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

   1 /*****************************************************************************
   2  * pixel.c: pixel metrics
   3  *****************************************************************************
   4  * Copyright (C) 2003-2011 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 #endif
  40 #if ARCH_UltraSPARC
  41 #   include "sparc/pixel.h"
  42 #endif
  43
  44
  45 /****************************************************************************
  46  * pixel_sad_WxH
  47  ****************************************************************************/
  48 #define PIXEL_SAD_C( name, lx, ly ) \
  49 static int name( pixel *pix1, int i_stride_pix1,  \
  50                  pixel *pix2, int i_stride_pix2 ) \
  51 {                                                   \
  52     int i_sum = 0;                                  \
  53     for( int y = 0; y < ly; y++ )                   \
  54     {                                               \
  55         for( int x = 0; x < lx; x++ )               \
  56         {                                           \
  57             i_sum += abs( pix1[x] - pix2[x] );      \
  58         }                                           \
  59         pix1 += i_stride_pix1;                      \
  60         pix2 += i_stride_pix2;                      \
  61     }                                               \
  62     return i_sum;                                   \
  63 }
  64
  65
  66 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  67 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  68 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  69 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  70 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  71 PIXEL_SAD_C( x264_pixel_sad_4x16,   4, 16 )
  72 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  73 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  74
  75 /****************************************************************************
  76  * pixel_ssd_WxH
  77  ****************************************************************************/
  78 #define PIXEL_SSD_C( name, lx, ly ) \
  79 static int name( pixel *pix1, int i_stride_pix1,  \
  80                  pixel *pix2, int i_stride_pix2 ) \
  81 {                                                   \
  82     int i_sum = 0;                                  \
  83     for( int y = 0; y < ly; y++ )                   \
  84     {                                               \
  85         for( int x = 0; x < lx; x++ )               \
  86         {                                           \
  87             int d = pix1[x] - pix2[x];              \
  88             i_sum += d*d;                           \
  89         }                                           \
  90         pix1 += i_stride_pix1;                      \
  91         pix2 += i_stride_pix2;                      \
  92     }                                               \
  93     return i_sum;                                   \
  94 }
  95
  96 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
  97 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
  98 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
  99 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
 100 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
 101 PIXEL_SSD_C( x264_pixel_ssd_4x16,   4, 16 )
 102 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
 103 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
 104
 105 uint64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, pixel *pix1, int i_pix1, pixel *pix2, int i_pix2, int i_width, int i_height )
 106 {
 107     uint64_t i_ssd = 0;
 108     int y;
 109     int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
 110
 111 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 112                                           pix2 + y*i_pix2 + x, i_pix2 );
 113     for( y = 0; y < i_height-15; y += 16 )
 114     {
 115         int x = 0;
 116         if( align )
 117             for( ; x < i_width-15; x += 16 )
 118                 SSD(PIXEL_16x16);
 119         for( ; x < i_width-7; x += 8 )
 120             SSD(PIXEL_8x16);
 121     }
 122     if( y < i_height-7 )
 123         for( int x = 0; x < i_width-7; x += 8 )
 124             SSD(PIXEL_8x8);
 125 #undef SSD
 126
 127 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 128     if( i_width & 7 )
 129     {
 130         for( y = 0; y < (i_height & ~7); y++ )
 131             for( int x = i_width & ~7; x < i_width; x++ )
 132                 SSD1;
 133     }
 134     if( i_height & 7 )
 135     {
 136         for( y = i_height & ~7; y < i_height; y++ )
 137             for( int x = 0; x < i_width; x++ )
 138                 SSD1;
 139     }
 140 #undef SSD1
 141
 142     return i_ssd;
 143 }
 144
 145 static void pixel_ssd_nv12_core( pixel *pixuv1, int stride1, pixel *pixuv2, int stride2, int width, int height, uint64_t *ssd_u, uint64_t *ssd_v )
 146 {
 147     *ssd_u = 0, *ssd_v = 0;
 148     for( int y = 0; y < height; y++, pixuv1+=stride1, pixuv2+=stride2 )
 149         for( int x = 0; x < width; x++ )
 150         {
 151             int du = pixuv1[2*x]   - pixuv2[2*x];
 152             int dv = pixuv1[2*x+1] - pixuv2[2*x+1];
 153             *ssd_u += du*du;
 154             *ssd_v += dv*dv;
 155         }
 156 }
 157
 158 void x264_pixel_ssd_nv12( x264_pixel_function_t *pf, pixel *pix1, int i_pix1, pixel *pix2, int i_pix2, int i_width, int i_height, uint64_t *ssd_u, uint64_t *ssd_v )
 159 {
 160     pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height, ssd_u, ssd_v );
 161     if( i_width&7 )
 162     {
 163         uint64_t tmp[2];
 164         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] );
 165         *ssd_u += tmp[0];
 166         *ssd_v += tmp[1];
 167     }
 168 }
 169
 170 /****************************************************************************
 171  * pixel_var_wxh
 172  ****************************************************************************/
 173 #define PIXEL_VAR_C( name, w, h ) \
 174 static uint64_t name( pixel *pix, int i_stride ) \
 175 {                                             \
 176     uint32_t sum = 0, sqr = 0;                \
 177     for( int y = 0; y < h; y++ )              \
 178     {                                         \
 179         for( int x = 0; x < w; x++ )          \
 180         {                                     \
 181             sum += pix[x];                    \
 182             sqr += pix[x] * pix[x];           \
 183         }                                     \
 184         pix += i_stride;                      \
 185     }                                         \
 186     return sum + ((uint64_t)sqr << 32);       \
 187 }
 188
 189 PIXEL_VAR_C( x264_pixel_var_16x16, 16, 16 )
 190 PIXEL_VAR_C( x264_pixel_var_8x16,   8, 16 )
 191 PIXEL_VAR_C( x264_pixel_var_8x8,    8,  8 )
 192
 193 /****************************************************************************
 194  * pixel_var2_wxh
 195  ****************************************************************************/
 196 #define PIXEL_VAR2_C( name, w, h, shift ) \
 197 static int name( pixel *pix1, int i_stride1, pixel *pix2, int i_stride2, int *ssd ) \
 198 { \
 199     uint32_t var = 0, sum = 0, sqr = 0; \
 200     for( int y = 0; y < h; y++ ) \
 201     { \
 202         for( int x = 0; x < w; x++ ) \
 203         { \
 204             int diff = pix1[x] - pix2[x]; \
 205             sum += diff; \
 206             sqr += diff * diff; \
 207         } \
 208         pix1 += i_stride1; \
 209         pix2 += i_stride2; \
 210     } \
 211     sum = abs(sum); \
 212     var = sqr - ((uint64_t)sum * sum >> shift); \
 213     *ssd = sqr; \
 214     return var; \
 215 }
 216
 217 PIXEL_VAR2_C( x264_pixel_var2_8x16, 8, 16, 7 )
 218 PIXEL_VAR2_C( x264_pixel_var2_8x8,  8,  8, 6 )
 219
 220 #if BIT_DEPTH > 8
 221     typedef uint32_t sum_t;
 222     typedef uint64_t sum2_t;
 223 #else
 224     typedef uint16_t sum_t;
 225     typedef uint32_t sum2_t;
 226 #endif
 227 #define BITS_PER_SUM (8 * sizeof(sum_t))
 228
 229 #define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
 230     sum2_t t0 = s0 + s1;\
 231     sum2_t t1 = s0 - s1;\
 232     sum2_t t2 = s2 + s3;\
 233     sum2_t t3 = s2 - s3;\
 234     d0 = t0 + t2;\
 235     d2 = t0 - t2;\
 236     d1 = t1 + t3;\
 237     d3 = t1 - t3;\
 238 }
 239
 240 // in: a pseudo-simd number of the form x+(y<<16)
 241 // return: abs(x)+(abs(y)<<16)
 242 static ALWAYS_INLINE sum2_t abs2( sum2_t a )
 243 {
 244     sum2_t s = ((a>>(BITS_PER_SUM-1))&(((sum2_t)1<<BITS_PER_SUM)+1))*((sum_t)-1);
 245     return (a+s)^s;
 246 }
 247
 248 /****************************************************************************
 249  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 250  ****************************************************************************/
 251
 252 static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 253 {
 254     sum2_t tmp[4][2];
 255     sum2_t a0, a1, a2, a3, b0, b1;
 256     sum2_t sum = 0;
 257     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 258     {
 259         a0 = pix1[0] - pix2[0];
 260         a1 = pix1[1] - pix2[1];
 261         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 262         a2 = pix1[2] - pix2[2];
 263         a3 = pix1[3] - pix2[3];
 264         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 265         tmp[i][0] = b0 + b1;
 266         tmp[i][1] = b0 - b1;
 267     }
 268     for( int i = 0; i < 2; i++ )
 269     {
 270         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 271         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 272         sum += ((sum_t)a0) + (a0>>BITS_PER_SUM);
 273     }
 274     return sum >> 1;
 275 }
 276
 277 static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 278 {
 279     sum2_t tmp[4][4];
 280     sum2_t a0, a1, a2, a3;
 281     sum2_t sum = 0;
 282     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 283     {
 284         a0 = (pix1[0] - pix2[0]) + ((sum2_t)(pix1[4] - pix2[4]) << BITS_PER_SUM);
 285         a1 = (pix1[1] - pix2[1]) + ((sum2_t)(pix1[5] - pix2[5]) << BITS_PER_SUM);
 286         a2 = (pix1[2] - pix2[2]) + ((sum2_t)(pix1[6] - pix2[6]) << BITS_PER_SUM);
 287         a3 = (pix1[3] - pix2[3]) + ((sum2_t)(pix1[7] - pix2[7]) << BITS_PER_SUM);
 288         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 289     }
 290     for( int i = 0; i < 4; i++ )
 291     {
 292         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 293         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 294     }
 295     return (((sum_t)sum) + (sum>>BITS_PER_SUM)) >> 1;
 296 }
 297
 298 #define PIXEL_SATD_C( w, h, sub )\
 299 static int x264_pixel_satd_##w##x##h( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )\
 300 {\
 301     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 302             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 303     if( w==16 )\
 304         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 305             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 306     if( h==16 )\
 307         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 308             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 309     if( w==16 && h==16 )\
 310         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 311             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 312     return sum;\
 313 }
 314 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 315 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 316 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 317 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 318 PIXEL_SATD_C( 4,  16, x264_pixel_satd_4x4 )
 319 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 320
 321 static NOINLINE int sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 322 {
 323     sum2_t tmp[8][4];
 324     sum2_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
 325     sum2_t sum = 0;
 326     for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
 327     {
 328         a0 = pix1[0] - pix2[0];
 329         a1 = pix1[1] - pix2[1];
 330         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 331         a2 = pix1[2] - pix2[2];
 332         a3 = pix1[3] - pix2[3];
 333         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 334         a4 = pix1[4] - pix2[4];
 335         a5 = pix1[5] - pix2[5];
 336         b2 = (a4+a5) + ((a4-a5)<<BITS_PER_SUM);
 337         a6 = pix1[6] - pix2[6];
 338         a7 = pix1[7] - pix2[7];
 339         b3 = (a6+a7) + ((a6-a7)<<BITS_PER_SUM);
 340         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 341     }
 342     for( int i = 0; i < 4; i++ )
 343     {
 344         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 345         HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 346         b0  = abs2(a0+a4) + abs2(a0-a4);
 347         b0 += abs2(a1+a5) + abs2(a1-a5);
 348         b0 += abs2(a2+a6) + abs2(a2-a6);
 349         b0 += abs2(a3+a7) + abs2(a3-a7);
 350         sum += (sum_t)b0 + (b0>>BITS_PER_SUM);
 351     }
 352     return sum;
 353 }
 354
 355 static int x264_pixel_sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 356 {
 357     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
 358     return (sum+2)>>2;
 359 }
 360
 361 static int x264_pixel_sa8d_16x16( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 362 {
 363     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 364             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 365             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 366             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 367     return (sum+2)>>2;
 368 }
 369
 370
 371 static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, int stride )
 372 {
 373     sum2_t tmp[32];
 374     sum2_t a0, a1, a2, a3, dc;
 375     sum2_t sum4 = 0, sum8 = 0;
 376     for( int i = 0; i < 8; i++, pix+=stride )
 377     {
 378         sum2_t *t = tmp + (i&3) + (i&4)*4;
 379         a0 = (pix[0]+pix[1]) + ((sum2_t)(pix[0]-pix[1])<<BITS_PER_SUM);
 380         a1 = (pix[2]+pix[3]) + ((sum2_t)(pix[2]-pix[3])<<BITS_PER_SUM);
 381         t[0] = a0 + a1;
 382         t[4] = a0 - a1;
 383         a2 = (pix[4]+pix[5]) + ((sum2_t)(pix[4]-pix[5])<<BITS_PER_SUM);
 384         a3 = (pix[6]+pix[7]) + ((sum2_t)(pix[6]-pix[7])<<BITS_PER_SUM);
 385         t[8] = a2 + a3;
 386         t[12] = a2 - a3;
 387     }
 388     for( int i = 0; i < 8; i++ )
 389     {
 390         HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 391         tmp[i*4+0] = a0;
 392         tmp[i*4+1] = a1;
 393         tmp[i*4+2] = a2;
 394         tmp[i*4+3] = a3;
 395         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 396     }
 397     for( int i = 0; i < 8; i++ )
 398     {
 399         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 400         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 401     }
 402     dc = (sum_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 403     sum4 = (sum_t)sum4 + (sum4>>BITS_PER_SUM) - dc;
 404     sum8 = (sum_t)sum8 + (sum8>>BITS_PER_SUM) - dc;
 405     return ((uint64_t)sum8<<32) + sum4;
 406 }
 407
 408 #define HADAMARD_AC(w,h) \
 409 static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, int stride )\
 410 {\
 411     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 412     if( w==16 )\
 413         sum += pixel_hadamard_ac( pix+8, stride );\
 414     if( h==16 )\
 415         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 416     if( w==16 && h==16 )\
 417         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 418     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 419 }
 420 HADAMARD_AC( 16, 16 )
 421 HADAMARD_AC( 16, 8 )
 422 HADAMARD_AC( 8, 16 )
 423 HADAMARD_AC( 8, 8 )
 424
 425
 426 /****************************************************************************
 427  * pixel_sad_x4
 428  ****************************************************************************/
 429 #define SAD_X( size ) \
 430 static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
 431 {\
 432     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 433     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 434     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 435 }\
 436 static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
 437 {\
 438     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 439     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 440     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 441     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 442 }
 443
 444 SAD_X( 16x16 )
 445 SAD_X( 16x8 )
 446 SAD_X( 8x16 )
 447 SAD_X( 8x8 )
 448 SAD_X( 8x4 )
 449 SAD_X( 4x8 )
 450 SAD_X( 4x4 )
 451
 452 #if !HIGH_BIT_DEPTH
 453 #if ARCH_UltraSPARC
 454 SAD_X( 16x16_vis )
 455 SAD_X( 16x8_vis )
 456 SAD_X( 8x16_vis )
 457 SAD_X( 8x8_vis )
 458 #endif
 459 #endif // !HIGH_BIT_DEPTH
 460
 461 /****************************************************************************
 462  * pixel_satd_x4
 463  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 464  ****************************************************************************/
 465
 466 #define SATD_X( size, cpu ) \
 467 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int 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, int i_stride, int scores[4] )\
 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     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 479 }
 480 #define SATD_X_DECL6( cpu )\
 481 SATD_X( 16x16, cpu )\
 482 SATD_X( 16x8, cpu )\
 483 SATD_X( 8x16, cpu )\
 484 SATD_X( 8x8, cpu )\
 485 SATD_X( 8x4, cpu )\
 486 SATD_X( 4x8, cpu )
 487 #define SATD_X_DECL7( cpu )\
 488 SATD_X_DECL6( cpu )\
 489 SATD_X( 4x4, cpu )
 490
 491 SATD_X_DECL7()
 492 #if HAVE_MMX
 493 SATD_X_DECL7( _mmx2 )
 494 #if !HIGH_BIT_DEPTH
 495 SATD_X_DECL6( _sse2 )
 496 SATD_X_DECL7( _ssse3 )
 497 SATD_X_DECL7( _sse4 )
 498 SATD_X_DECL7( _avx )
 499 SATD_X_DECL7( _xop )
 500 #endif // !HIGH_BIT_DEPTH
 501 #endif
 502
 503 #if !HIGH_BIT_DEPTH
 504 #if HAVE_ARMV6
 505 SATD_X_DECL7( _neon )
 506 #endif
 507 #endif // !HIGH_BIT_DEPTH
 508
 509 #define INTRA_MBCMP_8x8( mbcmp, cpu, cpu2 )\
 510 void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[36], int res[3] )\
 511 {\
 512     ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
 513     x264_predict_8x8_v##cpu2( pix, edge );\
 514     res[0] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 515     x264_predict_8x8_h##cpu2( pix, edge );\
 516     res[1] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 517     x264_predict_8x8_dc##cpu2( pix, edge );\
 518     res[2] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 519 }
 520
 521 INTRA_MBCMP_8x8( sad,, _c )
 522 INTRA_MBCMP_8x8(sa8d,, _c )
 523 #if HIGH_BIT_DEPTH && HAVE_MMX
 524 INTRA_MBCMP_8x8( sad, _mmx2,  _c )
 525 INTRA_MBCMP_8x8( sad, _sse2,  _sse2 )
 526 INTRA_MBCMP_8x8( sad, _ssse3, _sse2 )
 527 INTRA_MBCMP_8x8(sa8d, _sse2,  _sse2 )
 528 #endif
 529
 530 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu, cpu2 )\
 531 void x264_intra_##mbcmp##_x3_##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
 532 {\
 533     x264_predict_##size##chroma##_##pred1##cpu2( fdec );\
 534     res[0] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 535     x264_predict_##size##chroma##_##pred2##cpu2( fdec );\
 536     res[1] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 537     x264_predict_##size##chroma##_##pred3##cpu2( fdec );\
 538     res[2] = x264_pixel_##mbcmp##_##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 539 }
 540
 541 INTRA_MBCMP( sad,  4x4,   v, h, dc,  ,, _c )
 542 INTRA_MBCMP(satd,  4x4,   v, h, dc,  ,, _c )
 543 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c,, _c )
 544 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c,, _c )
 545 INTRA_MBCMP( sad,  8x16, dc, h,  v, c,, _c )
 546 INTRA_MBCMP(satd,  8x16, dc, h,  v, c,, _c )
 547 INTRA_MBCMP( sad, 16x16,  v, h, dc,  ,, _c )
 548 INTRA_MBCMP(satd, 16x16,  v, h, dc,  ,, _c )
 549
 550 #if HIGH_BIT_DEPTH && HAVE_MMX
 551 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _mmx2, _c )
 552 INTRA_MBCMP(satd,  4x4,   v, h, dc,  , _mmx2, _c )
 553 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _mmx2, _c )
 554 INTRA_MBCMP(satd,  8x8,  dc, h,  v, c, _mmx2, _c )
 555 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _mmx2, _mmx2 )
 556 INTRA_MBCMP(satd, 16x16,  v, h, dc,  , _mmx2, _mmx2 )
 557 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _sse2, _sse2 )
 558 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _sse2, _sse2 )
 559 INTRA_MBCMP( sad,  4x4,   v, h, dc,  , _ssse3, _c )
 560 INTRA_MBCMP( sad,  8x8,  dc, h,  v, c, _ssse3, _sse2 )
 561 INTRA_MBCMP( sad, 16x16,  v, h, dc,  , _ssse3, _sse2 )
 562 #endif
 563
 564 // No C implementation of intra_satd_x9. See checkasm for its behavior,
 565 // or see x264_mb_analyse_intra for the entirely different algorithm we
 566 // use when lacking an asm implementation of it.
 567
 568
 569
 570 /****************************************************************************
 571  * structural similarity metric
 572  ****************************************************************************/
 573 static void ssim_4x4x2_core( const pixel *pix1, int stride1,
 574                              const pixel *pix2, int stride2,
 575                              int sums[2][4])
 576 {
 577     for( int z = 0; z < 2; z++ )
 578     {
 579         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 580         for( int y = 0; y < 4; y++ )
 581             for( int x = 0; x < 4; x++ )
 582             {
 583                 int a = pix1[x+y*stride1];
 584                 int b = pix2[x+y*stride2];
 585                 s1  += a;
 586                 s2  += b;
 587                 ss  += a*a;
 588                 ss  += b*b;
 589                 s12 += a*b;
 590             }
 591         sums[z][0] = s1;
 592         sums[z][1] = s2;
 593         sums[z][2] = ss;
 594         sums[z][3] = s12;
 595         pix1 += 4;
 596         pix2 += 4;
 597     }
 598 }
 599
 600 static float ssim_end1( int s1, int s2, int ss, int s12 )
 601 {
 602 /* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
 603  * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
 604  * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
 605 #if BIT_DEPTH > 9
 606 #define type float
 607     static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
 608     static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
 609 #else
 610 #define type int
 611     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 612     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 613 #endif
 614     type fs1 = s1;
 615     type fs2 = s2;
 616     type fss = ss;
 617     type fs12 = s12;
 618     type vars = fss*64 - fs1*fs1 - fs2*fs2;
 619     type covar = fs12*64 - fs1*fs2;
 620     return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
 621          / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
 622 #undef type
 623 }
 624
 625 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 626 {
 627     float ssim = 0.0;
 628     for( int i = 0; i < width; i++ )
 629         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 630                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 631                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 632                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 633     return ssim;
 634 }
 635
 636 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 637                            pixel *pix1, int stride1,
 638                            pixel *pix2, int stride2,
 639                            int width, int height, void *buf, int *cnt )
 640 {
 641     int z = 0;
 642     float ssim = 0.0;
 643     int (*sum0)[4] = buf;
 644     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 645     width >>= 2;
 646     height >>= 2;
 647     for( int y = 1; y < height; y++ )
 648     {
 649         for( ; z <= y; z++ )
 650         {
 651             XCHG( void*, sum0, sum1 );
 652             for( int x = 0; x < width; x+=2 )
 653                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 654         }
 655         for( int x = 0; x < width-1; x += 4 )
 656             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 657     }
 658     *cnt = (height-1) * (width-1);
 659     return ssim;
 660 }
 661
 662 static int pixel_vsad( pixel *src, int stride, int height )
 663 {
 664     int score = 0;
 665     for( int i = 1; i < height; i++, src += stride )
 666         for( int j = 0; j < 16; j++ )
 667             score += abs(src[j] - src[j+stride]);
 668     return score;
 669 }
 670
 671 int x264_field_vsad( x264_t *h, int mb_x, int mb_y )
 672 {
 673     int score_field, score_frame;
 674     int stride = h->fenc->i_stride[0];
 675     int mb_stride = h->mb.i_mb_stride;
 676     pixel *fenc = h->fenc->plane[0] + 16 * (mb_x + mb_y * stride);
 677     int mb_xy = mb_x + mb_y*mb_stride;
 678
 679     /* We don't want to analyze pixels outside the frame, as it gives inaccurate results. */
 680     int mbpair_height = X264_MIN( h->param.i_height - mb_y * 16, 32 );
 681     score_frame  = h->pixf.vsad( fenc,          stride, mbpair_height );
 682     score_field  = h->pixf.vsad( fenc,        stride*2, mbpair_height >> 1 );
 683     score_field += h->pixf.vsad( fenc+stride, stride*2, mbpair_height >> 1 );
 684
 685     if( mb_x > 0 )
 686         score_field += 512 - h->mb.field[mb_xy        -1]*1024;
 687     if( mb_y > 0 )
 688         score_field += 512 - h->mb.field[mb_xy-mb_stride]*1024;
 689
 690     return (score_field < score_frame);
 691 }
 692
 693 /****************************************************************************
 694  * successive elimination
 695  ****************************************************************************/
 696 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 697                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 698 {
 699     int nmv = 0;
 700     for( int i = 0; i < width; i++, sums++ )
 701     {
 702         int ads = abs( enc_dc[0] - sums[0] )
 703                 + abs( enc_dc[1] - sums[8] )
 704                 + abs( enc_dc[2] - sums[delta] )
 705                 + abs( enc_dc[3] - sums[delta+8] )
 706                 + cost_mvx[i];
 707         if( ads < thresh )
 708             mvs[nmv++] = i;
 709     }
 710     return nmv;
 711 }
 712
 713 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 714                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 715 {
 716     int nmv = 0;
 717     for( int i = 0; i < width; i++, sums++ )
 718     {
 719         int ads = abs( enc_dc[0] - sums[0] )
 720                 + abs( enc_dc[1] - sums[delta] )
 721                 + cost_mvx[i];
 722         if( ads < thresh )
 723             mvs[nmv++] = i;
 724     }
 725     return nmv;
 726 }
 727
 728 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 729                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 730 {
 731     int nmv = 0;
 732     for( int i = 0; i<width; i++, sums++ )
 733     {
 734         int ads = abs( enc_dc[0] - sums[0] )
 735                 + cost_mvx[i];
 736         if( ads < thresh )
 737             mvs[nmv++] = i;
 738     }
 739     return nmv;
 740 }
 741
 742
 743 /****************************************************************************
 744  * x264_pixel_init:
 745  ****************************************************************************/
 746 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 747 {
 748     memset( pixf, 0, sizeof(*pixf) );
 749
 750 #define INIT2_NAME( name1, name2, cpu ) \
 751     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 752     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 753 #define INIT4_NAME( name1, name2, cpu ) \
 754     INIT2_NAME( name1, name2, cpu ) \
 755     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 756     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 757 #define INIT5_NAME( name1, name2, cpu ) \
 758     INIT4_NAME( name1, name2, cpu ) \
 759     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 760 #define INIT6_NAME( name1, name2, cpu ) \
 761     INIT5_NAME( name1, name2, cpu ) \
 762     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 763 #define INIT7_NAME( name1, name2, cpu ) \
 764     INIT6_NAME( name1, name2, cpu ) \
 765     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 766 #define INIT8_NAME( name1, name2, cpu ) \
 767     INIT7_NAME( name1, name2, cpu ) \
 768     pixf->name1[PIXEL_4x16]  = x264_pixel_##name2##_4x16##cpu;
 769 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 770 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 771 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 772 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 773 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 774 #define INIT8( name, cpu ) INIT8_NAME( name, name, cpu )
 775
 776 #define INIT_ADS( cpu ) \
 777     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 778     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 779     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 780
 781     INIT8( sad, );
 782     INIT8_NAME( sad_aligned, sad, );
 783     INIT7( sad_x3, );
 784     INIT7( sad_x4, );
 785     INIT8( ssd, );
 786     INIT8( satd, );
 787     INIT7( satd_x3, );
 788     INIT7( satd_x4, );
 789     INIT4( hadamard_ac, );
 790     INIT_ADS( );
 791
 792     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 793     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 794     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 795     pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16;
 796     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 797     pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16;
 798     pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8;
 799
 800     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 801     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 802     pixf->ssim_end4 = ssim_end4;
 803     pixf->vsad = pixel_vsad;
 804
 805     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 806     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 807     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 808     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 809     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 810     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 811     pixf->intra_sad_x3_8x16c  = x264_intra_sad_x3_8x16c;
 812     pixf->intra_satd_x3_8x16c = x264_intra_satd_x3_8x16c;
 813     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 814     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 815
 816 #if HIGH_BIT_DEPTH
 817 #if HAVE_MMX
 818     if( cpu&X264_CPU_MMX2 )
 819     {
 820         INIT7( sad, _mmx2 );
 821         INIT7( sad_x3, _mmx2 );
 822         INIT7( sad_x4, _mmx2 );
 823         INIT7( satd, _mmx2 );
 824         INIT7( satd_x3, _mmx2 );
 825         INIT7( satd_x4, _mmx2 );
 826         INIT4( hadamard_ac, _mmx2 );
 827         INIT7( ssd, _mmx2 );
 828         INIT_ADS( _mmx2 );
 829
 830         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_mmx2;
 831         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
 832         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
 833         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_mmx2;
 834
 835         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
 836         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
 837         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
 838         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
 839         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
 840         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
 841         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
 842     }
 843     if( cpu&X264_CPU_SSE2 )
 844     {
 845         INIT4_NAME( sad_aligned, sad, _sse2_aligned );
 846         INIT5( ssd, _sse2 );
 847
 848         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 849         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 850 #if ARCH_X86_64
 851         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 852 #endif
 853         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_sse2;
 854         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 855         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 856         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 857         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_sse2;
 858         pixf->var2[PIXEL_8x8]  = x264_pixel_var2_8x8_sse2;
 859     }
 860     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 861     {
 862         INIT5( sad, _sse2 );
 863         INIT2( sad_x3, _sse2 );
 864         INIT2( sad_x4, _sse2 );
 865         INIT_ADS( _sse2 );
 866
 867         if( !(cpu&X264_CPU_STACK_MOD4) )
 868         {
 869             INIT4( hadamard_ac, _sse2 );
 870         }
 871
 872         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_sse2;
 873         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_sse2;
 874         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_sse2;
 875     }
 876     if( cpu&X264_CPU_SSE2_IS_FAST )
 877     {
 878         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 879         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 880         pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_sse2;
 881         pixf->sad_x3[PIXEL_8x4]  = x264_pixel_sad_x3_8x4_sse2;
 882         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 883         pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_sse2;
 884         pixf->sad_x4[PIXEL_8x4]  = x264_pixel_sad_x4_8x4_sse2;
 885     }
 886     if( cpu&X264_CPU_SSSE3 )
 887     {
 888         INIT4_NAME( sad_aligned, sad, _ssse3_aligned );
 889         INIT7( sad, _ssse3 );
 890         INIT7( sad_x3, _ssse3 );
 891         INIT7( sad_x4, _ssse3 );
 892         INIT_ADS( _ssse3 );
 893
 894         if( !(cpu&X264_CPU_STACK_MOD4) )
 895         {
 896             INIT4( hadamard_ac, _ssse3 );
 897         }
 898
 899         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 900         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 901         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_ssse3;
 902         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_ssse3;
 903         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 904         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 905     }
 906     if( cpu&X264_CPU_SSE4 )
 907     {
 908         if( !(cpu&X264_CPU_STACK_MOD4) )
 909         {
 910             INIT4( hadamard_ac, _sse4 );
 911         }
 912         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 913         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 914     }
 915     if( cpu&X264_CPU_AVX )
 916     {
 917         INIT_ADS( _avx );
 918         if( !(cpu&X264_CPU_STACK_MOD4) )
 919         {
 920             INIT4( hadamard_ac, _avx );
 921         }
 922         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
 923         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
 924         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
 925         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
 926         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
 927         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
 928         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
 929     }
 930 #endif // HAVE_MMX
 931 #else // !HIGH_BIT_DEPTH
 932 #if HAVE_MMX
 933     if( cpu&X264_CPU_MMX )
 934     {
 935         INIT8( ssd, _mmx );
 936     }
 937
 938     if( cpu&X264_CPU_MMX2 )
 939     {
 940         INIT8( sad, _mmx2 );
 941         INIT8_NAME( sad_aligned, sad, _mmx2 );
 942         INIT7( sad_x3, _mmx2 );
 943         INIT7( sad_x4, _mmx2 );
 944         INIT7( satd, _mmx2 );
 945         INIT7( satd_x3, _mmx2 );
 946         INIT7( satd_x4, _mmx2 );
 947         INIT4( hadamard_ac, _mmx2 );
 948         INIT_ADS( _mmx2 );
 949         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmx2;
 950         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_mmx2;
 951         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmx2;
 952         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmx2;
 953 #if ARCH_X86
 954         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmx2;
 955         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmx2;
 956         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmx2;
 957         pixf->ssim_4x4x2_core = x264_pixel_ssim_4x4x2_core_mmx2;
 958         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_mmx2;
 959         pixf->vsad = x264_pixel_vsad_mmx2;
 960
 961         if( cpu&X264_CPU_CACHELINE_32 )
 962         {
 963             INIT5( sad, _cache32_mmx2 );
 964             INIT4( sad_x3, _cache32_mmx2 );
 965             INIT4( sad_x4, _cache32_mmx2 );
 966         }
 967         else if( cpu&X264_CPU_CACHELINE_64 )
 968         {
 969             INIT5( sad, _cache64_mmx2 );
 970             INIT4( sad_x3, _cache64_mmx2 );
 971             INIT4( sad_x4, _cache64_mmx2 );
 972         }
 973 #else
 974         if( cpu&X264_CPU_CACHELINE_64 )
 975         {
 976             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmx2;
 977             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmx2;
 978             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmx2;
 979             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmx2;
 980             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmx2;
 981             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmx2;
 982             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmx2;
 983         }
 984 #endif
 985         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmx2;
 986         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmx2;
 987         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmx2;
 988         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmx2;
 989         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmx2;
 990         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmx2;
 991         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmx2;
 992     }
 993
 994     if( cpu&X264_CPU_SSE2 )
 995     {
 996         INIT5( ssd, _sse2slow );
 997         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
 998         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 999         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
1000         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
1001         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
1002         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
1003         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
1004 #if ARCH_X86_64
1005         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
1006 #endif
1007         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_sse2;
1008         pixf->vsad = x264_pixel_vsad_sse2;
1009     }
1010
1011     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
1012     {
1013         INIT2( sad, _sse2 );
1014         INIT2( sad_x3, _sse2 );
1015         INIT2( sad_x4, _sse2 );
1016         INIT6( satd, _sse2 );
1017         INIT6( satd_x3, _sse2 );
1018         INIT6( satd_x4, _sse2 );
1019         if( !(cpu&X264_CPU_STACK_MOD4) )
1020         {
1021             INIT4( hadamard_ac, _sse2 );
1022         }
1023         INIT_ADS( _sse2 );
1024         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
1025         pixf->var[PIXEL_8x16] = x264_pixel_var_8x16_sse2;
1026         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
1027         if( cpu&X264_CPU_CACHELINE_64 )
1028         {
1029             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
1030 #if ARCH_X86
1031             INIT2( sad, _cache64_sse2 );
1032             INIT2( sad_x3, _cache64_sse2 );
1033             INIT2( sad_x4, _cache64_sse2 );
1034 #endif
1035            if( cpu&X264_CPU_SSE2_IS_FAST )
1036            {
1037                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
1038                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
1039            }
1040         }
1041
1042         if( cpu&X264_CPU_SSE_MISALIGN )
1043         {
1044             INIT2( sad_x3, _sse2_misalign );
1045             INIT2( sad_x4, _sse2_misalign );
1046         }
1047     }
1048
1049     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
1050     {
1051         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1052         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1053         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
1054         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
1055         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
1056         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
1057         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
1058         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
1059     }
1060
1061     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
1062     {
1063         INIT2( sad, _sse3 );
1064         INIT2( sad_x3, _sse3 );
1065         INIT2( sad_x4, _sse3 );
1066     }
1067
1068     if( cpu&X264_CPU_SSSE3 )
1069     {
1070         if( !(cpu&X264_CPU_STACK_MOD4) )
1071         {
1072             INIT4( hadamard_ac, _ssse3 );
1073             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_ssse3;
1074             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_ssse3;
1075             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_ssse3;
1076 #if ARCH_X86_64
1077             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_ssse3;
1078 #endif
1079         }
1080         INIT_ADS( _ssse3 );
1081         if( !(cpu&X264_CPU_SLOW_ATOM) )
1082         {
1083             INIT8( ssd, _ssse3 );
1084             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
1085             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
1086             INIT7( satd, _ssse3 );
1087             INIT7( satd_x3, _ssse3 );
1088             INIT7( satd_x4, _ssse3 );
1089         }
1090         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
1091         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
1092         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
1093         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
1094         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
1095 #if ARCH_X86_64
1096         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
1097 #endif
1098         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_ssse3;
1099         if( cpu&X264_CPU_SHUFFLE_IS_FAST )
1100             pixf->intra_sad_x3_8x8  = x264_intra_sad_x3_8x8_ssse3;
1101         if( cpu&X264_CPU_CACHELINE_64 )
1102         {
1103             INIT2( sad, _cache64_ssse3 );
1104             INIT2( sad_x3, _cache64_ssse3 );
1105             INIT2( sad_x4, _cache64_ssse3 );
1106         }
1107         if( cpu&X264_CPU_SLOW_ATOM || !(cpu&X264_CPU_SHUFFLE_IS_FAST) )
1108         {
1109             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
1110         }
1111     }
1112
1113     if( cpu&X264_CPU_SSE4 )
1114     {
1115         INIT7( satd, _sse4 );
1116         INIT7( satd_x3, _sse4 );
1117         INIT7( satd_x4, _sse4 );
1118         if( !(cpu&X264_CPU_STACK_MOD4) )
1119         {
1120             INIT4( hadamard_ac, _sse4 );
1121             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_sse4;
1122             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_sse4;
1123             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_sse4;
1124 #if ARCH_X86_64
1125             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_sse4;
1126 #endif
1127         }
1128         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
1129         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
1130         pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_sse4;
1131     }
1132
1133     if( cpu&X264_CPU_AVX )
1134     {
1135         INIT7( satd, _avx );
1136         INIT7( satd_x3, _avx );
1137         INIT7( satd_x4, _avx );
1138         INIT_ADS( _avx );
1139         if( !(cpu&X264_CPU_STACK_MOD4) )
1140         {
1141             INIT4( hadamard_ac, _avx );
1142             pixf->intra_sad_x9_4x4  = x264_intra_sad_x9_4x4_avx;
1143             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_avx;
1144             pixf->intra_sad_x9_8x8  = x264_intra_sad_x9_8x8_avx;
1145 #if ARCH_X86_64
1146             pixf->intra_sa8d_x9_8x8 = x264_intra_sa8d_x9_8x8_avx;
1147 #endif
1148         }
1149         INIT5( ssd, _avx );
1150         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1151         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1152 #if ARCH_X86_64
1153         pixf->intra_sa8d_x3_8x8= x264_intra_sa8d_x3_8x8_avx;
1154 #endif
1155         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1156         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1157         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_avx;
1158         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1159         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1160         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1161         pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_avx;
1162         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_avx;
1163     }
1164
1165     if( cpu&X264_CPU_XOP )
1166     {
1167         INIT7( satd, _xop );
1168         INIT7( satd_x3, _xop );
1169         INIT7( satd_x4, _xop );
1170         if( !(cpu&X264_CPU_STACK_MOD4) )
1171         {
1172             INIT4( hadamard_ac, _xop );
1173             pixf->intra_satd_x9_4x4 = x264_intra_satd_x9_4x4_xop;
1174         }
1175         INIT5( ssd, _xop );
1176         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_xop;
1177         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_xop;
1178 #if ARCH_X86_64
1179         pixf->intra_sa8d_x3_8x8= x264_intra_sa8d_x3_8x8_xop;
1180 #endif
1181         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_xop;
1182         pixf->var[PIXEL_8x16]  = x264_pixel_var_8x16_xop;
1183         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_xop;
1184         pixf->var2[PIXEL_8x8] = x264_pixel_var2_8x8_xop;
1185     }
1186 #endif //HAVE_MMX
1187
1188 #if HAVE_ARMV6
1189     if( cpu&X264_CPU_ARMV6 )
1190     {
1191         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1192         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1193         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1194         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1195     }
1196     if( cpu&X264_CPU_NEON )
1197     {
1198         INIT5( sad, _neon );
1199         INIT5( sad_aligned, _neon );
1200         INIT7( sad_x3, _neon );
1201         INIT7( sad_x4, _neon );
1202         INIT7( ssd, _neon );
1203         INIT7( satd, _neon );
1204         INIT7( satd_x3, _neon );
1205         INIT7( satd_x4, _neon );
1206         INIT4( hadamard_ac, _neon );
1207         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1208         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1209         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1210         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1211         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
1212
1213         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1214         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1215
1216         if( cpu&X264_CPU_FAST_NEON_MRC )
1217         {
1218             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
1219             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
1220             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
1221             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
1222         }
1223         else    // really just scheduled for dual issue / A8
1224         {
1225             INIT5( sad_aligned, _neon_dual );
1226         }
1227     }
1228 #endif
1229 #endif // HIGH_BIT_DEPTH
1230 #if HAVE_ALTIVEC
1231     if( cpu&X264_CPU_ALTIVEC )
1232     {
1233         x264_pixel_altivec_init( pixf );
1234     }
1235 #endif
1236 #if !HIGH_BIT_DEPTH
1237 #if ARCH_UltraSPARC
1238     INIT4( sad, _vis );
1239     INIT4( sad_x3, _vis );
1240     INIT4( sad_x4, _vis );
1241 #endif
1242 #endif // !HIGH_BIT_DEPTH
1243
1244     pixf->ads[PIXEL_8x16] =
1245     pixf->ads[PIXEL_8x4] =
1246     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
1247     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
1248 }
1249