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 #if ARCH_AARCH64
  42 #   include "aarch64/pixel.h"
  43 #   include "aarch64/predict.h"
  44 #endif
  45
  46
  47 /****************************************************************************
  48  * pixel_sad_WxH
  49  ****************************************************************************/
  50 #define PIXEL_SAD_C( name, lx, ly ) \
  51 static int name( pixel *pix1, intptr_t i_stride_pix1,  \
  52                  pixel *pix2, intptr_t i_stride_pix2 ) \
  53 {                                                   \
  54     int i_sum = 0;                                  \
  55     for( int y = 0; y < ly; y++ )                   \
  56     {                                               \
  57         for( int x = 0; x < lx; x++ )               \
  58         {                                           \
  59             i_sum += abs( pix1[x] - pix2[x] );      \
  60         }                                           \
  61         pix1 += i_stride_pix1;                      \
  62         pix2 += i_stride_pix2;                      \
  63     }                                               \
  64     return i_sum;                                   \
  65 }
  66
  67
  68 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  69 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  70 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  71 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  72 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  73 PIXEL_SAD_C( x264_pixel_sad_4x16,   4, 16 )
  74 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  75 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  76
  77 /****************************************************************************
  78  * pixel_ssd_WxH
  79  ****************************************************************************/
  80 #define PIXEL_SSD_C( name, lx, ly ) \
  81 static int name( pixel *pix1, intptr_t i_stride_pix1,  \
  82                  pixel *pix2, intptr_t i_stride_pix2 ) \
  83 {                                                   \
  84     int i_sum = 0;                                  \
  85     for( int y = 0; y < ly; y++ )                   \
  86     {                                               \
  87         for( int x = 0; x < lx; x++ )               \
  88         {                                           \
  89             int d = pix1[x] - pix2[x];              \
  90             i_sum += d*d;                           \
  91         }                                           \
  92         pix1 += i_stride_pix1;                      \
  93         pix2 += i_stride_pix2;                      \
  94     }                                               \
  95     return i_sum;                                   \
  96 }
  97
  98 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
  99 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
 100 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
 101 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
 102 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
 103 PIXEL_SSD_C( x264_pixel_ssd_4x16,   4, 16 )
 104 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
 105 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
 106
 107 uint64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, pixel *pix1, intptr_t i_pix1,
 108                              pixel *pix2, intptr_t i_pix2, int i_width, int i_height )
 109 {
 110     uint64_t i_ssd = 0;
 111     int y;
 112     int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
 113
 114 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 115                                           pix2 + y*i_pix2 + x, i_pix2 );
 116     for( y = 0; y < i_height-15; y += 16 )
 117     {
 118         int x = 0;
 119         if( align )
 120             for( ; x < i_width-15; x += 16 )
 121                 SSD(PIXEL_16x16);
 122         for( ; x < i_width-7; x += 8 )
 123             SSD(PIXEL_8x16);
 124     }
 125     if( y < i_height-7 )
 126         for( int x = 0; x < i_width-7; x += 8 )
 127             SSD(PIXEL_8x8);
 128 #undef SSD
 129
 130 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 131     if( i_width & 7 )
 132     {
 133         for( y = 0; y < (i_height & ~7); y++ )
 134             for( int x = i_width & ~7; x < i_width; x++ )
 135                 SSD1;
 136     }
 137     if( i_height & 7 )
 138     {
 139         for( y = i_height & ~7; y < i_height; y++ )
 140             for( int x = 0; x < i_width; x++ )
 141                 SSD1;
 142     }
 143 #undef SSD1
 144
 145     return i_ssd;
 146 }
 147
 148 static void pixel_ssd_nv12_core( pixel *pixuv1, intptr_t stride1, pixel *pixuv2, intptr_t stride2,
 149                                  int width, int height, uint64_t *ssd_u, uint64_t *ssd_v )
 150 {
 151     *ssd_u = 0, *ssd_v = 0;
 152     for( int y = 0; y < height; y++, pixuv1+=stride1, pixuv2+=stride2 )
 153         for( int x = 0; x < width; x++ )
 154         {
 155             int du = pixuv1[2*x]   - pixuv2[2*x];
 156             int dv = pixuv1[2*x+1] - pixuv2[2*x+1];
 157             *ssd_u += du*du;
 158             *ssd_v += dv*dv;
 159         }
 160 }
 161
 162 void x264_pixel_ssd_nv12( x264_pixel_function_t *pf, pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2,
 163                           int i_width, int i_height, uint64_t *ssd_u, uint64_t *ssd_v )
 164 {
 165     pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height, ssd_u, ssd_v );
 166     if( i_width&7 )
 167     {
 168         uint64_t tmp[2];
 169         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] );
 170         *ssd_u += tmp[0];
 171         *ssd_v += tmp[1];
 172     }
 173 }
 174
 175 /****************************************************************************
 176  * pixel_var_wxh
 177  ****************************************************************************/
 178 #define PIXEL_VAR_C( name, w, h ) \
 179 static uint64_t name( pixel *pix, intptr_t i_stride ) \
 180 {                                             \
 181     uint32_t sum = 0, sqr = 0;                \
 182     for( int y = 0; y < h; y++ )              \
 183     {                                         \
 184         for( int x = 0; x < w; x++ )          \
 185         {                                     \
 186             sum += pix[x];                    \
 187             sqr += pix[x] * pix[x];           \
 188         }                                     \
 189         pix += i_stride;                      \
 190     }                                         \
 191     return sum + ((uint64_t)sqr << 32);       \
 192 }
 193
 194 PIXEL_VAR_C( x264_pixel_var_16x16, 16, 16 )
 195 PIXEL_VAR_C( x264_pixel_var_8x16,   8, 16 )
 196 PIXEL_VAR_C( x264_pixel_var_8x8,    8,  8 )
 197
 198 /****************************************************************************
 199  * pixel_var2_wxh
 200  ****************************************************************************/
 201 #define PIXEL_VAR2_C( name, w, h, shift ) \
 202 static int name( pixel *pix1, intptr_t i_stride1, pixel *pix2, intptr_t i_stride2, int *ssd ) \
 203 { \
 204     int var = 0, sum = 0, sqr = 0; \
 205     for( int y = 0; y < h; y++ ) \
 206     { \
 207         for( int x = 0; x < w; x++ ) \
 208         { \
 209             int diff = pix1[x] - pix2[x]; \
 210             sum += diff; \
 211             sqr += diff * diff; \
 212         } \
 213         pix1 += i_stride1; \
 214         pix2 += i_stride2; \
 215     } \
 216     var = sqr - ((int64_t)sum * sum >> shift); \
 217     *ssd = sqr; \
 218     return var; \
 219 }
 220
 221 PIXEL_VAR2_C( x264_pixel_var2_8x16, 8, 16, 7 )
 222 PIXEL_VAR2_C( x264_pixel_var2_8x8,  8,  8, 6 )
 223
 224 #if BIT_DEPTH > 8
 225     typedef uint32_t sum_t;
 226     typedef uint64_t sum2_t;
 227 #else
 228     typedef uint16_t sum_t;
 229     typedef uint32_t sum2_t;
 230 #endif
 231 #define BITS_PER_SUM (8 * sizeof(sum_t))
 232
 233 #define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
 234     sum2_t t0 = s0 + s1;\
 235     sum2_t t1 = s0 - s1;\
 236     sum2_t t2 = s2 + s3;\
 237     sum2_t t3 = s2 - s3;\
 238     d0 = t0 + t2;\
 239     d2 = t0 - t2;\
 240     d1 = t1 + t3;\
 241     d3 = t1 - t3;\
 242 }
 243
 244 // in: a pseudo-simd number of the form x+(y<<16)
 245 // return: abs(x)+(abs(y)<<16)
 246 static ALWAYS_INLINE sum2_t abs2( sum2_t a )
 247 {
 248     sum2_t s = ((a>>(BITS_PER_SUM-1))&(((sum2_t)1<<BITS_PER_SUM)+1))*((sum_t)-1);
 249     return (a+s)^s;
 250 }
 251
 252 /****************************************************************************
 253  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 254  ****************************************************************************/
 255
 256 static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 257 {
 258     sum2_t tmp[4][2];
 259     sum2_t a0, a1, a2, a3, b0, b1;
 260     sum2_t sum = 0;
 261     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 262     {
 263         a0 = pix1[0] - pix2[0];
 264         a1 = pix1[1] - pix2[1];
 265         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 266         a2 = pix1[2] - pix2[2];
 267         a3 = pix1[3] - pix2[3];
 268         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 269         tmp[i][0] = b0 + b1;
 270         tmp[i][1] = b0 - b1;
 271     }
 272     for( int i = 0; i < 2; i++ )
 273     {
 274         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 275         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 276         sum += ((sum_t)a0) + (a0>>BITS_PER_SUM);
 277     }
 278     return sum >> 1;
 279 }
 280
 281 static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 282 {
 283     sum2_t tmp[4][4];
 284     sum2_t a0, a1, a2, a3;
 285     sum2_t sum = 0;
 286     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 287     {
 288         a0 = (pix1[0] - pix2[0]) + ((sum2_t)(pix1[4] - pix2[4]) << BITS_PER_SUM);
 289         a1 = (pix1[1] - pix2[1]) + ((sum2_t)(pix1[5] - pix2[5]) << BITS_PER_SUM);
 290         a2 = (pix1[2] - pix2[2]) + ((sum2_t)(pix1[6] - pix2[6]) << BITS_PER_SUM);
 291         a3 = (pix1[3] - pix2[3]) + ((sum2_t)(pix1[7] - pix2[7]) << BITS_PER_SUM);
 292         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 293     }
 294     for( int i = 0; i < 4; i++ )
 295     {
 296         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 297         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 298     }
 299     return (((sum_t)sum) + (sum>>BITS_PER_SUM)) >> 1;
 300 }
 301
 302 #define PIXEL_SATD_C( w, h, sub )\
 303 static int x264_pixel_satd_##w##x##h( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )\
 304 {\
 305     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 306             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 307     if( w==16 )\
 308         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 309             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 310     if( h==16 )\
 311         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 312             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 313     if( w==16 && h==16 )\
 314         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 315             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 316     return sum;\
 317 }
 318 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 319 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 320 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 321 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 322 PIXEL_SATD_C( 4,  16, x264_pixel_satd_4x4 )
 323 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 324
 325 static NOINLINE int sa8d_8x8( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 326 {
 327     sum2_t tmp[8][4];
 328     sum2_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
 329     sum2_t sum = 0;
 330     for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
 331     {
 332         a0 = pix1[0] - pix2[0];
 333         a1 = pix1[1] - pix2[1];
 334         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 335         a2 = pix1[2] - pix2[2];
 336         a3 = pix1[3] - pix2[3];
 337         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 338         a4 = pix1[4] - pix2[4];
 339         a5 = pix1[5] - pix2[5];
 340         b2 = (a4+a5) + ((a4-a5)<<BITS_PER_SUM);
 341         a6 = pix1[6] - pix2[6];
 342         a7 = pix1[7] - pix2[7];
 343         b3 = (a6+a7) + ((a6-a7)<<BITS_PER_SUM);
 344         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 345     }
 346     for( int i = 0; i < 4; i++ )
 347     {
 348         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 349         HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 350         b0  = abs2(a0+a4) + abs2(a0-a4);
 351         b0 += abs2(a1+a5) + abs2(a1-a5);
 352         b0 += abs2(a2+a6) + abs2(a2-a6);
 353         b0 += abs2(a3+a7) + abs2(a3-a7);
 354         sum += (sum_t)b0 + (b0>>BITS_PER_SUM);
 355     }
 356     return sum;
 357 }
 358
 359 static int x264_pixel_sa8d_8x8( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 360 {
 361     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
 362     return (sum+2)>>2;
 363 }
 364
 365 static int x264_pixel_sa8d_16x16( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 366 {
 367     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 368             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 369             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 370             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 371     return (sum+2)>>2;
 372 }
 373
 374 static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, intptr_t stride )
 375 {
 376     sum2_t tmp[32];
 377     sum2_t a0, a1, a2, a3, dc;
 378     sum2_t sum4 = 0, sum8 = 0;
 379     for( int i = 0; i < 8; i++, pix+=stride )
 380     {
 381         sum2_t *t = tmp + (i&3) + (i&4)*4;
 382         a0 = (pix[0]+pix[1]) + ((sum2_t)(pix[0]-pix[1])<<BITS_PER_SUM);
 383         a1 = (pix[2]+pix[3]) + ((sum2_t)(pix[2]-pix[3])<<BITS_PER_SUM);
 384         t[0] = a0 + a1;
 385         t[4] = a0 - a1;
 386         a2 = (pix[4]+pix[5]) + ((sum2_t)(pix[4]-pix[5])<<BITS_PER_SUM);
 387         a3 = (pix[6]+pix[7]) + ((sum2_t)(pix[6]-pix[7])<<BITS_PER_SUM);
 388         t[8] = a2 + a3;
 389         t[12] = a2 - a3;
 390     }
 391     for( int i = 0; i < 8; i++ )
 392     {
 393         HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 394         tmp[i*4+0] = a0;
 395         tmp[i*4+1] = a1;
 396         tmp[i*4+2] = a2;
 397         tmp[i*4+3] = a3;
 398         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 399     }
 400     for( int i = 0; i < 8; i++ )
 401     {
 402         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 403         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 404     }
 405     dc = (sum_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 406     sum4 = (sum_t)sum4 + (sum4>>BITS_PER_SUM) - dc;
 407     sum8 = (sum_t)sum8 + (sum8>>BITS_PER_SUM) - dc;
 408     return ((uint64_t)sum8<<32) + sum4;
 409 }
 410
 411 #define HADAMARD_AC(w,h) \
 412 static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, intptr_t stride )\
 413 {\
 414     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 415     if( w==16 )\
 416         sum += pixel_hadamard_ac( pix+8, stride );\
 417     if( h==16 )\
 418         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 419     if( w==16 && h==16 )\
 420         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 421     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 422 }
 423 HADAMARD_AC( 16, 16 )
 424 HADAMARD_AC( 16, 8 )
 425 HADAMARD_AC( 8, 16 )
 426 HADAMARD_AC( 8, 8 )
 427
 428
 429 /****************************************************************************
 430  * pixel_sad_x4
 431  ****************************************************************************/
 432 #define SAD_X( size ) \
 433 static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 434                                       intptr_t i_stride, int scores[3] )\
 435 {\
 436     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 437     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 438     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 439 }\
 440 static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1,pixel *pix2, pixel *pix3,\
 441                                       intptr_t i_stride, int scores[4] )\
 442 {\
 443     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 444     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 445     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 446     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 447 }
 448
 449 SAD_X( 16x16 )
 450 SAD_X( 16x8 )
 451 SAD_X( 8x16 )
 452 SAD_X( 8x8 )
 453 SAD_X( 8x4 )
 454 SAD_X( 4x8 )
 455 SAD_X( 4x4 )
 456
 457 /****************************************************************************
 458  * pixel_satd_x4
 459  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 460  ****************************************************************************/
 461
 462 #define SATD_X( size, cpu ) \
 463 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 464                                             intptr_t i_stride, int scores[3] )\
 465 {\
 466     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 467     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 468     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 469 }\
 470 static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3,\
 471                                             intptr_t i_stride, int scores[4] )\
 472 {\
 473     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 474     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 475     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 476     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 477 }
 478 #define SATD_X_DECL6( cpu )\
 479 SATD_X( 16x16, cpu )\
 480 SATD_X( 16x8, cpu )\
 481 SATD_X( 8x16, cpu )\
 482 SATD_X( 8x8, cpu )\
 483 SATD_X( 8x4, cpu )\
 484 SATD_X( 4x8, cpu )
 485 #define SATD_X_DECL7( cpu )\
 486 SATD_X_DECL6( cpu )\
 487 SATD_X( 4x4, cpu )
 488
 489 SATD_X_DECL7()
 490 #if HAVE_MMX
 491 SATD_X_DECL7( _mmx2 )
 492 #if !HIGH_BIT_DEPTH
 493 SATD_X_DECL6( _sse2 )
 494 SATD_X_DECL7( _ssse3 )
 495 SATD_X_DECL6( _ssse3_atom )
 496 SATD_X_DECL7( _sse4 )
 497 SATD_X_DECL7( _avx )
 498 SATD_X_DECL7( _xop )
 499 #endif // !HIGH_BIT_DEPTH
 500 #endif
 501
 502 #if !HIGH_BIT_DEPTH
 503 #if HAVE_ARMV6 || ARCH_AARCH64
 504 SATD_X_DECL7( _neon )
 505 #endif
 506 #endif // !HIGH_BIT_DEPTH
 507
 508 #define INTRA_MBCMP_8x8( mbcmp, cpu, cpu2 )\
 509 void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[36], int res[3] )\
 510 {\
 511     ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
 512     x264_predict_8x8_v##cpu2( pix, edge );\
 513     res[0] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 514     x264_predict_8x8_h##cpu2( pix, edge );\
 515     res[1] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 516     x264_predict_8x8_dc##cpu2( pix, edge );\
 517     res[2] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 518 }
 519
 520 INTRA_MBCMP_8x8( sad,, _c )
 521 INTRA_MBCMP_8x8(sa8d,, _c )
 522 #if HIGH_BIT_DEPTH && HAVE_MMX
 523 #define x264_predict_8x8_v_sse2 x264_predict_8x8_v_sse
 524 INTRA_MBCMP_8x8( sad, _mmx2,  _c )
 525 INTRA_MBCMP_8x8(sa8d, _sse2,  _sse2 )
 526 #endif
 527 #if !HIGH_BIT_DEPTH && (HAVE_ARMV6 || ARCH_AARCH64)
 528 INTRA_MBCMP_8x8( sad, _neon, _neon )
 529 INTRA_MBCMP_8x8(sa8d, _neon, _neon )
 530 #endif
 531
 532 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu, cpu2 )\
 533 void x264_intra_##mbcmp##_x3_##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
 534 {\
 535     x264_predict_##size##chroma##_##pred1##cpu2( fdec );\
 536     res[0] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 537     x264_predict_##size##chroma##_##pred2##cpu2( fdec );\
 538     res[1] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 539     x264_predict_##size##chroma##_##pred3##cpu2( fdec );\
 540     res[2] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 541 }
 542
 543 INTRA_MBCMP( sad,  4x4,   v, h, dc,  ,, _c )
 544 INTRA_MBCMP(satd,  4x4,   v, h, dc,  ,, _c )
 545 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c,, _c )
 546 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c,, _c )
 547 INTRA_MBCMP( sad,  8x16, dc, h,  v, c,, _c )
 548 INTRA_MBCMP(satd,  8x16, dc, h,  v, c,, _c )
 549 INTRA_MBCMP( sad, 16x16,  v, h, dc,  ,, _c )
 550 INTRA_MBCMP(satd, 16x16,  v, h, dc,  ,, _c )
 551
 552 #if HAVE_MMX
 553 #if HIGH_BIT_DEPTH
 554 #define x264_predict_8x8c_v_mmx2 x264_predict_8x8c_v_mmx
 555 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_c
 556 #define x264_predict_8x8c_v_sse2 x264_predict_8x8c_v_sse
 557 #define x264_predict_8x16c_v_sse2 x264_predict_8x16c_v_sse
 558 #define x264_predict_16x16_v_sse2 x264_predict_16x16_v_sse
 559 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _mmx2, _c )
 560 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _mmx2, _mmx2 )
 561 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 562 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 563 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _mmx2, _mmx2 )
 564 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _sse2, _sse2 )
 565 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _sse2 )
 566 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _sse2 )
 567 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _sse2, _sse2 )
 568 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _ssse3, _sse2 )
 569 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 570 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 571 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _ssse3, _sse2 )
 572 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _sse2 )
 573 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _sse2 )
 574 #else
 575 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_mmx
 576 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 577 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 578 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 579 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 580 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _mmx2 )
 581 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _mmx2 )
 582 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _mmx2 )
 583 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _xop, _mmx2 )
 584 #endif
 585 #endif
 586 #if !HIGH_BIT_DEPTH && HAVE_ARMV6
 587 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _neon, _armv6 )
 588 INTRA_MBCMP(satd,  4x4,   v, h, dc,  , _neon, _armv6 )
 589 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _neon, _neon )
 590 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c, _neon, _neon )
 591 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _neon, _c )
 592 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _neon, _c )
 593 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _neon, _neon )
 594 INTRA_MBCMP(satd, 16x16,  v, h, dc,  , _neon, _neon )
 595 #endif
 596 #if !HIGH_BIT_DEPTH && ARCH_AARCH64
 597 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _neon, _neon )
 598 INTRA_MBCMP(satd,  4x4,   v, h, dc,  , _neon, _neon )
 599 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _neon, _neon )
 600 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c, _neon, _neon )
 601 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _neon, _c )
 602 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _neon, _c )
 603 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _neon, _neon )
 604 INTRA_MBCMP(satd, 16x16,  v, h, dc,  , _neon, _neon )
 605 #endif
 606
 607 // No C implementation of intra_satd_x9. See checkasm for its behavior,
 608 // or see x264_mb_analyse_intra for the entirely different algorithm we
 609 // use when lacking an asm implementation of it.
 610
 611
 612
 613 /****************************************************************************
 614  * structural similarity metric
 615  ****************************************************************************/
 616 static void ssim_4x4x2_core( const pixel *pix1, intptr_t stride1,
 617                              const pixel *pix2, intptr_t stride2,
 618                              int sums[2][4] )
 619 {
 620     for( int z = 0; z < 2; z++ )
 621     {
 622         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 623         for( int y = 0; y < 4; y++ )
 624             for( int x = 0; x < 4; x++ )
 625             {
 626                 int a = pix1[x+y*stride1];
 627                 int b = pix2[x+y*stride2];
 628                 s1  += a;
 629                 s2  += b;
 630                 ss  += a*a;
 631                 ss  += b*b;
 632                 s12 += a*b;
 633             }
 634         sums[z][0] = s1;
 635         sums[z][1] = s2;
 636         sums[z][2] = ss;
 637         sums[z][3] = s12;
 638         pix1 += 4;
 639         pix2 += 4;
 640     }
 641 }
 642
 643 static float ssim_end1( int s1, int s2, int ss, int s12 )
 644 {
 645 /* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
 646  * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
 647  * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
 648 #if BIT_DEPTH > 9
 649 #define type float
 650     static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
 651     static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
 652 #else
 653 #define type int
 654     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 655     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 656 #endif
 657     type fs1 = s1;
 658     type fs2 = s2;
 659     type fss = ss;
 660     type fs12 = s12;
 661     type vars = fss*64 - fs1*fs1 - fs2*fs2;
 662     type covar = fs12*64 - fs1*fs2;
 663     return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
 664          / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
 665 #undef type
 666 }
 667
 668 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 669 {
 670     float ssim = 0.0;
 671     for( int i = 0; i < width; i++ )
 672         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 673                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 674                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 675                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 676     return ssim;
 677 }
 678
 679 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 680                            pixel *pix1, intptr_t stride1,
 681                            pixel *pix2, intptr_t stride2,
 682                            int width, int height, void *buf, int *cnt )
 683 {
 684     int z = 0;
 685     float ssim = 0.0;
 686     int (*sum0)[4] = buf;
 687     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 688     width >>= 2;
 689     height >>= 2;
 690     for( int y = 1; y < height; y++ )
 691     {
 692         for( ; z <= y; z++ )
 693         {
 694             XCHG( void*, sum0, sum1 );
 695             for( int x = 0; x < width; x+=2 )
 696                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 697         }
 698         for( int x = 0; x < width-1; x += 4 )
 699             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 700     }
 701     *cnt = (height-1) * (width-1);
 702     return ssim;
 703 }
 704
 705 static int pixel_vsad( pixel *src, intptr_t stride, int height )
 706 {
 707     int score = 0;
 708     for( int i = 1; i < height; i++, src += stride )
 709         for( int j = 0; j < 16; j++ )
 710             score += abs(src[j] - src[j+stride]);
 711     return score;
 712 }
 713
 714 int x264_field_vsad( x264_t *h, int mb_x, int mb_y )
 715 {
 716     int score_field, score_frame;
 717     int stride = h->fenc->i_stride[0];
 718     int mb_stride = h->mb.i_mb_stride;
 719     pixel *fenc = h->fenc->plane[0] + 16 * (mb_x + mb_y * stride);
 720     int mb_xy = mb_x + mb_y*mb_stride;
 721
 722     /* We don't want to analyze pixels outside the frame, as it gives inaccurate results. */
 723     int mbpair_height = X264_MIN( h->param.i_height - mb_y * 16, 32 );
 724     score_frame  = h->pixf.vsad( fenc,          stride, mbpair_height );
 725     score_field  = h->pixf.vsad( fenc,        stride*2, mbpair_height >> 1 );
 726     score_field += h->pixf.vsad( fenc+stride, stride*2, mbpair_height >> 1 );
 727
 728     if( mb_x > 0 )
 729         score_field += 512 - h->mb.field[mb_xy        -1]*1024;
 730     if( mb_y > 0 )
 731         score_field += 512 - h->mb.field[mb_xy-mb_stride]*1024;
 732
 733     return (score_field < score_frame);
 734 }
 735
 736 static int pixel_asd8( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2, int height )
 737 {
 738     int sum = 0;
 739     for( int y = 0; y < height; y++, pix1 += stride1, pix2 += stride2 )
 740         for( int x = 0; x < 8; x++ )
 741             sum += pix1[x] - pix2[x];
 742     return abs( sum );
 743 }
 744
 745 /****************************************************************************
 746  * successive elimination
 747  ****************************************************************************/
 748 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 749                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 750 {
 751     int nmv = 0;
 752     for( int i = 0; i < width; i++, sums++ )
 753     {
 754         int ads = abs( enc_dc[0] - sums[0] )
 755                 + abs( enc_dc[1] - sums[8] )
 756                 + abs( enc_dc[2] - sums[delta] )
 757                 + abs( enc_dc[3] - sums[delta+8] )
 758                 + cost_mvx[i];
 759         if( ads < thresh )
 760             mvs[nmv++] = i;
 761     }
 762     return nmv;
 763 }
 764
 765 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 766                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 767 {
 768     int nmv = 0;
 769     for( int i = 0; i < width; i++, sums++ )
 770     {
 771         int ads = abs( enc_dc[0] - sums[0] )
 772                 + abs( enc_dc[1] - sums[delta] )
 773                 + cost_mvx[i];
 774         if( ads < thresh )
 775             mvs[nmv++] = i;
 776     }
 777     return nmv;
 778 }
 779
 780 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 781                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 782 {
 783     int nmv = 0;
 784     for( int i = 0; i<width; i++, sums++ )
 785     {
 786         int ads = abs( enc_dc[0] - sums[0] )
 787                 + cost_mvx[i];
 788         if( ads < thresh )
 789             mvs[nmv++] = i;
 790     }
 791     return nmv;
 792 }
 793
 794
 795 /****************************************************************************
 796  * x264_pixel_init:
 797  ****************************************************************************/
 798 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 799 {
 800     memset( pixf, 0, sizeof(*pixf) );
 801
 802 #define INIT2_NAME( name1, name2, cpu ) \
 803     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 804     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 805 #define INIT4_NAME( name1, name2, cpu ) \
 806     INIT2_NAME( name1, name2, cpu ) \
 807     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 808     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 809 #define INIT5_NAME( name1, name2, cpu ) \
 810     INIT4_NAME( name1, name2, cpu ) \
 811     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 812 #define INIT6_NAME( name1, name2, cpu ) \
 813     INIT5_NAME( name1, name2, cpu ) \
 814     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 815 #define INIT7_NAME( name1, name2, cpu ) \
 816     INIT6_NAME( name1, name2, cpu ) \
 817     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 818 #define INIT8_NAME( name1, name2, cpu ) \
 819     INIT7_NAME( name1, name2, cpu ) \
 820     pixf->name1[PIXEL_4x16]  = x264_pixel_##name2##_4x16##cpu;
 821 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 822 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 823 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 824 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 825 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 826 #define INIT8( name, cpu ) INIT8_NAME( name, name, cpu )
 827
 828 #define INIT_ADS( cpu ) \
 829     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 830     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 831     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 832
 833     INIT8( sad, );
 834     INIT8_NAME( sad_aligned, sad, );
 835     INIT7( sad_x3, );
 836     INIT7( sad_x4, );
 837     INIT8( ssd, );
 838     INIT8( satd, );
 839     INIT7( satd_x3, );
 840     INIT7( satd_x4, );
 841     INIT4( hadamard_ac, );
 842     INIT_ADS( );
 843
 844     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 845     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 846     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 847     pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16;
 848     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 849     pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16;
 850     pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8;
 851
 852     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 853     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 854     pixf->ssim_end4 = ssim_end4;
 855     pixf->vsad = pixel_vsad;
 856     pixf->asd8 = pixel_asd8;
 857
 858     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 859     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 860     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 861     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 862     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 863     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 864     pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c;
 865     pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c;
 866     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 867     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 868
 869 #if HIGH_BIT_DEPTH
 870 #if HAVE_MMX
 871     if( cpu&X264_CPU_MMX2 )
 872     {
 873         INIT7( sad, _mmx2 );
 874         INIT7_NAME( sad_aligned, sad, _mmx2 );
 875         INIT7( sad_x3, _mmx2 );
 876         INIT7( sad_x4, _mmx2 );
 877         INIT8( satd, _mmx2 );
 878         INIT7( satd_x3, _mmx2 );
 879         INIT7( satd_x4, _mmx2 );
 880         INIT4( hadamard_ac, _mmx2 );
 881         INIT8( ssd, _mmx2 );
 882         INIT_ADS( _mmx2 );
 883
 884         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_mmx2;
 885         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
 886         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
 887 #if ARCH_X86
 888         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_mmx2;
 889         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
 890 #endif
 891
 892         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
 893         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
 894         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
 895         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
 896         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
 897         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
 898         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
 899         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
 900         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
 901     }
 902     if( cpu&X264_CPU_SSE2 )
 903     {
 904         INIT4_NAME( sad_aligned, sad, _sse2_aligned );
 905         INIT5( ssd, _sse2 );
 906         INIT6( satd, _sse2 );
 907         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse2;
 908
 909         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 910         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 911 #if ARCH_X86_64
 912         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 913         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
 914 #endif
 915         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_sse2;
 916         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_sse2;
 917         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 918         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 919         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 920         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_sse2;
 921         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_sse2;
 922         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_sse2;
 923         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_sse2;
 924     }
 925     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 926     {
 927         INIT5( sad, _sse2 );
 928         INIT2( sad_x3, _sse2 );
 929         INIT2( sad_x4, _sse2 );
 930         INIT_ADS( _sse2 );
 931
 932         if( !(cpu&X264_CPU_STACK_MOD4) )
 933         {
 934             INIT4( hadamard_ac, _sse2 );
 935         }
 936         pixf->vsad = x264_pixel_vsad_sse2;
 937         pixf->asd8 = x264_pixel_asd8_sse2;
 938         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_sse2;
 939         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_sse2;
 940         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
 941         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
 942         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_sse2;
 943     }
 944     if( cpu&X264_CPU_SSE2_IS_FAST )
 945     {
 946         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 947         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 948         pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_sse2;
 949         pixf->sad_x3[PIXEL_8x4]  = x264_pixel_sad_x3_8x4_sse2;
 950         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 951         pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_sse2;
 952         pixf->sad_x4[PIXEL_8x4]  = x264_pixel_sad_x4_8x4_sse2;
 953     }
 954     if( cpu&X264_CPU_SSSE3 )
 955     {
 956         INIT4_NAME( sad_aligned, sad, _ssse3_aligned );
 957         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_ssse3;
 958         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_ssse3;
 959         INIT7( sad, _ssse3 );
 960         INIT7( sad_x3, _ssse3 );
 961         INIT7( sad_x4, _ssse3 );
 962         INIT_ADS( _ssse3 );
 963         INIT6( satd, _ssse3 );
 964         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_ssse3;
 965
 966         if( !(cpu&X264_CPU_STACK_MOD4) )
 967         {
 968             INIT4( hadamard_ac, _ssse3 );
 969         }
 970         pixf->vsad = x264_pixel_vsad_ssse3;
 971         pixf->asd8 = x264_pixel_asd8_ssse3;
 972         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_ssse3;
 973         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 974         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 975 #if ARCH_X86_64
 976         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
 977 #endif
 978         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_ssse3;
 979         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_ssse3;
 980         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 981         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_ssse3;
 982         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
 983         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 984     }
 985     if( cpu&X264_CPU_SSE4 )
 986     {
 987         INIT6( satd, _sse4 );
 988         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse4;
 989         if( !(cpu&X264_CPU_STACK_MOD4) )
 990         {
 991             INIT4( hadamard_ac, _sse4 );
 992         }
 993         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 994         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 995 #if ARCH_X86_64
 996         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
 997 #endif
 998         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
 999     }
1000     if( cpu&X264_CPU_AVX )
1001     {
1002         INIT5_NAME( sad_aligned, sad, _ssse3 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
1003         INIT_ADS( _avx );
1004         INIT6( satd, _avx );
1005         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_avx;
1006         if( !(cpu&X264_CPU_STACK_MOD4) )
1007         {
1008             INIT4( hadamard_ac, _avx );
1009         }
1010         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_avx;
1011         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1012         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1013         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1014         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1015         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1016         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1017         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1018 #if ARCH_X86_64
1019         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1020 #endif
1021         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1022     }
1023     if( cpu&X264_CPU_XOP )
1024     {
1025         INIT5( sad_x3, _xop );
1026         INIT5( sad_x4, _xop );
1027         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_xop;
1028         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1029         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1030         pixf->vsad = x264_pixel_vsad_xop;
1031         pixf->asd8 = x264_pixel_asd8_xop;
1032 #if ARCH_X86_64
1033         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1034 #endif
1035     }
1036     if( cpu&X264_CPU_AVX2 )
1037     {
1038         INIT2( ssd, _avx2 );
1039         INIT2( sad, _avx2 );
1040         INIT2_NAME( sad_aligned, sad, _avx2 );
1041         INIT2( sad_x3, _avx2 );
1042         INIT2( sad_x4, _avx2 );
1043         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx2;
1044         pixf->vsad = x264_pixel_vsad_avx2;
1045         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1046         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_avx2;
1047     }
1048 #endif // HAVE_MMX
1049 #else // !HIGH_BIT_DEPTH
1050 #if HAVE_MMX
1051     if( cpu&X264_CPU_MMX )
1052     {
1053         INIT8( ssd, _mmx );
1054     }
1055
1056     if( cpu&X264_CPU_MMX2 )
1057     {
1058         INIT8( sad, _mmx2 );
1059         INIT8_NAME( sad_aligned, sad, _mmx2 );
1060         INIT7( sad_x3, _mmx2 );
1061         INIT7( sad_x4, _mmx2 );
1062         INIT8( satd, _mmx2 );
1063         INIT7( satd_x3, _mmx2 );
1064         INIT7( satd_x4, _mmx2 );
1065         INIT4( hadamard_ac, _mmx2 );
1066         INIT_ADS( _mmx2 );
1067         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
1068         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_mmx2;
1069         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
1070         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmx2;
1071 #if ARCH_X86
1072         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmx2;
1073         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmx2;
1074         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmx2;
1075         pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_mmx2;
1076         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_mmx2;
1077         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
1078         pixf->vsad = x264_pixel_vsad_mmx2;
1079
1080         if( cpu&X264_CPU_CACHELINE_32 )
1081         {
1082             INIT5( sad, _cache32_mmx2 );
1083             INIT4( sad_x3, _cache32_mmx2 );
1084             INIT4( sad_x4, _cache32_mmx2 );
1085         }
1086         else if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1087         {
1088             INIT5( sad, _cache64_mmx2 );
1089             INIT4( sad_x3, _cache64_mmx2 );
1090             INIT4( sad_x4, _cache64_mmx2 );
1091         }
1092 #else
1093         if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1094         {
1095             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmx2;
1096             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmx2;
1097             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmx2;
1098             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmx2;
1099             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmx2;
1100             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmx2;
1101             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmx2;
1102         }
1103 #endif
1104         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
1105         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
1106         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
1107         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
1108         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
1109         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
1110         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
1111         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
1112         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
1113     }
1114
1115     if( cpu&X264_CPU_SSE2 )
1116     {
1117         INIT5( ssd, _sse2slow );
1118         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
1119         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
1120         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
1121         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
1122         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
1123         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
1124         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
1125 #if ARCH_X86_64
1126         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
1127         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
1128 #endif
1129         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_sse2;
1130         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_sse2;
1131         pixf->vsad = x264_pixel_vsad_sse2;
1132         pixf->asd8 = x264_pixel_asd8_sse2;
1133     }
1134
1135     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
1136     {
1137         INIT2( sad, _sse2 );
1138         INIT2( sad_x3, _sse2 );
1139         INIT2( sad_x4, _sse2 );
1140         INIT6( satd, _sse2 );
1141         pixf->satd[PIXEL_4x16]   = x264_pixel_satd_4x16_sse2;
1142         INIT6( satd_x3, _sse2 );
1143         INIT6( satd_x4, _sse2 );
1144         INIT4( hadamard_ac, _sse2 );
1145         INIT_ADS( _sse2 );
1146         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
1147         pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_sse2;
1148         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
1149         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
1150         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
1151         if( cpu&X264_CPU_CACHELINE_64 )
1152         {
1153             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
1154 #if ARCH_X86
1155             INIT2( sad, _cache64_sse2 );
1156             INIT2( sad_x3, _cache64_sse2 );
1157             INIT2( sad_x4, _cache64_sse2 );
1158 #endif
1159            if( cpu&X264_CPU_SSE2_IS_FAST )
1160            {
1161                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
1162                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
1163            }
1164         }
1165     }
1166
1167     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
1168     {
1169         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1170         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1171         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
1172         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
1173         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
1174         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
1175         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
1176         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
1177     }
1178
1179     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
1180     {
1181         INIT2( sad, _sse3 );
1182         INIT2( sad_x3, _sse3 );
1183         INIT2( sad_x4, _sse3 );
1184     }
1185
1186     if( cpu&X264_CPU_SSSE3 )
1187     {
1188         INIT4( hadamard_ac, _ssse3 );
1189         if( !(cpu&X264_CPU_STACK_MOD4) )
1190         {
1191             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_ssse3;
1192             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_ssse3;
1193             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_ssse3;
1194 #if ARCH_X86_64
1195             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_ssse3;
1196 #endif
1197         }
1198         INIT_ADS( _ssse3 );
1199         if( cpu&X264_CPU_SLOW_ATOM )
1200         {
1201             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3_atom;
1202             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3_atom;
1203             INIT6( satd, _ssse3_atom );
1204             pixf->satd[PIXEL_4x16]  = x264_pixel_satd_4x16_ssse3_atom;
1205             INIT6( satd_x3, _ssse3_atom );
1206             INIT6( satd_x4, _ssse3_atom );
1207             INIT4( hadamard_ac, _ssse3_atom );
1208 #if ARCH_X86_64
1209             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3_atom;
1210 #endif
1211         }
1212         else
1213         {
1214             INIT8( ssd, _ssse3 );
1215             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
1216             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
1217             INIT8( satd, _ssse3 );
1218             INIT7( satd_x3, _ssse3 );
1219             INIT7( satd_x4, _ssse3 );
1220 #if ARCH_X86_64
1221             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
1222 #endif
1223         }
1224         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
1225         if( !(cpu&X264_CPU_SLOW_PSHUFB) )
1226             pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
1227         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
1228         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
1229         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
1230         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_ssse3;
1231         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_ssse3;
1232         pixf->asd8 = x264_pixel_asd8_ssse3;
1233         if( cpu&X264_CPU_CACHELINE_64 )
1234         {
1235             INIT2( sad, _cache64_ssse3 );
1236             INIT2( sad_x3, _cache64_ssse3 );
1237             INIT2( sad_x4, _cache64_ssse3 );
1238         }
1239         else
1240         {
1241             INIT2( sad_x3, _ssse3 );
1242             INIT5( sad_x4, _ssse3 );
1243         }
1244         if( (cpu&X264_CPU_SLOW_ATOM) || (cpu&X264_CPU_SLOW_SHUFFLE) )
1245         {
1246             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
1247         }
1248     }
1249
1250     if( cpu&X264_CPU_SSE4 )
1251     {
1252         INIT8( satd, _sse4 );
1253         INIT7( satd_x3, _sse4 );
1254         INIT7( satd_x4, _sse4 );
1255         INIT4( hadamard_ac, _sse4 );
1256         if( !(cpu&X264_CPU_STACK_MOD4) )
1257         {
1258             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_sse4;
1259             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_sse4;
1260             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_sse4;
1261 #if ARCH_X86_64
1262             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_sse4;
1263 #endif
1264         }
1265         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
1266         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
1267         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
1268 #if ARCH_X86_64
1269         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
1270 #endif
1271     }
1272
1273     if( cpu&X264_CPU_AVX )
1274     {
1275         INIT2_NAME( sad_aligned, sad, _sse2 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
1276         INIT2( sad_x3, _avx );
1277         INIT2( sad_x4, _avx );
1278         INIT8( satd, _avx );
1279         INIT7( satd_x3, _avx );
1280         INIT7( satd_x4, _avx );
1281         INIT_ADS( _avx );
1282         INIT4( hadamard_ac, _avx );
1283         if( !(cpu&X264_CPU_STACK_MOD4) )
1284         {
1285             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_avx;
1286             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_avx;
1287             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx;
1288 #if ARCH_X86_64
1289             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_avx;
1290 #endif
1291         }
1292         INIT5( ssd, _avx );
1293         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1294         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1295         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1296         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1297         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1298         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_avx;
1299         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1300         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1301         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1302 #if ARCH_X86_64
1303         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1304 #endif
1305     }
1306
1307     if( cpu&X264_CPU_XOP )
1308     {
1309         INIT7( satd, _xop );
1310         INIT7( satd_x3, _xop );
1311         INIT7( satd_x4, _xop );
1312         INIT4( hadamard_ac, _xop );
1313         if( !(cpu&X264_CPU_STACK_MOD4) )
1314         {
1315             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_xop;
1316         }
1317         INIT5( ssd, _xop );
1318         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_xop;
1319         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_xop;
1320         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_xop;
1321         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_xop;
1322         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1323         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_xop;
1324         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1325         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_xop;
1326         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_xop;
1327 #if ARCH_X86_64
1328         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1329 #endif
1330     }
1331
1332     if( cpu&X264_CPU_AVX2 )
1333     {
1334         INIT2( ssd, _avx2 );
1335         INIT2( sad_x3, _avx2 );
1336         INIT2( sad_x4, _avx2 );
1337         INIT4( satd, _avx2 );
1338         INIT2( hadamard_ac, _avx2 );
1339         INIT_ADS( _avx2 );
1340         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx2;
1341         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx2;
1342         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_avx2;
1343         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_avx2;
1344         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_avx2;
1345         pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx2;
1346         pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_avx2;
1347         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1348 #if ARCH_X86_64
1349         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx2;
1350 #endif
1351     }
1352 #endif //HAVE_MMX
1353
1354 #if HAVE_ARMV6
1355     if( cpu&X264_CPU_ARMV6 )
1356     {
1357         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1358         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1359         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1360         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1361     }
1362     if( cpu&X264_CPU_NEON )
1363     {
1364         INIT5( sad, _neon );
1365         INIT5( sad_aligned, _neon );
1366         INIT7( sad_x3, _neon );
1367         INIT7( sad_x4, _neon );
1368         INIT7( ssd, _neon );
1369         INIT7( satd, _neon );
1370         INIT7( satd_x3, _neon );
1371         INIT7( satd_x4, _neon );
1372         INIT4( hadamard_ac, _neon );
1373         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1374         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1375         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1376         pixf->var[PIXEL_8x16]   = x264_pixel_var_8x16_neon;
1377         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1378         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
1379         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
1380
1381         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_neon;
1382         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_neon;
1383         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_neon;
1384         pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8_neon;
1385         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_neon;
1386         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_neon;
1387         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_neon;
1388         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
1389         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_neon;
1390         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
1391
1392         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1393         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1394
1395         if( cpu&X264_CPU_FAST_NEON_MRC )
1396         {
1397             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
1398             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
1399             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
1400             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
1401         }
1402         else    // really just scheduled for dual issue / A8
1403         {
1404             INIT5( sad_aligned, _neon_dual );
1405         }
1406     }
1407 #endif
1408
1409 #if ARCH_AARCH64
1410     if( cpu&X264_CPU_NEON )
1411     {
1412         INIT8( sad, _neon );
1413         // AArch64 has no distinct instructions for aligned load/store
1414         INIT8_NAME( sad_aligned, sad, _neon );
1415         INIT7( sad_x3, _neon );
1416         INIT7( sad_x4, _neon );
1417         INIT8( ssd, _neon );
1418         INIT8( satd, _neon );
1419         INIT7( satd_x3, _neon );
1420         INIT7( satd_x4, _neon );
1421         INIT4( hadamard_ac, _neon );
1422
1423         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1424         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1425         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_neon;
1426
1427         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1428         pixf->var[PIXEL_8x16]   = x264_pixel_var_8x16_neon;
1429         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1430         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
1431         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
1432         pixf->vsad = x264_pixel_vsad_neon;
1433
1434         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_neon;
1435         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_neon;
1436         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_neon;
1437         pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8_neon;
1438         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_neon;
1439         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_neon;
1440         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_neon;
1441         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
1442         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_neon;
1443         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
1444
1445         pixf->ssd_nv12_core     = x264_pixel_ssd_nv12_core_neon;
1446         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1447         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1448     }
1449 #endif // ARCH_AARCH64
1450
1451 #endif // HIGH_BIT_DEPTH
1452 #if HAVE_ALTIVEC
1453     if( cpu&X264_CPU_ALTIVEC )
1454     {
1455         x264_pixel_altivec_init( pixf );
1456     }
1457 #endif
1458
1459     pixf->ads[PIXEL_8x16] =
1460     pixf->ads[PIXEL_8x4] =
1461     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
1462     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
1463 }
1464