git.sesse.net Git - x264/blob - common/pixel.c

   1 /*****************************************************************************
   2  * pixel.c: h264 encoder
   3  *****************************************************************************
   4  * Copyright (C) 2003-2008 x264 project
   5  *
   6  * Authors: Loren Merritt <lorenm@u.washington.edu>
   7  *          Laurent Aimar <fenrir@via.ecp.fr>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  22  *****************************************************************************/
  23
  24 #include "common.h"
  25
  26 #ifdef HAVE_MMX
  27 #   include "x86/pixel.h"
  28 #endif
  29 #ifdef ARCH_PPC
  30 #   include "ppc/pixel.h"
  31 #endif
  32 #ifdef ARCH_UltraSparc
  33 #   include "sparc/pixel.h"
  34 #endif
  35
  36
  37 /****************************************************************************
  38  * pixel_sad_WxH
  39  ****************************************************************************/
  40 #define PIXEL_SAD_C( name, lx, ly ) \
  41 static int name( uint8_t *pix1, int i_stride_pix1,  \
  42                  uint8_t *pix2, int i_stride_pix2 ) \
  43 {                                                   \
  44     int i_sum = 0;                                  \
  45     int x, y;                                       \
  46     for( y = 0; y < ly; y++ )                       \
  47     {                                               \
  48         for( x = 0; x < lx; x++ )                   \
  49         {                                           \
  50             i_sum += abs( pix1[x] - pix2[x] );      \
  51         }                                           \
  52         pix1 += i_stride_pix1;                      \
  53         pix2 += i_stride_pix2;                      \
  54     }                                               \
  55     return i_sum;                                   \
  56 }
  57
  58
  59 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  60 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  61 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  62 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  63 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  64 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  65 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  66
  67
  68 /****************************************************************************
  69  * pixel_ssd_WxH
  70  ****************************************************************************/
  71 #define PIXEL_SSD_C( name, lx, ly ) \
  72 static int name( uint8_t *pix1, int i_stride_pix1,  \
  73                  uint8_t *pix2, int i_stride_pix2 ) \
  74 {                                                   \
  75     int i_sum = 0;                                  \
  76     int x, y;                                       \
  77     for( y = 0; y < ly; y++ )                       \
  78     {                                               \
  79         for( x = 0; x < lx; x++ )                   \
  80         {                                           \
  81             int d = pix1[x] - pix2[x];              \
  82             i_sum += d*d;                           \
  83         }                                           \
  84         pix1 += i_stride_pix1;                      \
  85         pix2 += i_stride_pix2;                      \
  86     }                                               \
  87     return i_sum;                                   \
  88 }
  89
  90 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
  91 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
  92 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
  93 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
  94 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
  95 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
  96 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
  97
  98 int64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2, int i_width, int i_height )
  99 {
 100     int64_t i_ssd = 0;
 101     int x, y;
 102     int align = !(((long)pix1 | (long)pix2 | i_pix1 | i_pix2) & 15);
 103
 104 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 105                                           pix2 + y*i_pix2 + x, i_pix2 );
 106     for( y = 0; y < i_height-15; y += 16 )
 107     {
 108         x = 0;
 109         if( align )
 110             for( ; x < i_width-15; x += 16 )
 111                 SSD(PIXEL_16x16);
 112         for( ; x < i_width-7; x += 8 )
 113             SSD(PIXEL_8x16);
 114     }
 115     if( y < i_height-7 )
 116         for( x = 0; x < i_width-7; x += 8 )
 117             SSD(PIXEL_8x8);
 118 #undef SSD
 119
 120 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 121     if( i_width % 8 != 0 )
 122     {
 123         for( y = 0; y < (i_height & ~7); y++ )
 124             for( x = i_width & ~7; x < i_width; x++ )
 125                 SSD1;
 126     }
 127     if( i_height % 8 != 0 )
 128     {
 129         for( y = i_height & ~7; y < i_height; y++ )
 130             for( x = 0; x < i_width; x++ )
 131                 SSD1;
 132     }
 133 #undef SSD1
 134
 135     return i_ssd;
 136 }
 137
 138
 139 static inline void pixel_sub_wxh( int16_t *diff, int i_size,
 140                                   uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
 141 {
 142     int y, x;
 143     for( y = 0; y < i_size; y++ )
 144     {
 145         for( x = 0; x < i_size; x++ )
 146         {
 147             diff[x + y*i_size] = pix1[x] - pix2[x];
 148         }
 149         pix1 += i_pix1;
 150         pix2 += i_pix2;
 151     }
 152 }
 153
 154
 155 /****************************************************************************
 156  * pixel_var_wxh
 157  ****************************************************************************/
 158 #define PIXEL_VAR_C( name, w, shift ) \
 159 static int name( uint8_t *pix, int i_stride, uint32_t *sad ) \
 160 {                                             \
 161     uint32_t var = 0, sum = 0, sqr = 0;       \
 162     int x, y;                                 \
 163     for( y = 0; y < w; y++ )                  \
 164     {                                         \
 165         for( x = 0; x < w; x++ )              \
 166         {                                     \
 167             sum += pix[x];                    \
 168             sqr += pix[x] * pix[x];           \
 169         }                                     \
 170         pix += i_stride;                      \
 171     }                                         \
 172     var = sqr - (sum * sum >> shift);         \
 173     *sad = sum;                               \
 174     return var;                               \
 175 }
 176
 177 PIXEL_VAR_C( x264_pixel_var_16x16, 16, 8 )
 178 PIXEL_VAR_C( x264_pixel_var_8x8,    8, 6 )
 179
 180
 181
 182 /****************************************************************************
 183  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 184  ****************************************************************************/
 185 static int pixel_satd_wxh( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2, int i_width, int i_height )
 186 {
 187     int16_t tmp[4][4];
 188     int16_t diff[4][4];
 189     int x, y;
 190     int i_satd = 0;
 191
 192     for( y = 0; y < i_height; y += 4 )
 193     {
 194         for( x = 0; x < i_width; x += 4 )
 195         {
 196             int d;
 197
 198             pixel_sub_wxh( (int16_t*)diff, 4, &pix1[x], i_pix1, &pix2[x], i_pix2 );
 199
 200             for( d = 0; d < 4; d++ )
 201             {
 202                 int s01, s23;
 203                 int d01, d23;
 204
 205                 s01 = diff[d][0] + diff[d][1]; s23 = diff[d][2] + diff[d][3];
 206                 d01 = diff[d][0] - diff[d][1]; d23 = diff[d][2] - diff[d][3];
 207
 208                 tmp[d][0] = s01 + s23;
 209                 tmp[d][1] = s01 - s23;
 210                 tmp[d][2] = d01 - d23;
 211                 tmp[d][3] = d01 + d23;
 212             }
 213             for( d = 0; d < 4; d++ )
 214             {
 215                 int s01, s23;
 216                 int d01, d23;
 217
 218                 s01 = tmp[0][d] + tmp[1][d]; s23 = tmp[2][d] + tmp[3][d];
 219                 d01 = tmp[0][d] - tmp[1][d]; d23 = tmp[2][d] - tmp[3][d];
 220
 221                 i_satd += abs( s01 + s23 ) + abs( s01 - s23 ) + abs( d01 - d23 ) + abs( d01 + d23 );
 222             }
 223
 224         }
 225         pix1 += 4 * i_pix1;
 226         pix2 += 4 * i_pix2;
 227     }
 228
 229     return i_satd / 2;
 230 }
 231 #define PIXEL_SATD_C( name, width, height ) \
 232 static int name( uint8_t *pix1, int i_stride_pix1, \
 233                  uint8_t *pix2, int i_stride_pix2 ) \
 234 { \
 235     return pixel_satd_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
 236 }
 237 PIXEL_SATD_C( x264_pixel_satd_16x16, 16, 16 )
 238 PIXEL_SATD_C( x264_pixel_satd_16x8,  16, 8 )
 239 PIXEL_SATD_C( x264_pixel_satd_8x16,  8, 16 )
 240 PIXEL_SATD_C( x264_pixel_satd_8x8,   8, 8 )
 241 PIXEL_SATD_C( x264_pixel_satd_8x4,   8, 4 )
 242 PIXEL_SATD_C( x264_pixel_satd_4x8,   4, 8 )
 243 PIXEL_SATD_C( x264_pixel_satd_4x4,   4, 4 )
 244
 245
 246 /****************************************************************************
 247  * pixel_sa8d_WxH: sum of 8x8 Hadamard transformed differences
 248  ****************************************************************************/
 249 #define SA8D_1D {\
 250     const int a0 = SRC(0) + SRC(4);\
 251     const int a4 = SRC(0) - SRC(4);\
 252     const int a1 = SRC(1) + SRC(5);\
 253     const int a5 = SRC(1) - SRC(5);\
 254     const int a2 = SRC(2) + SRC(6);\
 255     const int a6 = SRC(2) - SRC(6);\
 256     const int a3 = SRC(3) + SRC(7);\
 257     const int a7 = SRC(3) - SRC(7);\
 258     const int b0 = a0 + a2;\
 259     const int b2 = a0 - a2;\
 260     const int b1 = a1 + a3;\
 261     const int b3 = a1 - a3;\
 262     const int b4 = a4 + a6;\
 263     const int b6 = a4 - a6;\
 264     const int b5 = a5 + a7;\
 265     const int b7 = a5 - a7;\
 266     DST(0, b0 + b1);\
 267     DST(1, b0 - b1);\
 268     DST(2, b2 + b3);\
 269     DST(3, b2 - b3);\
 270     DST(4, b4 + b5);\
 271     DST(5, b4 - b5);\
 272     DST(6, b6 + b7);\
 273     DST(7, b6 - b7);\
 274 }
 275
 276 static inline int pixel_sa8d_wxh( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2,
 277                                   int i_width, int i_height )
 278 {
 279     int16_t diff[8][8];
 280     int i_satd = 0;
 281     int x, y;
 282
 283     for( y = 0; y < i_height; y += 8 )
 284     {
 285         for( x = 0; x < i_width; x += 8 )
 286         {
 287             int i;
 288             pixel_sub_wxh( (int16_t*)diff, 8, pix1+x, i_pix1, pix2+x, i_pix2 );
 289
 290 #define SRC(x)     diff[i][x]
 291 #define DST(x,rhs) diff[i][x] = (rhs)
 292             for( i = 0; i < 8; i++ )
 293                 SA8D_1D
 294 #undef SRC
 295 #undef DST
 296
 297 #define SRC(x)     diff[x][i]
 298 #define DST(x,rhs) i_satd += abs(rhs)
 299             for( i = 0; i < 8; i++ )
 300                 SA8D_1D
 301 #undef SRC
 302 #undef DST
 303         }
 304         pix1 += 8 * i_pix1;
 305         pix2 += 8 * i_pix2;
 306     }
 307
 308     return i_satd;
 309 }
 310
 311 #define PIXEL_SA8D_C( width, height ) \
 312 static int x264_pixel_sa8d_##width##x##height( uint8_t *pix1, int i_stride_pix1, \
 313                                                uint8_t *pix2, int i_stride_pix2 ) \
 314 { \
 315     return ( pixel_sa8d_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ) + 2 ) >> 2; \
 316 }
 317 PIXEL_SA8D_C( 16, 16 )
 318 PIXEL_SA8D_C( 16, 8 )
 319 PIXEL_SA8D_C( 8, 16 )
 320 PIXEL_SA8D_C( 8, 8 )
 321
 322 /****************************************************************************
 323  * pixel_sad_x4
 324  ****************************************************************************/
 325 #define SAD_X( size ) \
 326 static void x264_pixel_sad_x3_##size( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, int i_stride, int scores[3] )\
 327 {\
 328     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 329     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 330     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 331 }\
 332 static void x264_pixel_sad_x4_##size( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, uint8_t *pix3, int i_stride, int scores[4] )\
 333 {\
 334     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 335     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 336     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 337     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 338 }
 339
 340 SAD_X( 16x16 )
 341 SAD_X( 16x8 )
 342 SAD_X( 8x16 )
 343 SAD_X( 8x8 )
 344 SAD_X( 8x4 )
 345 SAD_X( 4x8 )
 346 SAD_X( 4x4 )
 347
 348 #ifdef ARCH_UltraSparc
 349 SAD_X( 16x16_vis )
 350 SAD_X( 16x8_vis )
 351 SAD_X( 8x16_vis )
 352 SAD_X( 8x8_vis )
 353 #endif
 354
 355 /****************************************************************************
 356  * pixel_satd_x4
 357  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 358  ****************************************************************************/
 359
 360 #define SATD_X( size, cpu ) \
 361 static void x264_pixel_satd_x3_##size##cpu( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, int i_stride, int scores[3] )\
 362 {\
 363     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 364     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 365     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 366 }\
 367 static void x264_pixel_satd_x4_##size##cpu( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, uint8_t *pix3, int i_stride, int scores[4] )\
 368 {\
 369     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 370     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 371     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 372     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 373 }
 374 #define SATD_X_DECL5( cpu )\
 375 SATD_X( 16x16, cpu )\
 376 SATD_X( 16x8, cpu )\
 377 SATD_X( 8x16, cpu )\
 378 SATD_X( 8x8, cpu )\
 379 SATD_X( 8x4, cpu )
 380 #define SATD_X_DECL7( cpu )\
 381 SATD_X_DECL5( cpu )\
 382 SATD_X( 4x8, cpu )\
 383 SATD_X( 4x4, cpu )
 384
 385 SATD_X_DECL7()
 386 #ifdef HAVE_MMX
 387 SATD_X_DECL7( _mmxext )
 388 SATD_X_DECL5( _sse2 )
 389 SATD_X_DECL7( _ssse3 )
 390 SATD_X_DECL5( _ssse3_phadd )
 391 #endif
 392
 393 /****************************************************************************
 394  * structural similarity metric
 395  ****************************************************************************/
 396 static void ssim_4x4x2_core( const uint8_t *pix1, int stride1,
 397                              const uint8_t *pix2, int stride2,
 398                              int sums[2][4])
 399 {
 400     int x, y, z;
 401     for(z=0; z<2; z++)
 402     {
 403         uint32_t s1=0, s2=0, ss=0, s12=0;
 404         for(y=0; y<4; y++)
 405             for(x=0; x<4; x++)
 406             {
 407                 int a = pix1[x+y*stride1];
 408                 int b = pix2[x+y*stride2];
 409                 s1  += a;
 410                 s2  += b;
 411                 ss  += a*a;
 412                 ss  += b*b;
 413                 s12 += a*b;
 414             }
 415         sums[z][0] = s1;
 416         sums[z][1] = s2;
 417         sums[z][2] = ss;
 418         sums[z][3] = s12;
 419         pix1 += 4;
 420         pix2 += 4;
 421     }
 422 }
 423
 424 static float ssim_end1( int s1, int s2, int ss, int s12 )
 425 {
 426     static const int ssim_c1 = (int)(.01*.01*255*255*64 + .5);
 427     static const int ssim_c2 = (int)(.03*.03*255*255*64*63 + .5);
 428     int vars = ss*64 - s1*s1 - s2*s2;
 429     int covar = s12*64 - s1*s2;
 430     return (float)(2*s1*s2 + ssim_c1) * (float)(2*covar + ssim_c2)\
 431            / ((float)(s1*s1 + s2*s2 + ssim_c1) * (float)(vars + ssim_c2));
 432 }
 433
 434 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 435 {
 436     int i;
 437     float ssim = 0.0;
 438     for( i = 0; i < width; i++ )
 439         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 440                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 441                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 442                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 443     return ssim;
 444 }
 445
 446 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 447                            uint8_t *pix1, int stride1,
 448                            uint8_t *pix2, int stride2,
 449                            int width, int height )
 450 {
 451     int x, y, z;
 452     float ssim = 0.0;
 453     int (*sum0)[4] = x264_malloc(4 * (width/4+3) * sizeof(int));
 454     int (*sum1)[4] = x264_malloc(4 * (width/4+3) * sizeof(int));
 455     width >>= 2;
 456     height >>= 2;
 457     z = 0;
 458     for( y = 1; y < height; y++ )
 459     {
 460         for( ; z <= y; z++ )
 461         {
 462             XCHG( void*, sum0, sum1 );
 463             for( x = 0; x < width; x+=2 )
 464                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 465         }
 466         for( x = 0; x < width-1; x += 4 )
 467             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 468     }
 469     x264_free(sum0);
 470     x264_free(sum1);
 471     return ssim;
 472 }
 473
 474
 475 /****************************************************************************
 476  * successive elimination
 477  ****************************************************************************/
 478 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 479                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 480 {
 481     int nmv=0, i;
 482     for( i=0; i<width; i++, sums++ )
 483     {
 484         int ads = abs( enc_dc[0] - sums[0] )
 485                 + abs( enc_dc[1] - sums[8] )
 486                 + abs( enc_dc[2] - sums[delta] )
 487                 + abs( enc_dc[3] - sums[delta+8] )
 488                 + cost_mvx[i];
 489         if( ads < thresh )
 490             mvs[nmv++] = i;
 491     }
 492     return nmv;
 493 }
 494
 495 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 496                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 497 {
 498     int nmv=0, i;
 499     for( i=0; i<width; i++, sums++ )
 500     {
 501         int ads = abs( enc_dc[0] - sums[0] )
 502                 + abs( enc_dc[1] - sums[delta] )
 503                 + cost_mvx[i];
 504         if( ads < thresh )
 505             mvs[nmv++] = i;
 506     }
 507     return nmv;
 508 }
 509
 510 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 511                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 512 {
 513     int nmv=0, i;
 514     for( i=0; i<width; i++, sums++ )
 515     {
 516         int ads = abs( enc_dc[0] - sums[0] )
 517                 + cost_mvx[i];
 518         if( ads < thresh )
 519             mvs[nmv++] = i;
 520     }
 521     return nmv;
 522 }
 523
 524
 525 /****************************************************************************
 526  * x264_pixel_init:
 527  ****************************************************************************/
 528 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 529 {
 530     memset( pixf, 0, sizeof(*pixf) );
 531
 532 #define INIT2( name, cpu ) \
 533     pixf->name[PIXEL_16x16] = x264_pixel_##name##_16x16##cpu;\
 534     pixf->name[PIXEL_16x8]  = x264_pixel_##name##_16x8##cpu;
 535 #define INIT4( name, cpu ) \
 536     INIT2( name, cpu ) \
 537     pixf->name[PIXEL_8x16]  = x264_pixel_##name##_8x16##cpu;\
 538     pixf->name[PIXEL_8x8]   = x264_pixel_##name##_8x8##cpu;
 539 #define INIT5( name, cpu ) \
 540     INIT4( name, cpu ) \
 541     pixf->name[PIXEL_8x4]   = x264_pixel_##name##_8x4##cpu;
 542 #define INIT7( name, cpu ) \
 543     INIT5( name, cpu ) \
 544     pixf->name[PIXEL_4x8]   = x264_pixel_##name##_4x8##cpu;\
 545     pixf->name[PIXEL_4x4]   = x264_pixel_##name##_4x4##cpu;
 546
 547 #define INIT_ADS( cpu ) \
 548     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 549     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 550     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 551
 552     INIT7( sad, );
 553     INIT7( sad_x3, );
 554     INIT7( sad_x4, );
 555     INIT7( ssd, );
 556     INIT7( satd, );
 557     INIT7( satd_x3, );
 558     INIT7( satd_x4, );
 559     INIT4( sa8d, );
 560     INIT_ADS( );
 561
 562     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 563     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 564
 565     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 566     pixf->ssim_end4 = ssim_end4;
 567
 568 #ifdef HAVE_MMX
 569     if( cpu&X264_CPU_MMX )
 570     {
 571         INIT7( ssd, _mmx );
 572     }
 573
 574     if( cpu&X264_CPU_MMXEXT )
 575     {
 576         INIT7( sad, _mmxext );
 577         INIT7( sad_x3, _mmxext );
 578         INIT7( sad_x4, _mmxext );
 579         INIT7( satd, _mmxext );
 580         INIT7( satd_x3, _mmxext );
 581         INIT7( satd_x4, _mmxext );
 582         INIT_ADS( _mmxext );
 583         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
 584         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
 585 #ifdef ARCH_X86
 586         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
 587         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmxext;
 588         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
 589         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_mmxext;
 590
 591         if( cpu&X264_CPU_CACHELINE_32 )
 592         {
 593             INIT5( sad, _cache32_mmxext );
 594             INIT4( sad_x3, _cache32_mmxext );
 595             INIT4( sad_x4, _cache32_mmxext );
 596         }
 597         else if( cpu&X264_CPU_CACHELINE_64 )
 598         {
 599             INIT5( sad, _cache64_mmxext );
 600             INIT4( sad_x3, _cache64_mmxext );
 601             INIT4( sad_x4, _cache64_mmxext );
 602         }
 603 #else
 604         if( cpu&X264_CPU_CACHELINE_64 )
 605         {
 606             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
 607             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmxext;
 608             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmxext;
 609             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
 610             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmxext;
 611             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
 612             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmxext;
 613         }
 614 #endif
 615         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
 616         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
 617         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
 618     }
 619
 620     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 621     {
 622         INIT2( sad, _sse2 );
 623         INIT2( sad_x3, _sse2 );
 624         INIT2( sad_x4, _sse2 );
 625         INIT_ADS( _sse2 );
 626         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
 627
 628 #ifdef ARCH_X86
 629         if( cpu&X264_CPU_CACHELINE_64 )
 630         {
 631             INIT2( sad, _cache64_sse2 );
 632             INIT2( sad_x3, _cache64_sse2 );
 633             INIT2( sad_x4, _cache64_sse2 );
 634         }
 635 #endif
 636     }
 637     if( cpu&X264_CPU_SSE2 )
 638     {
 639         INIT5( ssd, _sse2 );
 640         INIT5( satd, _sse2 );
 641         INIT5( satd_x3, _sse2 );
 642         INIT5( satd_x4, _sse2 );
 643         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 644         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 645         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 646         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 647         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 648 #ifdef ARCH_X86_64
 649         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 650 #endif
 651     }
 652
 653     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
 654     {
 655         INIT2( sad, _sse3 );
 656         INIT2( sad_x3, _sse3 );
 657         INIT2( sad_x4, _sse3 );
 658     }
 659
 660     if( cpu&X264_CPU_SSSE3 )
 661     {
 662         INIT7( satd, _ssse3 );
 663         INIT7( satd_x3, _ssse3 );
 664         INIT7( satd_x4, _ssse3 );
 665         INIT_ADS( _ssse3 );
 666         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 667         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 668         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
 669         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
 670         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
 671 #ifdef ARCH_X86_64
 672         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
 673 #endif
 674         if( cpu&X264_CPU_CACHELINE_64 )
 675         {
 676             INIT2( sad, _cache64_ssse3 );
 677             INIT2( sad_x3, _cache64_ssse3 );
 678             INIT2( sad_x4, _cache64_ssse3 );
 679         }
 680         if( cpu&X264_CPU_PHADD_IS_FAST )
 681         {
 682             INIT5( satd, _ssse3_phadd );
 683             INIT5( satd_x3, _ssse3_phadd );
 684             INIT5( satd_x4, _ssse3_phadd );
 685         }
 686     }
 687 #endif //HAVE_MMX
 688
 689 #ifdef ARCH_PPC
 690     if( cpu&X264_CPU_ALTIVEC )
 691     {
 692         x264_pixel_altivec_init( pixf );
 693     }
 694 #endif
 695 #ifdef ARCH_UltraSparc
 696     INIT4( sad, _vis );
 697     INIT4( sad_x3, _vis );
 698     INIT4( sad_x4, _vis );
 699 #endif
 700
 701     pixf->ads[PIXEL_8x16] =
 702     pixf->ads[PIXEL_8x4] =
 703     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
 704     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
 705 }
 706