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

   1 /*****************************************************************************
   2  * pixel.c: pixel metrics
   3  *****************************************************************************
   4  * Copyright (C) 2003-2014 x264 project
   5  *
   6  * Authors: Loren Merritt <lorenm@u.washington.edu>
   7  *          Laurent Aimar <fenrir@via.ecp.fr>
   8  *          Fiona Glaser <fiona@x264.com>
   9  *
  10  * This program is free software; you can redistribute it and/or modify
  11  * it under the terms of the GNU General Public License as published by
  12  * the Free Software Foundation; either version 2 of the License, or
  13  * (at your option) any later version.
  14  *
  15  * This program is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  18  * GNU General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU General Public License
  21  * along with this program; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  23  *
  24  * This program is also available under a commercial proprietary license.
  25  * For more information, contact us at licensing@x264.com.
  26  *****************************************************************************/
  27
  28 #include "common.h"
  29
  30 #if HAVE_MMX
  31 #   include "x86/pixel.h"
  32 #   include "x86/predict.h"
  33 #endif
  34 #if ARCH_PPC
  35 #   include "ppc/pixel.h"
  36 #endif
  37 #if ARCH_ARM
  38 #   include "arm/pixel.h"
  39 #   include "arm/predict.h"
  40 #endif
  41
  42
  43 /****************************************************************************
  44  * pixel_sad_WxH
  45  ****************************************************************************/
  46 #define PIXEL_SAD_C( name, lx, ly ) \
  47 static int name( pixel *pix1, intptr_t i_stride_pix1,  \
  48                  pixel *pix2, intptr_t i_stride_pix2 ) \
  49 {                                                   \
  50     int i_sum = 0;                                  \
  51     for( int y = 0; y < ly; y++ )                   \
  52     {                                               \
  53         for( int x = 0; x < lx; x++ )               \
  54         {                                           \
  55             i_sum += abs( pix1[x] - pix2[x] );      \
  56         }                                           \
  57         pix1 += i_stride_pix1;                      \
  58         pix2 += i_stride_pix2;                      \
  59     }                                               \
  60     return i_sum;                                   \
  61 }
  62
  63
  64 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  65 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  66 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  67 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  68 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  69 PIXEL_SAD_C( x264_pixel_sad_4x16,   4, 16 )
  70 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  71 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  72
  73 /****************************************************************************
  74  * pixel_ssd_WxH
  75  ****************************************************************************/
  76 #define PIXEL_SSD_C( name, lx, ly ) \
  77 static int name( pixel *pix1, intptr_t i_stride_pix1,  \
  78                  pixel *pix2, intptr_t i_stride_pix2 ) \
  79 {                                                   \
  80     int i_sum = 0;                                  \
  81     for( int y = 0; y < ly; y++ )                   \
  82     {                                               \
  83         for( int x = 0; x < lx; x++ )               \
  84         {                                           \
  85             int d = pix1[x] - pix2[x];              \
  86             i_sum += d*d;                           \
  87         }                                           \
  88         pix1 += i_stride_pix1;                      \
  89         pix2 += i_stride_pix2;                      \
  90     }                                               \
  91     return i_sum;                                   \
  92 }
  93
  94 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
  95 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
  96 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
  97 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
  98 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
  99 PIXEL_SSD_C( x264_pixel_ssd_4x16,   4, 16 )
 100 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
 101 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
 102
 103 uint64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, pixel *pix1, intptr_t i_pix1,
 104                              pixel *pix2, intptr_t i_pix2, int i_width, int i_height )
 105 {
 106     uint64_t i_ssd = 0;
 107     int y;
 108     int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
 109
 110 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 111                                           pix2 + y*i_pix2 + x, i_pix2 );
 112     for( y = 0; y < i_height-15; y += 16 )
 113     {
 114         int x = 0;
 115         if( align )
 116             for( ; x < i_width-15; x += 16 )
 117                 SSD(PIXEL_16x16);
 118         for( ; x < i_width-7; x += 8 )
 119             SSD(PIXEL_8x16);
 120     }
 121     if( y < i_height-7 )
 122         for( int x = 0; x < i_width-7; x += 8 )
 123             SSD(PIXEL_8x8);
 124 #undef SSD
 125
 126 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 127     if( i_width & 7 )
 128     {
 129         for( y = 0; y < (i_height & ~7); y++ )
 130             for( int x = i_width & ~7; x < i_width; x++ )
 131                 SSD1;
 132     }
 133     if( i_height & 7 )
 134     {
 135         for( y = i_height & ~7; y < i_height; y++ )
 136             for( int x = 0; x < i_width; x++ )
 137                 SSD1;
 138     }
 139 #undef SSD1
 140
 141     return i_ssd;
 142 }
 143
 144 static void pixel_ssd_nv12_core( pixel *pixuv1, intptr_t stride1, pixel *pixuv2, intptr_t stride2,
 145                                  int width, int height, uint64_t *ssd_u, uint64_t *ssd_v )
 146 {
 147     *ssd_u = 0, *ssd_v = 0;
 148     for( int y = 0; y < height; y++, pixuv1+=stride1, pixuv2+=stride2 )
 149         for( int x = 0; x < width; x++ )
 150         {
 151             int du = pixuv1[2*x]   - pixuv2[2*x];
 152             int dv = pixuv1[2*x+1] - pixuv2[2*x+1];
 153             *ssd_u += du*du;
 154             *ssd_v += dv*dv;
 155         }
 156 }
 157
 158 void x264_pixel_ssd_nv12( x264_pixel_function_t *pf, pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2,
 159                           int i_width, int i_height, uint64_t *ssd_u, uint64_t *ssd_v )
 160 {
 161     pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height, ssd_u, ssd_v );
 162     if( i_width&7 )
 163     {
 164         uint64_t tmp[2];
 165         pixel_ssd_nv12_core( pix1+(i_width&~7), i_pix1, pix2+(i_width&~7), i_pix2, i_width&7, i_height, &tmp[0], &tmp[1] );
 166         *ssd_u += tmp[0];
 167         *ssd_v += tmp[1];
 168     }
 169 }
 170
 171 /****************************************************************************
 172  * pixel_var_wxh
 173  ****************************************************************************/
 174 #define PIXEL_VAR_C( name, w, h ) \
 175 static uint64_t name( pixel *pix, intptr_t i_stride ) \
 176 {                                             \
 177     uint32_t sum = 0, sqr = 0;                \
 178     for( int y = 0; y < h; y++ )              \
 179     {                                         \
 180         for( int x = 0; x < w; x++ )          \
 181         {                                     \
 182             sum += pix[x];                    \
 183             sqr += pix[x] * pix[x];           \
 184         }                                     \
 185         pix += i_stride;                      \
 186     }                                         \
 187     return sum + ((uint64_t)sqr << 32);       \
 188 }
 189
 190 PIXEL_VAR_C( x264_pixel_var_16x16, 16, 16 )
 191 PIXEL_VAR_C( x264_pixel_var_8x16,   8, 16 )
 192 PIXEL_VAR_C( x264_pixel_var_8x8,    8,  8 )
 193
 194 /****************************************************************************
 195  * pixel_var2_wxh
 196  ****************************************************************************/
 197 #define PIXEL_VAR2_C( name, w, h, shift ) \
 198 static int name( pixel *pix1, intptr_t i_stride1, pixel *pix2, intptr_t i_stride2, int *ssd ) \
 199 { \
 200     int var = 0, sum = 0, sqr = 0; \
 201     for( int y = 0; y < h; y++ ) \
 202     { \
 203         for( int x = 0; x < w; x++ ) \
 204         { \
 205             int diff = pix1[x] - pix2[x]; \
 206             sum += diff; \
 207             sqr += diff * diff; \
 208         } \
 209         pix1 += i_stride1; \
 210         pix2 += i_stride2; \
 211     } \
 212     var = sqr - ((int64_t)sum * sum >> shift); \
 213     *ssd = sqr; \
 214     return var; \
 215 }
 216
 217 PIXEL_VAR2_C( x264_pixel_var2_8x16, 8, 16, 7 )
 218 PIXEL_VAR2_C( x264_pixel_var2_8x8,  8,  8, 6 )
 219
 220 #if BIT_DEPTH > 8
 221     typedef uint32_t sum_t;
 222     typedef uint64_t sum2_t;
 223 #else
 224     typedef uint16_t sum_t;
 225     typedef uint32_t sum2_t;
 226 #endif
 227 #define BITS_PER_SUM (8 * sizeof(sum_t))
 228
 229 #define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
 230     sum2_t t0 = s0 + s1;\
 231     sum2_t t1 = s0 - s1;\
 232     sum2_t t2 = s2 + s3;\
 233     sum2_t t3 = s2 - s3;\
 234     d0 = t0 + t2;\
 235     d2 = t0 - t2;\
 236     d1 = t1 + t3;\
 237     d3 = t1 - t3;\
 238 }
 239
 240 // in: a pseudo-simd number of the form x+(y<<16)
 241 // return: abs(x)+(abs(y)<<16)
 242 static ALWAYS_INLINE sum2_t abs2( sum2_t a )
 243 {
 244     sum2_t s = ((a>>(BITS_PER_SUM-1))&(((sum2_t)1<<BITS_PER_SUM)+1))*((sum_t)-1);
 245     return (a+s)^s;
 246 }
 247
 248 /****************************************************************************
 249  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 250  ****************************************************************************/
 251
 252 static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 253 {
 254     sum2_t tmp[4][2];
 255     sum2_t a0, a1, a2, a3, b0, b1;
 256     sum2_t sum = 0;
 257     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 258     {
 259         a0 = pix1[0] - pix2[0];
 260         a1 = pix1[1] - pix2[1];
 261         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 262         a2 = pix1[2] - pix2[2];
 263         a3 = pix1[3] - pix2[3];
 264         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 265         tmp[i][0] = b0 + b1;
 266         tmp[i][1] = b0 - b1;
 267     }
 268     for( int i = 0; i < 2; i++ )
 269     {
 270         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 271         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 272         sum += ((sum_t)a0) + (a0>>BITS_PER_SUM);
 273     }
 274     return sum >> 1;
 275 }
 276
 277 static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 278 {
 279     sum2_t tmp[4][4];
 280     sum2_t a0, a1, a2, a3;
 281     sum2_t sum = 0;
 282     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 283     {
 284         a0 = (pix1[0] - pix2[0]) + ((sum2_t)(pix1[4] - pix2[4]) << BITS_PER_SUM);
 285         a1 = (pix1[1] - pix2[1]) + ((sum2_t)(pix1[5] - pix2[5]) << BITS_PER_SUM);
 286         a2 = (pix1[2] - pix2[2]) + ((sum2_t)(pix1[6] - pix2[6]) << BITS_PER_SUM);
 287         a3 = (pix1[3] - pix2[3]) + ((sum2_t)(pix1[7] - pix2[7]) << BITS_PER_SUM);
 288         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 289     }
 290     for( int i = 0; i < 4; i++ )
 291     {
 292         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 293         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 294     }
 295     return (((sum_t)sum) + (sum>>BITS_PER_SUM)) >> 1;
 296 }
 297
 298 #define PIXEL_SATD_C( w, h, sub )\
 299 static int x264_pixel_satd_##w##x##h( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )\
 300 {\
 301     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 302             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 303     if( w==16 )\
 304         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 305             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 306     if( h==16 )\
 307         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 308             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 309     if( w==16 && h==16 )\
 310         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 311             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 312     return sum;\
 313 }
 314 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 315 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 316 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 317 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 318 PIXEL_SATD_C( 4,  16, x264_pixel_satd_4x4 )
 319 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 320
 321 static NOINLINE int sa8d_8x8( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 322 {
 323     sum2_t tmp[8][4];
 324     sum2_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
 325     sum2_t sum = 0;
 326     for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
 327     {
 328         a0 = pix1[0] - pix2[0];
 329         a1 = pix1[1] - pix2[1];
 330         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 331         a2 = pix1[2] - pix2[2];
 332         a3 = pix1[3] - pix2[3];
 333         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 334         a4 = pix1[4] - pix2[4];
 335         a5 = pix1[5] - pix2[5];
 336         b2 = (a4+a5) + ((a4-a5)<<BITS_PER_SUM);
 337         a6 = pix1[6] - pix2[6];
 338         a7 = pix1[7] - pix2[7];
 339         b3 = (a6+a7) + ((a6-a7)<<BITS_PER_SUM);
 340         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 341     }
 342     for( int i = 0; i < 4; i++ )
 343     {
 344         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 345         HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 346         b0  = abs2(a0+a4) + abs2(a0-a4);
 347         b0 += abs2(a1+a5) + abs2(a1-a5);
 348         b0 += abs2(a2+a6) + abs2(a2-a6);
 349         b0 += abs2(a3+a7) + abs2(a3-a7);
 350         sum += (sum_t)b0 + (b0>>BITS_PER_SUM);
 351     }
 352     return sum;
 353 }
 354
 355 static int x264_pixel_sa8d_8x8( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 356 {
 357     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
 358     return (sum+2)>>2;
 359 }
 360
 361 static int x264_pixel_sa8d_16x16( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 362 {
 363     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 364             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 365             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 366             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 367     return (sum+2)>>2;
 368 }
 369
 370 static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, intptr_t stride )
 371 {
 372     sum2_t tmp[32];
 373     sum2_t a0, a1, a2, a3, dc;
 374     sum2_t sum4 = 0, sum8 = 0;
 375     for( int i = 0; i < 8; i++, pix+=stride )
 376     {
 377         sum2_t *t = tmp + (i&3) + (i&4)*4;
 378         a0 = (pix[0]+pix[1]) + ((sum2_t)(pix[0]-pix[1])<<BITS_PER_SUM);
 379         a1 = (pix[2]+pix[3]) + ((sum2_t)(pix[2]-pix[3])<<BITS_PER_SUM);
 380         t[0] = a0 + a1;
 381         t[4] = a0 - a1;
 382         a2 = (pix[4]+pix[5]) + ((sum2_t)(pix[4]-pix[5])<<BITS_PER_SUM);
 383         a3 = (pix[6]+pix[7]) + ((sum2_t)(pix[6]-pix[7])<<BITS_PER_SUM);
 384         t[8] = a2 + a3;
 385         t[12] = a2 - a3;
 386     }
 387     for( int i = 0; i < 8; i++ )
 388     {
 389         HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 390         tmp[i*4+0] = a0;
 391         tmp[i*4+1] = a1;
 392         tmp[i*4+2] = a2;
 393         tmp[i*4+3] = a3;
 394         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 395     }
 396     for( int i = 0; i < 8; i++ )
 397     {
 398         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 399         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 400     }
 401     dc = (sum_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 402     sum4 = (sum_t)sum4 + (sum4>>BITS_PER_SUM) - dc;
 403     sum8 = (sum_t)sum8 + (sum8>>BITS_PER_SUM) - dc;
 404     return ((uint64_t)sum8<<32) + sum4;
 405 }
 406
 407 #define HADAMARD_AC(w,h) \
 408 static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, intptr_t stride )\
 409 {\
 410     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 411     if( w==16 )\
 412         sum += pixel_hadamard_ac( pix+8, stride );\
 413     if( h==16 )\
 414         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 415     if( w==16 && h==16 )\
 416         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 417     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 418 }
 419 HADAMARD_AC( 16, 16 )
 420 HADAMARD_AC( 16, 8 )
 421 HADAMARD_AC( 8, 16 )
 422 HADAMARD_AC( 8, 8 )
 423
 424
 425 /****************************************************************************
 426  * pixel_sad_x4
 427  ****************************************************************************/
 428 #define SAD_X( size ) \
 429 static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 430                                       intptr_t i_stride, int scores[3] )\
 431 {\
 432     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 433     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 434     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 435 }\
 436 static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1,pixel *pix2, pixel *pix3,\
 437                                       intptr_t i_stride, int scores[4] )\
 438 {\
 439     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 440     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 441     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 442     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 443 }
 444
 445 SAD_X( 16x16 )
 446 SAD_X( 16x8 )
 447 SAD_X( 8x16 )
 448 SAD_X( 8x8 )
 449 SAD_X( 8x4 )
 450 SAD_X( 4x8 )
 451 SAD_X( 4x4 )
 452
 453 /****************************************************************************
 454  * pixel_satd_x4
 455  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 456  ****************************************************************************/
 457
 458 #define SATD_X( size, cpu ) \
 459 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 460                                             intptr_t i_stride, int scores[3] )\
 461 {\
 462     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 463     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 464     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 465 }\
 466 static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3,\
 467                                             intptr_t i_stride, int scores[4] )\
 468 {\
 469     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 470     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 471     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 472     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 473 }
 474 #define SATD_X_DECL6( cpu )\
 475 SATD_X( 16x16, cpu )\
 476 SATD_X( 16x8, cpu )\
 477 SATD_X( 8x16, cpu )\
 478 SATD_X( 8x8, cpu )\
 479 SATD_X( 8x4, cpu )\
 480 SATD_X( 4x8, cpu )
 481 #define SATD_X_DECL7( cpu )\
 482 SATD_X_DECL6( cpu )\
 483 SATD_X( 4x4, cpu )
 484
 485 SATD_X_DECL7()
 486 #if HAVE_MMX
 487 SATD_X_DECL7( _mmx2 )
 488 #if !HIGH_BIT_DEPTH
 489 SATD_X_DECL6( _sse2 )
 490 SATD_X_DECL7( _ssse3 )
 491 SATD_X_DECL6( _ssse3_atom )
 492 SATD_X_DECL7( _sse4 )
 493 SATD_X_DECL7( _avx )
 494 SATD_X_DECL7( _xop )
 495 #endif // !HIGH_BIT_DEPTH
 496 #endif
 497
 498 #if !HIGH_BIT_DEPTH
 499 #if HAVE_ARMV6
 500 SATD_X_DECL7( _neon )
 501 #endif
 502 #endif // !HIGH_BIT_DEPTH
 503
 504 #define INTRA_MBCMP_8x8( mbcmp, cpu, cpu2 )\
 505 void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[36], int res[3] )\
 506 {\
 507     ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
 508     x264_predict_8x8_v##cpu2( pix, edge );\
 509     res[0] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 510     x264_predict_8x8_h##cpu2( pix, edge );\
 511     res[1] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 512     x264_predict_8x8_dc##cpu2( pix, edge );\
 513     res[2] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 514 }
 515
 516 INTRA_MBCMP_8x8( sad,, _c )
 517 INTRA_MBCMP_8x8(sa8d,, _c )
 518 #if HIGH_BIT_DEPTH && HAVE_MMX
 519 #define x264_predict_8x8_v_sse2 x264_predict_8x8_v_sse
 520 INTRA_MBCMP_8x8( sad, _mmx2,  _c )
 521 INTRA_MBCMP_8x8(sa8d, _sse2,  _sse2 )
 522 #endif
 523 #if !HIGH_BIT_DEPTH && HAVE_ARMV6
 524 INTRA_MBCMP_8x8( sad, _neon, _neon )
 525 INTRA_MBCMP_8x8(sa8d, _neon, _neon )
 526 #endif
 527
 528 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu, cpu2 )\
 529 void x264_intra_##mbcmp##_x3_##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
 530 {\
 531     x264_predict_##size##chroma##_##pred1##cpu2( fdec );\
 532     res[0] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 533     x264_predict_##size##chroma##_##pred2##cpu2( fdec );\
 534     res[1] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 535     x264_predict_##size##chroma##_##pred3##cpu2( fdec );\
 536     res[2] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 537 }
 538
 539 INTRA_MBCMP( sad,  4x4,   v, h, dc,  ,, _c )
 540 INTRA_MBCMP(satd,  4x4,   v, h, dc,  ,, _c )
 541 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c,, _c )
 542 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c,, _c )
 543 INTRA_MBCMP( sad,  8x16, dc, h,  v, c,, _c )
 544 INTRA_MBCMP(satd,  8x16, dc, h,  v, c,, _c )
 545 INTRA_MBCMP( sad, 16x16,  v, h, dc,  ,, _c )
 546 INTRA_MBCMP(satd, 16x16,  v, h, dc,  ,, _c )
 547
 548 #if HAVE_MMX
 549 #if HIGH_BIT_DEPTH
 550 #define x264_predict_8x8c_v_mmx2 x264_predict_8x8c_v_mmx
 551 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_c
 552 #define x264_predict_8x8c_v_sse2 x264_predict_8x8c_v_sse
 553 #define x264_predict_8x16c_v_sse2 x264_predict_8x16c_v_sse
 554 #define x264_predict_16x16_v_sse2 x264_predict_16x16_v_sse
 555 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _mmx2, _c )
 556 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _mmx2, _mmx2 )
 557 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 558 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 559 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _mmx2, _mmx2 )
 560 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _sse2, _sse2 )
 561 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _sse2 )
 562 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _sse2 )
 563 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _sse2, _sse2 )
 564 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _ssse3, _sse2 )
 565 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 566 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 567 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _ssse3, _sse2 )
 568 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _sse2 )
 569 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _sse2 )
 570 #else
 571 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_mmx
 572 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 573 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 574 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 575 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 576 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _mmx2 )
 577 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _mmx2 )
 578 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _mmx2 )
 579 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _xop, _mmx2 )
 580 #endif
 581 #endif
 582 #if !HIGH_BIT_DEPTH && HAVE_ARMV6
 583 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _neon, _armv6 )
 584 INTRA_MBCMP(satd,  4x4,   v, h, dc,  , _neon, _armv6 )
 585 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _neon, _neon )
 586 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c, _neon, _neon )
 587 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _neon, _c )
 588 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _neon, _c )
 589 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _neon, _neon )
 590 INTRA_MBCMP(satd, 16x16,  v, h, dc,  , _neon, _neon )
 591 #endif
 592
 593 // No C implementation of intra_satd_x9. See checkasm for its behavior,
 594 // or see x264_mb_analyse_intra for the entirely different algorithm we
 595 // use when lacking an asm implementation of it.
 596
 597
 598
 599 /****************************************************************************
 600  * structural similarity metric
 601  ****************************************************************************/
 602 static void ssim_4x4x2_core( const pixel *pix1, intptr_t stride1,
 603                              const pixel *pix2, intptr_t stride2,
 604                              int sums[2][4] )
 605 {
 606     for( int z = 0; z < 2; z++ )
 607     {
 608         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 609         for( int y = 0; y < 4; y++ )
 610             for( int x = 0; x < 4; x++ )
 611             {
 612                 int a = pix1[x+y*stride1];
 613                 int b = pix2[x+y*stride2];
 614                 s1  += a;
 615                 s2  += b;
 616                 ss  += a*a;
 617                 ss  += b*b;
 618                 s12 += a*b;
 619             }
 620         sums[z][0] = s1;
 621         sums[z][1] = s2;
 622         sums[z][2] = ss;
 623         sums[z][3] = s12;
 624         pix1 += 4;
 625         pix2 += 4;
 626     }
 627 }
 628
 629 static float ssim_end1( int s1, int s2, int ss, int s12 )
 630 {
 631 /* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
 632  * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
 633  * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
 634 #if BIT_DEPTH > 9
 635 #define type float
 636     static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
 637     static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
 638 #else
 639 #define type int
 640     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 641     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 642 #endif
 643     type fs1 = s1;
 644     type fs2 = s2;
 645     type fss = ss;
 646     type fs12 = s12;
 647     type vars = fss*64 - fs1*fs1 - fs2*fs2;
 648     type covar = fs12*64 - fs1*fs2;
 649     return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
 650          / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
 651 #undef type
 652 }
 653
 654 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 655 {
 656     float ssim = 0.0;
 657     for( int i = 0; i < width; i++ )
 658         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 659                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 660                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 661                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 662     return ssim;
 663 }
 664
 665 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 666                            pixel *pix1, intptr_t stride1,
 667                            pixel *pix2, intptr_t stride2,
 668                            int width, int height, void *buf, int *cnt )
 669 {
 670     int z = 0;
 671     float ssim = 0.0;
 672     int (*sum0)[4] = buf;
 673     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 674     width >>= 2;
 675     height >>= 2;
 676     for( int y = 1; y < height; y++ )
 677     {
 678         for( ; z <= y; z++ )
 679         {
 680             XCHG( void*, sum0, sum1 );
 681             for( int x = 0; x < width; x+=2 )
 682                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 683         }
 684         for( int x = 0; x < width-1; x += 4 )
 685             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 686     }
 687     *cnt = (height-1) * (width-1);
 688     return ssim;
 689 }
 690
 691 static int pixel_vsad( pixel *src, intptr_t stride, int height )
 692 {
 693     int score = 0;
 694     for( int i = 1; i < height; i++, src += stride )
 695         for( int j = 0; j < 16; j++ )
 696             score += abs(src[j] - src[j+stride]);
 697     return score;
 698 }
 699
 700 int x264_field_vsad( x264_t *h, int mb_x, int mb_y )
 701 {
 702     int score_field, score_frame;
 703     int stride = h->fenc->i_stride[0];
 704     int mb_stride = h->mb.i_mb_stride;
 705     pixel *fenc = h->fenc->plane[0] + 16 * (mb_x + mb_y * stride);
 706     int mb_xy = mb_x + mb_y*mb_stride;
 707
 708     /* We don't want to analyze pixels outside the frame, as it gives inaccurate results. */
 709     int mbpair_height = X264_MIN( h->param.i_height - mb_y * 16, 32 );
 710     score_frame  = h->pixf.vsad( fenc,          stride, mbpair_height );
 711     score_field  = h->pixf.vsad( fenc,        stride*2, mbpair_height >> 1 );
 712     score_field += h->pixf.vsad( fenc+stride, stride*2, mbpair_height >> 1 );
 713
 714     if( mb_x > 0 )
 715         score_field += 512 - h->mb.field[mb_xy        -1]*1024;
 716     if( mb_y > 0 )
 717         score_field += 512 - h->mb.field[mb_xy-mb_stride]*1024;
 718
 719     return (score_field < score_frame);
 720 }
 721
 722 static int pixel_asd8( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2, int height )
 723 {
 724     int sum = 0;
 725     for( int y = 0; y < height; y++, pix1 += stride1, pix2 += stride2 )
 726         for( int x = 0; x < 8; x++ )
 727             sum += pix1[x] - pix2[x];
 728     return abs( sum );
 729 }
 730
 731 /****************************************************************************
 732  * successive elimination
 733  ****************************************************************************/
 734 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 735                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 736 {
 737     int nmv = 0;
 738     for( int i = 0; i < width; i++, sums++ )
 739     {
 740         int ads = abs( enc_dc[0] - sums[0] )
 741                 + abs( enc_dc[1] - sums[8] )
 742                 + abs( enc_dc[2] - sums[delta] )
 743                 + abs( enc_dc[3] - sums[delta+8] )
 744                 + cost_mvx[i];
 745         if( ads < thresh )
 746             mvs[nmv++] = i;
 747     }
 748     return nmv;
 749 }
 750
 751 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 752                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 753 {
 754     int nmv = 0;
 755     for( int i = 0; i < width; i++, sums++ )
 756     {
 757         int ads = abs( enc_dc[0] - sums[0] )
 758                 + abs( enc_dc[1] - sums[delta] )
 759                 + cost_mvx[i];
 760         if( ads < thresh )
 761             mvs[nmv++] = i;
 762     }
 763     return nmv;
 764 }
 765
 766 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 767                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 768 {
 769     int nmv = 0;
 770     for( int i = 0; i<width; i++, sums++ )
 771     {
 772         int ads = abs( enc_dc[0] - sums[0] )
 773                 + cost_mvx[i];
 774         if( ads < thresh )
 775             mvs[nmv++] = i;
 776     }
 777     return nmv;
 778 }
 779
 780
 781 /****************************************************************************
 782  * x264_pixel_init:
 783  ****************************************************************************/
 784 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 785 {
 786     memset( pixf, 0, sizeof(*pixf) );
 787
 788 #define INIT2_NAME( name1, name2, cpu ) \
 789     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 790     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 791 #define INIT4_NAME( name1, name2, cpu ) \
 792     INIT2_NAME( name1, name2, cpu ) \
 793     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 794     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 795 #define INIT5_NAME( name1, name2, cpu ) \
 796     INIT4_NAME( name1, name2, cpu ) \
 797     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 798 #define INIT6_NAME( name1, name2, cpu ) \
 799     INIT5_NAME( name1, name2, cpu ) \
 800     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 801 #define INIT7_NAME( name1, name2, cpu ) \
 802     INIT6_NAME( name1, name2, cpu ) \
 803     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 804 #define INIT8_NAME( name1, name2, cpu ) \
 805     INIT7_NAME( name1, name2, cpu ) \
 806     pixf->name1[PIXEL_4x16]  = x264_pixel_##name2##_4x16##cpu;
 807 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 808 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 809 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 810 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 811 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 812 #define INIT8( name, cpu ) INIT8_NAME( name, name, cpu )
 813
 814 #define INIT_ADS( cpu ) \
 815     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 816     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 817     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 818
 819     INIT8( sad, );
 820     INIT8_NAME( sad_aligned, sad, );
 821     INIT7( sad_x3, );
 822     INIT7( sad_x4, );
 823     INIT8( ssd, );
 824     INIT8( satd, );
 825     INIT7( satd_x3, );
 826     INIT7( satd_x4, );
 827     INIT4( hadamard_ac, );
 828     INIT_ADS( );
 829
 830     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 831     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 832     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 833     pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16;
 834     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 835     pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16;
 836     pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8;
 837
 838     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 839     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 840     pixf->ssim_end4 = ssim_end4;
 841     pixf->vsad = pixel_vsad;
 842     pixf->asd8 = pixel_asd8;
 843
 844     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 845     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 846     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 847     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 848     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 849     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 850     pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c;
 851     pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c;
 852     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 853     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 854
 855 #if HIGH_BIT_DEPTH
 856 #if HAVE_MMX
 857     if( cpu&X264_CPU_MMX2 )
 858     {
 859         INIT7( sad, _mmx2 );
 860         INIT7_NAME( sad_aligned, sad, _mmx2 );
 861         INIT7( sad_x3, _mmx2 );
 862         INIT7( sad_x4, _mmx2 );
 863         INIT8( satd, _mmx2 );
 864         INIT7( satd_x3, _mmx2 );
 865         INIT7( satd_x4, _mmx2 );
 866         INIT4( hadamard_ac, _mmx2 );
 867         INIT8( ssd, _mmx2 );
 868         INIT_ADS( _mmx2 );
 869
 870         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_mmx2;
 871         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
 872         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
 873 #if ARCH_X86
 874         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_mmx2;
 875         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
 876 #endif
 877
 878         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
 879         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
 880         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
 881         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
 882         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
 883         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
 884         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
 885         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
 886         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
 887     }
 888     if( cpu&X264_CPU_SSE2 )
 889     {
 890         INIT4_NAME( sad_aligned, sad, _sse2_aligned );
 891         INIT5( ssd, _sse2 );
 892         INIT6( satd, _sse2 );
 893         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse2;
 894
 895         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 896         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 897 #if ARCH_X86_64
 898         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 899         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
 900 #endif
 901         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_sse2;
 902         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_sse2;
 903         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 904         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 905         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 906         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_sse2;
 907         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_sse2;
 908         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_sse2;
 909         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_sse2;
 910     }
 911     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 912     {
 913         INIT5( sad, _sse2 );
 914         INIT2( sad_x3, _sse2 );
 915         INIT2( sad_x4, _sse2 );
 916         INIT_ADS( _sse2 );
 917
 918         if( !(cpu&X264_CPU_STACK_MOD4) )
 919         {
 920             INIT4( hadamard_ac, _sse2 );
 921         }
 922         pixf->vsad = x264_pixel_vsad_sse2;
 923         pixf->asd8 = x264_pixel_asd8_sse2;
 924         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_sse2;
 925         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_sse2;
 926         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
 927         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
 928         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_sse2;
 929     }
 930     if( cpu&X264_CPU_SSE2_IS_FAST )
 931     {
 932         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 933         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 934         pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_sse2;
 935         pixf->sad_x3[PIXEL_8x4]  = x264_pixel_sad_x3_8x4_sse2;
 936         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 937         pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_sse2;
 938         pixf->sad_x4[PIXEL_8x4]  = x264_pixel_sad_x4_8x4_sse2;
 939     }
 940     if( cpu&X264_CPU_SSSE3 )
 941     {
 942         INIT4_NAME( sad_aligned, sad, _ssse3_aligned );
 943         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_ssse3;
 944         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_ssse3;
 945         INIT7( sad, _ssse3 );
 946         INIT7( sad_x3, _ssse3 );
 947         INIT7( sad_x4, _ssse3 );
 948         INIT_ADS( _ssse3 );
 949         INIT6( satd, _ssse3 );
 950         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_ssse3;
 951
 952         if( !(cpu&X264_CPU_STACK_MOD4) )
 953         {
 954             INIT4( hadamard_ac, _ssse3 );
 955         }
 956         pixf->vsad = x264_pixel_vsad_ssse3;
 957         pixf->asd8 = x264_pixel_asd8_ssse3;
 958         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_ssse3;
 959         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 960         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 961 #if ARCH_X86_64
 962         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
 963 #endif
 964         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_ssse3;
 965         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_ssse3;
 966         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 967         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_ssse3;
 968         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
 969         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 970     }
 971     if( cpu&X264_CPU_SSE4 )
 972     {
 973         INIT6( satd, _sse4 );
 974         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse4;
 975         if( !(cpu&X264_CPU_STACK_MOD4) )
 976         {
 977             INIT4( hadamard_ac, _sse4 );
 978         }
 979         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 980         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 981 #if ARCH_X86_64
 982         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
 983 #endif
 984         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
 985     }
 986     if( cpu&X264_CPU_AVX )
 987     {
 988         INIT5_NAME( sad_aligned, sad, _ssse3 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
 989         INIT_ADS( _avx );
 990         INIT6( satd, _avx );
 991         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_avx;
 992         if( !(cpu&X264_CPU_STACK_MOD4) )
 993         {
 994             INIT4( hadamard_ac, _avx );
 995         }
 996         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_avx;
 997         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
 998         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
 999         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1000         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1001         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1002         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1003         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1004 #if ARCH_X86_64
1005         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1006 #endif
1007         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1008     }
1009     if( cpu&X264_CPU_XOP )
1010     {
1011         INIT5( sad_x3, _xop );
1012         INIT5( sad_x4, _xop );
1013         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_xop;
1014         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1015         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1016         pixf->vsad = x264_pixel_vsad_xop;
1017         pixf->asd8 = x264_pixel_asd8_xop;
1018 #if ARCH_X86_64
1019         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1020 #endif
1021     }
1022     if( cpu&X264_CPU_AVX2 )
1023     {
1024         INIT2( ssd, _avx2 );
1025         INIT2( sad, _avx2 );
1026         INIT2_NAME( sad_aligned, sad, _avx2 );
1027         INIT2( sad_x3, _avx2 );
1028         INIT2( sad_x4, _avx2 );
1029         pixf->vsad = x264_pixel_vsad_avx2;
1030         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1031         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_avx2;
1032     }
1033 #endif // HAVE_MMX
1034 #else // !HIGH_BIT_DEPTH
1035 #if HAVE_MMX
1036     if( cpu&X264_CPU_MMX )
1037     {
1038         INIT8( ssd, _mmx );
1039     }
1040
1041     if( cpu&X264_CPU_MMX2 )
1042     {
1043         INIT8( sad, _mmx2 );
1044         INIT8_NAME( sad_aligned, sad, _mmx2 );
1045         INIT7( sad_x3, _mmx2 );
1046         INIT7( sad_x4, _mmx2 );
1047         INIT8( satd, _mmx2 );
1048         INIT7( satd_x3, _mmx2 );
1049         INIT7( satd_x4, _mmx2 );
1050         INIT4( hadamard_ac, _mmx2 );
1051         INIT_ADS( _mmx2 );
1052         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
1053         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_mmx2;
1054         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
1055         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmx2;
1056 #if ARCH_X86
1057         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmx2;
1058         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmx2;
1059         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmx2;
1060         pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_mmx2;
1061         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_mmx2;
1062         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
1063         pixf->vsad = x264_pixel_vsad_mmx2;
1064
1065         if( cpu&X264_CPU_CACHELINE_32 )
1066         {
1067             INIT5( sad, _cache32_mmx2 );
1068             INIT4( sad_x3, _cache32_mmx2 );
1069             INIT4( sad_x4, _cache32_mmx2 );
1070         }
1071         else if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1072         {
1073             INIT5( sad, _cache64_mmx2 );
1074             INIT4( sad_x3, _cache64_mmx2 );
1075             INIT4( sad_x4, _cache64_mmx2 );
1076         }
1077 #else
1078         if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1079         {
1080             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmx2;
1081             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmx2;
1082             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmx2;
1083             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmx2;
1084             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmx2;
1085             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmx2;
1086             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmx2;
1087         }
1088 #endif
1089         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
1090         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
1091         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
1092         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
1093         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
1094         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
1095         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
1096         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
1097         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
1098     }
1099
1100     if( cpu&X264_CPU_SSE2 )
1101     {
1102         INIT5( ssd, _sse2slow );
1103         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
1104         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
1105         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
1106         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
1107         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
1108         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
1109         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
1110 #if ARCH_X86_64
1111         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
1112         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
1113 #endif
1114         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_sse2;
1115         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_sse2;
1116         pixf->vsad = x264_pixel_vsad_sse2;
1117         pixf->asd8 = x264_pixel_asd8_sse2;
1118     }
1119
1120     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
1121     {
1122         INIT2( sad, _sse2 );
1123         INIT2( sad_x3, _sse2 );
1124         INIT2( sad_x4, _sse2 );
1125         INIT6( satd, _sse2 );
1126         pixf->satd[PIXEL_4x16]   = x264_pixel_satd_4x16_sse2;
1127         INIT6( satd_x3, _sse2 );
1128         INIT6( satd_x4, _sse2 );
1129         INIT4( hadamard_ac, _sse2 );
1130         INIT_ADS( _sse2 );
1131         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
1132         pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_sse2;
1133         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
1134         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
1135         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
1136         if( cpu&X264_CPU_CACHELINE_64 )
1137         {
1138             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
1139 #if ARCH_X86
1140             INIT2( sad, _cache64_sse2 );
1141             INIT2( sad_x3, _cache64_sse2 );
1142             INIT2( sad_x4, _cache64_sse2 );
1143 #endif
1144            if( cpu&X264_CPU_SSE2_IS_FAST )
1145            {
1146                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
1147                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
1148            }
1149         }
1150     }
1151
1152     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
1153     {
1154         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1155         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1156         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
1157         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
1158         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
1159         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
1160         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
1161         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
1162     }
1163
1164     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
1165     {
1166         INIT2( sad, _sse3 );
1167         INIT2( sad_x3, _sse3 );
1168         INIT2( sad_x4, _sse3 );
1169     }
1170
1171     if( cpu&X264_CPU_SSSE3 )
1172     {
1173         INIT4( hadamard_ac, _ssse3 );
1174         if( !(cpu&X264_CPU_STACK_MOD4) )
1175         {
1176             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_ssse3;
1177             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_ssse3;
1178             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_ssse3;
1179 #if ARCH_X86_64
1180             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_ssse3;
1181 #endif
1182         }
1183         INIT_ADS( _ssse3 );
1184         if( cpu&X264_CPU_SLOW_ATOM )
1185         {
1186             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3_atom;
1187             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3_atom;
1188             INIT6( satd, _ssse3_atom );
1189             pixf->satd[PIXEL_4x16]  = x264_pixel_satd_4x16_ssse3_atom;
1190             INIT6( satd_x3, _ssse3_atom );
1191             INIT6( satd_x4, _ssse3_atom );
1192             INIT4( hadamard_ac, _ssse3_atom );
1193 #if ARCH_X86_64
1194             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3_atom;
1195 #endif
1196         }
1197         else
1198         {
1199             INIT8( ssd, _ssse3 );
1200             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
1201             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
1202             INIT8( satd, _ssse3 );
1203             INIT7( satd_x3, _ssse3 );
1204             INIT7( satd_x4, _ssse3 );
1205 #if ARCH_X86_64
1206             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
1207 #endif
1208         }
1209         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
1210         if( !(cpu&X264_CPU_SLOW_PSHUFB) )
1211             pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
1212         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
1213         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
1214         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
1215         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_ssse3;
1216         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_ssse3;
1217         pixf->asd8 = x264_pixel_asd8_ssse3;
1218         if( cpu&X264_CPU_CACHELINE_64 )
1219         {
1220             INIT2( sad, _cache64_ssse3 );
1221             INIT2( sad_x3, _cache64_ssse3 );
1222             INIT2( sad_x4, _cache64_ssse3 );
1223         }
1224         else
1225         {
1226             INIT2( sad_x3, _ssse3 );
1227             INIT5( sad_x4, _ssse3 );
1228         }
1229         if( (cpu&X264_CPU_SLOW_ATOM) || (cpu&X264_CPU_SLOW_SHUFFLE) )
1230         {
1231             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
1232         }
1233     }
1234
1235     if( cpu&X264_CPU_SSE4 )
1236     {
1237         INIT8( satd, _sse4 );
1238         INIT7( satd_x3, _sse4 );
1239         INIT7( satd_x4, _sse4 );
1240         INIT4( hadamard_ac, _sse4 );
1241         if( !(cpu&X264_CPU_STACK_MOD4) )
1242         {
1243             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_sse4;
1244             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_sse4;
1245             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_sse4;
1246 #if ARCH_X86_64
1247             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_sse4;
1248 #endif
1249         }
1250         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
1251         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
1252         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
1253 #if ARCH_X86_64
1254         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
1255 #endif
1256     }
1257
1258     if( cpu&X264_CPU_AVX )
1259     {
1260         INIT2_NAME( sad_aligned, sad, _sse2 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
1261         INIT2( sad_x3, _avx );
1262         INIT2( sad_x4, _avx );
1263         INIT8( satd, _avx );
1264         INIT7( satd_x3, _avx );
1265         INIT7( satd_x4, _avx );
1266         INIT_ADS( _avx );
1267         INIT4( hadamard_ac, _avx );
1268         if( !(cpu&X264_CPU_STACK_MOD4) )
1269         {
1270             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_avx;
1271             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_avx;
1272             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx;
1273 #if ARCH_X86_64
1274             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_avx;
1275 #endif
1276         }
1277         INIT5( ssd, _avx );
1278         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1279         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1280         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1281         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1282         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1283         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_avx;
1284         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1285         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1286         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1287 #if ARCH_X86_64
1288         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1289 #endif
1290     }
1291
1292     if( cpu&X264_CPU_XOP )
1293     {
1294         INIT7( satd, _xop );
1295         INIT7( satd_x3, _xop );
1296         INIT7( satd_x4, _xop );
1297         INIT4( hadamard_ac, _xop );
1298         if( !(cpu&X264_CPU_STACK_MOD4) )
1299         {
1300             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_xop;
1301         }
1302         INIT5( ssd, _xop );
1303         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_xop;
1304         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_xop;
1305         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_xop;
1306         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_xop;
1307         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1308         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_xop;
1309         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1310         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_xop;
1311         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_xop;
1312 #if ARCH_X86_64
1313         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1314 #endif
1315     }
1316
1317     if( cpu&X264_CPU_AVX2 )
1318     {
1319         INIT2( ssd, _avx2 );
1320         INIT2( sad_x3, _avx2 );
1321         INIT2( sad_x4, _avx2 );
1322         INIT4( satd, _avx2 );
1323         INIT2( hadamard_ac, _avx2 );
1324         INIT_ADS( _avx2 );
1325         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx2;
1326         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx2;
1327         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_avx2;
1328         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_avx2;
1329         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_avx2;
1330         pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx2;
1331         pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_avx2;
1332         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1333 #if ARCH_X86_64
1334         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx2;
1335 #endif
1336     }
1337 #endif //HAVE_MMX
1338
1339 #if HAVE_ARMV6
1340     if( cpu&X264_CPU_ARMV6 )
1341     {
1342         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1343         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1344         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1345         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1346     }
1347     if( cpu&X264_CPU_NEON )
1348     {
1349         INIT5( sad, _neon );
1350         INIT5( sad_aligned, _neon );
1351         INIT7( sad_x3, _neon );
1352         INIT7( sad_x4, _neon );
1353         INIT7( ssd, _neon );
1354         INIT7( satd, _neon );
1355         INIT7( satd_x3, _neon );
1356         INIT7( satd_x4, _neon );
1357         INIT4( hadamard_ac, _neon );
1358         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1359         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1360         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1361         pixf->var[PIXEL_8x16]   = x264_pixel_var_8x16_neon;
1362         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1363         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
1364         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
1365
1366         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_neon;
1367         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_neon;
1368         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_neon;
1369         pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8_neon;
1370         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_neon;
1371         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_neon;
1372         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_neon;
1373         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
1374         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_neon;
1375         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
1376
1377         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1378         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1379
1380         if( cpu&X264_CPU_FAST_NEON_MRC )
1381         {
1382             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
1383             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
1384             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
1385             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
1386         }
1387         else    // really just scheduled for dual issue / A8
1388         {
1389             INIT5( sad_aligned, _neon_dual );
1390         }
1391     }
1392 #endif
1393 #endif // HIGH_BIT_DEPTH
1394 #if HAVE_ALTIVEC
1395     if( cpu&X264_CPU_ALTIVEC )
1396     {
1397         x264_pixel_altivec_init( pixf );
1398     }
1399 #endif
1400
1401     pixf->ads[PIXEL_8x16] =
1402     pixf->ads[PIXEL_8x4] =
1403     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
1404     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
1405 }
1406