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

   1 /*****************************************************************************
   2  * pixel.c: pixel metrics
   3  *****************************************************************************
   4  * Copyright (C) 2003-2016 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 #if ARCH_MIPS
  46 #   include "mips/pixel.h"
  47 #endif
  48
  49
  50 /****************************************************************************
  51  * pixel_sad_WxH
  52  ****************************************************************************/
  53 #define PIXEL_SAD_C( name, lx, ly ) \
  54 static int name( pixel *pix1, intptr_t i_stride_pix1,  \
  55                  pixel *pix2, intptr_t i_stride_pix2 ) \
  56 {                                                   \
  57     int i_sum = 0;                                  \
  58     for( int y = 0; y < ly; y++ )                   \
  59     {                                               \
  60         for( int x = 0; x < lx; x++ )               \
  61         {                                           \
  62             i_sum += abs( pix1[x] - pix2[x] );      \
  63         }                                           \
  64         pix1 += i_stride_pix1;                      \
  65         pix2 += i_stride_pix2;                      \
  66     }                                               \
  67     return i_sum;                                   \
  68 }
  69
  70
  71 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  72 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  73 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  74 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  75 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  76 PIXEL_SAD_C( x264_pixel_sad_4x16,   4, 16 )
  77 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  78 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  79
  80 /****************************************************************************
  81  * pixel_ssd_WxH
  82  ****************************************************************************/
  83 #define PIXEL_SSD_C( name, lx, ly ) \
  84 static int name( pixel *pix1, intptr_t i_stride_pix1,  \
  85                  pixel *pix2, intptr_t i_stride_pix2 ) \
  86 {                                                   \
  87     int i_sum = 0;                                  \
  88     for( int y = 0; y < ly; y++ )                   \
  89     {                                               \
  90         for( int x = 0; x < lx; x++ )               \
  91         {                                           \
  92             int d = pix1[x] - pix2[x];              \
  93             i_sum += d*d;                           \
  94         }                                           \
  95         pix1 += i_stride_pix1;                      \
  96         pix2 += i_stride_pix2;                      \
  97     }                                               \
  98     return i_sum;                                   \
  99 }
 100
 101 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
 102 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
 103 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
 104 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
 105 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
 106 PIXEL_SSD_C( x264_pixel_ssd_4x16,   4, 16 )
 107 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
 108 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
 109
 110 uint64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, pixel *pix1, intptr_t i_pix1,
 111                              pixel *pix2, intptr_t i_pix2, int i_width, int i_height )
 112 {
 113     uint64_t i_ssd = 0;
 114     int y;
 115     int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
 116
 117 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 118                                           pix2 + y*i_pix2 + x, i_pix2 );
 119     for( y = 0; y < i_height-15; y += 16 )
 120     {
 121         int x = 0;
 122         if( align )
 123             for( ; x < i_width-15; x += 16 )
 124                 SSD(PIXEL_16x16);
 125         for( ; x < i_width-7; x += 8 )
 126             SSD(PIXEL_8x16);
 127     }
 128     if( y < i_height-7 )
 129         for( int x = 0; x < i_width-7; x += 8 )
 130             SSD(PIXEL_8x8);
 131 #undef SSD
 132
 133 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 134     if( i_width & 7 )
 135     {
 136         for( y = 0; y < (i_height & ~7); y++ )
 137             for( int x = i_width & ~7; x < i_width; x++ )
 138                 SSD1;
 139     }
 140     if( i_height & 7 )
 141     {
 142         for( y = i_height & ~7; y < i_height; y++ )
 143             for( int x = 0; x < i_width; x++ )
 144                 SSD1;
 145     }
 146 #undef SSD1
 147
 148     return i_ssd;
 149 }
 150
 151 static void pixel_ssd_nv12_core( pixel *pixuv1, intptr_t stride1, pixel *pixuv2, intptr_t stride2,
 152                                  int width, int height, uint64_t *ssd_u, uint64_t *ssd_v )
 153 {
 154     *ssd_u = 0, *ssd_v = 0;
 155     for( int y = 0; y < height; y++, pixuv1+=stride1, pixuv2+=stride2 )
 156         for( int x = 0; x < width; x++ )
 157         {
 158             int du = pixuv1[2*x]   - pixuv2[2*x];
 159             int dv = pixuv1[2*x+1] - pixuv2[2*x+1];
 160             *ssd_u += du*du;
 161             *ssd_v += dv*dv;
 162         }
 163 }
 164
 165 void x264_pixel_ssd_nv12( x264_pixel_function_t *pf, pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2,
 166                           int i_width, int i_height, uint64_t *ssd_u, uint64_t *ssd_v )
 167 {
 168     pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height, ssd_u, ssd_v );
 169     if( i_width&7 )
 170     {
 171         uint64_t tmp[2];
 172         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] );
 173         *ssd_u += tmp[0];
 174         *ssd_v += tmp[1];
 175     }
 176 }
 177
 178 /****************************************************************************
 179  * pixel_var_wxh
 180  ****************************************************************************/
 181 #define PIXEL_VAR_C( name, w, h ) \
 182 static uint64_t name( pixel *pix, intptr_t i_stride ) \
 183 {                                             \
 184     uint32_t sum = 0, sqr = 0;                \
 185     for( int y = 0; y < h; y++ )              \
 186     {                                         \
 187         for( int x = 0; x < w; x++ )          \
 188         {                                     \
 189             sum += pix[x];                    \
 190             sqr += pix[x] * pix[x];           \
 191         }                                     \
 192         pix += i_stride;                      \
 193     }                                         \
 194     return sum + ((uint64_t)sqr << 32);       \
 195 }
 196
 197 PIXEL_VAR_C( x264_pixel_var_16x16, 16, 16 )
 198 PIXEL_VAR_C( x264_pixel_var_8x16,   8, 16 )
 199 PIXEL_VAR_C( x264_pixel_var_8x8,    8,  8 )
 200
 201 /****************************************************************************
 202  * pixel_var2_wxh
 203  ****************************************************************************/
 204 #define PIXEL_VAR2_C( name, w, h, shift ) \
 205 static int name( pixel *pix1, intptr_t i_stride1, pixel *pix2, intptr_t i_stride2, int *ssd ) \
 206 { \
 207     int var = 0, sum = 0, sqr = 0; \
 208     for( int y = 0; y < h; y++ ) \
 209     { \
 210         for( int x = 0; x < w; x++ ) \
 211         { \
 212             int diff = pix1[x] - pix2[x]; \
 213             sum += diff; \
 214             sqr += diff * diff; \
 215         } \
 216         pix1 += i_stride1; \
 217         pix2 += i_stride2; \
 218     } \
 219     var = sqr - ((int64_t)sum * sum >> shift); \
 220     *ssd = sqr; \
 221     return var; \
 222 }
 223
 224 PIXEL_VAR2_C( x264_pixel_var2_8x16, 8, 16, 7 )
 225 PIXEL_VAR2_C( x264_pixel_var2_8x8,  8,  8, 6 )
 226
 227 #if BIT_DEPTH > 8
 228     typedef uint32_t sum_t;
 229     typedef uint64_t sum2_t;
 230 #else
 231     typedef uint16_t sum_t;
 232     typedef uint32_t sum2_t;
 233 #endif
 234 #define BITS_PER_SUM (8 * sizeof(sum_t))
 235
 236 #define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
 237     sum2_t t0 = s0 + s1;\
 238     sum2_t t1 = s0 - s1;\
 239     sum2_t t2 = s2 + s3;\
 240     sum2_t t3 = s2 - s3;\
 241     d0 = t0 + t2;\
 242     d2 = t0 - t2;\
 243     d1 = t1 + t3;\
 244     d3 = t1 - t3;\
 245 }
 246
 247 // in: a pseudo-simd number of the form x+(y<<16)
 248 // return: abs(x)+(abs(y)<<16)
 249 static ALWAYS_INLINE sum2_t abs2( sum2_t a )
 250 {
 251     sum2_t s = ((a>>(BITS_PER_SUM-1))&(((sum2_t)1<<BITS_PER_SUM)+1))*((sum_t)-1);
 252     return (a+s)^s;
 253 }
 254
 255 /****************************************************************************
 256  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 257  ****************************************************************************/
 258
 259 static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 260 {
 261     sum2_t tmp[4][2];
 262     sum2_t a0, a1, a2, a3, b0, b1;
 263     sum2_t sum = 0;
 264     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 265     {
 266         a0 = pix1[0] - pix2[0];
 267         a1 = pix1[1] - pix2[1];
 268         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 269         a2 = pix1[2] - pix2[2];
 270         a3 = pix1[3] - pix2[3];
 271         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 272         tmp[i][0] = b0 + b1;
 273         tmp[i][1] = b0 - b1;
 274     }
 275     for( int i = 0; i < 2; i++ )
 276     {
 277         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 278         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 279         sum += ((sum_t)a0) + (a0>>BITS_PER_SUM);
 280     }
 281     return sum >> 1;
 282 }
 283
 284 static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 285 {
 286     sum2_t tmp[4][4];
 287     sum2_t a0, a1, a2, a3;
 288     sum2_t sum = 0;
 289     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 290     {
 291         a0 = (pix1[0] - pix2[0]) + ((sum2_t)(pix1[4] - pix2[4]) << BITS_PER_SUM);
 292         a1 = (pix1[1] - pix2[1]) + ((sum2_t)(pix1[5] - pix2[5]) << BITS_PER_SUM);
 293         a2 = (pix1[2] - pix2[2]) + ((sum2_t)(pix1[6] - pix2[6]) << BITS_PER_SUM);
 294         a3 = (pix1[3] - pix2[3]) + ((sum2_t)(pix1[7] - pix2[7]) << BITS_PER_SUM);
 295         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 296     }
 297     for( int i = 0; i < 4; i++ )
 298     {
 299         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 300         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 301     }
 302     return (((sum_t)sum) + (sum>>BITS_PER_SUM)) >> 1;
 303 }
 304
 305 #define PIXEL_SATD_C( w, h, sub )\
 306 static int x264_pixel_satd_##w##x##h( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )\
 307 {\
 308     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 309             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 310     if( w==16 )\
 311         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 312             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 313     if( h==16 )\
 314         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 315             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 316     if( w==16 && h==16 )\
 317         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 318             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 319     return sum;\
 320 }
 321 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 322 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 323 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 324 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 325 PIXEL_SATD_C( 4,  16, x264_pixel_satd_4x4 )
 326 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 327
 328 static NOINLINE int sa8d_8x8( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 329 {
 330     sum2_t tmp[8][4];
 331     sum2_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
 332     sum2_t sum = 0;
 333     for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
 334     {
 335         a0 = pix1[0] - pix2[0];
 336         a1 = pix1[1] - pix2[1];
 337         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 338         a2 = pix1[2] - pix2[2];
 339         a3 = pix1[3] - pix2[3];
 340         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 341         a4 = pix1[4] - pix2[4];
 342         a5 = pix1[5] - pix2[5];
 343         b2 = (a4+a5) + ((a4-a5)<<BITS_PER_SUM);
 344         a6 = pix1[6] - pix2[6];
 345         a7 = pix1[7] - pix2[7];
 346         b3 = (a6+a7) + ((a6-a7)<<BITS_PER_SUM);
 347         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 348     }
 349     for( int i = 0; i < 4; i++ )
 350     {
 351         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 352         HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 353         b0  = abs2(a0+a4) + abs2(a0-a4);
 354         b0 += abs2(a1+a5) + abs2(a1-a5);
 355         b0 += abs2(a2+a6) + abs2(a2-a6);
 356         b0 += abs2(a3+a7) + abs2(a3-a7);
 357         sum += (sum_t)b0 + (b0>>BITS_PER_SUM);
 358     }
 359     return sum;
 360 }
 361
 362 static int x264_pixel_sa8d_8x8( 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     return (sum+2)>>2;
 366 }
 367
 368 static int x264_pixel_sa8d_16x16( pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2 )
 369 {
 370     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 371             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 372             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 373             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 374     return (sum+2)>>2;
 375 }
 376
 377 static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, intptr_t stride )
 378 {
 379     sum2_t tmp[32];
 380     sum2_t a0, a1, a2, a3, dc;
 381     sum2_t sum4 = 0, sum8 = 0;
 382     for( int i = 0; i < 8; i++, pix+=stride )
 383     {
 384         sum2_t *t = tmp + (i&3) + (i&4)*4;
 385         a0 = (pix[0]+pix[1]) + ((sum2_t)(pix[0]-pix[1])<<BITS_PER_SUM);
 386         a1 = (pix[2]+pix[3]) + ((sum2_t)(pix[2]-pix[3])<<BITS_PER_SUM);
 387         t[0] = a0 + a1;
 388         t[4] = a0 - a1;
 389         a2 = (pix[4]+pix[5]) + ((sum2_t)(pix[4]-pix[5])<<BITS_PER_SUM);
 390         a3 = (pix[6]+pix[7]) + ((sum2_t)(pix[6]-pix[7])<<BITS_PER_SUM);
 391         t[8] = a2 + a3;
 392         t[12] = a2 - a3;
 393     }
 394     for( int i = 0; i < 8; i++ )
 395     {
 396         HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 397         tmp[i*4+0] = a0;
 398         tmp[i*4+1] = a1;
 399         tmp[i*4+2] = a2;
 400         tmp[i*4+3] = a3;
 401         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 402     }
 403     for( int i = 0; i < 8; i++ )
 404     {
 405         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 406         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 407     }
 408     dc = (sum_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 409     sum4 = (sum_t)sum4 + (sum4>>BITS_PER_SUM) - dc;
 410     sum8 = (sum_t)sum8 + (sum8>>BITS_PER_SUM) - dc;
 411     return ((uint64_t)sum8<<32) + sum4;
 412 }
 413
 414 #define HADAMARD_AC(w,h) \
 415 static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, intptr_t stride )\
 416 {\
 417     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 418     if( w==16 )\
 419         sum += pixel_hadamard_ac( pix+8, stride );\
 420     if( h==16 )\
 421         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 422     if( w==16 && h==16 )\
 423         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 424     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 425 }
 426 HADAMARD_AC( 16, 16 )
 427 HADAMARD_AC( 16, 8 )
 428 HADAMARD_AC( 8, 16 )
 429 HADAMARD_AC( 8, 8 )
 430
 431
 432 /****************************************************************************
 433  * pixel_sad_x4
 434  ****************************************************************************/
 435 #define SAD_X( size ) \
 436 static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 437                                       intptr_t i_stride, int scores[3] )\
 438 {\
 439     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 440     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 441     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 442 }\
 443 static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1,pixel *pix2, pixel *pix3,\
 444                                       intptr_t i_stride, int scores[4] )\
 445 {\
 446     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 447     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 448     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 449     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 450 }
 451
 452 SAD_X( 16x16 )
 453 SAD_X( 16x8 )
 454 SAD_X( 8x16 )
 455 SAD_X( 8x8 )
 456 SAD_X( 8x4 )
 457 SAD_X( 4x8 )
 458 SAD_X( 4x4 )
 459
 460 /****************************************************************************
 461  * pixel_satd_x4
 462  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 463  ****************************************************************************/
 464
 465 #define SATD_X( size, cpu ) \
 466 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 467                                             intptr_t i_stride, int scores[3] )\
 468 {\
 469     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 470     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 471     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 472 }\
 473 static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3,\
 474                                             intptr_t i_stride, int scores[4] )\
 475 {\
 476     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 477     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 478     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 479     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 480 }
 481 #define SATD_X_DECL6( cpu )\
 482 SATD_X( 16x16, cpu )\
 483 SATD_X( 16x8, cpu )\
 484 SATD_X( 8x16, cpu )\
 485 SATD_X( 8x8, cpu )\
 486 SATD_X( 8x4, cpu )\
 487 SATD_X( 4x8, cpu )
 488 #define SATD_X_DECL7( cpu )\
 489 SATD_X_DECL6( cpu )\
 490 SATD_X( 4x4, cpu )
 491
 492 SATD_X_DECL7()
 493 #if HAVE_MMX
 494 SATD_X_DECL7( _mmx2 )
 495 #if !HIGH_BIT_DEPTH
 496 SATD_X_DECL6( _sse2 )
 497 SATD_X_DECL7( _ssse3 )
 498 SATD_X_DECL6( _ssse3_atom )
 499 SATD_X_DECL7( _sse4 )
 500 SATD_X_DECL7( _avx )
 501 SATD_X_DECL7( _xop )
 502 #endif // !HIGH_BIT_DEPTH
 503 #endif
 504
 505 #if !HIGH_BIT_DEPTH
 506 #if HAVE_ARMV6 || ARCH_AARCH64
 507 SATD_X_DECL7( _neon )
 508 #endif
 509 #endif // !HIGH_BIT_DEPTH
 510
 511 #define INTRA_MBCMP_8x8( mbcmp, cpu, cpu2 )\
 512 void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[36], int res[3] )\
 513 {\
 514     ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
 515     x264_predict_8x8_v##cpu2( pix, edge );\
 516     res[0] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 517     x264_predict_8x8_h##cpu2( pix, edge );\
 518     res[1] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 519     x264_predict_8x8_dc##cpu2( pix, edge );\
 520     res[2] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 521 }
 522
 523 INTRA_MBCMP_8x8( sad,, _c )
 524 INTRA_MBCMP_8x8(sa8d,, _c )
 525 #if HIGH_BIT_DEPTH && HAVE_MMX
 526 #define x264_predict_8x8_v_sse2 x264_predict_8x8_v_sse
 527 INTRA_MBCMP_8x8( sad, _mmx2,  _c )
 528 INTRA_MBCMP_8x8(sa8d, _sse2,  _sse2 )
 529 #endif
 530 #if !HIGH_BIT_DEPTH && (HAVE_ARMV6 || ARCH_AARCH64)
 531 INTRA_MBCMP_8x8( sad, _neon, _neon )
 532 INTRA_MBCMP_8x8(sa8d, _neon, _neon )
 533 #endif
 534
 535 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu, cpu2 )\
 536 void x264_intra_##mbcmp##_x3_##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
 537 {\
 538     x264_predict_##size##chroma##_##pred1##cpu2( fdec );\
 539     res[0] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 540     x264_predict_##size##chroma##_##pred2##cpu2( fdec );\
 541     res[1] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 542     x264_predict_##size##chroma##_##pred3##cpu2( fdec );\
 543     res[2] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 544 }
 545
 546 INTRA_MBCMP( sad,  4x4,   v, h, dc,  ,, _c )
 547 INTRA_MBCMP(satd,  4x4,   v, h, dc,  ,, _c )
 548 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c,, _c )
 549 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c,, _c )
 550 INTRA_MBCMP( sad,  8x16, dc, h,  v, c,, _c )
 551 INTRA_MBCMP(satd,  8x16, dc, h,  v, c,, _c )
 552 INTRA_MBCMP( sad, 16x16,  v, h, dc,  ,, _c )
 553 INTRA_MBCMP(satd, 16x16,  v, h, dc,  ,, _c )
 554
 555 #if HAVE_MMX
 556 #if HIGH_BIT_DEPTH
 557 #define x264_predict_8x8c_v_mmx2 x264_predict_8x8c_v_mmx
 558 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_c
 559 #define x264_predict_8x8c_v_sse2 x264_predict_8x8c_v_sse
 560 #define x264_predict_8x16c_v_sse2 x264_predict_8x16c_v_sse
 561 #define x264_predict_16x16_v_sse2 x264_predict_16x16_v_sse
 562 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _mmx2, _c )
 563 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _mmx2, _mmx2 )
 564 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 565 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 566 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _mmx2, _mmx2 )
 567 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _sse2, _sse2 )
 568 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _sse2 )
 569 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _sse2 )
 570 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _sse2, _sse2 )
 571 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _ssse3, _sse2 )
 572 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 573 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 574 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _ssse3, _sse2 )
 575 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _sse2 )
 576 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _sse2 )
 577 #else
 578 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_mmx
 579 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 580 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 581 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 582 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 583 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _mmx2 )
 584 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _mmx2 )
 585 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _mmx2 )
 586 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _xop, _mmx2 )
 587 #endif
 588 #endif
 589 #if !HIGH_BIT_DEPTH && HAVE_ARMV6
 590 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _neon, _armv6 )
 591 INTRA_MBCMP(satd,  4x4,   v, h, dc,  , _neon, _armv6 )
 592 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _neon, _neon )
 593 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c, _neon, _neon )
 594 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _neon, _c )
 595 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _neon, _c )
 596 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _neon, _neon )
 597 INTRA_MBCMP(satd, 16x16,  v, h, dc,  , _neon, _neon )
 598 #endif
 599 #if !HIGH_BIT_DEPTH && ARCH_AARCH64
 600 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _neon, _neon )
 601 INTRA_MBCMP(satd,  4x4,   v, h, dc,  , _neon, _neon )
 602 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _neon, _neon )
 603 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c, _neon, _neon )
 604 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _neon, _neon )
 605 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _neon, _neon )
 606 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _neon, _neon )
 607 INTRA_MBCMP(satd, 16x16,  v, h, dc,  , _neon, _neon )
 608 #endif
 609
 610 // No C implementation of intra_satd_x9. See checkasm for its behavior,
 611 // or see x264_mb_analyse_intra for the entirely different algorithm we
 612 // use when lacking an asm implementation of it.
 613
 614
 615
 616 /****************************************************************************
 617  * structural similarity metric
 618  ****************************************************************************/
 619 static void ssim_4x4x2_core( const pixel *pix1, intptr_t stride1,
 620                              const pixel *pix2, intptr_t stride2,
 621                              int sums[2][4] )
 622 {
 623     for( int z = 0; z < 2; z++ )
 624     {
 625         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 626         for( int y = 0; y < 4; y++ )
 627             for( int x = 0; x < 4; x++ )
 628             {
 629                 int a = pix1[x+y*stride1];
 630                 int b = pix2[x+y*stride2];
 631                 s1  += a;
 632                 s2  += b;
 633                 ss  += a*a;
 634                 ss  += b*b;
 635                 s12 += a*b;
 636             }
 637         sums[z][0] = s1;
 638         sums[z][1] = s2;
 639         sums[z][2] = ss;
 640         sums[z][3] = s12;
 641         pix1 += 4;
 642         pix2 += 4;
 643     }
 644 }
 645
 646 static float ssim_end1( int s1, int s2, int ss, int s12 )
 647 {
 648 /* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
 649  * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
 650  * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
 651 #if BIT_DEPTH > 9
 652 #define type float
 653     static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
 654     static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
 655 #else
 656 #define type int
 657     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 658     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 659 #endif
 660     type fs1 = s1;
 661     type fs2 = s2;
 662     type fss = ss;
 663     type fs12 = s12;
 664     type vars = fss*64 - fs1*fs1 - fs2*fs2;
 665     type covar = fs12*64 - fs1*fs2;
 666     return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
 667          / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
 668 #undef type
 669 }
 670
 671 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 672 {
 673     float ssim = 0.0;
 674     for( int i = 0; i < width; i++ )
 675         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 676                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 677                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 678                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 679     return ssim;
 680 }
 681
 682 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 683                            pixel *pix1, intptr_t stride1,
 684                            pixel *pix2, intptr_t stride2,
 685                            int width, int height, void *buf, int *cnt )
 686 {
 687     int z = 0;
 688     float ssim = 0.0;
 689     int (*sum0)[4] = buf;
 690     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 691     width >>= 2;
 692     height >>= 2;
 693     for( int y = 1; y < height; y++ )
 694     {
 695         for( ; z <= y; z++ )
 696         {
 697             XCHG( void*, sum0, sum1 );
 698             for( int x = 0; x < width; x+=2 )
 699                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 700         }
 701         for( int x = 0; x < width-1; x += 4 )
 702             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 703     }
 704     *cnt = (height-1) * (width-1);
 705     return ssim;
 706 }
 707
 708 static int pixel_vsad( pixel *src, intptr_t stride, int height )
 709 {
 710     int score = 0;
 711     for( int i = 1; i < height; i++, src += stride )
 712         for( int j = 0; j < 16; j++ )
 713             score += abs(src[j] - src[j+stride]);
 714     return score;
 715 }
 716
 717 int x264_field_vsad( x264_t *h, int mb_x, int mb_y )
 718 {
 719     int score_field, score_frame;
 720     int stride = h->fenc->i_stride[0];
 721     int mb_stride = h->mb.i_mb_stride;
 722     pixel *fenc = h->fenc->plane[0] + 16 * (mb_x + mb_y * stride);
 723     int mb_xy = mb_x + mb_y*mb_stride;
 724
 725     /* We don't want to analyze pixels outside the frame, as it gives inaccurate results. */
 726     int mbpair_height = X264_MIN( h->param.i_height - mb_y * 16, 32 );
 727     score_frame  = h->pixf.vsad( fenc,          stride, mbpair_height );
 728     score_field  = h->pixf.vsad( fenc,        stride*2, mbpair_height >> 1 );
 729     score_field += h->pixf.vsad( fenc+stride, stride*2, mbpair_height >> 1 );
 730
 731     if( mb_x > 0 )
 732         score_field += 512 - h->mb.field[mb_xy        -1]*1024;
 733     if( mb_y > 0 )
 734         score_field += 512 - h->mb.field[mb_xy-mb_stride]*1024;
 735
 736     return (score_field < score_frame);
 737 }
 738
 739 static int pixel_asd8( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2, int height )
 740 {
 741     int sum = 0;
 742     for( int y = 0; y < height; y++, pix1 += stride1, pix2 += stride2 )
 743         for( int x = 0; x < 8; x++ )
 744             sum += pix1[x] - pix2[x];
 745     return abs( sum );
 746 }
 747
 748 /****************************************************************************
 749  * successive elimination
 750  ****************************************************************************/
 751 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 752                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 753 {
 754     int nmv = 0;
 755     for( int i = 0; i < width; i++, sums++ )
 756     {
 757         int ads = abs( enc_dc[0] - sums[0] )
 758                 + abs( enc_dc[1] - sums[8] )
 759                 + abs( enc_dc[2] - sums[delta] )
 760                 + abs( enc_dc[3] - sums[delta+8] )
 761                 + cost_mvx[i];
 762         if( ads < thresh )
 763             mvs[nmv++] = i;
 764     }
 765     return nmv;
 766 }
 767
 768 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 769                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 770 {
 771     int nmv = 0;
 772     for( int i = 0; i < width; i++, sums++ )
 773     {
 774         int ads = abs( enc_dc[0] - sums[0] )
 775                 + abs( enc_dc[1] - sums[delta] )
 776                 + cost_mvx[i];
 777         if( ads < thresh )
 778             mvs[nmv++] = i;
 779     }
 780     return nmv;
 781 }
 782
 783 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 784                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 785 {
 786     int nmv = 0;
 787     for( int i = 0; i<width; i++, sums++ )
 788     {
 789         int ads = abs( enc_dc[0] - sums[0] )
 790                 + cost_mvx[i];
 791         if( ads < thresh )
 792             mvs[nmv++] = i;
 793     }
 794     return nmv;
 795 }
 796
 797
 798 /****************************************************************************
 799  * x264_pixel_init:
 800  ****************************************************************************/
 801 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 802 {
 803     memset( pixf, 0, sizeof(*pixf) );
 804
 805 #define INIT2_NAME( name1, name2, cpu ) \
 806     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 807     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 808 #define INIT4_NAME( name1, name2, cpu ) \
 809     INIT2_NAME( name1, name2, cpu ) \
 810     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 811     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 812 #define INIT5_NAME( name1, name2, cpu ) \
 813     INIT4_NAME( name1, name2, cpu ) \
 814     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 815 #define INIT6_NAME( name1, name2, cpu ) \
 816     INIT5_NAME( name1, name2, cpu ) \
 817     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 818 #define INIT7_NAME( name1, name2, cpu ) \
 819     INIT6_NAME( name1, name2, cpu ) \
 820     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 821 #define INIT8_NAME( name1, name2, cpu ) \
 822     INIT7_NAME( name1, name2, cpu ) \
 823     pixf->name1[PIXEL_4x16]  = x264_pixel_##name2##_4x16##cpu;
 824 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 825 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 826 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 827 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 828 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 829 #define INIT8( name, cpu ) INIT8_NAME( name, name, cpu )
 830
 831 #define INIT_ADS( cpu ) \
 832     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 833     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 834     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 835
 836     INIT8( sad, );
 837     INIT8_NAME( sad_aligned, sad, );
 838     INIT7( sad_x3, );
 839     INIT7( sad_x4, );
 840     INIT8( ssd, );
 841     INIT8( satd, );
 842     INIT7( satd_x3, );
 843     INIT7( satd_x4, );
 844     INIT4( hadamard_ac, );
 845     INIT_ADS( );
 846
 847     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 848     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 849     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 850     pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16;
 851     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 852     pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16;
 853     pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8;
 854
 855     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 856     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 857     pixf->ssim_end4 = ssim_end4;
 858     pixf->vsad = pixel_vsad;
 859     pixf->asd8 = pixel_asd8;
 860
 861     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 862     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 863     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 864     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 865     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 866     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 867     pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c;
 868     pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c;
 869     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 870     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 871
 872 #if HIGH_BIT_DEPTH
 873 #if HAVE_MMX
 874     if( cpu&X264_CPU_MMX2 )
 875     {
 876         INIT7( sad, _mmx2 );
 877         INIT7_NAME( sad_aligned, sad, _mmx2 );
 878         INIT7( sad_x3, _mmx2 );
 879         INIT7( sad_x4, _mmx2 );
 880         INIT8( satd, _mmx2 );
 881         INIT7( satd_x3, _mmx2 );
 882         INIT7( satd_x4, _mmx2 );
 883         INIT4( hadamard_ac, _mmx2 );
 884         INIT8( ssd, _mmx2 );
 885         INIT_ADS( _mmx2 );
 886
 887         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_mmx2;
 888         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
 889         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
 890 #if ARCH_X86
 891         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_mmx2;
 892         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
 893 #endif
 894
 895         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
 896         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
 897         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
 898         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
 899         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
 900         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
 901         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
 902         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
 903         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
 904     }
 905     if( cpu&X264_CPU_SSE2 )
 906     {
 907         INIT4_NAME( sad_aligned, sad, _sse2_aligned );
 908         INIT5( ssd, _sse2 );
 909         INIT6( satd, _sse2 );
 910         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse2;
 911
 912         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 913         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 914 #if ARCH_X86_64
 915         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 916         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
 917 #endif
 918         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_sse2;
 919         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_sse2;
 920         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 921         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 922         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 923         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_sse2;
 924         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_sse2;
 925         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_sse2;
 926         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_sse2;
 927     }
 928     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 929     {
 930         INIT5( sad, _sse2 );
 931         INIT2( sad_x3, _sse2 );
 932         INIT2( sad_x4, _sse2 );
 933         INIT_ADS( _sse2 );
 934
 935         if( !(cpu&X264_CPU_STACK_MOD4) )
 936         {
 937             INIT4( hadamard_ac, _sse2 );
 938         }
 939         pixf->vsad = x264_pixel_vsad_sse2;
 940         pixf->asd8 = x264_pixel_asd8_sse2;
 941         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_sse2;
 942         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_sse2;
 943         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
 944         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
 945         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_sse2;
 946     }
 947     if( cpu&X264_CPU_SSE2_IS_FAST )
 948     {
 949         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 950         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 951         pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_sse2;
 952         pixf->sad_x3[PIXEL_8x4]  = x264_pixel_sad_x3_8x4_sse2;
 953         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 954         pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_sse2;
 955         pixf->sad_x4[PIXEL_8x4]  = x264_pixel_sad_x4_8x4_sse2;
 956     }
 957     if( cpu&X264_CPU_SSSE3 )
 958     {
 959         INIT4_NAME( sad_aligned, sad, _ssse3_aligned );
 960         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_ssse3;
 961         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_ssse3;
 962         INIT7( sad, _ssse3 );
 963         INIT7( sad_x3, _ssse3 );
 964         INIT7( sad_x4, _ssse3 );
 965         INIT_ADS( _ssse3 );
 966         INIT6( satd, _ssse3 );
 967         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_ssse3;
 968
 969         if( !(cpu&X264_CPU_STACK_MOD4) )
 970         {
 971             INIT4( hadamard_ac, _ssse3 );
 972         }
 973         pixf->vsad = x264_pixel_vsad_ssse3;
 974         pixf->asd8 = x264_pixel_asd8_ssse3;
 975         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_ssse3;
 976         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 977         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 978 #if ARCH_X86_64
 979         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
 980 #endif
 981         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_ssse3;
 982         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_ssse3;
 983         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 984         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_ssse3;
 985         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
 986         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 987     }
 988     if( cpu&X264_CPU_SSE4 )
 989     {
 990         INIT6( satd, _sse4 );
 991         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse4;
 992         if( !(cpu&X264_CPU_STACK_MOD4) )
 993         {
 994             INIT4( hadamard_ac, _sse4 );
 995         }
 996         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 997         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 998 #if ARCH_X86_64
 999         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
1000 #endif
1001         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
1002     }
1003     if( cpu&X264_CPU_AVX )
1004     {
1005         INIT5_NAME( sad_aligned, sad, _ssse3 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
1006         INIT_ADS( _avx );
1007         INIT6( satd, _avx );
1008         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_avx;
1009         if( !(cpu&X264_CPU_STACK_MOD4) )
1010         {
1011             INIT4( hadamard_ac, _avx );
1012         }
1013         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_avx;
1014         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1015         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1016         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1017         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1018         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1019         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1020         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1021 #if ARCH_X86_64
1022         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1023 #endif
1024         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1025     }
1026     if( cpu&X264_CPU_XOP )
1027     {
1028         INIT5( sad_x3, _xop );
1029         INIT5( sad_x4, _xop );
1030         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_xop;
1031         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1032         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1033         pixf->vsad = x264_pixel_vsad_xop;
1034         pixf->asd8 = x264_pixel_asd8_xop;
1035 #if ARCH_X86_64
1036         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1037 #endif
1038     }
1039     if( cpu&X264_CPU_AVX2 )
1040     {
1041         INIT2( ssd, _avx2 );
1042         INIT2( sad, _avx2 );
1043         INIT2_NAME( sad_aligned, sad, _avx2 );
1044         INIT2( sad_x3, _avx2 );
1045         INIT2( sad_x4, _avx2 );
1046         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx2;
1047         pixf->vsad = x264_pixel_vsad_avx2;
1048         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1049         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_avx2;
1050     }
1051 #endif // HAVE_MMX
1052 #else // !HIGH_BIT_DEPTH
1053 #if HAVE_MMX
1054     if( cpu&X264_CPU_MMX )
1055     {
1056         INIT8( ssd, _mmx );
1057     }
1058
1059     if( cpu&X264_CPU_MMX2 )
1060     {
1061         INIT8( sad, _mmx2 );
1062         INIT8_NAME( sad_aligned, sad, _mmx2 );
1063         INIT7( sad_x3, _mmx2 );
1064         INIT7( sad_x4, _mmx2 );
1065         INIT8( satd, _mmx2 );
1066         INIT7( satd_x3, _mmx2 );
1067         INIT7( satd_x4, _mmx2 );
1068         INIT4( hadamard_ac, _mmx2 );
1069         INIT_ADS( _mmx2 );
1070         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
1071         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_mmx2;
1072         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
1073         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmx2;
1074 #if ARCH_X86
1075         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmx2;
1076         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmx2;
1077         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmx2;
1078         pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_mmx2;
1079         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_mmx2;
1080         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
1081         pixf->vsad = x264_pixel_vsad_mmx2;
1082
1083         if( cpu&X264_CPU_CACHELINE_32 )
1084         {
1085             INIT5( sad, _cache32_mmx2 );
1086             INIT4( sad_x3, _cache32_mmx2 );
1087             INIT4( sad_x4, _cache32_mmx2 );
1088         }
1089         else if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1090         {
1091             INIT5( sad, _cache64_mmx2 );
1092             INIT4( sad_x3, _cache64_mmx2 );
1093             INIT4( sad_x4, _cache64_mmx2 );
1094         }
1095 #else
1096         if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1097         {
1098             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmx2;
1099             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmx2;
1100             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmx2;
1101             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmx2;
1102             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmx2;
1103             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmx2;
1104             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmx2;
1105         }
1106 #endif
1107         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
1108         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
1109         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
1110         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
1111         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
1112         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
1113         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
1114         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
1115         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
1116     }
1117
1118     if( cpu&X264_CPU_SSE2 )
1119     {
1120         INIT5( ssd, _sse2slow );
1121         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
1122         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
1123         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
1124         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
1125         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
1126         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
1127         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
1128 #if ARCH_X86_64
1129         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
1130         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
1131 #endif
1132         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_sse2;
1133         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_sse2;
1134         pixf->vsad = x264_pixel_vsad_sse2;
1135         pixf->asd8 = x264_pixel_asd8_sse2;
1136     }
1137
1138     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
1139     {
1140         INIT2( sad, _sse2 );
1141         INIT2( sad_x3, _sse2 );
1142         INIT2( sad_x4, _sse2 );
1143         INIT6( satd, _sse2 );
1144         pixf->satd[PIXEL_4x16]   = x264_pixel_satd_4x16_sse2;
1145         INIT6( satd_x3, _sse2 );
1146         INIT6( satd_x4, _sse2 );
1147         INIT4( hadamard_ac, _sse2 );
1148         INIT_ADS( _sse2 );
1149         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
1150         pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_sse2;
1151         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
1152         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
1153         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
1154         if( cpu&X264_CPU_CACHELINE_64 )
1155         {
1156             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
1157 #if ARCH_X86
1158             INIT2( sad, _cache64_sse2 );
1159             INIT2( sad_x3, _cache64_sse2 );
1160             INIT2( sad_x4, _cache64_sse2 );
1161 #endif
1162            if( cpu&X264_CPU_SSE2_IS_FAST )
1163            {
1164                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
1165                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
1166            }
1167         }
1168     }
1169
1170     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
1171     {
1172         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1173         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1174         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
1175         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
1176         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
1177         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
1178         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
1179         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
1180     }
1181
1182     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
1183     {
1184         INIT2( sad, _sse3 );
1185         INIT2( sad_x3, _sse3 );
1186         INIT2( sad_x4, _sse3 );
1187     }
1188
1189     if( cpu&X264_CPU_SSSE3 )
1190     {
1191         INIT4( hadamard_ac, _ssse3 );
1192         if( !(cpu&X264_CPU_STACK_MOD4) )
1193         {
1194             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_ssse3;
1195             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_ssse3;
1196             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_ssse3;
1197 #if ARCH_X86_64
1198             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_ssse3;
1199 #endif
1200         }
1201         INIT_ADS( _ssse3 );
1202         if( cpu&X264_CPU_SLOW_ATOM )
1203         {
1204             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3_atom;
1205             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3_atom;
1206             INIT6( satd, _ssse3_atom );
1207             pixf->satd[PIXEL_4x16]  = x264_pixel_satd_4x16_ssse3_atom;
1208             INIT6( satd_x3, _ssse3_atom );
1209             INIT6( satd_x4, _ssse3_atom );
1210             INIT4( hadamard_ac, _ssse3_atom );
1211 #if ARCH_X86_64
1212             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3_atom;
1213 #endif
1214         }
1215         else
1216         {
1217             INIT8( ssd, _ssse3 );
1218             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
1219             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
1220             INIT8( satd, _ssse3 );
1221             INIT7( satd_x3, _ssse3 );
1222             INIT7( satd_x4, _ssse3 );
1223 #if ARCH_X86_64
1224             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
1225 #endif
1226         }
1227         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
1228         if( !(cpu&X264_CPU_SLOW_PSHUFB) )
1229             pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
1230         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
1231         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
1232         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
1233         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_ssse3;
1234         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_ssse3;
1235         pixf->asd8 = x264_pixel_asd8_ssse3;
1236         if( cpu&X264_CPU_CACHELINE_64 )
1237         {
1238             INIT2( sad, _cache64_ssse3 );
1239             INIT2( sad_x3, _cache64_ssse3 );
1240             INIT2( sad_x4, _cache64_ssse3 );
1241         }
1242         else
1243         {
1244             INIT2( sad_x3, _ssse3 );
1245             INIT5( sad_x4, _ssse3 );
1246         }
1247         if( (cpu&X264_CPU_SLOW_ATOM) || (cpu&X264_CPU_SLOW_SHUFFLE) )
1248         {
1249             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
1250         }
1251     }
1252
1253     if( cpu&X264_CPU_SSE4 )
1254     {
1255         INIT8( satd, _sse4 );
1256         INIT7( satd_x3, _sse4 );
1257         INIT7( satd_x4, _sse4 );
1258         INIT4( hadamard_ac, _sse4 );
1259         if( !(cpu&X264_CPU_STACK_MOD4) )
1260         {
1261             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_sse4;
1262             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_sse4;
1263             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_sse4;
1264 #if ARCH_X86_64
1265             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_sse4;
1266 #endif
1267         }
1268         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
1269         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
1270         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
1271 #if ARCH_X86_64
1272         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
1273 #endif
1274     }
1275
1276     if( cpu&X264_CPU_AVX )
1277     {
1278         INIT2_NAME( sad_aligned, sad, _sse2 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
1279         INIT2( sad_x3, _avx );
1280         INIT2( sad_x4, _avx );
1281         INIT8( satd, _avx );
1282         INIT7( satd_x3, _avx );
1283         INIT7( satd_x4, _avx );
1284         INIT_ADS( _avx );
1285         INIT4( hadamard_ac, _avx );
1286         if( !(cpu&X264_CPU_STACK_MOD4) )
1287         {
1288             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_avx;
1289             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_avx;
1290             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx;
1291 #if ARCH_X86_64
1292             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_avx;
1293 #endif
1294         }
1295         INIT5( ssd, _avx );
1296         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1297         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1298         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1299         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1300         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1301         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_avx;
1302         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1303         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1304         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1305 #if ARCH_X86_64
1306         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1307 #endif
1308     }
1309
1310     if( cpu&X264_CPU_XOP )
1311     {
1312         INIT7( satd, _xop );
1313         INIT7( satd_x3, _xop );
1314         INIT7( satd_x4, _xop );
1315         INIT4( hadamard_ac, _xop );
1316         if( !(cpu&X264_CPU_STACK_MOD4) )
1317         {
1318             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_xop;
1319         }
1320         INIT5( ssd, _xop );
1321         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_xop;
1322         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_xop;
1323         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_xop;
1324         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_xop;
1325         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1326         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_xop;
1327         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1328         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_xop;
1329         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_xop;
1330 #if ARCH_X86_64
1331         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1332 #endif
1333     }
1334
1335     if( cpu&X264_CPU_AVX2 )
1336     {
1337         INIT2( ssd, _avx2 );
1338         INIT2( sad_x3, _avx2 );
1339         INIT2( sad_x4, _avx2 );
1340         INIT4( satd, _avx2 );
1341         INIT2( hadamard_ac, _avx2 );
1342         INIT_ADS( _avx2 );
1343         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx2;
1344         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx2;
1345         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_avx2;
1346         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_avx2;
1347         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_avx2;
1348         pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx2;
1349         pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_avx2;
1350         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1351 #if ARCH_X86_64
1352         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx2;
1353 #endif
1354     }
1355 #endif //HAVE_MMX
1356
1357 #if HAVE_ARMV6
1358     if( cpu&X264_CPU_ARMV6 )
1359     {
1360         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1361         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1362         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1363         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1364     }
1365     if( cpu&X264_CPU_NEON )
1366     {
1367         INIT5( sad, _neon );
1368         INIT5( sad_aligned, _neon );
1369         INIT7( sad_x3, _neon );
1370         INIT7( sad_x4, _neon );
1371         INIT7( ssd, _neon );
1372         INIT7( satd, _neon );
1373         INIT7( satd_x3, _neon );
1374         INIT7( satd_x4, _neon );
1375         INIT4( hadamard_ac, _neon );
1376         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1377         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1378         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_neon;
1379         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1380         pixf->var[PIXEL_8x16]   = x264_pixel_var_8x16_neon;
1381         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1382         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
1383         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
1384         pixf->vsad = x264_pixel_vsad_neon;
1385         pixf->asd8 = x264_pixel_asd8_neon;
1386
1387         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_neon;
1388         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_neon;
1389         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_neon;
1390         pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8_neon;
1391         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_neon;
1392         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_neon;
1393         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_neon;
1394         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
1395         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_neon;
1396         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
1397
1398         pixf->ssd_nv12_core     = x264_pixel_ssd_nv12_core_neon;
1399         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1400         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1401
1402         if( cpu&X264_CPU_FAST_NEON_MRC )
1403         {
1404             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
1405             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
1406             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
1407             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
1408         }
1409         else    // really just scheduled for dual issue / A8
1410         {
1411             INIT5( sad_aligned, _neon_dual );
1412         }
1413     }
1414 #endif
1415
1416 #if ARCH_AARCH64
1417     if( cpu&X264_CPU_NEON )
1418     {
1419         INIT8( sad, _neon );
1420         // AArch64 has no distinct instructions for aligned load/store
1421         INIT8_NAME( sad_aligned, sad, _neon );
1422         INIT7( sad_x3, _neon );
1423         INIT7( sad_x4, _neon );
1424         INIT8( ssd, _neon );
1425         INIT8( satd, _neon );
1426         INIT7( satd_x3, _neon );
1427         INIT7( satd_x4, _neon );
1428         INIT4( hadamard_ac, _neon );
1429
1430         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1431         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1432         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_neon;
1433
1434         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1435         pixf->var[PIXEL_8x16]   = x264_pixel_var_8x16_neon;
1436         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1437         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
1438         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
1439         pixf->vsad = x264_pixel_vsad_neon;
1440         pixf->asd8 = x264_pixel_asd8_neon;
1441
1442         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_neon;
1443         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_neon;
1444         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_neon;
1445         pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8_neon;
1446         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_neon;
1447         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_neon;
1448         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_neon;
1449         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
1450         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_neon;
1451         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
1452
1453         pixf->ssd_nv12_core     = x264_pixel_ssd_nv12_core_neon;
1454         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1455         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1456     }
1457 #endif // ARCH_AARCH64
1458
1459 #if HAVE_MSA
1460     if( cpu&X264_CPU_MSA )
1461     {
1462         INIT8( sad, _msa );
1463         INIT8_NAME( sad_aligned, sad, _msa );
1464         INIT8( ssd, _msa );
1465         INIT7( sad_x3, _msa );
1466         INIT7( sad_x4, _msa );
1467         INIT8( satd, _msa );
1468         INIT4( hadamard_ac, _msa );
1469
1470         pixf->intra_sad_x3_4x4   = x264_intra_sad_x3_4x4_msa;
1471         pixf->intra_sad_x3_8x8   = x264_intra_sad_x3_8x8_msa;
1472         pixf->intra_sad_x3_8x8c  = x264_intra_sad_x3_8x8c_msa;
1473         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_msa;
1474         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_msa;
1475         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_msa;
1476         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_msa;
1477         pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8_msa;
1478
1479         pixf->ssim_4x4x2_core = x264_ssim_4x4x2_core_msa;
1480
1481         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_msa;
1482         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_msa;
1483         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_msa;
1484         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_msa;
1485         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_msa;
1486         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
1487         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
1488     }
1489 #endif // HAVE_MSA
1490
1491 #endif // HIGH_BIT_DEPTH
1492 #if HAVE_ALTIVEC
1493     if( cpu&X264_CPU_ALTIVEC )
1494     {
1495         x264_pixel_altivec_init( pixf );
1496     }
1497 #endif
1498
1499     pixf->ads[PIXEL_8x16] =
1500     pixf->ads[PIXEL_8x4] =
1501     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
1502     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
1503 }
1504