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

   1 /*****************************************************************************
   2  * pixel.c: h264 encoder
   3  *****************************************************************************
   4  * Copyright (C) 2003-2008 x264 project
   5  *
   6  * Authors: Loren Merritt <lorenm@u.washington.edu>
   7  *          Laurent Aimar <fenrir@via.ecp.fr>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  22  *****************************************************************************/
  23
  24 #include "common.h"
  25
  26 #if HAVE_MMX
  27 #   include "x86/pixel.h"
  28 #endif
  29 #if ARCH_PPC
  30 #   include "ppc/pixel.h"
  31 #endif
  32 #if ARCH_ARM
  33 #   include "arm/pixel.h"
  34 #endif
  35 #if ARCH_UltraSparc
  36 #   include "sparc/pixel.h"
  37 #endif
  38
  39
  40 /****************************************************************************
  41  * pixel_sad_WxH
  42  ****************************************************************************/
  43 #define PIXEL_SAD_C( name, lx, ly ) \
  44 static int name( pixel *pix1, int i_stride_pix1,  \
  45                  pixel *pix2, int i_stride_pix2 ) \
  46 {                                                   \
  47     int i_sum = 0;                                  \
  48     for( int y = 0; y < ly; y++ )                   \
  49     {                                               \
  50         for( int x = 0; x < lx; x++ )               \
  51         {                                           \
  52             i_sum += abs( pix1[x] - pix2[x] );      \
  53         }                                           \
  54         pix1 += i_stride_pix1;                      \
  55         pix2 += i_stride_pix2;                      \
  56     }                                               \
  57     return i_sum;                                   \
  58 }
  59
  60
  61 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  62 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  63 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  64 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  65 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  66 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  67 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  68
  69
  70 /****************************************************************************
  71  * pixel_ssd_WxH
  72  ****************************************************************************/
  73 #define PIXEL_SSD_C( name, lx, ly ) \
  74 static int name( pixel *pix1, int i_stride_pix1,  \
  75                  pixel *pix2, int i_stride_pix2 ) \
  76 {                                                   \
  77     int i_sum = 0;                                  \
  78     for( int y = 0; y < ly; y++ )                   \
  79     {                                               \
  80         for( int x = 0; x < lx; x++ )               \
  81         {                                           \
  82             int d = pix1[x] - pix2[x];              \
  83             i_sum += d*d;                           \
  84         }                                           \
  85         pix1 += i_stride_pix1;                      \
  86         pix2 += i_stride_pix2;                      \
  87     }                                               \
  88     return i_sum;                                   \
  89 }
  90
  91 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
  92 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
  93 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
  94 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
  95 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
  96 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
  97 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
  98
  99 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 )
 100 {
 101     uint64_t i_ssd = 0;
 102     int y;
 103     int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
 104
 105 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 106                                           pix2 + y*i_pix2 + x, i_pix2 );
 107     for( y = 0; y < i_height-15; y += 16 )
 108     {
 109         int x = 0;
 110         if( align )
 111             for( ; x < i_width-15; x += 16 )
 112                 SSD(PIXEL_16x16);
 113         for( ; x < i_width-7; x += 8 )
 114             SSD(PIXEL_8x16);
 115     }
 116     if( y < i_height-7 )
 117         for( int x = 0; x < i_width-7; x += 8 )
 118             SSD(PIXEL_8x8);
 119 #undef SSD
 120
 121 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 122     if( i_width & 7 )
 123     {
 124         for( y = 0; y < (i_height & ~7); y++ )
 125             for( int x = i_width & ~7; x < i_width; x++ )
 126                 SSD1;
 127     }
 128     if( i_height & 7 )
 129     {
 130         for( y = i_height & ~7; y < i_height; y++ )
 131             for( int x = 0; x < i_width; x++ )
 132                 SSD1;
 133     }
 134 #undef SSD1
 135
 136     return i_ssd;
 137 }
 138
 139 static uint64_t pixel_ssd_nv12_core( pixel *pixuv1, int stride1, pixel *pixuv2, int stride2, int width, int height )
 140 {
 141     uint32_t ssd_u=0, ssd_v=0;
 142     for( int y = 0; y < height; y++, pixuv1+=stride1, pixuv2+=stride2 )
 143         for( int x = 0; x < width; x++ )
 144         {
 145             int du = pixuv1[2*x]   - pixuv2[2*x];
 146             int dv = pixuv1[2*x+1] - pixuv2[2*x+1];
 147             ssd_u += du*du;
 148             ssd_v += dv*dv;
 149         }
 150     return ssd_u + ((uint64_t)ssd_v<<32);
 151 }
 152
 153 // SSD in uint32 (i.e. packing two into uint64) can potentially overflow on
 154 // image widths >= 11008 (or 6604 if interlaced), since this is called on blocks
 155 // of height up to 12 (resp 20). Though it will probably take significantly more
 156 // than that at sane distortion levels.
 157 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 )
 158 {
 159     uint64_t ssd = pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height );
 160     if( i_width&7 )
 161         ssd += pixel_ssd_nv12_core( pix1+(i_width&~7), i_pix1, pix2+(i_width&~7), i_pix2, i_width&7, i_height );
 162     return ssd;
 163 }
 164
 165 /****************************************************************************
 166  * pixel_var_wxh
 167  ****************************************************************************/
 168 #define PIXEL_VAR_C( name, w ) \
 169 static uint64_t name( pixel *pix, int i_stride ) \
 170 {                                             \
 171     uint32_t sum = 0, sqr = 0;                \
 172     for( int y = 0; y < w; y++ )              \
 173     {                                         \
 174         for( int x = 0; x < w; x++ )          \
 175         {                                     \
 176             sum += pix[x];                    \
 177             sqr += pix[x] * pix[x];           \
 178         }                                     \
 179         pix += i_stride;                      \
 180     }                                         \
 181     return sum + ((uint64_t)sqr << 32);       \
 182 }
 183
 184 PIXEL_VAR_C( x264_pixel_var_16x16, 16 )
 185 PIXEL_VAR_C( x264_pixel_var_8x8,    8 )
 186
 187 /****************************************************************************
 188  * pixel_var2_wxh
 189  ****************************************************************************/
 190 static int pixel_var2_8x8( pixel *pix1, int i_stride1, pixel *pix2, int i_stride2, int *ssd )
 191 {
 192     uint32_t var = 0, sum = 0, sqr = 0;
 193     for( int y = 0; y < 8; y++ )
 194     {
 195         for( int x = 0; x < 8; x++ )
 196         {
 197             int diff = pix1[x] - pix2[x];
 198             sum += diff;
 199             sqr += diff * diff;
 200         }
 201         pix1 += i_stride1;
 202         pix2 += i_stride2;
 203     }
 204     sum = abs(sum);
 205     var = sqr - ((uint64_t)sum * sum >> 6);
 206     *ssd = sqr;
 207     return var;
 208 }
 209
 210
 211 #define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
 212     int t0 = s0 + s1;\
 213     int t1 = s0 - s1;\
 214     int t2 = s2 + s3;\
 215     int t3 = s2 - s3;\
 216     d0 = t0 + t2;\
 217     d2 = t0 - t2;\
 218     d1 = t1 + t3;\
 219     d3 = t1 - t3;\
 220 }
 221
 222 // in: a pseudo-simd number of the form x+(y<<16)
 223 // return: abs(x)+(abs(y)<<16)
 224 static ALWAYS_INLINE uint32_t abs2( uint32_t a )
 225 {
 226     uint32_t s = ((a>>15)&0x10001)*0xffff;
 227     return (a+s)^s;
 228 }
 229
 230 /****************************************************************************
 231  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 232  ****************************************************************************/
 233
 234 static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 235 {
 236     uint32_t tmp[4][2];
 237     uint32_t a0, a1, a2, a3, b0, b1;
 238     int sum = 0;
 239     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 240     {
 241         a0 = pix1[0] - pix2[0];
 242         a1 = pix1[1] - pix2[1];
 243         b0 = (a0+a1) + ((a0-a1)<<16);
 244         a2 = pix1[2] - pix2[2];
 245         a3 = pix1[3] - pix2[3];
 246         b1 = (a2+a3) + ((a2-a3)<<16);
 247         tmp[i][0] = b0 + b1;
 248         tmp[i][1] = b0 - b1;
 249     }
 250     for( int i = 0; i < 2; i++ )
 251     {
 252         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 253         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 254         sum += ((uint16_t)a0) + (a0>>16);
 255     }
 256     return sum >> 1;
 257 }
 258
 259 static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 260 {
 261     uint32_t tmp[4][4];
 262     uint32_t a0, a1, a2, a3;
 263     int sum = 0;
 264     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 265     {
 266         a0 = (pix1[0] - pix2[0]) + ((pix1[4] - pix2[4]) << 16);
 267         a1 = (pix1[1] - pix2[1]) + ((pix1[5] - pix2[5]) << 16);
 268         a2 = (pix1[2] - pix2[2]) + ((pix1[6] - pix2[6]) << 16);
 269         a3 = (pix1[3] - pix2[3]) + ((pix1[7] - pix2[7]) << 16);
 270         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 271     }
 272     for( int i = 0; i < 4; i++ )
 273     {
 274         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 275         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 276     }
 277     return (((uint16_t)sum) + ((uint32_t)sum>>16)) >> 1;
 278 }
 279
 280 #define PIXEL_SATD_C( w, h, sub )\
 281 static int x264_pixel_satd_##w##x##h( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )\
 282 {\
 283     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 284             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 285     if( w==16 )\
 286         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 287             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 288     if( h==16 )\
 289         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 290             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 291     if( w==16 && h==16 )\
 292         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 293             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 294     return sum;\
 295 }
 296 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 297 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 298 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 299 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 300 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 301
 302
 303 static NOINLINE int sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 304 {
 305     uint32_t tmp[8][4];
 306     uint32_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
 307     int sum = 0;
 308     for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
 309     {
 310         a0 = pix1[0] - pix2[0];
 311         a1 = pix1[1] - pix2[1];
 312         b0 = (a0+a1) + ((a0-a1)<<16);
 313         a2 = pix1[2] - pix2[2];
 314         a3 = pix1[3] - pix2[3];
 315         b1 = (a2+a3) + ((a2-a3)<<16);
 316         a4 = pix1[4] - pix2[4];
 317         a5 = pix1[5] - pix2[5];
 318         b2 = (a4+a5) + ((a4-a5)<<16);
 319         a6 = pix1[6] - pix2[6];
 320         a7 = pix1[7] - pix2[7];
 321         b3 = (a6+a7) + ((a6-a7)<<16);
 322         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 323     }
 324     for( int i = 0; i < 4; i++ )
 325     {
 326         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 327         HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 328         b0  = abs2(a0+a4) + abs2(a0-a4);
 329         b0 += abs2(a1+a5) + abs2(a1-a5);
 330         b0 += abs2(a2+a6) + abs2(a2-a6);
 331         b0 += abs2(a3+a7) + abs2(a3-a7);
 332         sum += (uint16_t)b0 + (b0>>16);
 333     }
 334     return sum;
 335 }
 336
 337 static int x264_pixel_sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 338 {
 339     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
 340     return (sum+2)>>2;
 341 }
 342
 343 static int x264_pixel_sa8d_16x16( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 344 {
 345     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 346             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 347             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 348             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 349     return (sum+2)>>2;
 350 }
 351
 352
 353 static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, int stride )
 354 {
 355     uint32_t tmp[32];
 356     uint32_t a0, a1, a2, a3, dc;
 357     int sum4 = 0, sum8 = 0;
 358     for( int i = 0; i < 8; i++, pix+=stride )
 359     {
 360         uint32_t *t = tmp + (i&3) + (i&4)*4;
 361         a0 = (pix[0]+pix[1]) + ((pix[0]-pix[1])<<16);
 362         a1 = (pix[2]+pix[3]) + ((pix[2]-pix[3])<<16);
 363         t[0] = a0 + a1;
 364         t[4] = a0 - a1;
 365         a2 = (pix[4]+pix[5]) + ((pix[4]-pix[5])<<16);
 366         a3 = (pix[6]+pix[7]) + ((pix[6]-pix[7])<<16);
 367         t[8] = a2 + a3;
 368         t[12] = a2 - a3;
 369     }
 370     for( int i = 0; i < 8; i++ )
 371     {
 372         HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 373         tmp[i*4+0] = a0;
 374         tmp[i*4+1] = a1;
 375         tmp[i*4+2] = a2;
 376         tmp[i*4+3] = a3;
 377         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 378     }
 379     for( int i = 0; i < 8; i++ )
 380     {
 381         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 382         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 383     }
 384     dc = (uint16_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 385     sum4 = (uint16_t)sum4 + ((uint32_t)sum4>>16) - dc;
 386     sum8 = (uint16_t)sum8 + ((uint32_t)sum8>>16) - dc;
 387     return ((uint64_t)sum8<<32) + sum4;
 388 }
 389
 390 #define HADAMARD_AC(w,h) \
 391 static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, int stride )\
 392 {\
 393     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 394     if( w==16 )\
 395         sum += pixel_hadamard_ac( pix+8, stride );\
 396     if( h==16 )\
 397         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 398     if( w==16 && h==16 )\
 399         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 400     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 401 }
 402 HADAMARD_AC( 16, 16 )
 403 HADAMARD_AC( 16, 8 )
 404 HADAMARD_AC( 8, 16 )
 405 HADAMARD_AC( 8, 8 )
 406
 407
 408 /****************************************************************************
 409  * pixel_sad_x4
 410  ****************************************************************************/
 411 #define SAD_X( size ) \
 412 static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
 413 {\
 414     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 415     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 416     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 417 }\
 418 static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
 419 {\
 420     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 421     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 422     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 423     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 424 }
 425
 426 SAD_X( 16x16 )
 427 SAD_X( 16x8 )
 428 SAD_X( 8x16 )
 429 SAD_X( 8x8 )
 430 SAD_X( 8x4 )
 431 SAD_X( 4x8 )
 432 SAD_X( 4x4 )
 433
 434 #if !X264_HIGH_BIT_DEPTH
 435 #if ARCH_UltraSparc
 436 SAD_X( 16x16_vis )
 437 SAD_X( 16x8_vis )
 438 SAD_X( 8x16_vis )
 439 SAD_X( 8x8_vis )
 440 #endif
 441 #endif // !X264_HIGH_BIT_DEPTH
 442
 443 /****************************************************************************
 444  * pixel_satd_x4
 445  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 446  ****************************************************************************/
 447
 448 #define SATD_X( size, cpu ) \
 449 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
 450 {\
 451     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 452     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 453     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 454 }\
 455 static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
 456 {\
 457     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 458     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 459     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 460     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 461 }
 462 #define SATD_X_DECL6( cpu )\
 463 SATD_X( 16x16, cpu )\
 464 SATD_X( 16x8, cpu )\
 465 SATD_X( 8x16, cpu )\
 466 SATD_X( 8x8, cpu )\
 467 SATD_X( 8x4, cpu )\
 468 SATD_X( 4x8, cpu )
 469 #define SATD_X_DECL7( cpu )\
 470 SATD_X_DECL6( cpu )\
 471 SATD_X( 4x4, cpu )
 472
 473 SATD_X_DECL7()
 474 #if !X264_HIGH_BIT_DEPTH
 475 #if HAVE_MMX
 476 SATD_X_DECL7( _mmxext )
 477 SATD_X_DECL6( _sse2 )
 478 SATD_X_DECL7( _ssse3 )
 479 SATD_X_DECL7( _sse4 )
 480 #endif
 481
 482 #if HAVE_ARMV6
 483 SATD_X_DECL7( _neon )
 484 #endif
 485 #endif // !X264_HIGH_BIT_DEPTH
 486
 487 #define INTRA_MBCMP_8x8( mbcmp )\
 488 void x264_intra_##mbcmp##_x3_8x8( pixel *fenc, pixel edge[33], int res[3] )\
 489 {\
 490     pixel pix[8*FDEC_STRIDE];\
 491     x264_predict_8x8_v_c( pix, edge );\
 492     res[0] = x264_pixel_##mbcmp##_8x8( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 493     x264_predict_8x8_h_c( pix, edge );\
 494     res[1] = x264_pixel_##mbcmp##_8x8( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 495     x264_predict_8x8_dc_c( pix, edge );\
 496     res[2] = x264_pixel_##mbcmp##_8x8( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 497 }
 498
 499 INTRA_MBCMP_8x8(sad)
 500 INTRA_MBCMP_8x8(sa8d)
 501
 502 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma )\
 503 void x264_intra_##mbcmp##_x3_##size##x##size##chroma( pixel *fenc, pixel *fdec, int res[3] )\
 504 {\
 505     x264_predict_##size##x##size##chroma##_##pred1##_c( fdec );\
 506     res[0] = x264_pixel_##mbcmp##_##size##x##size( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 507     x264_predict_##size##x##size##chroma##_##pred2##_c( fdec );\
 508     res[1] = x264_pixel_##mbcmp##_##size##x##size( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 509     x264_predict_##size##x##size##chroma##_##pred3##_c( fdec );\
 510     res[2] = x264_pixel_##mbcmp##_##size##x##size( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 511 }
 512
 513 INTRA_MBCMP(sad, 4, v, h, dc, )
 514 INTRA_MBCMP(satd, 4, v, h, dc, )
 515 INTRA_MBCMP(sad, 8, dc, h, v, c )
 516 INTRA_MBCMP(satd, 8, dc, h, v, c )
 517 INTRA_MBCMP(sad, 16, v, h, dc, )
 518 INTRA_MBCMP(satd, 16, v, h, dc, )
 519
 520 /****************************************************************************
 521  * structural similarity metric
 522  ****************************************************************************/
 523 static void ssim_4x4x2_core( const pixel *pix1, int stride1,
 524                              const pixel *pix2, int stride2,
 525                              int sums[2][4])
 526 {
 527     for( int z = 0; z < 2; z++ )
 528     {
 529         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 530         for( int y = 0; y < 4; y++ )
 531             for( int x = 0; x < 4; x++ )
 532             {
 533                 int a = pix1[x+y*stride1];
 534                 int b = pix2[x+y*stride2];
 535                 s1  += a;
 536                 s2  += b;
 537                 ss  += a*a;
 538                 ss  += b*b;
 539                 s12 += a*b;
 540             }
 541         sums[z][0] = s1;
 542         sums[z][1] = s2;
 543         sums[z][2] = ss;
 544         sums[z][3] = s12;
 545         pix1 += 4;
 546         pix2 += 4;
 547     }
 548 }
 549
 550 static float ssim_end1( int s1, int s2, int ss, int s12 )
 551 {
 552     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 553     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 554     int vars = ss*64 - s1*s1 - s2*s2;
 555     int covar = s12*64 - s1*s2;
 556     return (float)(2*s1*s2 + ssim_c1) * (float)(2*covar + ssim_c2)
 557          / ((float)(s1*s1 + s2*s2 + ssim_c1) * (float)(vars + ssim_c2));
 558 }
 559
 560 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 561 {
 562     float ssim = 0.0;
 563     for( int i = 0; i < width; i++ )
 564         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 565                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 566                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 567                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 568     return ssim;
 569 }
 570
 571 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 572                            pixel *pix1, int stride1,
 573                            pixel *pix2, int stride2,
 574                            int width, int height, void *buf )
 575 {
 576     int z = 0;
 577     float ssim = 0.0;
 578     int (*sum0)[4] = buf;
 579     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 580     width >>= 2;
 581     height >>= 2;
 582     for( int y = 1; y < height; y++ )
 583     {
 584         for( ; z <= y; z++ )
 585         {
 586             XCHG( void*, sum0, sum1 );
 587             for( int x = 0; x < width; x+=2 )
 588                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 589         }
 590         for( int x = 0; x < width-1; x += 4 )
 591             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 592     }
 593     return ssim;
 594 }
 595
 596
 597 /****************************************************************************
 598  * successive elimination
 599  ****************************************************************************/
 600 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 601                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 602 {
 603     int nmv = 0;
 604     for( int i = 0; i < width; i++, sums++ )
 605     {
 606         int ads = abs( enc_dc[0] - sums[0] )
 607                 + abs( enc_dc[1] - sums[8] )
 608                 + abs( enc_dc[2] - sums[delta] )
 609                 + abs( enc_dc[3] - sums[delta+8] )
 610                 + cost_mvx[i];
 611         if( ads < thresh )
 612             mvs[nmv++] = i;
 613     }
 614     return nmv;
 615 }
 616
 617 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 618                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 619 {
 620     int nmv = 0;
 621     for( int i = 0; i < width; i++, sums++ )
 622     {
 623         int ads = abs( enc_dc[0] - sums[0] )
 624                 + abs( enc_dc[1] - sums[delta] )
 625                 + cost_mvx[i];
 626         if( ads < thresh )
 627             mvs[nmv++] = i;
 628     }
 629     return nmv;
 630 }
 631
 632 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 633                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 634 {
 635     int nmv = 0;
 636     for( int i = 0; i<width; i++, sums++ )
 637     {
 638         int ads = abs( enc_dc[0] - sums[0] )
 639                 + cost_mvx[i];
 640         if( ads < thresh )
 641             mvs[nmv++] = i;
 642     }
 643     return nmv;
 644 }
 645
 646
 647 /****************************************************************************
 648  * x264_pixel_init:
 649  ****************************************************************************/
 650 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 651 {
 652     memset( pixf, 0, sizeof(*pixf) );
 653
 654 #define INIT2_NAME( name1, name2, cpu ) \
 655     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 656     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 657 #define INIT4_NAME( name1, name2, cpu ) \
 658     INIT2_NAME( name1, name2, cpu ) \
 659     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 660     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 661 #define INIT5_NAME( name1, name2, cpu ) \
 662     INIT4_NAME( name1, name2, cpu ) \
 663     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 664 #define INIT6_NAME( name1, name2, cpu ) \
 665     INIT5_NAME( name1, name2, cpu ) \
 666     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 667 #define INIT7_NAME( name1, name2, cpu ) \
 668     INIT6_NAME( name1, name2, cpu ) \
 669     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 670 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 671 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 672 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 673 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 674 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 675
 676 #define INIT_ADS( cpu ) \
 677     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 678     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 679     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 680
 681     INIT7( sad, );
 682     INIT7_NAME( sad_aligned, sad, );
 683     INIT7( sad_x3, );
 684     INIT7( sad_x4, );
 685     INIT7( ssd, );
 686     INIT7( satd, );
 687     INIT7( satd_x3, );
 688     INIT7( satd_x4, );
 689     INIT4( hadamard_ac, );
 690     INIT_ADS( );
 691
 692     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 693     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 694     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 695     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 696
 697     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 698     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 699     pixf->ssim_end4 = ssim_end4;
 700     pixf->var2_8x8 = pixel_var2_8x8;
 701
 702     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 703     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 704     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 705     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 706     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 707     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 708     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 709     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 710
 711 #if !X264_HIGH_BIT_DEPTH
 712 #if HAVE_MMX
 713     if( cpu&X264_CPU_MMX )
 714     {
 715         INIT7( ssd, _mmx );
 716     }
 717
 718     if( cpu&X264_CPU_MMXEXT )
 719     {
 720         INIT7( sad, _mmxext );
 721         INIT7_NAME( sad_aligned, sad, _mmxext );
 722         INIT7( sad_x3, _mmxext );
 723         INIT7( sad_x4, _mmxext );
 724         INIT7( satd, _mmxext );
 725         INIT7( satd_x3, _mmxext );
 726         INIT7( satd_x4, _mmxext );
 727         INIT4( hadamard_ac, _mmxext );
 728         INIT_ADS( _mmxext );
 729         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
 730         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
 731         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmxext;
 732 #if ARCH_X86
 733         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
 734         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmxext;
 735         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
 736         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_mmxext;
 737         pixf->var2_8x8 = x264_pixel_var2_8x8_mmxext;
 738
 739         if( cpu&X264_CPU_CACHELINE_32 )
 740         {
 741             INIT5( sad, _cache32_mmxext );
 742             INIT4( sad_x3, _cache32_mmxext );
 743             INIT4( sad_x4, _cache32_mmxext );
 744         }
 745         else if( cpu&X264_CPU_CACHELINE_64 )
 746         {
 747             INIT5( sad, _cache64_mmxext );
 748             INIT4( sad_x3, _cache64_mmxext );
 749             INIT4( sad_x4, _cache64_mmxext );
 750         }
 751 #else
 752         if( cpu&X264_CPU_CACHELINE_64 )
 753         {
 754             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
 755             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmxext;
 756             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmxext;
 757             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
 758             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmxext;
 759             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
 760             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmxext;
 761         }
 762 #endif
 763         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
 764         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmxext;
 765         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
 766         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmxext;
 767         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmxext;
 768         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
 769         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmxext;
 770     }
 771
 772     if( cpu&X264_CPU_SSE2 )
 773     {
 774         INIT5( ssd, _sse2slow );
 775         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
 776         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 777         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
 778         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 779         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 780         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 781         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 782 #if ARCH_X86_64
 783         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 784 #endif
 785         pixf->var2_8x8 = x264_pixel_var2_8x8_sse2;
 786     }
 787
 788     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 789     {
 790         INIT2( sad, _sse2 );
 791         INIT2( sad_x3, _sse2 );
 792         INIT2( sad_x4, _sse2 );
 793         INIT6( satd, _sse2 );
 794         INIT6( satd_x3, _sse2 );
 795         INIT6( satd_x4, _sse2 );
 796         if( !(cpu&X264_CPU_STACK_MOD4) )
 797         {
 798             INIT4( hadamard_ac, _sse2 );
 799         }
 800         INIT_ADS( _sse2 );
 801         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
 802         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
 803         if( cpu&X264_CPU_CACHELINE_64 )
 804         {
 805             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
 806 #if ARCH_X86
 807             INIT2( sad, _cache64_sse2 );
 808             INIT2( sad_x3, _cache64_sse2 );
 809             INIT2( sad_x4, _cache64_sse2 );
 810 #endif
 811            if( cpu&X264_CPU_SSE2_IS_FAST )
 812            {
 813                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
 814                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
 815            }
 816         }
 817
 818         if( cpu&X264_CPU_SSE_MISALIGN )
 819         {
 820             INIT2( sad_x3, _sse2_misalign );
 821             INIT2( sad_x4, _sse2_misalign );
 822         }
 823     }
 824
 825     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
 826     {
 827         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 828         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 829         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 830         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
 831         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
 832         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 833         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
 834         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
 835     }
 836
 837     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
 838     {
 839         INIT2( sad, _sse3 );
 840         INIT2( sad_x3, _sse3 );
 841         INIT2( sad_x4, _sse3 );
 842     }
 843
 844     if( cpu&X264_CPU_SSSE3 )
 845     {
 846         if( !(cpu&X264_CPU_STACK_MOD4) )
 847         {
 848             INIT4( hadamard_ac, _ssse3 );
 849         }
 850         INIT_ADS( _ssse3 );
 851         if( !(cpu&X264_CPU_SLOW_ATOM) )
 852         {
 853             INIT7( ssd, _ssse3 );
 854             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 855             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 856             INIT7( satd, _ssse3 );
 857             INIT7( satd_x3, _ssse3 );
 858             INIT7( satd_x4, _ssse3 );
 859         }
 860         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
 861         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 862         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
 863         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 864         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
 865 #if ARCH_X86_64
 866         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
 867 #endif
 868         pixf->var2_8x8 = x264_pixel_var2_8x8_ssse3;
 869         if( cpu&X264_CPU_CACHELINE_64 )
 870         {
 871             INIT2( sad, _cache64_ssse3 );
 872             INIT2( sad_x3, _cache64_ssse3 );
 873             INIT2( sad_x4, _cache64_ssse3 );
 874         }
 875         if( cpu&X264_CPU_SLOW_ATOM || !(cpu&X264_CPU_SHUFFLE_IS_FAST) )
 876         {
 877             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
 878         }
 879     }
 880
 881     if( cpu&X264_CPU_SSE4 )
 882     {
 883         INIT7( satd, _sse4 );
 884         INIT7( satd_x3, _sse4 );
 885         INIT7( satd_x4, _sse4 );
 886         if( !(cpu&X264_CPU_STACK_MOD4) )
 887         {
 888             INIT4( hadamard_ac, _sse4 );
 889         }
 890         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 891         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 892         pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_sse4;
 893         /* Slower on Conroe, so only enable under SSE4 */
 894         pixf->intra_sad_x3_8x8  = x264_intra_sad_x3_8x8_ssse3;
 895     }
 896 #endif //HAVE_MMX
 897
 898 #if HAVE_ARMV6
 899     if( cpu&X264_CPU_ARMV6 )
 900     {
 901         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
 902         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
 903         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
 904         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
 905     }
 906     if( cpu&X264_CPU_NEON )
 907     {
 908         INIT5( sad, _neon );
 909         INIT5( sad_aligned, _neon );
 910         INIT7( sad_x3, _neon );
 911         INIT7( sad_x4, _neon );
 912         INIT7( ssd, _neon );
 913         INIT7( satd, _neon );
 914         INIT7( satd_x3, _neon );
 915         INIT7( satd_x4, _neon );
 916         INIT4( hadamard_ac, _neon );
 917         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
 918         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
 919         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
 920         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
 921         pixf->var2_8x8          = x264_pixel_var2_8x8_neon;
 922
 923         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
 924         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
 925
 926         if( cpu&X264_CPU_FAST_NEON_MRC )
 927         {
 928             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
 929             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
 930             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
 931             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
 932         }
 933         else    // really just scheduled for dual issue / A8
 934         {
 935             INIT5( sad_aligned, _neon_dual );
 936         }
 937     }
 938 #endif
 939 #endif // !X264_HIGH_BIT_DEPTH
 940 #if HAVE_ALTIVEC
 941     if( cpu&X264_CPU_ALTIVEC )
 942     {
 943         x264_pixel_altivec_init( pixf );
 944     }
 945 #endif
 946 #if !X264_HIGH_BIT_DEPTH
 947 #if ARCH_UltraSparc
 948     INIT4( sad, _vis );
 949     INIT4( sad_x3, _vis );
 950     INIT4( sad_x4, _vis );
 951 #endif
 952 #endif // !X264_HIGH_BIT_DEPTH
 953
 954     pixf->ads[PIXEL_8x16] =
 955     pixf->ads[PIXEL_8x4] =
 956     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
 957     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
 958 }
 959