINE(8)
 214     LOADCHROMA(0);
 215     PUTRGBA(dst_1,py_1,pa_1,0,0,0);
 216     PUTRGBA(dst_2,py_2,pa_2,0,1,0);
 217
 218     LOADCHROMA(1);
 219     PUTRGBA(dst_2,py_2,pa_2,1,1,0);
 220     PUTRGBA(dst_1,py_1,pa_1,1,0,0);
 221 ENDYUV2RGBFUNC()
 222
 223 YUV2RGBFUNC(yuv2rgb_c_24_rgb, uint8_t, 0)
 224     LOADCHROMA(0);
 225     PUTRGB24(dst_1,py_1,0);
 226     PUTRGB24(dst_2,py_2,0);
 227
 228     LOADCHROMA(1);
 229     PUTRGB24(dst_2,py_2,1);
 230     PUTRGB24(dst_1,py_1,1);
 231
 232     LOADCHROMA(2);
 233     PUTRGB24(dst_1,py_1,2);
 234     PUTRGB24(dst_2,py_2,2);
 235
 236     LOADCHROMA(3);
 237     PUTRGB24(dst_2,py_2,3);
 238     PUTRGB24(dst_1,py_1,3);
 239 ENDYUV2RGBLINE(24)
 240     LOADCHROMA(0);
 241     PUTRGB24(dst_1,py_1,0);
 242     PUTRGB24(dst_2,py_2,0);
 243
 244     LOADCHROMA(1);
 245     PUTRGB24(dst_2,py_2,1);
 246     PUTRGB24(dst_1,py_1,1);
 247 ENDYUV2RGBFUNC()
 248
 249 // only trivial mods from yuv2rgb_c_24_rgb
 250 YUV2RGBFUNC(yuv2rgb_c_24_bgr, uint8_t, 0)
 251     LOADCHROMA(0);
 252     PUTBGR24(dst_1,py_1,0);
 253     PUTBGR24(dst_2,py_2,0);
 254
 255     LOADCHROMA(1);
 256     PUTBGR24(dst_2,py_2,1);
 257     PUTBGR24(dst_1,py_1,1);
 258
 259     LOADCHROMA(2);
 260     PUTBGR24(dst_1,py_1,2);
 261     PUTBGR24(dst_2,py_2,2);
 262
 263     LOADCHROMA(3);
 264     PUTBGR24(dst_2,py_2,3);
 265     PUTBGR24(dst_1,py_1,3);
 266 ENDYUV2RGBLINE(24)
 267     LOADCHROMA(0);
 268     PUTBGR24(dst_1,py_1,0);
 269     PUTBGR24(dst_2,py_2,0);
 270
 271     LOADCHROMA(1);
 272     PUTBGR24(dst_2,py_2,1);
 273     PUTBGR24(dst_1,py_1,1);
 274 ENDYUV2RGBFUNC()
 275
 276 // This is exactly the same code as yuv2rgb_c_32 except for the types of
 277 // r, g, b, dst_1, dst_2
 278 YUV2RGBFUNC(yuv2rgb_c_16, uint16_t, 0)
 279     LOADCHROMA(0);
 280     PUTRGB(dst_1,py_1,0,0);
 281     PUTRGB(dst_2,py_2,0,1);
 282
 283     LOADCHROMA(1);
 284     PUTRGB(dst_2,py_2,1,1);
 285     PUTRGB(dst_1,py_1,1,0);
 286
 287     LOADCHROMA(2);
 288     PUTRGB(dst_1,py_1,2,0);
 289     PUTRGB(dst_2,py_2,2,1);
 290
 291     LOADCHROMA(3);
 292     PUTRGB(dst_2,py_2,3,1);
 293     PUTRGB(dst_1,py_1,3,0);
 294 CLOSEYUV2RGBFUNC(8)
 295
 296 // This is exactly the same code as yuv2rgb_c_32 except for the types of
 297 // r, g, b, dst_1, dst_2
 298 YUV2RGBFUNC(yuv2rgb_c_8, uint8_t, 0)
 299     LOADCHROMA(0);
 300     PUTRGB(dst_1,py_1,0,0);
 301     PUTRGB(dst_2,py_2,0,1);
 302
 303     LOADCHROMA(1);
 304     PUTRGB(dst_2,py_2,1,1);
 305     PUTRGB(dst_1,py_1,1,0);
 306
 307     LOADCHROMA(2);
 308     PUTRGB(dst_1,py_1,2,0);
 309     PUTRGB(dst_2,py_2,2,1);
 310
 311     LOADCHROMA(3);
 312     PUTRGB(dst_2,py_2,3,1);
 313     PUTRGB(dst_1,py_1,3,0);
 314 CLOSEYUV2RGBFUNC(8)
 315
 316 // r, g, b, dst_1, dst_2
 317 YUV2RGBFUNC(yuv2rgb_c_8_ordered_dither, uint8_t, 0)
 318     const uint8_t *d32 = dither_8x8_32[y&7];
 319     const uint8_t *d64 = dither_8x8_73[y&7];
 320 #define PUTRGB8(dst,src,i,o)                                    \
 321     Y = src[2*i];                                               \
 322     dst[2*i]   = r[Y+d32[0+o]] + g[Y+d32[0+o]] + b[Y+d64[0+o]]; \
 323     Y = src[2*i+1];                                             \
 324     dst[2*i+1] = r[Y+d32[1+o]] + g[Y+d32[1+o]] + b[Y+d64[1+o]];
 325
 326     LOADCHROMA(0);
 327     PUTRGB8(dst_1,py_1,0,0);
 328     PUTRGB8(dst_2,py_2,0,0+8);
 329
 330     LOADCHROMA(1);
 331     PUTRGB8(dst_2,py_2,1,2+8);
 332     PUTRGB8(dst_1,py_1,1,2);
 333
 334     LOADCHROMA(2);
 335     PUTRGB8(dst_1,py_1,2,4);
 336     PUTRGB8(dst_2,py_2,2,4+8);
 337
 338     LOADCHROMA(3);
 339     PUTRGB8(dst_2,py_2,3,6+8);
 340     PUTRGB8(dst_1,py_1,3,6);
 341 CLOSEYUV2RGBFUNC(8)
 342
 343
 344 // This is exactly the same code as yuv2rgb_c_32 except for the types of
 345 // r, g, b, dst_1, dst_2
 346 YUV2RGBFUNC(yuv2rgb_c_4, uint8_t, 0)
 347     int acc;
 348 #define PUTRGB4(dst,src,i)          \
 349     Y = src[2*i];                   \
 350     acc = r[Y] + g[Y] + b[Y];       \
 351     Y = src[2*i+1];                 \
 352     acc |= (r[Y] + g[Y] + b[Y])<<4; \
 353     dst[i] = acc;
 354
 355     LOADCHROMA(0);
 356     PUTRGB4(dst_1,py_1,0);
 357     PUTRGB4(dst_2,py_2,0);
 358
 359     LOADCHROMA(1);
 360     PUTRGB4(dst_2,py_2,1);
 361     PUTRGB4(dst_1,py_1,1);
 362
 363     LOADCHROMA(2);
 364     PUTRGB4(dst_1,py_1,2);
 365     PUTRGB4(dst_2,py_2,2);
 366
 367     LOADCHROMA(3);
 368     PUTRGB4(dst_2,py_2,3);
 369     PUTRGB4(dst_1,py_1,3);
 370 CLOSEYUV2RGBFUNC(4)
 371
 372 YUV2RGBFUNC(yuv2rgb_c_4_ordered_dither, uint8_t, 0)
 373     const uint8_t *d64 =  dither_8x8_73[y&7];
 374     const uint8_t *d128 = dither_8x8_220[y&7];
 375     int acc;
 376
 377 #define PUTRGB4D(dst,src,i,o)                                     \
 378     Y = src[2*i];                                                 \
 379     acc = r[Y+d128[0+o]] + g[Y+d64[0+o]] + b[Y+d128[0+o]];        \
 380     Y = src[2*i+1];                                               \
 381     acc |= (r[Y+d128[1+o]] + g[Y+d64[1+o]] + b[Y+d128[1+o]])<<4;  \
 382     dst[i]= acc;
 383
 384     LOADCHROMA(0);
 385     PUTRGB4D(dst_1,py_1,0,0);
 386     PUTRGB4D(dst_2,py_2,0,0+8);
 387
 388     LOADCHROMA(1);
 389     PUTRGB4D(dst_2,py_2,1,2+8);
 390     PUTRGB4D(dst_1,py_1,1,2);
 391
 392     LOADCHROMA(2);
 393     PUTRGB4D(dst_1,py_1,2,4);
 394     PUTRGB4D(dst_2,py_2,2,4+8);
 395
 396     LOADCHROMA(3);
 397     PUTRGB4D(dst_2,py_2,3,6+8);
 398     PUTRGB4D(dst_1,py_1,3,6);
 399 CLOSEYUV2RGBFUNC(4)
 400
 401 // This is exactly the same code as yuv2rgb_c_32 except for the types of
 402 // r, g, b, dst_1, dst_2
 403 YUV2RGBFUNC(yuv2rgb_c_4b, uint8_t, 0)
 404     LOADCHROMA(0);
 405     PUTRGB(dst_1,py_1,0,0);
 406     PUTRGB(dst_2,py_2,0,1);
 407
 408     LOADCHROMA(1);
 409     PUTRGB(dst_2,py_2,1,1);
 410     PUTRGB(dst_1,py_1,1,0);
 411
 412     LOADCHROMA(2);
 413     PUTRGB(dst_1,py_1,2,0);
 414     PUTRGB(dst_2,py_2,2,1);
 415
 416     LOADCHROMA(3);
 417     PUTRGB(dst_2,py_2,3,1);
 418     PUTRGB(dst_1,py_1,3,0);
 419 CLOSEYUV2RGBFUNC(8)
 420
 421 YUV2RGBFUNC(yuv2rgb_c_4b_ordered_dither, uint8_t, 0)
 422     const uint8_t *d64 =  dither_8x8_73[y&7];
 423     const uint8_t *d128 = dither_8x8_220[y&7];
 424
 425 #define PUTRGB4DB(dst,src,i,o)                                    \
 426     Y = src[2*i];                                                 \
 427     dst[2*i]   = r[Y+d128[0+o]] + g[Y+d64[0+o]] + b[Y+d128[0+o]]; \
 428     Y = src[2*i+1];                                               \
 429     dst[2*i+1] = r[Y+d128[1+o]] + g[Y+d64[1+o]] + b[Y+d128[1+o]];
 430
 431     LOADCHROMA(0);
 432     PUTRGB4DB(dst_1,py_1,0,0);
 433     PUTRGB4DB(dst_2,py_2,0,0+8);
 434
 435     LOADCHROMA(1);
 436     PUTRGB4DB(dst_2,py_2,1,2+8);
 437     PUTRGB4DB(dst_1,py_1,1,2);
 438
 439     LOADCHROMA(2);
 440     PUTRGB4DB(dst_1,py_1,2,4);
 441     PUTRGB4DB(dst_2,py_2,2,4+8);
 442
 443     LOADCHROMA(3);
 444     PUTRGB4DB(dst_2,py_2,3,6+8);
 445     PUTRGB4DB(dst_1,py_1,3,6);
 446 CLOSEYUV2RGBFUNC(8)
 447
 448 YUV2RGBFUNC(yuv2rgb_c_1_ordered_dither, uint8_t, 0)
 449         const uint8_t *d128 = dither_8x8_220[y&7];
 450         char out_1 = 0, out_2 = 0;
 451         g= c->table_gU[128] + c->table_gV[128];
 452
 453 #define PUTRGB1(out,src,i,o)    \
 454     Y = src[2*i];               \
 455     out+= out + g[Y+d128[0+o]]; \
 456     Y = src[2*i+1];             \
 457     out+= out + g[Y+d128[1+o]];
 458
 459     PUTRGB1(out_1,py_1,0,0);
 460     PUTRGB1(out_2,py_2,0,0+8);
 461
 462     PUTRGB1(out_2,py_2,1,2+8);
 463     PUTRGB1(out_1,py_1,1,2);
 464
 465     PUTRGB1(out_1,py_1,2,4);
 466     PUTRGB1(out_2,py_2,2,4+8);
 467
 468     PUTRGB1(out_2,py_2,3,6+8);
 469     PUTRGB1(out_1,py_1,3,6);
 470
 471     dst_1[0]= out_1;
 472     dst_2[0]= out_2;
 473 CLOSEYUV2RGBFUNC(1)
 474
 475 SwsFunc ff_yuv2rgb_get_func_ptr(SwsContext *c)
 476 {
 477     SwsFunc t = NULL;
 478 #if (HAVE_MMX2 || HAVE_MMX) && CONFIG_GPL
 479      t = ff_yuv2rgb_init_mmx(c);
 480 #endif
 481 #if HAVE_VIS
 482     t = ff_yuv2rgb_init_vis(c);
 483 #endif
 484 #if CONFIG_MLIB
 485     t = ff_yuv2rgb_init_mlib(c);
 486 #endif
 487 #if HAVE_ALTIVEC && CONFIG_GPL
 488     if (c->flags & SWS_CPU_CAPS_ALTIVEC)
 489         t = ff_yuv2rgb_init_altivec(c);
 490 #endif
 491
 492 #if ARCH_BFIN
 493     if (c->flags & SWS_CPU_CAPS_BFIN)
 494         t = ff_yuv2rgb_get_func_ptr_bfin(c);
 495 #endif
 496
 497     if (t)
 498         return t;
 499
 500     av_log(c, AV_LOG_WARNING, "No accelerated colorspace conversion found.\n");
 501
 502     switch (c->dstFormat) {
 503     case PIX_FMT_ARGB:
 504     case PIX_FMT_ABGR:       if (CONFIG_SWSCALE_ALPHA && c->srcFormat == PIX_FMT_YUVA420P) return yuva2argb_c;
 505     case PIX_FMT_RGBA:
 506     case PIX_FMT_BGRA:       return (CONFIG_SWSCALE_ALPHA && c->srcFormat == PIX_FMT_YUVA420P) ? yuva2rgba_c : yuv2rgb_c_32;
 507     case PIX_FMT_RGB24:      return yuv2rgb_c_24_rgb;
 508     case PIX_FMT_BGR24:      return yuv2rgb_c_24_bgr;
 509     case PIX_FMT_RGB565:
 510     case PIX_FMT_BGR565:
 511     case PIX_FMT_RGB555:
 512     case PIX_FMT_BGR555:     return yuv2rgb_c_16;
 513     case PIX_FMT_RGB8:
 514     case PIX_FMT_BGR8:       return yuv2rgb_c_8_ordered_dither;
 515     case PIX_FMT_RGB4:
 516     case PIX_FMT_BGR4:       return yuv2rgb_c_4_ordered_dither;
 517     case PIX_FMT_RGB4_BYTE:
 518     case PIX_FMT_BGR4_BYTE:  return yuv2rgb_c_4b_ordered_dither;
 519     case PIX_FMT_MONOBLACK:  return yuv2rgb_c_1_ordered_dither;
 520     default:
 521         assert(0);
 522     }
 523     return NULL;
 524 }
 525
 526 static void fill_table(uint8_t* table[256], const int elemsize, const int inc, uint8_t *y_table)
 527 {
 528     int i;
 529     int64_t cb = 0;
 530
 531     y_table -= elemsize * (inc >> 9);
 532
 533     for (i = 0; i < 256; i++) {
 534         table[i] = y_table + elemsize * (cb >> 16);
 535         cb += inc;
 536     }
 537 }
 538
 539 static void fill_gv_table(int table[256], const int elemsize, const int inc)
 540 {
 541     int i;
 542     int64_t cb = 0;
 543     int off = -(inc >> 9);
 544
 545     for (i = 0; i < 256; i++) {
 546         table[i] = elemsize * (off + (cb >> 16));
 547         cb += inc;
 548     }
 549 }
 550
 551 av_cold int ff_yuv2rgb_c_init_tables(SwsContext *c, const int inv_table[4], int fullRange,
 552                                      int brightness, int contrast, int saturation)
 553 {
 554     const int isRgb =      c->dstFormat==PIX_FMT_RGB32
 555                         || c->dstFormat==PIX_FMT_RGB32_1
 556                         || c->dstFormat==PIX_FMT_BGR24
 557                         || c->dstFormat==PIX_FMT_RGB565
 558                         || c->dstFormat==PIX_FMT_RGB555
 559                         || c->dstFormat==PIX_FMT_RGB8
 560                         || c->dstFormat==PIX_FMT_RGB4
 561                         || c->dstFormat==PIX_FMT_RGB4_BYTE
 562                         || c->dstFormat==PIX_FMT_MONOBLACK;
 563     const int bpp = fmt_depth(c->dstFormat);
 564     uint8_t *y_table;
 565     uint16_t *y_table16;
 566     uint32_t *y_table32;
 567     int i, base, rbase, gbase, bbase, abase, needAlpha;
 568     const int yoffs = fullRange ? 384 : 326;
 569
 570     int64_t crv =  inv_table[0];
 571     int64_t cbu =  inv_table[1];
 572     int64_t cgu = -inv_table[2];
 573     int64_t cgv = -inv_table[3];
 574     int64_t cy  = 1<<16;
 575     int64_t oy  = 0;
 576
 577     int64_t yb = 0;
 578
 579     if (!fullRange) {
 580         cy = (cy*255) / 219;
 581         oy = 16<<16;
 582     } else {
 583         crv = (crv*224) / 255;
 584         cbu = (cbu*224) / 255;
 585         cgu = (cgu*224) / 255;
 586         cgv = (cgv*224) / 255;
 587     }
 588
 589     cy  = (cy *contrast             ) >> 16;
 590     crv = (crv*contrast * saturation) >> 32;
 591     cbu = (cbu*contrast * saturation) >> 32;
 592     cgu = (cgu*contrast * saturation) >> 32;
 593     cgv = (cgv*contrast * saturation) >> 32;
 594     oy -= 256*brightness;
 595
 596     //scale coefficients by cy
 597     crv = ((crv << 16) + 0x8000) / cy;
 598     cbu = ((cbu << 16) + 0x8000) / cy;
 599     cgu = ((cgu << 16) + 0x8000) / cy;
 600     cgv = ((cgv << 16) + 0x8000) / cy;
 601
 602     av_free(c->yuvTable);
 603
 604     switch (bpp) {
 605     case 1:
 606         c->yuvTable = av_malloc(1024);
 607         y_table = c->yuvTable;
 608         yb = -(384<<16) - oy;
 609         for (i = 0; i < 1024-110; i++) {
 610             y_table[i+110] = av_clip_uint8((yb + 0x8000) >> 16) >> 7;
 611             yb += cy;
 612         }
 613         fill_table(c->table_gU, 1, cgu, y_table + yoffs);
 614         fill_gv_table(c->table_gV, 1, cgv);
 615         break;
 616     case 4:
 617     case 4|128:
 618         rbase = isRgb ? 3 : 0;
 619         gbase = 1;
 620         bbase = isRgb ? 0 : 3;
 621         c->yuvTable = av_malloc(1024*3);
 622         y_table = c->yuvTable;
 623         yb = -(384<<16) - oy;
 624         for (i = 0; i < 1024-110; i++) {
 625             int yval = av_clip_uint8((yb + 0x8000) >> 16);
 626             y_table[i+110     ] =  (yval >> 7)       << rbase;
 627             y_table[i+ 37+1024] = ((yval + 43) / 85) << gbase;
 628             y_table[i+110+2048] =  (yval >> 7)       << bbase;
 629             yb += cy;
 630         }
 631         fill_table(c->table_rV, 1, crv, y_table + yoffs);
 632         fill_table(c->table_gU, 1, cgu, y_table + yoffs + 1024);
 633         fill_table(c->table_bU, 1, cbu, y_table + yoffs + 2048);
 634         fill_gv_table(c->table_gV, 1, cgv);
 635         break;
 636     case 8:
 637         rbase = isRgb ? 5 : 0;
 638         gbase = isRgb ? 2 : 3;
 639         bbase = isRgb ? 0 : 6;
 640         c->yuvTable = av_malloc(1024*3);
 641         y_table = c->yuvTable;
 642         yb = -(384<<16) - oy;
 643         for (i = 0; i < 1024-38; i++) {
 644             int yval = av_clip_uint8((yb + 0x8000) >> 16);
 645             y_table[i+16     ] = ((yval + 18) / 36) << rbase;
 646             y_table[i+16+1024] = ((yval + 18) / 36) << gbase;
 647             y_table[i+37+2048] = ((yval + 43) / 85) << bbase;
 648             yb += cy;
 649         }
 650         fill_table(c->table_rV, 1, crv, y_table + yoffs);
 651         fill_table(c->table_gU, 1, cgu, y_table + yoffs + 1024);
 652         fill_table(c->table_bU, 1, cbu, y_table + yoffs + 2048);
 653         fill_gv_table(c->table_gV, 1, cgv);
 654         break;
 655     case 15:
 656     case 16:
 657         rbase = isRgb ? bpp - 5 : 0;
 658         gbase = 5;
 659         bbase = isRgb ? 0 : (bpp - 5);
 660         c->yuvTable = av_malloc(1024*3*2);
 661         y_table16 = c->yuvTable;
 662         yb = -(384<<16) - oy;
 663         for (i = 0; i < 1024; i++) {
 664             uint8_t yval = av_clip_uint8((yb + 0x8000) >> 16);
 665             y_table16[i     ] = (yval >> 3)          << rbase;
 666             y_table16[i+1024] = (yval >> (18 - bpp)) << gbase;
 667             y_table16[i+2048] = (yval >> 3)          << bbase;
 668             yb += cy;
 669         }
 670         fill_table(c->table_rV, 2, crv, y_table16 + yoffs);
 671         fill_table(c->table_gU, 2, cgu, y_table16 + yoffs + 1024);
 672         fill_table(c->table_bU, 2, cbu, y_table16 + yoffs + 2048);
 673         fill_gv_table(c->table_gV, 2, cgv);
 674         break;
 675     case 24:
 676         c->yuvTable = av_malloc(1024);
 677         y_table = c->yuvTable;
 678         yb = -(384<<16) - oy;
 679         for (i = 0; i < 1024; i++) {
 680             y_table[i] = av_clip_uint8((yb + 0x8000) >> 16);
 681             yb += cy;
 682         }
 683         fill_table(c->table_rV, 1, crv, y_table + yoffs);
 684         fill_table(c->table_gU, 1, cgu, y_table + yoffs);
 685         fill_table(c->table_bU, 1, cbu, y_table + yoffs);
 686         fill_gv_table(c->table_gV, 1, cgv);
 687         break;
 688     case 32:
 689         base = (c->dstFormat == PIX_FMT_RGB32_1 || c->dstFormat == PIX_FMT_BGR32_1) ? 8 : 0;
 690         rbase = base + (isRgb ? 16 : 0);
 691         gbase = base + 8;
 692         bbase = base + (isRgb ? 0 : 16);
 693         needAlpha = CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat);
 694         if (!needAlpha)
 695             abase = (base + 24) & 31;
 696         c->yuvTable = av_malloc(1024*3*4);
 697         y_table32 = c->yuvTable;
 698         yb = -(384<<16) - oy;
 699         for (i = 0; i < 1024; i++) {
 700             uint8_t yval = av_clip_uint8((yb + 0x8000) >> 16);
 701             y_table32[i     ] = (yval << rbase) + (needAlpha ? 0 : (255 << abase));
 702             y_table32[i+1024] = yval << gbase;
 703             y_table32[i+2048] = yval << bbase;
 704             yb += cy;
 705         }
 706         fill_table(c->table_rV, 4, crv, y_table32 + yoffs);
 707         fill_table(c->table_gU, 4, cgu, y_table32 + yoffs + 1024);
 708         fill_table(c->table_bU, 4, cbu, y_table32 + yoffs + 2048);
 709         fill_gv_table(c->table_gV, 4, cgv);
 710         break;
 711     default:
 712         c->yuvTable = NULL;
 713         av_log(c, AV_LOG_ERROR, "%ibpp not supported by yuv2rgb\n", bpp);
 714         return -1;
 715     }
 716     return 0;
 717 }