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

   1 /*****************************************************************************
   2  * pixel.c: pixel metrics
   3  *****************************************************************************
   4  * Copyright (C) 2003-2010 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 #endif
  33 #if ARCH_PPC
  34 #   include "ppc/pixel.h"
  35 #endif
  36 #if ARCH_ARM
  37 #   include "arm/pixel.h"
  38 #endif
  39 #if ARCH_UltraSparc
  40 #   include "sparc/pixel.h"
  41 #endif
  42
  43
  44 /****************************************************************************
  45  * pixel_sad_WxH
  46  ****************************************************************************/
  47 #define PIXEL_SAD_C( name, lx, ly ) \
  48 static int name( pixel *pix1, int i_stride_pix1,  \
  49                  pixel *pix2, int i_stride_pix2 ) \
  50 {                                                   \
  51     int i_sum = 0;                                  \
  52     for( int y = 0; y < ly; y++ )                   \
  53     {                                               \
  54         for( int x = 0; x < lx; x++ )               \
  55         {                                           \
  56             i_sum += abs( pix1[x] - pix2[x] );      \
  57         }                                           \
  58         pix1 += i_stride_pix1;                      \
  59         pix2 += i_stride_pix2;                      \
  60     }                                               \
  61     return i_sum;                                   \
  62 }
  63
  64
  65 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  66 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  67 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  68 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  69 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  70 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  71 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  72
  73
  74 /****************************************************************************
  75  * pixel_ssd_WxH
  76  ****************************************************************************/
  77 #define PIXEL_SSD_C( name, lx, ly ) \
  78 static int name( pixel *pix1, int i_stride_pix1,  \
  79                  pixel *pix2, int i_stride_pix2 ) \
  80 {                                                   \
  81     int i_sum = 0;                                  \
  82     for( int y = 0; y < ly; y++ )                   \
  83     {                                               \
  84         for( int x = 0; x < lx; x++ )               \
  85         {                                           \
  86             int d = pix1[x] - pix2[x];              \
  87             i_sum += d*d;                           \
  88         }                                           \
  89         pix1 += i_stride_pix1;                      \
  90         pix2 += i_stride_pix2;                      \
  91     }                                               \
  92     return i_sum;                                   \
  93 }
  94
  95 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
  96 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
  97 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
  98 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
  99 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
 100 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
 101 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
 102
 103 uint64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, pixel *pix1, int i_pix1, pixel *pix2, int i_pix2, int i_width, int i_height )
 104 {
 105     uint64_t i_ssd = 0;
 106     int y;
 107     int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
 108
 109 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 110                                           pix2 + y*i_pix2 + x, i_pix2 );
 111     for( y = 0; y < i_height-15; y += 16 )
 112     {
 113         int x = 0;
 114         if( align )
 115             for( ; x < i_width-15; x += 16 )
 116                 SSD(PIXEL_16x16);
 117         for( ; x < i_width-7; x += 8 )
 118             SSD(PIXEL_8x16);
 119     }
 120     if( y < i_height-7 )
 121         for( int x = 0; x < i_width-7; x += 8 )
 122             SSD(PIXEL_8x8);
 123 #undef SSD
 124
 125 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 126     if( i_width & 7 )
 127     {
 128         for( y = 0; y < (i_height & ~7); y++ )
 129             for( int x = i_width & ~7; x < i_width; x++ )
 130                 SSD1;
 131     }
 132     if( i_height & 7 )
 133     {
 134         for( y = i_height & ~7; y < i_height; y++ )
 135             for( int x = 0; x < i_width; x++ )
 136                 SSD1;
 137     }
 138 #undef SSD1
 139
 140     return i_ssd;
 141 }
 142
 143 static uint64_t pixel_ssd_nv12_core( pixel *pixuv1, int stride1, pixel *pixuv2, int stride2, int width, int height )
 144 {
 145     uint32_t ssd_u=0, ssd_v=0;
 146     for( int y = 0; y < height; y++, pixuv1+=stride1, pixuv2+=stride2 )
 147         for( int x = 0; x < width; x++ )
 148         {
 149             int du = pixuv1[2*x]   - pixuv2[2*x];
 150             int dv = pixuv1[2*x+1] - pixuv2[2*x+1];
 151             ssd_u += du*du;
 152             ssd_v += dv*dv;
 153         }
 154     return ssd_u + ((uint64_t)ssd_v<<32);
 155 }
 156
 157 // SSD in uint32 (i.e. packing two into uint64) can potentially overflow on
 158 // image widths >= 11008 (or 6604 if interlaced), since this is called on blocks
 159 // of height up to 12 (resp 20). Though it will probably take significantly more
 160 // than that at sane distortion levels.
 161 uint64_t 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 )
 162 {
 163     uint64_t ssd = pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height );
 164     if( i_width&7 )
 165         ssd += pixel_ssd_nv12_core( pix1+(i_width&~7), i_pix1, pix2+(i_width&~7), i_pix2, i_width&7, i_height );
 166     return ssd;
 167 }
 168
 169 /****************************************************************************
 170  * pixel_var_wxh
 171  ****************************************************************************/
 172 #define PIXEL_VAR_C( name, w ) \
 173 static uint64_t name( pixel *pix, int i_stride ) \
 174 {                                             \
 175     uint32_t sum = 0, sqr = 0;                \
 176     for( int y = 0; y < w; y++ )              \
 177     {                                         \
 178         for( int x = 0; x < w; x++ )          \
 179         {                                     \
 180             sum += pix[x];                    \
 181             sqr += pix[x] * pix[x];           \
 182         }                                     \
 183         pix += i_stride;                      \
 184     }                                         \
 185     return sum + ((uint64_t)sqr << 32);       \
 186 }
 187
 188 PIXEL_VAR_C( x264_pixel_var_16x16, 16 )
 189 PIXEL_VAR_C( x264_pixel_var_8x8,    8 )
 190
 191 /****************************************************************************
 192  * pixel_var2_wxh
 193  ****************************************************************************/
 194 static int pixel_var2_8x8( pixel *pix1, int i_stride1, pixel *pix2, int i_stride2, int *ssd )
 195 {
 196     uint32_t var = 0, sum = 0, sqr = 0;
 197     for( int y = 0; y < 8; y++ )
 198     {
 199         for( int x = 0; x < 8; x++ )
 200         {
 201             int diff = pix1[x] - pix2[x];
 202             sum += diff;
 203             sqr += diff * diff;
 204         }
 205         pix1 += i_stride1;
 206         pix2 += i_stride2;
 207     }
 208     sum = abs(sum);
 209     var = sqr - ((uint64_t)sum * sum >> 6);
 210     *ssd = sqr;
 211     return var;
 212 }
 213
 214
 215 #define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
 216     int t0 = s0 + s1;\
 217     int t1 = s0 - s1;\
 218     int t2 = s2 + s3;\
 219     int t3 = s2 - s3;\
 220     d0 = t0 + t2;\
 221     d2 = t0 - t2;\
 222     d1 = t1 + t3;\
 223     d3 = t1 - t3;\
 224 }
 225
 226 // in: a pseudo-simd number of the form x+(y<<16)
 227 // return: abs(x)+(abs(y)<<16)
 228 static ALWAYS_INLINE uint32_t abs2( uint32_t a )
 229 {
 230     uint32_t s = ((a>>15)&0x10001)*0xffff;
 231     return (a+s)^s;
 232 }
 233
 234 /****************************************************************************
 235  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 236  ****************************************************************************/
 237
 238 static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 239 {
 240     uint32_t tmp[4][2];
 241     uint32_t a0, a1, a2, a3, b0, b1;
 242     int sum = 0;
 243     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 244     {
 245         a0 = pix1[0] - pix2[0];
 246         a1 = pix1[1] - pix2[1];
 247         b0 = (a0+a1) + ((a0-a1)<<16);
 248         a2 = pix1[2] - pix2[2];
 249         a3 = pix1[3] - pix2[3];
 250         b1 = (a2+a3) + ((a2-a3)<<16);
 251         tmp[i][0] = b0 + b1;
 252         tmp[i][1] = b0 - b1;
 253     }
 254     for( int i = 0; i < 2; i++ )
 255     {
 256         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 257         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 258         sum += ((uint16_t)a0) + (a0>>16);
 259     }
 260     return sum >> 1;
 261 }
 262
 263 static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 264 {
 265     uint32_t tmp[4][4];
 266     uint32_t a0, a1, a2, a3;
 267     int sum = 0;
 268     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 269     {
 270         a0 = (pix1[0] - pix2[0]) + ((pix1[4] - pix2[4]) << 16);
 271         a1 = (pix1[1] - pix2[1]) + ((pix1[5] - pix2[5]) << 16);
 272         a2 = (pix1[2] - pix2[2]) + ((pix1[6] - pix2[6]) << 16);
 273         a3 = (pix1[3] - pix2[3]) + ((pix1[7] - pix2[7]) << 16);
 274         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 275     }
 276     for( int i = 0; i < 4; i++ )
 277     {
 278         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 279         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 280     }
 281     return (((uint16_t)sum) + ((uint32_t)sum>>16)) >> 1;
 282 }
 283
 284 #define PIXEL_SATD_C( w, h, sub )\
 285 static int x264_pixel_satd_##w##x##h( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )\
 286 {\
 287     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 288             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 289     if( w==16 )\
 290         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 291             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 292     if( h==16 )\
 293         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 294             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 295     if( w==16 && h==16 )\
 296         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 297             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 298     return sum;\
 299 }
 300 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 301 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 302 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 303 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 304 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 305
 306
 307 static NOINLINE int sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 308 {
 309     uint32_t tmp[8][4];
 310     uint32_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
 311     int sum = 0;
 312     for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
 313     {
 314         a0 = pix1[0] - pix2[0];
 315         a1 = pix1[1] - pix2[1];
 316         b0 = (a0+a1) + ((a0-a1)<<16);
 317         a2 = pix1[2] - pix2[2];
 318         a3 = pix1[3] - pix2[3];
 319         b1 = (a2+a3) + ((a2-a3)<<16);
 320         a4 = pix1[4] - pix2[4];
 321         a5 = pix1[5] - pix2[5];
 322         b2 = (a4+a5) + ((a4-a5)<<16);
 323         a6 = pix1[6] - pix2[6];
 324         a7 = pix1[7] - pix2[7];
 325         b3 = (a6+a7) + ((a6-a7)<<16);
 326         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 327     }
 328     for( int i = 0; i < 4; i++ )
 329     {
 330         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 331         HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 332         b0  = abs2(a0+a4) + abs2(a0-a4);
 333         b0 += abs2(a1+a5) + abs2(a1-a5);
 334         b0 += abs2(a2+a6) + abs2(a2-a6);
 335         b0 += abs2(a3+a7) + abs2(a3-a7);
 336         sum += (uint16_t)b0 + (b0>>16);
 337     }
 338     return sum;
 339 }
 340
 341 static int x264_pixel_sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 342 {
 343     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
 344     return (sum+2)>>2;
 345 }
 346
 347 static int x264_pixel_sa8d_16x16( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 348 {
 349     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 350             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 351             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 352             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 353     return (sum+2)>>2;
 354 }
 355
 356
 357 static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, int stride )
 358 {
 359     uint32_t tmp[32];
 360     uint32_t a0, a1, a2, a3, dc;
 361     int sum4 = 0, sum8 = 0;
 362     for( int i = 0; i < 8; i++, pix+=stride )
 363     {
 364         uint32_t *t = tmp + (i&3) + (i&4)*4;
 365         a0 = (pix[0]+pix[1]) + ((pix[0]-pix[1])<<16);
 366         a1 = (pix[2]+pix[3]) + ((pix[2]-pix[3])<<16);
 367         t[0] = a0 + a1;
 368         t[4] = a0 - a1;
 369         a2 = (pix[4]+pix[5]) + ((pix[4]-pix[5])<<16);
 370         a3 = (pix[6]+pix[7]) + ((pix[6]-pix[7])<<16);
 371         t[8] = a2 + a3;
 372         t[12] = a2 - a3;
 373     }
 374     for( int i = 0; i < 8; i++ )
 375     {
 376         HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 377         tmp[i*4+0] = a0;
 378         tmp[i*4+1] = a1;
 379         tmp[i*4+2] = a2;
 380         tmp[i*4+3] = a3;
 381         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 382     }
 383     for( int i = 0; i < 8; i++ )
 384     {
 385         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 386         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 387     }
 388     dc = (uint16_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 389     sum4 = (uint16_t)sum4 + ((uint32_t)sum4>>16) - dc;
 390     sum8 = (uint16_t)sum8 + ((uint32_t)sum8>>16) - dc;
 391     return ((uint64_t)sum8<<32) + sum4;
 392 }
 393
 394 #define HADAMARD_AC(w,h) \
 395 static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, int stride )\
 396 {\
 397     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 398     if( w==16 )\
 399         sum += pixel_hadamard_ac( pix+8, stride );\
 400     if( h==16 )\
 401         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 402     if( w==16 && h==16 )\
 403         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 404     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 405 }
 406 HADAMARD_AC( 16, 16 )
 407 HADAMARD_AC( 16, 8 )
 408 HADAMARD_AC( 8, 16 )
 409 HADAMARD_AC( 8, 8 )
 410
 411
 412 /****************************************************************************
 413  * pixel_sad_x4
 414  ****************************************************************************/
 415 #define SAD_X( size ) \
 416 static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
 417 {\
 418     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 419     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 420     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 421 }\
 422 static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
 423 {\
 424     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 425     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 426     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 427     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 428 }
 429
 430 SAD_X( 16x16 )
 431 SAD_X( 16x8 )
 432 SAD_X( 8x16 )
 433 SAD_X( 8x8 )
 434 SAD_X( 8x4 )
 435 SAD_X( 4x8 )
 436 SAD_X( 4x4 )
 437
 438 #if !X264_HIGH_BIT_DEPTH
 439 #if ARCH_UltraSparc
 440 SAD_X( 16x16_vis )
 441 SAD_X( 16x8_vis )
 442 SAD_X( 8x16_vis )
 443 SAD_X( 8x8_vis )
 444 #endif
 445 #endif // !X264_HIGH_BIT_DEPTH
 446
 447 /****************************************************************************
 448  * pixel_satd_x4
 449  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 450  ****************************************************************************/
 451
 452 #define SATD_X( size, cpu ) \
 453 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
 454 {\
 455     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 456     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 457     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 458 }\
 459 static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
 460 {\
 461     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 462     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 463     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 464     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 465 }
 466 #define SATD_X_DECL6( cpu )\
 467 SATD_X( 16x16, cpu )\
 468 SATD_X( 16x8, cpu )\
 469 SATD_X( 8x16, cpu )\
 470 SATD_X( 8x8, cpu )\
 471 SATD_X( 8x4, cpu )\
 472 SATD_X( 4x8, cpu )
 473 #define SATD_X_DECL7( cpu )\
 474 SATD_X_DECL6( cpu )\
 475 SATD_X( 4x4, cpu )
 476
 477 SATD_X_DECL7()
 478 #if !X264_HIGH_BIT_DEPTH
 479 #if HAVE_MMX
 480 SATD_X_DECL7( _mmxext )
 481 SATD_X_DECL6( _sse2 )
 482 SATD_X_DECL7( _ssse3 )
 483 SATD_X_DECL7( _sse4 )
 484 #endif
 485
 486 #if HAVE_ARMV6
 487 SATD_X_DECL7( _neon )
 488 #endif
 489 #endif // !X264_HIGH_BIT_DEPTH
 490
 491 #define INTRA_MBCMP_8x8( mbcmp )\
 492 void x264_intra_##mbcmp##_x3_8x8( pixel *fenc, pixel edge[33], int res[3] )\
 493 {\
 494     pixel pix[8*FDEC_STRIDE];\
 495     x264_predict_8x8_v_c( pix, edge );\
 496     res[0] = x264_pixel_##mbcmp##_8x8( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 497     x264_predict_8x8_h_c( pix, edge );\
 498     res[1] = x264_pixel_##mbcmp##_8x8( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 499     x264_predict_8x8_dc_c( pix, edge );\
 500     res[2] = x264_pixel_##mbcmp##_8x8( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 501 }
 502
 503 INTRA_MBCMP_8x8(sad)
 504 INTRA_MBCMP_8x8(sa8d)
 505
 506 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma )\
 507 void x264_intra_##mbcmp##_x3_##size##x##size##chroma( pixel *fenc, pixel *fdec, int res[3] )\
 508 {\
 509     x264_predict_##size##x##size##chroma##_##pred1##_c( fdec );\
 510     res[0] = x264_pixel_##mbcmp##_##size##x##size( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 511     x264_predict_##size##x##size##chroma##_##pred2##_c( fdec );\
 512     res[1] = x264_pixel_##mbcmp##_##size##x##size( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 513     x264_predict_##size##x##size##chroma##_##pred3##_c( fdec );\
 514     res[2] = x264_pixel_##mbcmp##_##size##x##size( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 515 }
 516
 517 INTRA_MBCMP(sad, 4, v, h, dc, )
 518 INTRA_MBCMP(satd, 4, v, h, dc, )
 519 INTRA_MBCMP(sad, 8, dc, h, v, c )
 520 INTRA_MBCMP(satd, 8, dc, h, v, c )
 521 INTRA_MBCMP(sad, 16, v, h, dc, )
 522 INTRA_MBCMP(satd, 16, v, h, dc, )
 523
 524 /****************************************************************************
 525  * structural similarity metric
 526  ****************************************************************************/
 527 static void ssim_4x4x2_core( const pixel *pix1, int stride1,
 528                              const pixel *pix2, int stride2,
 529                              int sums[2][4])
 530 {
 531     for( int z = 0; z < 2; z++ )
 532     {
 533         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 534         for( int y = 0; y < 4; y++ )
 535             for( int x = 0; x < 4; x++ )
 536             {
 537                 int a = pix1[x+y*stride1];
 538                 int b = pix2[x+y*stride2];
 539                 s1  += a;
 540                 s2  += b;
 541                 ss  += a*a;
 542                 ss  += b*b;
 543                 s12 += a*b;
 544             }
 545         sums[z][0] = s1;
 546         sums[z][1] = s2;
 547         sums[z][2] = ss;
 548         sums[z][3] = s12;
 549         pix1 += 4;
 550         pix2 += 4;
 551     }
 552 }
 553
 554 static float ssim_end1( int s1, int s2, int ss, int s12 )
 555 {
 556     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 557     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 558     int vars = ss*64 - s1*s1 - s2*s2;
 559     int covar = s12*64 - s1*s2;
 560     return (float)(2*s1*s2 + ssim_c1) * (float)(2*covar + ssim_c2)
 561          / ((float)(s1*s1 + s2*s2 + ssim_c1) * (float)(vars + ssim_c2));
 562 }
 563
 564 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 565 {
 566     float ssim = 0.0;
 567     for( int i = 0; i < width; i++ )
 568         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 569                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 570                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 571                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 572     return ssim;
 573 }
 574
 575 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 576                            pixel *pix1, int stride1,
 577                            pixel *pix2, int stride2,
 578                            int width, int height, void *buf )
 579 {
 580     int z = 0;
 581     float ssim = 0.0;
 582     int (*sum0)[4] = buf;
 583     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 584     width >>= 2;
 585     height >>= 2;
 586     for( int y = 1; y < height; y++ )
 587     {
 588         for( ; z <= y; z++ )
 589         {
 590             XCHG( void*, sum0, sum1 );
 591             for( int x = 0; x < width; x+=2 )
 592                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 593         }
 594         for( int x = 0; x < width-1; x += 4 )
 595             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 596     }
 597     return ssim;
 598 }
 599
 600
 601 /****************************************************************************
 602  * successive elimination
 603  ****************************************************************************/
 604 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 605                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 606 {
 607     int nmv = 0;
 608     for( int i = 0; i < width; i++, sums++ )
 609     {
 610         int ads = abs( enc_dc[0] - sums[0] )
 611                 + abs( enc_dc[1] - sums[8] )
 612                 + abs( enc_dc[2] - sums[delta] )
 613                 + abs( enc_dc[3] - sums[delta+8] )
 614                 + cost_mvx[i];
 615         if( ads < thresh )
 616             mvs[nmv++] = i;
 617     }
 618     return nmv;
 619 }
 620
 621 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 622                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 623 {
 624     int nmv = 0;
 625     for( int i = 0; i < width; i++, sums++ )
 626     {
 627         int ads = abs( enc_dc[0] - sums[0] )
 628                 + abs( enc_dc[1] - sums[delta] )
 629                 + cost_mvx[i];
 630         if( ads < thresh )
 631             mvs[nmv++] = i;
 632     }
 633     return nmv;
 634 }
 635
 636 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 637                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 638 {
 639     int nmv = 0;
 640     for( int i = 0; i<width; i++, sums++ )
 641     {
 642         int ads = abs( enc_dc[0] - sums[0] )
 643                 + cost_mvx[i];
 644         if( ads < thresh )
 645             mvs[nmv++] = i;
 646     }
 647     return nmv;
 648 }
 649
 650
 651 /****************************************************************************
 652  * x264_pixel_init:
 653  ****************************************************************************/
 654 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 655 {
 656     memset( pixf, 0, sizeof(*pixf) );
 657
 658 #define INIT2_NAME( name1, name2, cpu ) \
 659     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 660     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 661 #define INIT4_NAME( name1, name2, cpu ) \
 662     INIT2_NAME( name1, name2, cpu ) \
 663     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 664     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 665 #define INIT5_NAME( name1, name2, cpu ) \
 666     INIT4_NAME( name1, name2, cpu ) \
 667     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 668 #define INIT6_NAME( name1, name2, cpu ) \
 669     INIT5_NAME( name1, name2, cpu ) \
 670     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 671 #define INIT7_NAME( name1, name2, cpu ) \
 672     INIT6_NAME( name1, name2, cpu ) \
 673     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 674 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 675 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 676 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 677 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 678 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 679
 680 #define INIT_ADS( cpu ) \
 681     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 682     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 683     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 684
 685     INIT7( sad, );
 686     INIT7_NAME( sad_aligned, sad, );
 687     INIT7( sad_x3, );
 688     INIT7( sad_x4, );
 689     INIT7( ssd, );
 690     INIT7( satd, );
 691     INIT7( satd_x3, );
 692     INIT7( satd_x4, );
 693     INIT4( hadamard_ac, );
 694     INIT_ADS( );
 695
 696     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 697     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 698     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 699     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 700
 701     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 702     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 703     pixf->ssim_end4 = ssim_end4;
 704     pixf->var2_8x8 = pixel_var2_8x8;
 705
 706     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 707     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 708     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 709     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 710     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 711     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 712     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 713     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 714
 715 #if !X264_HIGH_BIT_DEPTH
 716 #if HAVE_MMX
 717     if( cpu&X264_CPU_MMX )
 718     {
 719         INIT7( ssd, _mmx );
 720     }
 721
 722     if( cpu&X264_CPU_MMXEXT )
 723     {
 724         INIT7( sad, _mmxext );
 725         INIT7_NAME( sad_aligned, sad, _mmxext );
 726         INIT7( sad_x3, _mmxext );
 727         INIT7( sad_x4, _mmxext );
 728         INIT7( satd, _mmxext );
 729         INIT7( satd_x3, _mmxext );
 730         INIT7( satd_x4, _mmxext );
 731         INIT4( hadamard_ac, _mmxext );
 732         INIT_ADS( _mmxext );
 733         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
 734         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
 735         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmxext;
 736 #if ARCH_X86
 737         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
 738         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmxext;
 739         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
 740         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_mmxext;
 741         pixf->var2_8x8 = x264_pixel_var2_8x8_mmxext;
 742
 743         if( cpu&X264_CPU_CACHELINE_32 )
 744         {
 745             INIT5( sad, _cache32_mmxext );
 746             INIT4( sad_x3, _cache32_mmxext );
 747             INIT4( sad_x4, _cache32_mmxext );
 748         }
 749         else if( cpu&X264_CPU_CACHELINE_64 )
 750         {
 751             INIT5( sad, _cache64_mmxext );
 752             INIT4( sad_x3, _cache64_mmxext );
 753             INIT4( sad_x4, _cache64_mmxext );
 754         }
 755 #else
 756         if( cpu&X264_CPU_CACHELINE_64 )
 757         {
 758             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
 759             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmxext;
 760             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmxext;
 761             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
 762             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmxext;
 763             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
 764             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmxext;
 765         }
 766 #endif
 767         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
 768         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmxext;
 769         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
 770         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmxext;
 771         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmxext;
 772         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
 773         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmxext;
 774     }
 775
 776     if( cpu&X264_CPU_SSE2 )
 777     {
 778         INIT5( ssd, _sse2slow );
 779         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
 780         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 781         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
 782         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 783         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 784         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 785         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 786 #if ARCH_X86_64
 787         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 788 #endif
 789         pixf->var2_8x8 = x264_pixel_var2_8x8_sse2;
 790     }
 791
 792     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 793     {
 794         INIT2( sad, _sse2 );
 795         INIT2( sad_x3, _sse2 );
 796         INIT2( sad_x4, _sse2 );
 797         INIT6( satd, _sse2 );
 798         INIT6( satd_x3, _sse2 );
 799         INIT6( satd_x4, _sse2 );
 800         if( !(cpu&X264_CPU_STACK_MOD4) )
 801         {
 802             INIT4( hadamard_ac, _sse2 );
 803         }
 804         INIT_ADS( _sse2 );
 805         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
 806         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
 807         if( cpu&X264_CPU_CACHELINE_64 )
 808         {
 809             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
 810 #if ARCH_X86
 811             INIT2( sad, _cache64_sse2 );
 812             INIT2( sad_x3, _cache64_sse2 );
 813             INIT2( sad_x4, _cache64_sse2 );
 814 #endif
 815            if( cpu&X264_CPU_SSE2_IS_FAST )
 816            {
 817                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
 818                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
 819            }
 820         }
 821
 822         if( cpu&X264_CPU_SSE_MISALIGN )
 823         {
 824             INIT2( sad_x3, _sse2_misalign );
 825             INIT2( sad_x4, _sse2_misalign );
 826         }
 827     }
 828
 829     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
 830     {
 831         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 832         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 833         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 834         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
 835         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
 836         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 837         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
 838         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
 839     }
 840
 841     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
 842     {
 843         INIT2( sad, _sse3 );
 844         INIT2( sad_x3, _sse3 );
 845         INIT2( sad_x4, _sse3 );
 846     }
 847
 848     if( cpu&X264_CPU_SSSE3 )
 849     {
 850         if( !(cpu&X264_CPU_STACK_MOD4) )
 851         {
 852             INIT4( hadamard_ac, _ssse3 );
 853         }
 854         INIT_ADS( _ssse3 );
 855         if( !(cpu&X264_CPU_SLOW_ATOM) )
 856         {
 857             INIT7( ssd, _ssse3 );
 858             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 859             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 860             INIT7( satd, _ssse3 );
 861             INIT7( satd_x3, _ssse3 );
 862             INIT7( satd_x4, _ssse3 );
 863         }
 864         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
 865         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 866         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
 867         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 868         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
 869 #if ARCH_X86_64
 870         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
 871 #endif
 872         pixf->var2_8x8 = x264_pixel_var2_8x8_ssse3;
 873         if( cpu&X264_CPU_CACHELINE_64 )
 874         {
 875             INIT2( sad, _cache64_ssse3 );
 876             INIT2( sad_x3, _cache64_ssse3 );
 877             INIT2( sad_x4, _cache64_ssse3 );
 878         }
 879         if( cpu&X264_CPU_SLOW_ATOM || !(cpu&X264_CPU_SHUFFLE_IS_FAST) )
 880         {
 881             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
 882         }
 883     }
 884
 885     if( cpu&X264_CPU_SSE4 )
 886     {
 887         INIT7( satd, _sse4 );
 888         INIT7( satd_x3, _sse4 );
 889         INIT7( satd_x4, _sse4 );
 890         if( !(cpu&X264_CPU_STACK_MOD4) )
 891         {
 892             INIT4( hadamard_ac, _sse4 );
 893         }
 894         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 895         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 896         pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_sse4;
 897         /* Slower on Conroe, so only enable under SSE4 */
 898         pixf->intra_sad_x3_8x8  = x264_intra_sad_x3_8x8_ssse3;
 899     }
 900 #endif //HAVE_MMX
 901
 902 #if HAVE_ARMV6
 903     if( cpu&X264_CPU_ARMV6 )
 904     {
 905         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
 906         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
 907         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
 908         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
 909     }
 910     if( cpu&X264_CPU_NEON )
 911     {
 912         INIT5( sad, _neon );
 913         INIT5( sad_aligned, _neon );
 914         INIT7( sad_x3, _neon );
 915         INIT7( sad_x4, _neon );
 916         INIT7( ssd, _neon );
 917         INIT7( satd, _neon );
 918         INIT7( satd_x3, _neon );
 919         INIT7( satd_x4, _neon );
 920         INIT4( hadamard_ac, _neon );
 921         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
 922         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
 923         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
 924         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
 925         pixf->var2_8x8          = x264_pixel_var2_8x8_neon;
 926
 927         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
 928         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
 929
 930         if( cpu&X264_CPU_FAST_NEON_MRC )
 931         {
 932             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
 933             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
 934             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
 935             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
 936         }
 937         else    // really just scheduled for dual issue / A8
 938         {
 939             INIT5( sad_aligned, _neon_dual );
 940         }
 941     }
 942 #endif
 943 #endif // !X264_HIGH_BIT_DEPTH
 944 #if HAVE_ALTIVEC
 945     if( cpu&X264_CPU_ALTIVEC )
 946     {
 947         x264_pixel_altivec_init( pixf );
 948     }
 949 #endif
 950 #if !X264_HIGH_BIT_DEPTH
 951 #if ARCH_UltraSparc
 952     INIT4( sad, _vis );
 953     INIT4( sad_x3, _vis );
 954     INIT4( sad_x4, _vis );
 955 #endif
 956 #endif // !X264_HIGH_BIT_DEPTH
 957
 958     pixf->ads[PIXEL_8x16] =
 959     pixf->ads[PIXEL_8x4] =
 960     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
 961     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
 962 }
 963