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 /****************************************************************************
 140  * pixel_var_wxh
 141  ****************************************************************************/
 142 #define PIXEL_VAR_C( name, w, shift ) \
 143 static int name( uint8_t *pix, int i_stride ) \
 144 {                                             \
 145     uint32_t var = 0, sum = 0, sqr = 0;       \
 146     int x, y;                                 \
 147     for( y = 0; y < w; y++ )                  \
 148     {                                         \
 149         for( x = 0; x < w; x++ )              \
 150         {                                     \
 151             sum += pix[x];                    \
 152             sqr += pix[x] * pix[x];           \
 153         }                                     \
 154         pix += i_stride;                      \
 155     }                                         \
 156     var = sqr - (sum * sum >> shift);         \
 157     return var;                               \
 158 }
 159
 160 PIXEL_VAR_C( x264_pixel_var_16x16, 16, 8 )
 161 PIXEL_VAR_C( x264_pixel_var_8x8,    8, 6 )
 162
 163
 164 #define HADAMARD4(d0,d1,d2,d3,s0,s1,s2,s3) {\
 165     int t0 = s0 + s1;\
 166     int t1 = s0 - s1;\
 167     int t2 = s2 + s3;\
 168     int t3 = s2 - s3;\
 169     d0 = t0 + t2;\
 170     d2 = t0 - t2;\
 171     d1 = t1 + t3;\
 172     d3 = t1 - t3;\
 173 }
 174
 175 /****************************************************************************
 176  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 177  ****************************************************************************/
 178 static int pixel_satd_wxh( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2, int i_width, int i_height )
 179 {
 180     int16_t tmp[4][4];
 181     int x, y;
 182     int i_satd = 0;
 183
 184     for( y = 0; y < i_height; y += 4 )
 185     {
 186         for( x = 0; x < i_width; x += 4 )
 187         {
 188             int i;
 189             uint8_t *p1 = pix1+x, *p2 = pix2+x;
 190
 191             for( i=0; i<4; i++, p1+=i_pix1, p2+=i_pix2 )
 192             {
 193                 int a0 = p1[0] - p2[0];
 194                 int a1 = p1[1] - p2[1];
 195                 int a2 = p1[2] - p2[2];
 196                 int a3 = p1[3] - p2[3];
 197                 HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 198             }
 199             for( i=0; i<4; i++ )
 200             {
 201                 int a0,a1,a2,a3;
 202                 HADAMARD4( a0,a1,a2,a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 203                 i_satd += abs(a0) + abs(a1) + abs(a2) + abs(a3);
 204             }
 205
 206         }
 207         pix1 += 4 * i_pix1;
 208         pix2 += 4 * i_pix2;
 209     }
 210
 211     return i_satd / 2;
 212 }
 213 #define PIXEL_SATD_C( name, width, height ) \
 214 static int name( uint8_t *pix1, int i_stride_pix1, \
 215                  uint8_t *pix2, int i_stride_pix2 ) \
 216 { \
 217     return pixel_satd_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ); \
 218 }
 219 PIXEL_SATD_C( x264_pixel_satd_16x16, 16, 16 )
 220 PIXEL_SATD_C( x264_pixel_satd_16x8,  16, 8 )
 221 PIXEL_SATD_C( x264_pixel_satd_8x16,  8, 16 )
 222 PIXEL_SATD_C( x264_pixel_satd_8x8,   8, 8 )
 223 PIXEL_SATD_C( x264_pixel_satd_8x4,   8, 4 )
 224 PIXEL_SATD_C( x264_pixel_satd_4x8,   4, 8 )
 225 PIXEL_SATD_C( x264_pixel_satd_4x4,   4, 4 )
 226
 227
 228 /****************************************************************************
 229  * pixel_sa8d_WxH: sum of 8x8 Hadamard transformed differences
 230  ****************************************************************************/
 231 #define SA8D_1D {\
 232     int b0,b1,b2,b3,b4,b5,b6,b7;\
 233     HADAMARD4( b0,b1,b2,b3, SRC(0), SRC(1), SRC(2), SRC(3) );\
 234     HADAMARD4( b4,b5,b6,b7, SRC(4), SRC(5), SRC(6), SRC(7) );\
 235     DST(0, b0 + b4);\
 236     DST(4, b0 - b4);\
 237     DST(1, b1 + b5);\
 238     DST(5, b1 - b5);\
 239     DST(2, b2 + b6);\
 240     DST(6, b2 - b6);\
 241     DST(3, b3 + b7);\
 242     DST(7, b3 - b7);\
 243 }
 244
 245 static inline int pixel_sa8d_wxh( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2,
 246                                   int i_width, int i_height )
 247 {
 248     int16_t diff[8][8];
 249     int i_satd = 0;
 250     int x, y;
 251
 252     for( y = 0; y < i_height; y += 8 )
 253     {
 254         for( x = 0; x < i_width; x += 8 )
 255         {
 256             int i;
 257             uint8_t *p1 = pix1+x, *p2 = pix2+x;
 258
 259 #define SRC(x)     a##x
 260 #define DST(x,rhs) diff[i][x] = (rhs)
 261             for( i=0; i<8; i++, p1+=i_pix1, p2+=i_pix2 )
 262             {
 263                 int a0 = p1[0] - p2[0];
 264                 int a1 = p1[1] - p2[1];
 265                 int a2 = p1[2] - p2[2];
 266                 int a3 = p1[3] - p2[3];
 267                 int a4 = p1[4] - p2[4];
 268                 int a5 = p1[5] - p2[5];
 269                 int a6 = p1[6] - p2[6];
 270                 int a7 = p1[7] - p2[7];
 271                 SA8D_1D
 272             }
 273 #undef SRC
 274 #undef DST
 275
 276 #define SRC(x)     diff[x][i]
 277 #define DST(x,rhs) i_satd += abs(rhs)
 278             for( i=0; i<8; i++ )
 279                 SA8D_1D
 280 #undef SRC
 281 #undef DST
 282         }
 283         pix1 += 8 * i_pix1;
 284         pix2 += 8 * i_pix2;
 285     }
 286
 287     return i_satd;
 288 }
 289
 290 #define PIXEL_SA8D_C( width, height ) \
 291 static int x264_pixel_sa8d_##width##x##height( uint8_t *pix1, int i_stride_pix1, \
 292                                                uint8_t *pix2, int i_stride_pix2 ) \
 293 { \
 294     return ( pixel_sa8d_wxh( pix1, i_stride_pix1, pix2, i_stride_pix2, width, height ) + 2 ) >> 2; \
 295 }
 296 PIXEL_SA8D_C( 16, 16 )
 297 PIXEL_SA8D_C( 16, 8 )
 298 PIXEL_SA8D_C( 8, 16 )
 299 PIXEL_SA8D_C( 8, 8 )
 300
 301
 302 static uint64_t pixel_hadamard_ac( uint8_t *pix, int stride )
 303 {
 304     int16_t tmp[8][8];
 305     int sum4=0, sum8=0;
 306     int i;
 307     for( i=0; i<8; i++, pix+=stride )
 308     {
 309         HADAMARD4( tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i],
 310                    pix[0], pix[1], pix[2], pix[3] );
 311         HADAMARD4( tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i],
 312                    pix[4], pix[5], pix[6], pix[7] );
 313     }
 314     for( i=0; i<8; i++ )
 315     {
 316         int a0,a1,a2,a3,a4,a5,a6,a7;
 317         HADAMARD4( a0,a1,a2,a3, tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3] );
 318         sum4 += abs(a0) + abs(a1) + abs(a2) + abs(a3);
 319         HADAMARD4( a4,a5,a6,a7, tmp[i][4], tmp[i][5], tmp[i][6], tmp[i][7] );
 320         sum4 += abs(a4) + abs(a5) + abs(a6) + abs(a7);
 321         tmp[i][0] = a0 + a4;
 322         tmp[i][4] = a0 - a4;
 323         tmp[i][1] = a1 + a5;
 324         tmp[i][5] = a1 - a5;
 325         tmp[i][2] = a2 + a6;
 326         tmp[i][6] = a2 - a6;
 327         tmp[i][3] = a3 + a7;
 328         tmp[i][7] = a3 - a7;
 329     }
 330     for( i=0; i<8; i++ )
 331     {
 332         sum8 += abs( tmp[0][i] + tmp[4][i] )
 333               + abs( tmp[0][i] - tmp[4][i] )
 334               + abs( tmp[1][i] + tmp[5][i] )
 335               + abs( tmp[1][i] - tmp[5][i] )
 336               + abs( tmp[2][i] + tmp[6][i] )
 337               + abs( tmp[2][i] - tmp[6][i] )
 338               + abs( tmp[3][i] + tmp[7][i] )
 339               + abs( tmp[3][i] - tmp[7][i] );
 340     }
 341     sum4 -= tmp[0][0]+tmp[4][0];
 342     sum8 -= tmp[0][0]+tmp[4][0];
 343     return ((uint64_t)sum8<<32) + sum4;
 344 }
 345
 346 #define HADAMARD_AC(w,h) \
 347 static uint64_t x264_pixel_hadamard_ac_##w##x##h( uint8_t *pix, int stride )\
 348 {\
 349     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 350     if( w==16 )\
 351         sum += pixel_hadamard_ac( pix+8, stride );\
 352     if( h==16 )\
 353         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 354     if( w==16 && h==16 )\
 355         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 356     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 357 }
 358 HADAMARD_AC( 16, 16 )
 359 HADAMARD_AC( 16, 8 )
 360 HADAMARD_AC( 8, 16 )
 361 HADAMARD_AC( 8, 8 )
 362
 363
 364 /****************************************************************************
 365  * pixel_sad_x4
 366  ****************************************************************************/
 367 #define SAD_X( size ) \
 368 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] )\
 369 {\
 370     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 371     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 372     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 373 }\
 374 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] )\
 375 {\
 376     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 377     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 378     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 379     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 380 }
 381
 382 SAD_X( 16x16 )
 383 SAD_X( 16x8 )
 384 SAD_X( 8x16 )
 385 SAD_X( 8x8 )
 386 SAD_X( 8x4 )
 387 SAD_X( 4x8 )
 388 SAD_X( 4x4 )
 389
 390 #ifdef ARCH_UltraSparc
 391 SAD_X( 16x16_vis )
 392 SAD_X( 16x8_vis )
 393 SAD_X( 8x16_vis )
 394 SAD_X( 8x8_vis )
 395 #endif
 396
 397 /****************************************************************************
 398  * pixel_satd_x4
 399  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 400  ****************************************************************************/
 401
 402 #define SATD_X( size, cpu ) \
 403 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] )\
 404 {\
 405     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 406     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 407     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 408 }\
 409 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] )\
 410 {\
 411     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 412     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 413     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 414     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 415 }
 416 #define SATD_X_DECL6( cpu )\
 417 SATD_X( 16x16, cpu )\
 418 SATD_X( 16x8, cpu )\
 419 SATD_X( 8x16, cpu )\
 420 SATD_X( 8x8, cpu )\
 421 SATD_X( 8x4, cpu )\
 422 SATD_X( 4x8, cpu )
 423 #define SATD_X_DECL7( cpu )\
 424 SATD_X_DECL6( cpu )\
 425 SATD_X( 4x4, cpu )
 426
 427 SATD_X_DECL7()
 428 #ifdef HAVE_MMX
 429 SATD_X_DECL7( _mmxext )
 430 SATD_X_DECL6( _sse2 )
 431 SATD_X_DECL7( _ssse3 )
 432 SATD_X_DECL6( _ssse3_phadd )
 433 #endif
 434
 435 /****************************************************************************
 436  * structural similarity metric
 437  ****************************************************************************/
 438 static void ssim_4x4x2_core( const uint8_t *pix1, int stride1,
 439                              const uint8_t *pix2, int stride2,
 440                              int sums[2][4])
 441 {
 442     int x, y, z;
 443     for(z=0; z<2; z++)
 444     {
 445         uint32_t s1=0, s2=0, ss=0, s12=0;
 446         for(y=0; y<4; y++)
 447             for(x=0; x<4; x++)
 448             {
 449                 int a = pix1[x+y*stride1];
 450                 int b = pix2[x+y*stride2];
 451                 s1  += a;
 452                 s2  += b;
 453                 ss  += a*a;
 454                 ss  += b*b;
 455                 s12 += a*b;
 456             }
 457         sums[z][0] = s1;
 458         sums[z][1] = s2;
 459         sums[z][2] = ss;
 460         sums[z][3] = s12;
 461         pix1 += 4;
 462         pix2 += 4;
 463     }
 464 }
 465
 466 static float ssim_end1( int s1, int s2, int ss, int s12 )
 467 {
 468     static const int ssim_c1 = (int)(.01*.01*255*255*64 + .5);
 469     static const int ssim_c2 = (int)(.03*.03*255*255*64*63 + .5);
 470     int vars = ss*64 - s1*s1 - s2*s2;
 471     int covar = s12*64 - s1*s2;
 472     return (float)(2*s1*s2 + ssim_c1) * (float)(2*covar + ssim_c2)\
 473            / ((float)(s1*s1 + s2*s2 + ssim_c1) * (float)(vars + ssim_c2));
 474 }
 475
 476 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 477 {
 478     int i;
 479     float ssim = 0.0;
 480     for( i = 0; i < width; i++ )
 481         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 482                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 483                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 484                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 485     return ssim;
 486 }
 487
 488 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 489                            uint8_t *pix1, int stride1,
 490                            uint8_t *pix2, int stride2,
 491                            int width, int height )
 492 {
 493     int x, y, z;
 494     float ssim = 0.0;
 495     int (*sum0)[4] = x264_malloc(4 * (width/4+3) * sizeof(int));
 496     int (*sum1)[4] = x264_malloc(4 * (width/4+3) * sizeof(int));
 497     width >>= 2;
 498     height >>= 2;
 499     z = 0;
 500     for( y = 1; y < height; y++ )
 501     {
 502         for( ; z <= y; z++ )
 503         {
 504             XCHG( void*, sum0, sum1 );
 505             for( x = 0; x < width; x+=2 )
 506                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 507         }
 508         for( x = 0; x < width-1; x += 4 )
 509             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 510     }
 511     x264_free(sum0);
 512     x264_free(sum1);
 513     return ssim;
 514 }
 515
 516
 517 /****************************************************************************
 518  * successive elimination
 519  ****************************************************************************/
 520 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 521                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 522 {
 523     int nmv=0, i;
 524     for( i=0; i<width; i++, sums++ )
 525     {
 526         int ads = abs( enc_dc[0] - sums[0] )
 527                 + abs( enc_dc[1] - sums[8] )
 528                 + abs( enc_dc[2] - sums[delta] )
 529                 + abs( enc_dc[3] - sums[delta+8] )
 530                 + cost_mvx[i];
 531         if( ads < thresh )
 532             mvs[nmv++] = i;
 533     }
 534     return nmv;
 535 }
 536
 537 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 538                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 539 {
 540     int nmv=0, i;
 541     for( i=0; i<width; i++, sums++ )
 542     {
 543         int ads = abs( enc_dc[0] - sums[0] )
 544                 + abs( enc_dc[1] - sums[delta] )
 545                 + cost_mvx[i];
 546         if( ads < thresh )
 547             mvs[nmv++] = i;
 548     }
 549     return nmv;
 550 }
 551
 552 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 553                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 554 {
 555     int nmv=0, i;
 556     for( i=0; i<width; i++, sums++ )
 557     {
 558         int ads = abs( enc_dc[0] - sums[0] )
 559                 + cost_mvx[i];
 560         if( ads < thresh )
 561             mvs[nmv++] = i;
 562     }
 563     return nmv;
 564 }
 565
 566
 567 /****************************************************************************
 568  * x264_pixel_init:
 569  ****************************************************************************/
 570 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 571 {
 572     memset( pixf, 0, sizeof(*pixf) );
 573
 574 #define INIT2_NAME( name1, name2, cpu ) \
 575     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 576     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 577 #define INIT4_NAME( name1, name2, cpu ) \
 578     INIT2_NAME( name1, name2, cpu ) \
 579     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 580     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 581 #define INIT5_NAME( name1, name2, cpu ) \
 582     INIT4_NAME( name1, name2, cpu ) \
 583     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 584 #define INIT6_NAME( name1, name2, cpu ) \
 585     INIT5_NAME( name1, name2, cpu ) \
 586     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 587 #define INIT7_NAME( name1, name2, cpu ) \
 588     INIT6_NAME( name1, name2, cpu ) \
 589     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 590 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 591 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 592 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 593 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 594 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 595
 596 #define INIT_ADS( cpu ) \
 597     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 598     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 599     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 600
 601     INIT7( sad, );
 602     INIT7_NAME( sad_aligned, sad, );
 603     INIT7( sad_x3, );
 604     INIT7( sad_x4, );
 605     INIT7( ssd, );
 606     INIT7( satd, );
 607     INIT7( satd_x3, );
 608     INIT7( satd_x4, );
 609     INIT4( sa8d, );
 610     INIT4( hadamard_ac, );
 611     INIT_ADS( );
 612
 613     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 614     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 615
 616     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 617     pixf->ssim_end4 = ssim_end4;
 618
 619 #ifdef HAVE_MMX
 620     if( cpu&X264_CPU_MMX )
 621     {
 622         INIT7( ssd, _mmx );
 623     }
 624
 625     if( cpu&X264_CPU_MMXEXT )
 626     {
 627         INIT7( sad, _mmxext );
 628         INIT7_NAME( sad_aligned, sad, _mmxext );
 629         INIT7( sad_x3, _mmxext );
 630         INIT7( sad_x4, _mmxext );
 631         INIT7( satd, _mmxext );
 632         INIT7( satd_x3, _mmxext );
 633         INIT7( satd_x4, _mmxext );
 634         INIT4( hadamard_ac, _mmxext );
 635         INIT_ADS( _mmxext );
 636         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
 637         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
 638 #ifdef ARCH_X86
 639         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
 640         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmxext;
 641         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
 642         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_mmxext;
 643
 644         if( cpu&X264_CPU_CACHELINE_32 )
 645         {
 646             INIT5( sad, _cache32_mmxext );
 647             INIT4( sad_x3, _cache32_mmxext );
 648             INIT4( sad_x4, _cache32_mmxext );
 649         }
 650         else if( cpu&X264_CPU_CACHELINE_64 )
 651         {
 652             INIT5( sad, _cache64_mmxext );
 653             INIT4( sad_x3, _cache64_mmxext );
 654             INIT4( sad_x4, _cache64_mmxext );
 655         }
 656 #else
 657         if( cpu&X264_CPU_CACHELINE_64 )
 658         {
 659             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
 660             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmxext;
 661             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmxext;
 662             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
 663             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmxext;
 664             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
 665             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmxext;
 666         }
 667 #endif
 668         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
 669         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_mmxext;
 670         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
 671         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
 672     }
 673
 674     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 675     {
 676         INIT2( sad, _sse2 );
 677         INIT2( sad_x3, _sse2 );
 678         INIT2( sad_x4, _sse2 );
 679         if( !(cpu&X264_CPU_STACK_MOD4) )
 680         {
 681             INIT4( hadamard_ac, _sse2 );
 682         }
 683         INIT_ADS( _sse2 );
 684         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
 685         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
 686 #ifdef ARCH_X86
 687         if( cpu&X264_CPU_CACHELINE_64 )
 688         {
 689             INIT2( sad, _cache64_sse2 );
 690             INIT2( sad_x3, _cache64_sse2 );
 691             INIT2( sad_x4, _cache64_sse2 );
 692         }
 693 #endif
 694         if( cpu&X264_CPU_SSE_MISALIGN )
 695         {
 696             INIT2( sad_x3, _sse2_misalign );
 697             INIT2( sad_x4, _sse2_misalign );
 698         }
 699     }
 700     if( cpu&X264_CPU_SSE2 )
 701     {
 702         INIT5( ssd, _sse2 );
 703         if( cpu&X264_CPU_SSE2_IS_FAST )
 704         {
 705             INIT6( satd, _sse2 );
 706             INIT6( satd_x3, _sse2 );
 707             INIT6( satd_x4, _sse2 );
 708         }
 709         else
 710         {
 711             INIT5( satd, _sse2 );
 712             INIT5( satd_x3, _sse2 );
 713             INIT5( satd_x4, _sse2 );
 714         }
 715         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
 716         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 717         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 718         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 719         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 720         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 721 #ifdef ARCH_X86_64
 722         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 723 #endif
 724     }
 725
 726     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
 727     {
 728         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 729         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 730         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 731         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
 732         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
 733         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 734         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
 735         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
 736     }
 737
 738     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
 739     {
 740         INIT2( sad, _sse3 );
 741         INIT2( sad_x3, _sse3 );
 742         INIT2( sad_x4, _sse3 );
 743     }
 744
 745     if( cpu&X264_CPU_SSSE3 )
 746     {
 747         INIT7( satd, _ssse3 );
 748         INIT7( satd_x3, _ssse3 );
 749         INIT7( satd_x4, _ssse3 );
 750         if( !(cpu&X264_CPU_STACK_MOD4) )
 751         {
 752             INIT4( hadamard_ac, _ssse3 );
 753         }
 754         INIT_ADS( _ssse3 );
 755         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 756         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 757         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
 758         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 759         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
 760         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
 761 #ifdef ARCH_X86_64
 762         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
 763 #endif
 764         if( cpu&X264_CPU_CACHELINE_64 )
 765         {
 766             INIT2( sad, _cache64_ssse3 );
 767             INIT2( sad_x3, _cache64_ssse3 );
 768             INIT2( sad_x4, _cache64_ssse3 );
 769         }
 770         if( cpu&X264_CPU_PHADD_IS_FAST )
 771         {
 772             INIT6( satd, _ssse3_phadd );
 773             INIT6( satd_x3, _ssse3_phadd );
 774             INIT6( satd_x4, _ssse3_phadd );
 775         }
 776     }
 777
 778     if( cpu&X264_CPU_SSE4 )
 779     {
 780         pixf->ssd[PIXEL_4x8] = x264_pixel_ssd_4x8_sse4;
 781         pixf->ssd[PIXEL_4x4] = x264_pixel_ssd_4x4_sse4;
 782     }
 783 #endif //HAVE_MMX
 784
 785 #ifdef ARCH_PPC
 786     if( cpu&X264_CPU_ALTIVEC )
 787     {
 788         x264_pixel_altivec_init( pixf );
 789     }
 790 #endif
 791 #ifdef ARCH_UltraSparc
 792     INIT4( sad, _vis );
 793     INIT4( sad_x3, _vis );
 794     INIT4( sad_x4, _vis );
 795 #endif
 796
 797     pixf->ads[PIXEL_8x16] =
 798     pixf->ads[PIXEL_8x4] =
 799     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
 800     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
 801 }
 802