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_UltraSPARC
  42 #   include "sparc/pixel.h"
  43 #endif
  44
  45
  46 /****************************************************************************
  47  * pixel_sad_WxH
  48  ****************************************************************************/
  49 #define PIXEL_SAD_C( name, lx, ly ) \
  50 static int name( pixel *pix1, intptr_t i_stride_pix1,  \
  51                  pixel *pix2, intptr_t i_stride_pix2 ) \
  52 {                                                   \
  53     int i_sum = 0;                                  \
  54     for( int y = 0; y < ly; y++ )                   \
  55     {                                               \
  56         for( int x = 0; x < lx; x++ )               \
  57         {                                           \
  58             i_sum += abs( pix1[x] - pix2[x] );      \
  59         }                                           \
  60         pix1 += i_stride_pix1;                      \
  61         pix2 += i_stride_pix2;                      \
  62     }                                               \
  63     return i_sum;                                   \
  64 }
  65
  66
  67 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  68 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  69 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  70 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  71 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  72 PIXEL_SAD_C( x264_pixel_sad_4x16,   4, 16 )
  73 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  74 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  75
  76 /****************************************************************************
  77  * pixel_ssd_WxH
  78  ****************************************************************************/
  79 #define PIXEL_SSD_C( name, lx, ly ) \
  80 static int name( pixel *pix1, intptr_t i_stride_pix1,  \
  81                  pixel *pix2, intptr_t i_stride_pix2 ) \
  82 {                                                   \
  83     int i_sum = 0;                                  \
  84     for( int y = 0; y < ly; y++ )                   \
  85     {                                               \
  86         for( int x = 0; x < lx; x++ )               \
  87         {                                           \
  88             int d = pix1[x] - pix2[x];              \
  89             i_sum += d*d;                           \
  90         }                                           \
  91         pix1 += i_stride_pix1;                      \
  92         pix2 += i_stride_pix2;                      \
  93     }                                               \
  94     return i_sum;                                   \
  95 }
  96
  97 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
  98 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
  99 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
 100 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
 101 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
 102 PIXEL_SSD_C( x264_pixel_ssd_4x16,   4, 16 )
 103 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
 104 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
 105
 106 uint64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, pixel *pix1, intptr_t i_pix1,
 107                              pixel *pix2, intptr_t i_pix2, int i_width, int i_height )
 108 {
 109     uint64_t i_ssd = 0;
 110     int y;
 111     int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
 112
 113 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 114                                           pix2 + y*i_pix2 + x, i_pix2 );
 115     for( y = 0; y < i_height-15; y += 16 )
 116     {
 117         int x = 0;
 118         if( align )
 119             for( ; x < i_width-15; x += 16 )
 120                 SSD(PIXEL_16x16);
 121         for( ; x < i_width-7; x += 8 )
 122             SSD(PIXEL_8x16);
 123     }
 124     if( y < i_height-7 )
 125         for( int x = 0; x < i_width-7; x += 8 )
 126             SSD(PIXEL_8x8);
 127 #undef SSD
 128
 129 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 130     if( i_width & 7 )
 131     {
 132         for( y = 0; y < (i_height & ~7); y++ )
 133             for( int x = i_width & ~7; x < i_width; x++ )
 134                 SSD1;
 135     }
 136     if( i_height & 7 )
 137     {
 138         for( y = i_height & ~7; y < i_height; y++ )
 139             for( int x = 0; x < i_width; x++ )
 140                 SSD1;
 141     }
 142 #undef SSD1
 143
 144     return i_ssd;
 145 }
 146
 147 static void pixel_ssd_nv12_core( pixel *pixuv1, intptr_t stride1, pixel *pixuv2, intptr_t stride2,
 148                                  int width, int height, uint64_t *ssd_u, uint64_t *ssd_v )
 149 {
 150     *ssd_u = 0, *ssd_v = 0;
 151     for( int y = 0; y < height; y++, pixuv1+=stride1, pixuv2+=stride2 )
 152         for( int x = 0; x < width; x++ )
 153         {
 154             int du = pixuv1[2*x]   - pixuv2[2*x];
 155             int dv = pixuv1[2*x+1] - pixuv2[2*x+1];
 156             *ssd_u += du*du;
 157             *ssd_v += dv*dv;
 158         }
 159 }
 160
 161 void x264_pixel_ssd_nv12( x264_pixel_function_t *pf, pixel *pix1, intptr_t i_pix1, pixel *pix2, intptr_t i_pix2,
 162                           int i_width, int i_height, uint64_t *ssd_u, uint64_t *ssd_v )
 163 {
 164     pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height, ssd_u, ssd_v );
 165     if( i_width&7 )
 166     {
 167         uint64_t tmp[2];
 168         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] );
 169         *ssd_u += tmp[0];
 170         *ssd_v += tmp[1];
 171     }
 172 }
 173
 174 /****************************************************************************
 175  * pixel_var_wxh
 176  ****************************************************************************/
 177 #define PIXEL_VAR_C( name, w, h ) \
 178 static uint64_t name( pixel *pix, intptr_t i_stride ) \
 179 {                                             \
 180     uint32_t sum = 0, sqr = 0;                \
 181     for( int y = 0; y < h; y++ )              \
 182     {                                         \
 183         for( int x = 0; x < w; x++ )          \
 184         {                                     \
 185             sum += pix[x];                    \
 186             sqr += pix[x] * pix[x];           \
 187         }                                     \
 188         pix += i_stride;                      \
 189     }                                         \
 190     return sum + ((uint64_t)sqr << 32);       \
 191 }
 192
 193 PIXEL_VAR_C( x264_pixel_var_16x16, 16, 16 )
 194 PIXEL_VAR_C( x264_pixel_var_8x16,   8, 16 )
 195 PIXEL_VAR_C( x264_pixel_var_8x8,    8,  8 )
 196
 197 /****************************************************************************
 198  * pixel_var2_wxh
 199  ****************************************************************************/
 200 #define PIXEL_VAR2_C( name, w, h, shift ) \
 201 static int name( pixel *pix1, intptr_t i_stride1, pixel *pix2, intptr_t i_stride2, int *ssd ) \
 202 { \
 203     uint32_t var = 0, sum = 0, sqr = 0; \
 204     for( int y = 0; y < h; y++ ) \
 205     { \
 206         for( int x = 0; x < w; x++ ) \
 207         { \
 208             int diff = pix1[x] - pix2[x]; \
 209             sum += diff; \
 210             sqr += diff * diff; \
 211         } \
 212         pix1 += i_stride1; \
 213         pix2 += i_stride2; \
 214     } \
 215     sum = abs(sum); \
 216     var = sqr - ((uint64_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 #if !HIGH_BIT_DEPTH
 458 #if ARCH_UltraSPARC
 459 SAD_X( 16x16_vis )
 460 SAD_X( 16x8_vis )
 461 SAD_X( 8x16_vis )
 462 SAD_X( 8x8_vis )
 463 #endif
 464 #endif // !HIGH_BIT_DEPTH
 465
 466 /****************************************************************************
 467  * pixel_satd_x4
 468  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 469  ****************************************************************************/
 470
 471 #define SATD_X( size, cpu ) \
 472 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2,\
 473                                             intptr_t i_stride, int scores[3] )\
 474 {\
 475     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 476     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 477     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 478 }\
 479 static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3,\
 480                                             intptr_t i_stride, int scores[4] )\
 481 {\
 482     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 483     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 484     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 485     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 486 }
 487 #define SATD_X_DECL6( cpu )\
 488 SATD_X( 16x16, cpu )\
 489 SATD_X( 16x8, cpu )\
 490 SATD_X( 8x16, cpu )\
 491 SATD_X( 8x8, cpu )\
 492 SATD_X( 8x4, cpu )\
 493 SATD_X( 4x8, cpu )
 494 #define SATD_X_DECL7( cpu )\
 495 SATD_X_DECL6( cpu )\
 496 SATD_X( 4x4, cpu )
 497
 498 SATD_X_DECL7()
 499 #if HAVE_MMX
 500 SATD_X_DECL7( _mmx2 )
 501 #if !HIGH_BIT_DEPTH
 502 SATD_X_DECL6( _sse2 )
 503 SATD_X_DECL7( _ssse3 )
 504 SATD_X_DECL6( _ssse3_atom )
 505 SATD_X_DECL7( _sse4 )
 506 SATD_X_DECL7( _avx )
 507 SATD_X_DECL7( _xop )
 508 #endif // !HIGH_BIT_DEPTH
 509 #endif
 510
 511 #if !HIGH_BIT_DEPTH
 512 #if HAVE_ARMV6
 513 SATD_X_DECL7( _neon )
 514 #endif
 515 #endif // !HIGH_BIT_DEPTH
 516
 517 #define INTRA_MBCMP_8x8( mbcmp, cpu, cpu2 )\
 518 void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[36], int res[3] )\
 519 {\
 520     ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
 521     x264_predict_8x8_v##cpu2( pix, edge );\
 522     res[0] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 523     x264_predict_8x8_h##cpu2( pix, edge );\
 524     res[1] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 525     x264_predict_8x8_dc##cpu2( pix, edge );\
 526     res[2] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 527 }
 528
 529 INTRA_MBCMP_8x8( sad,, _c )
 530 INTRA_MBCMP_8x8(sa8d,, _c )
 531 #if HIGH_BIT_DEPTH && HAVE_MMX
 532 #define x264_predict_8x8_v_sse2 x264_predict_8x8_v_sse
 533 INTRA_MBCMP_8x8( sad, _mmx2,  _c )
 534 INTRA_MBCMP_8x8(sa8d, _sse2,  _sse2 )
 535 #endif
 536 #if !HIGH_BIT_DEPTH && HAVE_ARMV6
 537 INTRA_MBCMP_8x8( sad, _neon, _neon )
 538 INTRA_MBCMP_8x8(sa8d, _neon, _neon )
 539 #endif
 540
 541 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu, cpu2 )\
 542 void x264_intra_##mbcmp##_x3_##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
 543 {\
 544     x264_predict_##size##chroma##_##pred1##cpu2( fdec );\
 545     res[0] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 546     x264_predict_##size##chroma##_##pred2##cpu2( fdec );\
 547     res[1] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 548     x264_predict_##size##chroma##_##pred3##cpu2( fdec );\
 549     res[2] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 550 }
 551
 552 INTRA_MBCMP( sad,  4x4,   v, h, dc,  ,, _c )
 553 INTRA_MBCMP(satd,  4x4,   v, h, dc,  ,, _c )
 554 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c,, _c )
 555 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c,, _c )
 556 INTRA_MBCMP( sad,  8x16, dc, h,  v, c,, _c )
 557 INTRA_MBCMP(satd,  8x16, dc, h,  v, c,, _c )
 558 INTRA_MBCMP( sad, 16x16,  v, h, dc,  ,, _c )
 559 INTRA_MBCMP(satd, 16x16,  v, h, dc,  ,, _c )
 560
 561 #if HAVE_MMX
 562 #if HIGH_BIT_DEPTH
 563 #define x264_predict_8x8c_v_mmx2 x264_predict_8x8c_v_mmx
 564 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_c
 565 #define x264_predict_8x8c_v_sse2 x264_predict_8x8c_v_sse
 566 #define x264_predict_8x16c_v_sse2 x264_predict_8x16c_v_sse
 567 #define x264_predict_16x16_v_sse2 x264_predict_16x16_v_sse
 568 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _mmx2, _c )
 569 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _mmx2, _mmx2 )
 570 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 571 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 572 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _mmx2, _mmx2 )
 573 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _sse2, _sse2 )
 574 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _sse2 )
 575 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _sse2 )
 576 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _sse2, _sse2 )
 577 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _ssse3, _sse2 )
 578 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 579 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _sse2 )
 580 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _ssse3, _sse2 )
 581 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _sse2 )
 582 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _sse2 )
 583 #else
 584 #define x264_predict_8x16c_v_mmx2 x264_predict_8x16c_v_mmx
 585 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 586 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _mmx2, _mmx2 )
 587 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 588 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse2, _mmx2 )
 589 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _ssse3, _mmx2 )
 590 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _sse4, _mmx2 )
 591 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _avx, _mmx2 )
 592 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _xop, _mmx2 )
 593 #endif
 594 #endif
 595 #if !HIGH_BIT_DEPTH && HAVE_ARMV6
 596 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _neon, _c )
 597 INTRA_MBCMP(satd,  4x4,   v, h, dc,  , _neon, _c )
 598 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _neon, _neon )
 599 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c, _neon, _neon )
 600 INTRA_MBCMP( sad,  8x16, dc, h,  v, c, _neon, _c )
 601 INTRA_MBCMP(satd,  8x16, dc, h,  v, c, _neon, _c )
 602 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _neon, _neon )
 603 INTRA_MBCMP(satd, 16x16,  v, h, dc,  , _neon, _neon )
 604 #endif
 605
 606 // No C implementation of intra_satd_x9. See checkasm for its behavior,
 607 // or see x264_mb_analyse_intra for the entirely different algorithm we
 608 // use when lacking an asm implementation of it.
 609
 610
 611
 612 /****************************************************************************
 613  * structural similarity metric
 614  ****************************************************************************/
 615 static void ssim_4x4x2_core( const pixel *pix1, intptr_t stride1,
 616                              const pixel *pix2, intptr_t stride2,
 617                              int sums[2][4] )
 618 {
 619     for( int z = 0; z < 2; z++ )
 620     {
 621         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 622         for( int y = 0; y < 4; y++ )
 623             for( int x = 0; x < 4; x++ )
 624             {
 625                 int a = pix1[x+y*stride1];
 626                 int b = pix2[x+y*stride2];
 627                 s1  += a;
 628                 s2  += b;
 629                 ss  += a*a;
 630                 ss  += b*b;
 631                 s12 += a*b;
 632             }
 633         sums[z][0] = s1;
 634         sums[z][1] = s2;
 635         sums[z][2] = ss;
 636         sums[z][3] = s12;
 637         pix1 += 4;
 638         pix2 += 4;
 639     }
 640 }
 641
 642 static float ssim_end1( int s1, int s2, int ss, int s12 )
 643 {
 644 /* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
 645  * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
 646  * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
 647 #if BIT_DEPTH > 9
 648 #define type float
 649     static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
 650     static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
 651 #else
 652 #define type int
 653     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 654     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 655 #endif
 656     type fs1 = s1;
 657     type fs2 = s2;
 658     type fss = ss;
 659     type fs12 = s12;
 660     type vars = fss*64 - fs1*fs1 - fs2*fs2;
 661     type covar = fs12*64 - fs1*fs2;
 662     return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
 663          / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
 664 #undef type
 665 }
 666
 667 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 668 {
 669     float ssim = 0.0;
 670     for( int i = 0; i < width; i++ )
 671         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 672                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 673                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 674                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 675     return ssim;
 676 }
 677
 678 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 679                            pixel *pix1, intptr_t stride1,
 680                            pixel *pix2, intptr_t stride2,
 681                            int width, int height, void *buf, int *cnt )
 682 {
 683     int z = 0;
 684     float ssim = 0.0;
 685     int (*sum0)[4] = buf;
 686     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 687     width >>= 2;
 688     height >>= 2;
 689     for( int y = 1; y < height; y++ )
 690     {
 691         for( ; z <= y; z++ )
 692         {
 693             XCHG( void*, sum0, sum1 );
 694             for( int x = 0; x < width; x+=2 )
 695                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 696         }
 697         for( int x = 0; x < width-1; x += 4 )
 698             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 699     }
 700     *cnt = (height-1) * (width-1);
 701     return ssim;
 702 }
 703
 704 static int pixel_vsad( pixel *src, intptr_t stride, int height )
 705 {
 706     int score = 0;
 707     for( int i = 1; i < height; i++, src += stride )
 708         for( int j = 0; j < 16; j++ )
 709             score += abs(src[j] - src[j+stride]);
 710     return score;
 711 }
 712
 713 int x264_field_vsad( x264_t *h, int mb_x, int mb_y )
 714 {
 715     int score_field, score_frame;
 716     int stride = h->fenc->i_stride[0];
 717     int mb_stride = h->mb.i_mb_stride;
 718     pixel *fenc = h->fenc->plane[0] + 16 * (mb_x + mb_y * stride);
 719     int mb_xy = mb_x + mb_y*mb_stride;
 720
 721     /* We don't want to analyze pixels outside the frame, as it gives inaccurate results. */
 722     int mbpair_height = X264_MIN( h->param.i_height - mb_y * 16, 32 );
 723     score_frame  = h->pixf.vsad( fenc,          stride, mbpair_height );
 724     score_field  = h->pixf.vsad( fenc,        stride*2, mbpair_height >> 1 );
 725     score_field += h->pixf.vsad( fenc+stride, stride*2, mbpair_height >> 1 );
 726
 727     if( mb_x > 0 )
 728         score_field += 512 - h->mb.field[mb_xy        -1]*1024;
 729     if( mb_y > 0 )
 730         score_field += 512 - h->mb.field[mb_xy-mb_stride]*1024;
 731
 732     return (score_field < score_frame);
 733 }
 734
 735 static int pixel_asd8( pixel *pix1, intptr_t stride1, pixel *pix2, intptr_t stride2, int height )
 736 {
 737     int sum = 0;
 738     for( int y = 0; y < height; y++, pix1 += stride1, pix2 += stride2 )
 739         for( int x = 0; x < 8; x++ )
 740             sum += pix1[x] - pix2[x];
 741     return abs( sum );
 742 }
 743
 744 /****************************************************************************
 745  * successive elimination
 746  ****************************************************************************/
 747 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 748                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 749 {
 750     int nmv = 0;
 751     for( int i = 0; i < width; i++, sums++ )
 752     {
 753         int ads = abs( enc_dc[0] - sums[0] )
 754                 + abs( enc_dc[1] - sums[8] )
 755                 + abs( enc_dc[2] - sums[delta] )
 756                 + abs( enc_dc[3] - sums[delta+8] )
 757                 + cost_mvx[i];
 758         if( ads < thresh )
 759             mvs[nmv++] = i;
 760     }
 761     return nmv;
 762 }
 763
 764 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 765                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 766 {
 767     int nmv = 0;
 768     for( int i = 0; i < width; i++, sums++ )
 769     {
 770         int ads = abs( enc_dc[0] - sums[0] )
 771                 + abs( enc_dc[1] - sums[delta] )
 772                 + cost_mvx[i];
 773         if( ads < thresh )
 774             mvs[nmv++] = i;
 775     }
 776     return nmv;
 777 }
 778
 779 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 780                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 781 {
 782     int nmv = 0;
 783     for( int i = 0; i<width; i++, sums++ )
 784     {
 785         int ads = abs( enc_dc[0] - sums[0] )
 786                 + cost_mvx[i];
 787         if( ads < thresh )
 788             mvs[nmv++] = i;
 789     }
 790     return nmv;
 791 }
 792
 793
 794 /****************************************************************************
 795  * x264_pixel_init:
 796  ****************************************************************************/
 797 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 798 {
 799     memset( pixf, 0, sizeof(*pixf) );
 800
 801 #define INIT2_NAME( name1, name2, cpu ) \
 802     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 803     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 804 #define INIT4_NAME( name1, name2, cpu ) \
 805     INIT2_NAME( name1, name2, cpu ) \
 806     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 807     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 808 #define INIT5_NAME( name1, name2, cpu ) \
 809     INIT4_NAME( name1, name2, cpu ) \
 810     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 811 #define INIT6_NAME( name1, name2, cpu ) \
 812     INIT5_NAME( name1, name2, cpu ) \
 813     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 814 #define INIT7_NAME( name1, name2, cpu ) \
 815     INIT6_NAME( name1, name2, cpu ) \
 816     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 817 #define INIT8_NAME( name1, name2, cpu ) \
 818     INIT7_NAME( name1, name2, cpu ) \
 819     pixf->name1[PIXEL_4x16]  = x264_pixel_##name2##_4x16##cpu;
 820 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 821 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 822 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 823 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 824 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 825 #define INIT8( name, cpu ) INIT8_NAME( name, name, cpu )
 826
 827 #define INIT_ADS( cpu ) \
 828     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 829     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 830     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 831
 832     INIT8( sad, );
 833     INIT8_NAME( sad_aligned, sad, );
 834     INIT7( sad_x3, );
 835     INIT7( sad_x4, );
 836     INIT8( ssd, );
 837     INIT8( satd, );
 838     INIT7( satd_x3, );
 839     INIT7( satd_x4, );
 840     INIT4( hadamard_ac, );
 841     INIT_ADS( );
 842
 843     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 844     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 845     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 846     pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16;
 847     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 848     pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16;
 849     pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8;
 850
 851     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 852     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 853     pixf->ssim_end4 = ssim_end4;
 854     pixf->vsad = pixel_vsad;
 855     pixf->asd8 = pixel_asd8;
 856
 857     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 858     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 859     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 860     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 861     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 862     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 863     pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c;
 864     pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c;
 865     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 866     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 867
 868 #if HIGH_BIT_DEPTH
 869 #if HAVE_MMX
 870     if( cpu&X264_CPU_MMX2 )
 871     {
 872         INIT7( sad, _mmx2 );
 873         INIT7_NAME( sad_aligned, sad, _mmx2 );
 874         INIT7( sad_x3, _mmx2 );
 875         INIT7( sad_x4, _mmx2 );
 876         INIT8( satd, _mmx2 );
 877         INIT7( satd_x3, _mmx2 );
 878         INIT7( satd_x4, _mmx2 );
 879         INIT4( hadamard_ac, _mmx2 );
 880         INIT8( ssd, _mmx2 );
 881         INIT_ADS( _mmx2 );
 882
 883         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_mmx2;
 884         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
 885         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
 886 #if ARCH_X86
 887         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_mmx2;
 888         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
 889 #endif
 890
 891         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
 892         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
 893         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
 894         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
 895         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
 896         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
 897         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
 898         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
 899         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
 900     }
 901     if( cpu&X264_CPU_SSE2 )
 902     {
 903         INIT4_NAME( sad_aligned, sad, _sse2_aligned );
 904         INIT5( ssd, _sse2 );
 905         INIT6( satd, _sse2 );
 906         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse2;
 907
 908         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 909         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 910 #if ARCH_X86_64
 911         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 912         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
 913 #endif
 914         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_sse2;
 915         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_sse2;
 916         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 917         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 918         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 919         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_sse2;
 920         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_sse2;
 921         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_sse2;
 922         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_sse2;
 923     }
 924     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 925     {
 926         INIT5( sad, _sse2 );
 927         INIT2( sad_x3, _sse2 );
 928         INIT2( sad_x4, _sse2 );
 929         INIT_ADS( _sse2 );
 930
 931         if( !(cpu&X264_CPU_STACK_MOD4) )
 932         {
 933             INIT4( hadamard_ac, _sse2 );
 934         }
 935         pixf->vsad = x264_pixel_vsad_sse2;
 936         pixf->asd8 = x264_pixel_asd8_sse2;
 937         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_sse2;
 938         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_sse2;
 939         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
 940         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
 941         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_sse2;
 942     }
 943     if( cpu&X264_CPU_SSE2_IS_FAST )
 944     {
 945         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 946         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 947         pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_sse2;
 948         pixf->sad_x3[PIXEL_8x4]  = x264_pixel_sad_x3_8x4_sse2;
 949         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 950         pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_sse2;
 951         pixf->sad_x4[PIXEL_8x4]  = x264_pixel_sad_x4_8x4_sse2;
 952     }
 953     if( cpu&X264_CPU_SSSE3 )
 954     {
 955         INIT4_NAME( sad_aligned, sad, _ssse3_aligned );
 956         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_ssse3;
 957         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_ssse3;
 958         INIT7( sad, _ssse3 );
 959         INIT7( sad_x3, _ssse3 );
 960         INIT7( sad_x4, _ssse3 );
 961         INIT_ADS( _ssse3 );
 962         INIT6( satd, _ssse3 );
 963         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_ssse3;
 964
 965         if( !(cpu&X264_CPU_STACK_MOD4) )
 966         {
 967             INIT4( hadamard_ac, _ssse3 );
 968         }
 969         pixf->vsad = x264_pixel_vsad_ssse3;
 970         pixf->asd8 = x264_pixel_asd8_ssse3;
 971         pixf->intra_sad_x3_4x4  = x264_intra_sad_x3_4x4_ssse3;
 972         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 973         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 974 #if ARCH_X86_64
 975         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
 976 #endif
 977         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_ssse3;
 978         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_ssse3;
 979         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 980         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_ssse3;
 981         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
 982         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 983     }
 984     if( cpu&X264_CPU_SSE4 )
 985     {
 986         INIT6( satd, _sse4 );
 987         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_sse4;
 988         if( !(cpu&X264_CPU_STACK_MOD4) )
 989         {
 990             INIT4( hadamard_ac, _sse4 );
 991         }
 992         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 993         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 994 #if ARCH_X86_64
 995         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
 996 #endif
 997         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
 998     }
 999     if( cpu&X264_CPU_AVX )
1000     {
1001         INIT5_NAME( sad_aligned, sad, _ssse3 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
1002         INIT_ADS( _avx );
1003         INIT6( satd, _avx );
1004         pixf->satd[PIXEL_4x16] = x264_pixel_satd_4x16_avx;
1005         if( !(cpu&X264_CPU_STACK_MOD4) )
1006         {
1007             INIT4( hadamard_ac, _avx );
1008         }
1009         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_avx;
1010         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1011         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1012         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1013         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1014         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1015         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1016         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1017 #if ARCH_X86_64
1018         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1019 #endif
1020         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1021     }
1022     if( cpu&X264_CPU_XOP )
1023     {
1024         pixf->vsad = x264_pixel_vsad_xop;
1025         pixf->asd8 = x264_pixel_asd8_xop;
1026     }
1027     if( cpu&X264_CPU_AVX2 )
1028     {
1029         INIT2( ssd, _avx2 );
1030         INIT2( sad, _avx2 );
1031         INIT2_NAME( sad_aligned, sad, _avx2 );
1032         INIT2( sad_x3, _avx2 );
1033         INIT2( sad_x4, _avx2 );
1034         pixf->vsad = x264_pixel_vsad_avx2;
1035         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1036         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_avx2;
1037     }
1038 #endif // HAVE_MMX
1039 #else // !HIGH_BIT_DEPTH
1040 #if HAVE_MMX
1041     if( cpu&X264_CPU_MMX )
1042     {
1043         INIT8( ssd, _mmx );
1044     }
1045
1046     if( cpu&X264_CPU_MMX2 )
1047     {
1048         INIT8( sad, _mmx2 );
1049         INIT8_NAME( sad_aligned, sad, _mmx2 );
1050         INIT7( sad_x3, _mmx2 );
1051         INIT7( sad_x4, _mmx2 );
1052         INIT8( satd, _mmx2 );
1053         INIT7( satd_x3, _mmx2 );
1054         INIT7( satd_x4, _mmx2 );
1055         INIT4( hadamard_ac, _mmx2 );
1056         INIT_ADS( _mmx2 );
1057         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
1058         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_mmx2;
1059         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
1060         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmx2;
1061 #if ARCH_X86
1062         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmx2;
1063         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmx2;
1064         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmx2;
1065         pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_mmx2;
1066         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_mmx2;
1067         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_mmx2;
1068         pixf->vsad = x264_pixel_vsad_mmx2;
1069
1070         if( cpu&X264_CPU_CACHELINE_32 )
1071         {
1072             INIT5( sad, _cache32_mmx2 );
1073             INIT4( sad_x3, _cache32_mmx2 );
1074             INIT4( sad_x4, _cache32_mmx2 );
1075         }
1076         else if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1077         {
1078             INIT5( sad, _cache64_mmx2 );
1079             INIT4( sad_x3, _cache64_mmx2 );
1080             INIT4( sad_x4, _cache64_mmx2 );
1081         }
1082 #else
1083         if( cpu&X264_CPU_CACHELINE_64 && !(cpu&X264_CPU_SLOW_ATOM) )
1084         {
1085             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmx2;
1086             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmx2;
1087             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmx2;
1088             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmx2;
1089             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmx2;
1090             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmx2;
1091             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmx2;
1092         }
1093 #endif
1094         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
1095         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
1096         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_mmx2;
1097         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_mmx2;
1098         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
1099         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
1100         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
1101         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
1102         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
1103     }
1104
1105     if( cpu&X264_CPU_SSE2 )
1106     {
1107         INIT5( ssd, _sse2slow );
1108         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
1109         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
1110         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
1111         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
1112         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
1113         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
1114         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
1115 #if ARCH_X86_64
1116         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
1117         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse2;
1118 #endif
1119         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_sse2;
1120         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_sse2;
1121         pixf->vsad = x264_pixel_vsad_sse2;
1122         pixf->asd8 = x264_pixel_asd8_sse2;
1123     }
1124
1125     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
1126     {
1127         INIT2( sad, _sse2 );
1128         INIT2( sad_x3, _sse2 );
1129         INIT2( sad_x4, _sse2 );
1130         INIT6( satd, _sse2 );
1131         pixf->satd[PIXEL_4x16]   = x264_pixel_satd_4x16_sse2;
1132         INIT6( satd_x3, _sse2 );
1133         INIT6( satd_x4, _sse2 );
1134         INIT4( hadamard_ac, _sse2 );
1135         INIT_ADS( _sse2 );
1136         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
1137         pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_sse2;
1138         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
1139         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse2;
1140         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_sse2;
1141         if( cpu&X264_CPU_CACHELINE_64 )
1142         {
1143             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
1144 #if ARCH_X86
1145             INIT2( sad, _cache64_sse2 );
1146             INIT2( sad_x3, _cache64_sse2 );
1147             INIT2( sad_x4, _cache64_sse2 );
1148 #endif
1149            if( cpu&X264_CPU_SSE2_IS_FAST )
1150            {
1151                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
1152                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
1153            }
1154         }
1155     }
1156
1157     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
1158     {
1159         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1160         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1161         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
1162         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
1163         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
1164         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
1165         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
1166         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
1167     }
1168
1169     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
1170     {
1171         INIT2( sad, _sse3 );
1172         INIT2( sad_x3, _sse3 );
1173         INIT2( sad_x4, _sse3 );
1174     }
1175
1176     if( cpu&X264_CPU_SSSE3 )
1177     {
1178         INIT4( hadamard_ac, _ssse3 );
1179         if( !(cpu&X264_CPU_STACK_MOD4) )
1180         {
1181             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_ssse3;
1182             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_ssse3;
1183             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_ssse3;
1184 #if ARCH_X86_64
1185             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_ssse3;
1186 #endif
1187         }
1188         INIT_ADS( _ssse3 );
1189         if( cpu&X264_CPU_SLOW_ATOM )
1190         {
1191             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3_atom;
1192             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3_atom;
1193             INIT6( satd, _ssse3_atom );
1194             pixf->satd[PIXEL_4x16]  = x264_pixel_satd_4x16_ssse3_atom;
1195             INIT6( satd_x3, _ssse3_atom );
1196             INIT6( satd_x4, _ssse3_atom );
1197             INIT4( hadamard_ac, _ssse3_atom );
1198 #if ARCH_X86_64
1199             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3_atom;
1200 #endif
1201         }
1202         else
1203         {
1204             INIT8( ssd, _ssse3 );
1205             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
1206             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
1207             INIT8( satd, _ssse3 );
1208             INIT7( satd_x3, _ssse3 );
1209             INIT7( satd_x4, _ssse3 );
1210 #if ARCH_X86_64
1211             pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_ssse3;
1212 #endif
1213         }
1214         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
1215         if( !(cpu&X264_CPU_SLOW_PSHUFB) )
1216             pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
1217         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_ssse3;
1218         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
1219         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
1220         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_ssse3;
1221         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_ssse3;
1222         pixf->asd8 = x264_pixel_asd8_ssse3;
1223         if( cpu&X264_CPU_CACHELINE_64 )
1224         {
1225             INIT2( sad, _cache64_ssse3 );
1226             INIT2( sad_x3, _cache64_ssse3 );
1227             INIT2( sad_x4, _cache64_ssse3 );
1228         }
1229         else
1230         {
1231             INIT2( sad_x3, _ssse3 );
1232             INIT5( sad_x4, _ssse3 );
1233         }
1234         if( (cpu&X264_CPU_SLOW_ATOM) || (cpu&X264_CPU_SLOW_SHUFFLE) )
1235         {
1236             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
1237         }
1238     }
1239
1240     if( cpu&X264_CPU_SSE4 )
1241     {
1242         INIT8( satd, _sse4 );
1243         INIT7( satd_x3, _sse4 );
1244         INIT7( satd_x4, _sse4 );
1245         INIT4( hadamard_ac, _sse4 );
1246         if( !(cpu&X264_CPU_STACK_MOD4) )
1247         {
1248             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_sse4;
1249             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_sse4;
1250             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_sse4;
1251 #if ARCH_X86_64
1252             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_sse4;
1253 #endif
1254         }
1255         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
1256         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
1257         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_sse4;
1258 #if ARCH_X86_64
1259         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_sse4;
1260 #endif
1261     }
1262
1263     if( cpu&X264_CPU_AVX )
1264     {
1265         INIT2_NAME( sad_aligned, sad, _sse2 ); /* AVX-capable CPUs doesn't benefit from an aligned version */
1266         INIT2( sad_x3, _avx );
1267         INIT2( sad_x4, _avx );
1268         INIT8( satd, _avx );
1269         INIT7( satd_x3, _avx );
1270         INIT7( satd_x4, _avx );
1271         INIT_ADS( _avx );
1272         INIT4( hadamard_ac, _avx );
1273         if( !(cpu&X264_CPU_STACK_MOD4) )
1274         {
1275             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_avx;
1276             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_avx;
1277             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx;
1278 #if ARCH_X86_64
1279             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_avx;
1280 #endif
1281         }
1282         INIT5( ssd, _avx );
1283         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1284         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1285         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_avx;
1286         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1287         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1288         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_avx;
1289         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1290         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1291         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1292 #if ARCH_X86_64
1293         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx;
1294 #endif
1295     }
1296
1297     if( cpu&X264_CPU_XOP )
1298     {
1299         INIT7( satd, _xop );
1300         INIT7( satd_x3, _xop );
1301         INIT7( satd_x4, _xop );
1302         INIT4( hadamard_ac, _xop );
1303         if( !(cpu&X264_CPU_STACK_MOD4) )
1304         {
1305             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_xop;
1306         }
1307         INIT5( ssd, _xop );
1308         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_xop;
1309         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_xop;
1310         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_xop;
1311         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1312         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_xop;
1313         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1314         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_xop;
1315         pixf->var2[PIXEL_8x16] = x264_pixel_var2_8x16_xop;
1316 #if ARCH_X86_64
1317         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_xop;
1318 #endif
1319     }
1320
1321     if( cpu&X264_CPU_AVX2 )
1322     {
1323         INIT2( ssd, _avx2 );
1324         INIT2( sad_x3, _avx2 );
1325         INIT2( sad_x4, _avx2 );
1326         INIT4( satd, _avx2 );
1327         INIT2( hadamard_ac, _avx2 );
1328         INIT_ADS( _avx2 );
1329         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx2;
1330         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx2;
1331         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_avx2;
1332         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_avx2;
1333         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_avx2;
1334         pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx2;
1335         pixf->intra_sad_x3_8x8c = x264_intra_sad_x3_8x8c_avx2;
1336         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_avx2;
1337 #if ARCH_X86_64
1338         pixf->sa8d_satd[PIXEL_16x16] = x264_pixel_sa8d_satd_16x16_avx2;
1339 #endif
1340     }
1341 #endif //HAVE_MMX
1342
1343 #if HAVE_ARMV6
1344     if( cpu&X264_CPU_ARMV6 )
1345     {
1346         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1347         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1348         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1349         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1350     }
1351     if( cpu&X264_CPU_NEON )
1352     {
1353         INIT5( sad, _neon );
1354         INIT5( sad_aligned, _neon );
1355         INIT7( sad_x3, _neon );
1356         INIT7( sad_x4, _neon );
1357         INIT7( ssd, _neon );
1358         INIT7( satd, _neon );
1359         INIT7( satd_x3, _neon );
1360         INIT7( satd_x4, _neon );
1361         INIT4( hadamard_ac, _neon );
1362         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1363         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1364         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1365         pixf->var[PIXEL_8x16]   = x264_pixel_var_8x16_neon;
1366         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1367         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
1368         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
1369
1370         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_neon;
1371         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_neon;
1372         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_neon;
1373         pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8_neon;
1374         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_neon;
1375         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_neon;
1376         pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c_neon;
1377         pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c_neon;
1378         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_neon;
1379         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_neon;
1380
1381         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1382         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1383
1384         if( cpu&X264_CPU_FAST_NEON_MRC )
1385         {
1386             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
1387             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
1388             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
1389             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
1390         }
1391         else    // really just scheduled for dual issue / A8
1392         {
1393             INIT5( sad_aligned, _neon_dual );
1394         }
1395     }
1396 #endif
1397 #endif // HIGH_BIT_DEPTH
1398 #if HAVE_ALTIVEC
1399     if( cpu&X264_CPU_ALTIVEC )
1400     {
1401         x264_pixel_altivec_init( pixf );
1402     }
1403 #endif
1404 #if !HIGH_BIT_DEPTH
1405 #if ARCH_UltraSPARC
1406     INIT4( sad, _vis );
1407     INIT4( sad_x3, _vis );
1408     INIT4( sad_x4, _vis );
1409 #endif
1410 #endif // !HIGH_BIT_DEPTH
1411
1412     pixf->ads[PIXEL_8x16] =
1413     pixf->ads[PIXEL_8x4] =
1414     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
1415     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
1416 }
1417