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     uint32_t 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     sum = abs(sum); \
 213     var = sqr - ((uint64_t)sum * sum >> shift); \
 214     *ssd = sqr; \
 215     return var; \
 216 }
 217
 218 PIXEL_VAR2_C( x264_pixel_var2_8x16, 8, 16, 7 )
 219 PIXEL_VAR2_C( x264_pixel_var2_8x8,  8,  8, 6 )
 220
 221 #if BIT_DEPTH > 8
 222     typedef uint32_t sum_t;
 223     typedef uint64_t sum2_t;
 224 #else
 225     typedef uint16_t sum_t;
 226     typedef uint32_t sum2_t;
 227 #endif
 228 #define BITS_PER_SUM (8 * sizeof(sum_t))
 229
 230 #define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
 231     sum2_t t0 = s0 + s1;\
 232     sum2_t t1 = s0 - s1;\
 233     sum2_t t2 = s2 + s3;\
 234     sum2_t t3 = s2 - s3;\
 235     d0 = t0 + t2;\
 236     d2 = t0 - t2;\
 237     d1 = t1 + t3;\
 238     d3 = t1 - t3;\
 239 }
 240
 241 // in: a pseudo-simd number of the form x+(y<<16)
 242 // return: abs(x)+(abs(y)<<16)
 243 static ALWAYS_INLINE sum2_t abs2( sum2_t a )
 244 {
 245     sum2_t s = ((a>>(BITS_PER_SUM-1))&(((sum2_t)1<<BITS_PER_SUM)+1))*((sum_t)-1);
 246     return (a+s)^s;
 247 }
 248
 249 /****************************************************************************
 250  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 251  ****************************************************************************/
 252
 253 static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 254 {
 255     sum2_t tmp[4][2];
 256     sum2_t a0, a1, a2, a3, b0, b1;
 257     sum2_t sum = 0;
 258     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 259     {
 260         a0 = pix1[0] - pix2[0];
 261         a1 = pix1[1] - pix2[1];
 262         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 263         a2 = pix1[2] - pix2[2];
 264         a3 = pix1[3] - pix2[3];
 265         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 266         tmp[i][0] = b0 + b1;
 267         tmp[i][1] = b0 - b1;
 268     }
 269     for( int i = 0; i < 2; i++ )
 270     {
 271         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 272         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 273         sum += ((sum_t)a0) + (a0>>BITS_PER_SUM);
 274     }
 275     return sum >> 1;
 276 }
 277
 278 static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 279 {
 280     sum2_t tmp[4][4];
 281     sum2_t a0, a1, a2, a3;
 282     sum2_t sum = 0;
 283     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 284     {
 285         a0 = (pix1[0] - pix2[0]) + ((sum2_t)(pix1[4] - pix2[4]) << BITS_PER_SUM);
 286         a1 = (pix1[1] - pix2[1]) + ((sum2_t)(pix1[5] - pix2[5]) << BITS_PER_SUM);
 287         a2 = (pix1[2] - pix2[2]) + ((sum2_t)(pix1[6] - pix2[6]) << BITS_PER_SUM);
 288         a3 = (pix1[3] - pix2[3]) + ((sum2_t)(pix1[7] - pix2[7]) << BITS_PER_SUM);
 289         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 290     }
 291     for( int i = 0; i < 4; i++ )
 292     {
 293         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 294         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 295     }
 296     return (((sum_t)sum) + (sum>>BITS_PER_SUM)) >> 1;
 297 }
 298
 299 #define PIXEL_SATD_C( w, h, sub )\
 300 static int x264_pixel_satd_##w##x##h( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )\
 301 {\
 302     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 303             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 304     if( w==16 )\
 305         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 306             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 307     if( h==16 )\
 308         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 309             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 310     if( w==16 && h==16 )\
 311         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 312             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 313     return sum;\
 314 }
 315 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 316 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 317 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 318 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 319 PIXEL_SATD_C( 4,  16, x264_pixel_satd_4x4 )
 320 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 321
 322 static NOINLINE int sa8d_8x8( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 323 {
 324     sum2_t tmp[8][4];
 325     sum2_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
 326     sum2_t sum = 0;
 327     for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
 328     {
 329         a0 = pix1[0] - pix2[0];
 330         a1 = pix1[1] - pix2[1];
 331         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 332         a2 = pix1[2] - pix2[2];
 333         a3 = pix1[3] - pix2[3];
 334         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 335         a4 = pix1[4] - pix2[4];
 336         a5 = pix1[5] - pix2[5];
 337         b2 = (a4+a5) + ((a4-a5)<<BITS_PER_SUM);
 338         a6 = pix1[6] - pix2[6];
 339         a7 = pix1[7] - pix2[7];
 340         b3 = (a6+a7) + ((a6-a7)<<BITS_PER_SUM);
 341         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 342     }
 343     for( int i = 0; i < 4; i++ )
 344     {
 345         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 346         HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 347         b0  = abs2(a0+a4) + abs2(a0-a4);
 348         b0 += abs2(a1+a5) + abs2(a1-a5);
 349         b0 += abs2(a2+a6) + abs2(a2-a6);
 350         b0 += abs2(a3+a7) + abs2(a3-a7);
 351         sum += (sum_t)b0 + (b0>>BITS_PER_SUM);
 352     }
 353     return sum;
 354 }
 355
 356 static int x264_pixel_sa8d_8x8( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 357 {
 358     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
 359     return (sum+2)>>2;
 360 }
 361
 362 static int x264_pixel_sa8d_16x16( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 363 {
 364     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 365             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 366             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 367             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 368     return (sum+2)>>2;
 369 }
 370
 371 static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, intptr_t stride )
 372 {
 373     sum2_t tmp[32];
 374     sum2_t a0, a1, a2, a3, dc;
 375     sum2_t sum4 = 0, sum8 = 0;
 376     for( int i = 0; i < 8; i++, pix+=stride )
 377     {
 378         sum2_t *t = tmp + (i&3) + (i&4)*4;
 379         a0 = (pix[0]+pix[1]) + ((sum2_t)(pix[0]-pix[1])<<BITS_PER_SUM);
 380         a1 = (pix[2]+pix[3]) + ((sum2_t)(pix[2]-pix[3])<<BITS_PER_SUM);
 381         t[0] = a0 + a1;
 382         t[4] = a0 - a1;
 383         a2 = (pix[4]+pix[5]) + ((sum2_t)(pix[4]-pix[5])<<BITS_PER_SUM);
 384         a3 = (pix[6]+pix[7]) + ((sum2_t)(pix[6]-pix[7])<<BITS_PER_SUM);
 385         t[8] = a2 + a3;
 386         t[12] = a2 - a3;
 387     }
 388     for( int i = 0; i < 8; i++ )
 389     {
 390         HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 391         tmp[i*4+0] = a0;
 392         tmp[i*4+1] = a1;
 393         tmp[i*4+2] = a2;
 394         tmp[i*4+3] = a3;
 395         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 396     }
 397     for( int i = 0; i < 8; i++ )
 398     {
 399         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 400         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 401     }
 402     dc = (sum_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 403     sum4 = (sum_t)sum4 + (sum4>>BITS_PER_SUM) - dc;
 404     sum8 = (sum_t)sum8 + (sum8>>BITS_PER_SUM) - dc;
 405     return ((uint64_t)sum8<<32) + sum4;
 406 }
 407
 408 #define HADAMARD_AC(w,h) \
 409 static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, intptr_t stride )\
 410 {\
 411     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 412     if( w==16 )\
 413         sum += pixel_hadamard_ac( pix+8, stride );\
 414     if( h==16 )\
 415         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 416     if( w==16 && h==16 )\
 417         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 418     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 419 }
 420 HADAMARD_AC( 16, 16 )
 421 HADAMARD_AC( 16, 8 )
 422 HADAMARD_AC( 8, 16 )
 423 HADAMARD_AC( 8, 8 )
 424
 425
 426 /****************************************************************************
 427  * pixel_sad_x4
 428  ****************************************************************************/
 429 #define SAD_X( size ) \
 430 static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 431                                       intptr_t i_stride, int scores[3] )\
 432 {\
 433     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 434     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 435     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 436 }\
 437 static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1,pixel *pix2, pixel *pix3,\
 438                                       intptr_t i_stride, int scores[4] )\
 439 {\
 440     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 441     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 442     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 443     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 444 }
 445
 446 SAD_X( 16x16 )
 447 SAD_X( 16x8 )
 448 SAD_X( 8x16 )
 449 SAD_X( 8x8 )
 450 SAD_X( 8x4 )
 451 SAD_X( 4x8 )
 452 SAD_X( 4x4 )
 453
 454 /****************************************************************************
 455  * pixel_satd_x4
 456  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 457  ****************************************************************************/
 458
 459 #define SATD_X( size, cpu ) \
 460 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 461                                             intptr_t i_stride, int scores[3] )\
 462 {\
 463     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 464     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 465     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 466 }\
 467 static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3,\
 468                                             intptr_t i_stride, int scores[4] )\
 469 {\
 470     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 471     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 472     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 473     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 474 }
 475 #define SATD_X_DECL6( cpu )\
 476 SATD_X( 16x16, cpu )\
 477 SATD_X( 16x8, cpu )\
 478 SATD_X( 8x16, cpu )\
 479 SATD_X( 8x8, cpu )\
 480 SATD_X( 8x4, cpu )\
 481 SATD_X( 4x8, cpu )
 482 #define SATD_X_DECL7( cpu )\
 483 SATD_X_DECL6( cpu )\
 484 SATD_X( 4x4, cpu )
 485
 486 SATD_X_DECL7()
 487 #if HAVE_MMX
 488 SATD_X_DECL7( _mmx2 )
 489 #if !HIGH_BIT_DEPTH
 490 SATD_X_DECL6( _sse2 )
 491 SATD_X_DECL7( _ssse3 )
 492 SATD_X_DECL6( _ssse3_atom )
 493 SATD_X_DECL7( _sse4 )
 494 SATD_X_DECL7( _avx )
 495 SATD_X_DECL7( _xop )
 496 #endif // !HIGH_BIT_DEPTH
 497 #endif
 498
 499 #if !HIGH_BIT_DEPTH
 500 #if HAVE_ARMV6
 501 SATD_X_DECL7( _neon )
 502 #endif
 503 #endif // !HIGH_BIT_DEPTH
 504
 505 #define INTRA_MBCMP_8x8( mbcmp, cpu, cpu2 )\
 506 void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[36], int res[3] )\
 507 {\
 508     ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
 509     x264_predict_8x8_v##cpu2( pix, edge );\
 510     res[0] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 511     x264_predict_8x8_h##cpu2( pix, edge );\
 512     res[1] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 513     x264_predict_8x8_dc##cpu2( pix, edge );\
 514     res[2] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 515 }
 516
 517 INTRA_MBCMP_8x8( sad,, _c )
 518 INTRA_MBCMP_8x8(sa8d,, _c )
 519 #if HIGH_BIT_DEPTH && HAVE_MMX
 520 #define x264_predict_8x8_v_sse2 x264_predict_8x8_v_sse
 521 INTRA_MBCMP_8x8( sad, _mmx2,  _c )
 522 INTRA_MBCMP_8x8(sa8d, _sse2,  _sse2 )
 523 #endif
 524 #if !HIGH_BIT_DEPTH && HAVE_ARMV6
 525 INTRA_MBCMP_8x8( sad, _neon, _neon )
 526 INTRA_MBCMP_8x8(sa8d, _neon, _neon )
 527 #endif
 528
 529 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu, cpu2 )\
 530 void x264_intra_##mbcmp##_x3_##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
 531 {\
 532     x264_predict_##size##chroma##_##pred1##cpu2( fdec );\
 533     res[0] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 534     x264_predict_##size##chroma##_##pred2##cpu2( fdec );\
 535     res[1] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 536     x264_predict_##size##chroma##_##pred3##cpu2( fdec );\
 537     res[2] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 538 }
 539
 540 INTRA_MBCMP( sad,  4x4,   v, h, dc,  ,, _c )
 541 INTRA_MBCMP(satd,  4x4,   v, h, dc,  ,, _c )
 542 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c,, _c )
 543 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c,, _c )
 544 INTRA_MBCMP( sad,  8x16, dc, h,  v, c,, _c )
 545 INTRA_MBCMP(satd,  8x16, dc, h,  v, c,, _c )
 546 INTRA_MBCMP( sad, 16x16,  v, h, dc,  ,, _c )
 547 INTRA_MBCMP(satd, 16x16,  v, h, dc,  ,, _c )
 548
 549 #if HAVE_MMX
 550 #if HIGH_BIT_DEPTH
 551 #define x264_predict_8x8c_v_mmx2 x264_predict_8x8c_v_mmx
 552 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_c
 553 #define x264_predict_8x8c_v_sse2 x264_predict_8x8c_v_sse
 554 #define x264_predict_8x16c_v_sse2 x264_predict_8x16c_v_sse
 555 #define x264_predict_16x16_v_sse2 x264_predict_16x16_v_sse
 556 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _mmx2, _c )
 557 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _mmx2, _mmx2 )
 558 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 559 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 560 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _mmx2, _mmx2 )
 561 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _sse2, _sse2 )
 562 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _sse2 )
 563 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _sse2 )
 564 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _sse2, _sse2 )
 565 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _ssse3, _sse2 )
 566 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 567 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 568 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _ssse3, _sse2 )
 569 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _sse2 )
 570 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _sse2 )
 571 #else
 572 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_mmx
 573 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 574 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 575 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 576 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 577 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _mmx2 )
 578 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _mmx2 )
 579 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _mmx2 )
 580 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _xop, _mmx2 )
 581 #endif
 582 #endif
 583 #if !HIGH_BIT_DEPTH && HAVE_ARMV6
 584 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _neon, _armv6 )
 585 INTRA_MBCMP(satd,  4x4,   v, h, dc,  , _neon, _armv6 )
 586 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _neon, _neon )
 587 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c, _neon, _neon )
 588 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _neon, _c )
 589 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _neon, _c )
 590 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _neon, _neon )
 591 INTRA_MBCMP(satd, 16x16,  v, h, dc,  , _neon, _neon )
 592 #endif
 593
 594 // No C implementation of intra_satd_x9. See checkasm for its behavior,
 595 // or see x264_mb_analyse_intra for the entirely different algorithm we
 596 // use when lacking an asm implementation of it.
 597
 598
 599
 600 /****************************************************************************
 601  * structural similarity metric
 602  ****************************************************************************/
 603 static void ssim_4x4x2_core( const pixel *pix1, intptr_t stride1,
 604                              const pixel *pix2, intptr_t stride2,
 605                              int sums[2][4] )
 606 {
 607     for( int z = 0; z < 2; z++ )
 608     {
 609         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 610         for( int y = 0; y < 4; y++ )
 611             for( int x = 0; x < 4; x++ )
 612             {
 613                 int a = pix1[x+y*stride1];
 614                 int b = pix2[x+y*stride2];
 615                 s1  += a;
 616                 s2  += b;
 617                 ss  += a*a;
 618                 ss  += b*b;
 619                 s12 += a*b;
 620             }
 621         sums[z][0] = s1;
 622         sums[z][1] = s2;
 623         sums[z][2] = ss;
 624         sums[z][3] = s12;
 625         pix1 += 4;
 626         pix2 += 4;
 627     }
 628 }
 629
 630 static float ssim_end1( int s1, int s2, int ss, int s12 )
 631 {
 632 /* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
 633  * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
 634  * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
 635 #if BIT_DEPTH > 9
 636 #define type float
 637     static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
 638     static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
 639 #else
 640 #define type int
 641     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 642     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 643 #endif
 644     type fs1 = s1;
 645     type fs2 = s2;
 646     type fss = ss;
 647     type fs12 = s12;
 648     type vars = fss*64 - fs1*fs1 - fs2*fs2;
 649     type covar = fs12*64 - fs1*fs2;
 650     return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
 651          / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
 652 #undef type
 653 }
 654
 655 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 656 {
 657     float ssim = 0.0;
 658     for( int i = 0; i < width; i++ )
 659         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 660                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 661                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 662                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 663     return ssim;
 664 }
 665
 666 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 667                            pixel *pix1, intptr_t stride1,
 668                            pixel *pix2, intptr_t stride2,
 669                            int width, int height, void *buf, int *cnt )
 670 {
 671     int z = 0;
 672     float ssim = 0.0;
 673     int (*sum0)[4] = buf;
 674     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 675     width >>= 2;
 676     height >>= 2;
 677     for( int y = 1; y < height; y++ )
 678     {
 679         for( ; z <= y; z++ )
 680         {
 681             XCHG( void*, sum0, sum1 );
 682             for( int x = 0; x < width; x+=2 )
 683                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 684         }
 685         for( int x = 0; x < width-1; x += 4 )
 686             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 687     }
 688     *cnt = (height-1) * (width-1);
 689     return ssim;
 690 }
 691
 692 static int pixel_vsad( pixel *src, intptr_t stride, int height )
 693 {
 694     int score = 0;
 695     for( int i = 1; i < height; i++, src += stride )
 696         for( int j = 0; j < 16; j++ )
 697             score += abs(src[j] - src[j+stride]);
 698     return score;
 699 }
 700
 701 int x264_field_vsad( x264_t *h, int mb_x, int mb_y )
 702 {
 703     int score_field, score_frame;
 704     int stride = h->fenc->i_stride[0];
 705     int mb_stride = h->mb.i_mb_stride;
 706     pixel *fenc = h->fenc->plane[0] + 16 * (mb_x + mb_y * stride);
 707     int mb_xy = mb_x + mb_y*mb_stride;
 708
 709     /* We don't want to analyze pixels outside the frame, as it gives inaccurate results. */
 710     int mbpair_height = X264_MIN( h->param.i_height - mb_y * 16, 32 );
 711     score_frame  = h->pixf.vsad( fenc,          stride, mbpair_height );
 712     score_field  = h->pixf.vsad( fenc,        stride*2, mbpair_height >> 1 );
 713     score_field += h->pixf.vsad( fenc+stride, stride*2, mbpair_height >> 1 );
 714
 715     if( mb_x > 0 )
 716         score_field += 512 - h->mb.field[mb_xy        -1]*1024;
 717     if( mb_y > 0 )
 718         score_field += 512 - h->mb.field[mb_xy-mb_stride]*1024;
 719
 720     return (score_field < score_frame);
 721 }
 722
 723 static int pixel_asd8( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2, int height )
 724 {
 725     int sum = 0;
 726     for( int y = 0; y < height; y++, pix1 += stride1, pix2 += stride2 )
 727         for( int x = 0; x < 8; x++ )
 728             sum += pix1[x] - pix2[x];
 729     return abs( sum );
 730 }
 731
 732 /****************************************************************************
 733  * successive elimination
 734  ****************************************************************************/
 735 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 736                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 737 {
 738     int nmv = 0;
 739     for( int i = 0; i < width; i++, sums++ )
 740     {
 741         int ads = abs( enc_dc[0] - sums[0] )
 742                 + abs( enc_dc[1] - sums[8] )
 743                 + abs( enc_dc[2] - sums[delta] )
 744                 + abs( enc_dc[3] - sums[delta+8] )
 745                 + cost_mvx[i];
 746         if( ads < thresh )
 747             mvs[nmv++] = i;
 748     }
 749     return nmv;
 750 }
 751
 752 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 753                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 754 {
 755     int nmv = 0;
 756     for( int i = 0; i < width; i++, sums++ )
 757     {
 758         int ads = abs( enc_dc[0] - sums[0] )
 759                 + abs( enc_dc[1] - sums[delta] )
 760                 + cost_mvx[i];
 761         if( ads < thresh )
 762             mvs[nmv++] = i;
 763     }
 764     return nmv;
 765 }
 766
 767 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 768                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 769 {
 770     int nmv = 0;
 771     for( int i = 0; i<width; i++, sums++ )
 772     {
 773         int ads = abs( enc_dc[0] - sums[0] )
 774                 + cost_mvx[i];
 775         if( ads < thresh )
 776             mvs[nmv++] = i;
 777     }
 778     return nmv;
 779 }
 780
 781
 782 /****************************************************************************
 783  * x264_pixel_init:
 784  ****************************************************************************/
 785 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 786 {
 787     memset( pixf, 0, sizeof(*pixf) );
 788
 789 #define INIT2_NAME( name1, name2, cpu ) \
 790     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 791     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 792 #define INIT4_NAME( name1, name2, cpu ) \
 793     INIT2_NAME( name1, name2, cpu ) \
 794     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 795     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 796 #define INIT5_NAME( name1, name2, cpu ) \
 797     INIT4_NAME( name1, name2, cpu ) \
 798     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 799 #define INIT6_NAME( name1, name2, cpu ) \
 800     INIT5_NAME( name1, name2, cpu ) \
 801     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 802 #define INIT7_NAME( name1, name2, cpu ) \
 803     INIT6_NAME( name1, name2, cpu ) \
 804     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 805 #define INIT8_NAME( name1, name2, cpu ) \
 806     INIT7_NAME( name1, name2, cpu ) \
 807     pixf->name1[PIXEL_4x16]  = x264_pixel_##name2##_4x16##cpu;
 808 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 809 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 810 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 811 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 812 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 813 #define INIT8( name, cpu ) INIT8_NAME( name, name, cpu )
 814
 815 #define INIT_ADS( cpu ) \
 816     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 817     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 818     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 819
 820     INIT8( sad, );
 821     INIT8_NAME( sad_aligned, sad, );
 822     INIT7( sad_x3, );
 823     INIT7( sad_x4, );
 824     INIT8( ssd, );
 825     INIT8( satd, );
 826     INIT7( satd_x3, );
 827     INIT7( satd_x4, );
 828     INIT4( hadamard_ac, );
 829     INIT_ADS( );
 830
 831     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 832     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 833     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 834     pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16;
 835     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 836     pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16;
 837     pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8;
 838
 839     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 840     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 841     pixf->ssim_end4 = ssim_end4;
 842     pixf->vsad = pixel_vsad;
 843     pixf->asd8 = pixel_asd8;
 844
 845     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 846     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 847     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 848     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 849     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 850     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 851     pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c;
 852     pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c;
 853     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 854     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 855
 856 #if HIGH_BIT_DEPTH
 857 #if HAVE_MMX
 858     if( cpu&X264_CPU_MMX2 )
 859     {
 860         INIT7( sad, _mmx2 );
 861         INIT7_NAME( sad_aligned, sad, _mmx2 );
 862         INIT7( sad_x3, _mmx2 );
 863         INIT7( sad_x4, _mmx2 );
 864         INIT8( satd, _mmx2 );
 865         INIT7( satd_x3, _mmx2 );
 866         INIT7( satd_x4, _mmx2 );
 867         INIT4( hadamard_ac, _mmx2 );
 868         INIT8( ssd, _mmx2 );
 869         INIT_ADS( _mmx2 );
 870
 871         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_mmx2;
 872         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
 873         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
 874 #if ARCH_X86
 875         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_mmx2;
 876         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
 877 #endif
 878
 879         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
 880         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
 881         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
 882         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
 883         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
 884         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
 885         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
 886         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
 887         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
 888     }
 889     if( cpu&X264_CPU_SSE2 )
 890     {
 891         INIT4_NAME( sad_aligned, sad, _sse2_aligned );
 892         INIT5( ssd, _sse2 );
 893         INIT6( satd, _sse2 );
 894         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse2;
 895
 896         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 897         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 898 #if ARCH_X86_64
 899         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 900         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
 901 #endif
 902         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_sse2;
 903         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_sse2;
 904         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 905         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 906         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 907         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_sse2;
 908         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_sse2;
 909         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_sse2;
 910         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_sse2;
 911     }
 912     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 913     {
 914         INIT5( sad, _sse2 );
 915         INIT2( sad_x3, _sse2 );
 916         INIT2( sad_x4, _sse2 );
 917         INIT_ADS( _sse2 );
 918
 919         if( !(cpu&X264_CPU_STACK_MOD4) )
 920         {
 921             INIT4( hadamard_ac, _sse2 );
 922         }
 923         pixf->vsad = x264_pixel_vsad_sse2;
 924         pixf->asd8 = x264_pixel_asd8_sse2;
 925         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_sse2;
 926         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_sse2;
 927         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
 928         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
 929         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_sse2;
 930     }
 931     if( cpu&X264_CPU_SSE2_IS_FAST )
 932     {
 933         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 934         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 935         pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_sse2;
 936         pixf->sad_x3[PIXEL_8x4]  = x264_pixel_sad_x3_8x4_sse2;
 937         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 938         pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_sse2;
 939         pixf->sad_x4[PIXEL_8x4]  = x264_pixel_sad_x4_8x4_sse2;
 940     }
 941     if( cpu&X264_CPU_SSSE3 )
 942     {
 943         INIT4_NAME( sad_aligned, sad, _ssse3_aligned );
 944         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_ssse3;
 945         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_ssse3;
 946         INIT7( sad, _ssse3 );
 947         INIT7( sad_x3, _ssse3 );
 948         INIT7( sad_x4, _ssse3 );
 949         INIT_ADS( _ssse3 );
 950         INIT6( satd, _ssse3 );
 951         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_ssse3;
 952
 953         if( !(cpu&X264_CPU_STACK_MOD4) )
 954         {
 955             INIT4( hadamard_ac, _ssse3 );
 956         }
 957         pixf->vsad = x264_pixel_vsad_ssse3;
 958         pixf->asd8 = x264_pixel_asd8_ssse3;
 959         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_ssse3;
 960         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 961         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 962 #if ARCH_X86_64
 963         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
 964 #endif
 965         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_ssse3;
 966         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_ssse3;
 967         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 968         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_ssse3;
 969         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
 970         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 971     }
 972     if( cpu&X264_CPU_SSE4 )
 973     {
 974         INIT6( satd, _sse4 );
 975         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse4;
 976         if( !(cpu&X264_CPU_STACK_MOD4) )
 977         {
 978             INIT4( hadamard_ac, _sse4 );
 979         }
 980         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 981         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 982 #if ARCH_X86_64
 983         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
 984 #endif
 985         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
 986     }
 987     if( cpu&X264_CPU_AVX )
 988     {
 989         INIT5_NAME( sad_aligned, sad, _ssse3 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
 990         INIT_ADS( _avx );
 991         INIT6( satd, _avx );
 992         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_avx;
 993         if( !(cpu&X264_CPU_STACK_MOD4) )
 994         {
 995             INIT4( hadamard_ac, _avx );
 996         }
 997         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_avx;
 998         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
 999         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1000         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1001         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1002         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1003         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1004         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1005 #if ARCH_X86_64
1006         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1007 #endif
1008         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1009     }
1010     if( cpu&X264_CPU_XOP )
1011     {
1012         INIT5( sad_x3, _xop );
1013         INIT5( sad_x4, _xop );
1014         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_xop;
1015         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1016         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1017         pixf->vsad = x264_pixel_vsad_xop;
1018         pixf->asd8 = x264_pixel_asd8_xop;
1019 #if ARCH_X86_64
1020         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1021 #endif
1022     }
1023     if( cpu&X264_CPU_AVX2 )
1024     {
1025         INIT2( ssd, _avx2 );
1026         INIT2( sad, _avx2 );
1027         INIT2_NAME( sad_aligned, sad, _avx2 );
1028         INIT2( sad_x3, _avx2 );
1029         INIT2( sad_x4, _avx2 );
1030         pixf->vsad = x264_pixel_vsad_avx2;
1031         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1032         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_avx2;
1033     }
1034 #endif // HAVE_MMX
1035 #else // !HIGH_BIT_DEPTH
1036 #if HAVE_MMX
1037     if( cpu&X264_CPU_MMX )
1038     {
1039         INIT8( ssd, _mmx );
1040     }
1041
1042     if( cpu&X264_CPU_MMX2 )
1043     {
1044         INIT8( sad, _mmx2 );
1045         INIT8_NAME( sad_aligned, sad, _mmx2 );
1046         INIT7( sad_x3, _mmx2 );
1047         INIT7( sad_x4, _mmx2 );
1048         INIT8( satd, _mmx2 );
1049         INIT7( satd_x3, _mmx2 );
1050         INIT7( satd_x4, _mmx2 );
1051         INIT4( hadamard_ac, _mmx2 );
1052         INIT_ADS( _mmx2 );
1053         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
1054         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_mmx2;
1055         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
1056         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmx2;
1057 #if ARCH_X86
1058         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmx2;
1059         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmx2;
1060         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmx2;
1061         pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_mmx2;
1062         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_mmx2;
1063         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
1064         pixf->vsad = x264_pixel_vsad_mmx2;
1065
1066         if( cpu&X264_CPU_CACHELINE_32 )
1067         {
1068             INIT5( sad, _cache32_mmx2 );
1069             INIT4( sad_x3, _cache32_mmx2 );
1070             INIT4( sad_x4, _cache32_mmx2 );
1071         }
1072         else if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1073         {
1074             INIT5( sad, _cache64_mmx2 );
1075             INIT4( sad_x3, _cache64_mmx2 );
1076             INIT4( sad_x4, _cache64_mmx2 );
1077         }
1078 #else
1079         if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1080         {
1081             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmx2;
1082             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmx2;
1083             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmx2;
1084             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmx2;
1085             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmx2;
1086             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmx2;
1087             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmx2;
1088         }
1089 #endif
1090         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
1091         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
1092         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
1093         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
1094         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
1095         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
1096         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
1097         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
1098         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
1099     }
1100
1101     if( cpu&X264_CPU_SSE2 )
1102     {
1103         INIT5( ssd, _sse2slow );
1104         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
1105         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
1106         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
1107         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
1108         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
1109         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
1110         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
1111 #if ARCH_X86_64
1112         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
1113         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
1114 #endif
1115         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_sse2;
1116         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_sse2;
1117         pixf->vsad = x264_pixel_vsad_sse2;
1118         pixf->asd8 = x264_pixel_asd8_sse2;
1119     }
1120
1121     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
1122     {
1123         INIT2( sad, _sse2 );
1124         INIT2( sad_x3, _sse2 );
1125         INIT2( sad_x4, _sse2 );
1126         INIT6( satd, _sse2 );
1127         pixf->satd[PIXEL_4x16]   = x264_pixel_satd_4x16_sse2;
1128         INIT6( satd_x3, _sse2 );
1129         INIT6( satd_x4, _sse2 );
1130         INIT4( hadamard_ac, _sse2 );
1131         INIT_ADS( _sse2 );
1132         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
1133         pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_sse2;
1134         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
1135         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
1136         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
1137         if( cpu&X264_CPU_CACHELINE_64 )
1138         {
1139             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
1140 #if ARCH_X86
1141             INIT2( sad, _cache64_sse2 );
1142             INIT2( sad_x3, _cache64_sse2 );
1143             INIT2( sad_x4, _cache64_sse2 );
1144 #endif
1145            if( cpu&X264_CPU_SSE2_IS_FAST )
1146            {
1147                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
1148                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
1149            }
1150         }
1151     }
1152
1153     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
1154     {
1155         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1156         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1157         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
1158         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
1159         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
1160         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
1161         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
1162         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
1163     }
1164
1165     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
1166     {
1167         INIT2( sad, _sse3 );
1168         INIT2( sad_x3, _sse3 );
1169         INIT2( sad_x4, _sse3 );
1170     }
1171
1172     if( cpu&X264_CPU_SSSE3 )
1173     {
1174         INIT4( hadamard_ac, _ssse3 );
1175         if( !(cpu&X264_CPU_STACK_MOD4) )
1176         {
1177             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_ssse3;
1178             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_ssse3;
1179             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_ssse3;
1180 #if ARCH_X86_64
1181             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_ssse3;
1182 #endif
1183         }
1184         INIT_ADS( _ssse3 );
1185         if( cpu&X264_CPU_SLOW_ATOM )
1186         {
1187             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3_atom;
1188             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3_atom;
1189             INIT6( satd, _ssse3_atom );
1190             pixf->satd[PIXEL_4x16]  = x264_pixel_satd_4x16_ssse3_atom;
1191             INIT6( satd_x3, _ssse3_atom );
1192             INIT6( satd_x4, _ssse3_atom );
1193             INIT4( hadamard_ac, _ssse3_atom );
1194 #if ARCH_X86_64
1195             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3_atom;
1196 #endif
1197         }
1198         else
1199         {
1200             INIT8( ssd, _ssse3 );
1201             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
1202             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
1203             INIT8( satd, _ssse3 );
1204             INIT7( satd_x3, _ssse3 );
1205             INIT7( satd_x4, _ssse3 );
1206 #if ARCH_X86_64
1207             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
1208 #endif
1209         }
1210         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
1211         if( !(cpu&X264_CPU_SLOW_PSHUFB) )
1212             pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
1213         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
1214         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
1215         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
1216         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_ssse3;
1217         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_ssse3;
1218         pixf->asd8 = x264_pixel_asd8_ssse3;
1219         if( cpu&X264_CPU_CACHELINE_64 )
1220         {
1221             INIT2( sad, _cache64_ssse3 );
1222             INIT2( sad_x3, _cache64_ssse3 );
1223             INIT2( sad_x4, _cache64_ssse3 );
1224         }
1225         else
1226         {
1227             INIT2( sad_x3, _ssse3 );
1228             INIT5( sad_x4, _ssse3 );
1229         }
1230         if( (cpu&X264_CPU_SLOW_ATOM) || (cpu&X264_CPU_SLOW_SHUFFLE) )
1231         {
1232             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
1233         }
1234     }
1235
1236     if( cpu&X264_CPU_SSE4 )
1237     {
1238         INIT8( satd, _sse4 );
1239         INIT7( satd_x3, _sse4 );
1240         INIT7( satd_x4, _sse4 );
1241         INIT4( hadamard_ac, _sse4 );
1242         if( !(cpu&X264_CPU_STACK_MOD4) )
1243         {
1244             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_sse4;
1245             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_sse4;
1246             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_sse4;
1247 #if ARCH_X86_64
1248             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_sse4;
1249 #endif
1250         }
1251         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
1252         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
1253         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
1254 #if ARCH_X86_64
1255         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
1256 #endif
1257     }
1258
1259     if( cpu&X264_CPU_AVX )
1260     {
1261         INIT2_NAME( sad_aligned, sad, _sse2 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
1262         INIT2( sad_x3, _avx );
1263         INIT2( sad_x4, _avx );
1264         INIT8( satd, _avx );
1265         INIT7( satd_x3, _avx );
1266         INIT7( satd_x4, _avx );
1267         INIT_ADS( _avx );
1268         INIT4( hadamard_ac, _avx );
1269         if( !(cpu&X264_CPU_STACK_MOD4) )
1270         {
1271             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_avx;
1272             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_avx;
1273             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx;
1274 #if ARCH_X86_64
1275             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_avx;
1276 #endif
1277         }
1278         INIT5( ssd, _avx );
1279         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1280         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1281         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1282         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1283         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1284         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_avx;
1285         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1286         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1287         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1288 #if ARCH_X86_64
1289         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1290 #endif
1291     }
1292
1293     if( cpu&X264_CPU_XOP )
1294     {
1295         INIT7( satd, _xop );
1296         INIT7( satd_x3, _xop );
1297         INIT7( satd_x4, _xop );
1298         INIT4( hadamard_ac, _xop );
1299         if( !(cpu&X264_CPU_STACK_MOD4) )
1300         {
1301             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_xop;
1302         }
1303         INIT5( ssd, _xop );
1304         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_xop;
1305         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_xop;
1306         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_xop;
1307         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_xop;
1308         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1309         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_xop;
1310         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1311         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_xop;
1312         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_xop;
1313 #if ARCH_X86_64
1314         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1315 #endif
1316     }
1317
1318     if( cpu&X264_CPU_AVX2 )
1319     {
1320         INIT2( ssd, _avx2 );
1321         INIT2( sad_x3, _avx2 );
1322         INIT2( sad_x4, _avx2 );
1323         INIT4( satd, _avx2 );
1324         INIT2( hadamard_ac, _avx2 );
1325         INIT_ADS( _avx2 );
1326         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx2;
1327         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx2;
1328         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_avx2;
1329         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_avx2;
1330         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_avx2;
1331         pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx2;
1332         pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_avx2;
1333         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1334 #if ARCH_X86_64
1335         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx2;
1336 #endif
1337     }
1338 #endif //HAVE_MMX
1339
1340 #if HAVE_ARMV6
1341     if( cpu&X264_CPU_ARMV6 )
1342     {
1343         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1344         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1345         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1346         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1347     }
1348     if( cpu&X264_CPU_NEON )
1349     {
1350         INIT5( sad, _neon );
1351         INIT5( sad_aligned, _neon );
1352         INIT7( sad_x3, _neon );
1353         INIT7( sad_x4, _neon );
1354         INIT7( ssd, _neon );
1355         INIT7( satd, _neon );
1356         INIT7( satd_x3, _neon );
1357         INIT7( satd_x4, _neon );
1358         INIT4( hadamard_ac, _neon );
1359         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1360         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1361         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1362         pixf->var[PIXEL_8x16]   = x264_pixel_var_8x16_neon;
1363         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1364         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
1365         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
1366
1367         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_neon;
1368         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_neon;
1369         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_neon;
1370         pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8_neon;
1371         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_neon;
1372         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_neon;
1373         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_neon;
1374         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
1375         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_neon;
1376         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
1377
1378         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1379         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1380
1381         if( cpu&X264_CPU_FAST_NEON_MRC )
1382         {
1383             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
1384             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
1385             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
1386             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
1387         }
1388         else    // really just scheduled for dual issue / A8
1389         {
1390             INIT5( sad_aligned, _neon_dual );
1391         }
1392     }
1393 #endif
1394 #endif // HIGH_BIT_DEPTH
1395 #if HAVE_ALTIVEC
1396     if( cpu&X264_CPU_ALTIVEC )
1397     {
1398         x264_pixel_altivec_init( pixf );
1399     }
1400 #endif
1401
1402     pixf->ads[PIXEL_8x16] =
1403     pixf->ads[PIXEL_8x4] =
1404     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
1405     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
1406 }
1407