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 #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 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 )
 144 {
 145     *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 }
 155
 156 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 )
 157 {
 158     pf->ssd_nv12_core( pix1, i_pix1, pix2, i_pix2, i_width&~7, i_height, ssd_u, ssd_v );
 159     if( i_width&7 )
 160     {
 161         uint64_t tmp[2];
 162         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] );
 163         *ssd_u += tmp[0];
 164         *ssd_v += tmp[1];
 165     }
 166 }
 167
 168 /****************************************************************************
 169  * pixel_var_wxh
 170  ****************************************************************************/
 171 #define PIXEL_VAR_C( name, w ) \
 172 static uint64_t name( pixel *pix, int i_stride ) \
 173 {                                             \
 174     uint32_t sum = 0, sqr = 0;                \
 175     for( int y = 0; y < w; y++ )              \
 176     {                                         \
 177         for( int x = 0; x < w; x++ )          \
 178         {                                     \
 179             sum += pix[x];                    \
 180             sqr += pix[x] * pix[x];           \
 181         }                                     \
 182         pix += i_stride;                      \
 183     }                                         \
 184     return sum + ((uint64_t)sqr << 32);       \
 185 }
 186
 187 PIXEL_VAR_C( x264_pixel_var_16x16, 16 )
 188 PIXEL_VAR_C( x264_pixel_var_8x8,    8 )
 189
 190 /****************************************************************************
 191  * pixel_var2_wxh
 192  ****************************************************************************/
 193 static int pixel_var2_8x8( pixel *pix1, int i_stride1, pixel *pix2, int i_stride2, int *ssd )
 194 {
 195     uint32_t var = 0, sum = 0, sqr = 0;
 196     for( int y = 0; y < 8; y++ )
 197     {
 198         for( int x = 0; x < 8; x++ )
 199         {
 200             int diff = pix1[x] - pix2[x];
 201             sum += diff;
 202             sqr += diff * diff;
 203         }
 204         pix1 += i_stride1;
 205         pix2 += i_stride2;
 206     }
 207     sum = abs(sum);
 208     var = sqr - ((uint64_t)sum * sum >> 6);
 209     *ssd = sqr;
 210     return var;
 211 }
 212
 213 #if BIT_DEPTH > 8
 214     typedef uint32_t sum_t;
 215     typedef uint64_t sum2_t;
 216 #else
 217     typedef uint16_t sum_t;
 218     typedef uint32_t sum2_t;
 219 #endif
 220 #define BITS_PER_SUM (8 * sizeof(sum_t))
 221
 222 #define HADAMARD4(d0, d1, d2, d3, s0, s1, s2, s3) {\
 223     sum2_t t0 = s0 + s1;\
 224     sum2_t t1 = s0 - s1;\
 225     sum2_t t2 = s2 + s3;\
 226     sum2_t t3 = s2 - s3;\
 227     d0 = t0 + t2;\
 228     d2 = t0 - t2;\
 229     d1 = t1 + t3;\
 230     d3 = t1 - t3;\
 231 }
 232
 233 // in: a pseudo-simd number of the form x+(y<<16)
 234 // return: abs(x)+(abs(y)<<16)
 235 static ALWAYS_INLINE sum2_t abs2( sum2_t a )
 236 {
 237     sum2_t s = ((a>>(BITS_PER_SUM-1))&(((sum2_t)1<<BITS_PER_SUM)+1))*((sum_t)-1);
 238     return (a+s)^s;
 239 }
 240
 241 /****************************************************************************
 242  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 243  ****************************************************************************/
 244
 245 static NOINLINE int x264_pixel_satd_4x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 246 {
 247     sum2_t tmp[4][2];
 248     sum2_t a0, a1, a2, a3, b0, b1;
 249     sum2_t sum = 0;
 250     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 251     {
 252         a0 = pix1[0] - pix2[0];
 253         a1 = pix1[1] - pix2[1];
 254         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 255         a2 = pix1[2] - pix2[2];
 256         a3 = pix1[3] - pix2[3];
 257         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 258         tmp[i][0] = b0 + b1;
 259         tmp[i][1] = b0 - b1;
 260     }
 261     for( int i = 0; i < 2; i++ )
 262     {
 263         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 264         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 265         sum += ((sum_t)a0) + (a0>>BITS_PER_SUM);
 266     }
 267     return sum >> 1;
 268 }
 269
 270 static NOINLINE int x264_pixel_satd_8x4( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 271 {
 272     sum2_t tmp[4][4];
 273     sum2_t a0, a1, a2, a3;
 274     sum2_t sum = 0;
 275     for( int i = 0; i < 4; i++, pix1 += i_pix1, pix2 += i_pix2 )
 276     {
 277         a0 = (pix1[0] - pix2[0]) + ((sum2_t)(pix1[4] - pix2[4]) << BITS_PER_SUM);
 278         a1 = (pix1[1] - pix2[1]) + ((sum2_t)(pix1[5] - pix2[5]) << BITS_PER_SUM);
 279         a2 = (pix1[2] - pix2[2]) + ((sum2_t)(pix1[6] - pix2[6]) << BITS_PER_SUM);
 280         a3 = (pix1[3] - pix2[3]) + ((sum2_t)(pix1[7] - pix2[7]) << BITS_PER_SUM);
 281         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 282     }
 283     for( int i = 0; i < 4; i++ )
 284     {
 285         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 286         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 287     }
 288     return (((sum_t)sum) + (sum>>BITS_PER_SUM)) >> 1;
 289 }
 290
 291 #define PIXEL_SATD_C( w, h, sub )\
 292 static int x264_pixel_satd_##w##x##h( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )\
 293 {\
 294     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 295             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 296     if( w==16 )\
 297         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 298             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 299     if( h==16 )\
 300         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 301             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 302     if( w==16 && h==16 )\
 303         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 304             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 305     return sum;\
 306 }
 307 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 308 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 309 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 310 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 311 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 312
 313
 314 static NOINLINE int sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 315 {
 316     sum2_t tmp[8][4];
 317     sum2_t a0, a1, a2, a3, a4, a5, a6, a7, b0, b1, b2, b3;
 318     sum2_t sum = 0;
 319     for( int i = 0; i < 8; i++, pix1 += i_pix1, pix2 += i_pix2 )
 320     {
 321         a0 = pix1[0] - pix2[0];
 322         a1 = pix1[1] - pix2[1];
 323         b0 = (a0+a1) + ((a0-a1)<<BITS_PER_SUM);
 324         a2 = pix1[2] - pix2[2];
 325         a3 = pix1[3] - pix2[3];
 326         b1 = (a2+a3) + ((a2-a3)<<BITS_PER_SUM);
 327         a4 = pix1[4] - pix2[4];
 328         a5 = pix1[5] - pix2[5];
 329         b2 = (a4+a5) + ((a4-a5)<<BITS_PER_SUM);
 330         a6 = pix1[6] - pix2[6];
 331         a7 = pix1[7] - pix2[7];
 332         b3 = (a6+a7) + ((a6-a7)<<BITS_PER_SUM);
 333         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 334     }
 335     for( int i = 0; i < 4; i++ )
 336     {
 337         HADAMARD4( a0, a1, a2, a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 338         HADAMARD4( a4, a5, a6, a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 339         b0  = abs2(a0+a4) + abs2(a0-a4);
 340         b0 += abs2(a1+a5) + abs2(a1-a5);
 341         b0 += abs2(a2+a6) + abs2(a2-a6);
 342         b0 += abs2(a3+a7) + abs2(a3-a7);
 343         sum += (sum_t)b0 + (b0>>BITS_PER_SUM);
 344     }
 345     return sum;
 346 }
 347
 348 static int x264_pixel_sa8d_8x8( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 349 {
 350     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
 351     return (sum+2)>>2;
 352 }
 353
 354 static int x264_pixel_sa8d_16x16( pixel *pix1, int i_pix1, pixel *pix2, int i_pix2 )
 355 {
 356     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 357             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 358             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 359             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 360     return (sum+2)>>2;
 361 }
 362
 363
 364 static NOINLINE uint64_t pixel_hadamard_ac( pixel *pix, int stride )
 365 {
 366     sum2_t tmp[32];
 367     sum2_t a0, a1, a2, a3, dc;
 368     sum2_t sum4 = 0, sum8 = 0;
 369     for( int i = 0; i < 8; i++, pix+=stride )
 370     {
 371         sum2_t *t = tmp + (i&3) + (i&4)*4;
 372         a0 = (pix[0]+pix[1]) + ((sum2_t)(pix[0]-pix[1])<<BITS_PER_SUM);
 373         a1 = (pix[2]+pix[3]) + ((sum2_t)(pix[2]-pix[3])<<BITS_PER_SUM);
 374         t[0] = a0 + a1;
 375         t[4] = a0 - a1;
 376         a2 = (pix[4]+pix[5]) + ((sum2_t)(pix[4]-pix[5])<<BITS_PER_SUM);
 377         a3 = (pix[6]+pix[7]) + ((sum2_t)(pix[6]-pix[7])<<BITS_PER_SUM);
 378         t[8] = a2 + a3;
 379         t[12] = a2 - a3;
 380     }
 381     for( int i = 0; i < 8; i++ )
 382     {
 383         HADAMARD4( a0, a1, a2, a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 384         tmp[i*4+0] = a0;
 385         tmp[i*4+1] = a1;
 386         tmp[i*4+2] = a2;
 387         tmp[i*4+3] = a3;
 388         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 389     }
 390     for( int i = 0; i < 8; i++ )
 391     {
 392         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 393         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 394     }
 395     dc = (sum_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 396     sum4 = (sum_t)sum4 + (sum4>>BITS_PER_SUM) - dc;
 397     sum8 = (sum_t)sum8 + (sum8>>BITS_PER_SUM) - dc;
 398     return ((uint64_t)sum8<<32) + sum4;
 399 }
 400
 401 #define HADAMARD_AC(w,h) \
 402 static uint64_t x264_pixel_hadamard_ac_##w##x##h( pixel *pix, int stride )\
 403 {\
 404     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 405     if( w==16 )\
 406         sum += pixel_hadamard_ac( pix+8, stride );\
 407     if( h==16 )\
 408         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 409     if( w==16 && h==16 )\
 410         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 411     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 412 }
 413 HADAMARD_AC( 16, 16 )
 414 HADAMARD_AC( 16, 8 )
 415 HADAMARD_AC( 8, 16 )
 416 HADAMARD_AC( 8, 8 )
 417
 418
 419 /****************************************************************************
 420  * pixel_sad_x4
 421  ****************************************************************************/
 422 #define SAD_X( size ) \
 423 static void x264_pixel_sad_x3_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
 424 {\
 425     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 426     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 427     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 428 }\
 429 static void x264_pixel_sad_x4_##size( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
 430 {\
 431     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 432     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 433     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 434     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 435 }
 436
 437 SAD_X( 16x16 )
 438 SAD_X( 16x8 )
 439 SAD_X( 8x16 )
 440 SAD_X( 8x8 )
 441 SAD_X( 8x4 )
 442 SAD_X( 4x8 )
 443 SAD_X( 4x4 )
 444
 445 #if !HIGH_BIT_DEPTH
 446 #if ARCH_UltraSPARC
 447 SAD_X( 16x16_vis )
 448 SAD_X( 16x8_vis )
 449 SAD_X( 8x16_vis )
 450 SAD_X( 8x8_vis )
 451 #endif
 452 #endif // !HIGH_BIT_DEPTH
 453
 454 /****************************************************************************
 455  * pixel_satd_x4
 456  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 457  ****************************************************************************/
 458
 459 #define SATD_X( size, cpu ) \
 460 static void x264_pixel_satd_x3_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, int i_stride, int scores[3] )\
 461 {\
 462     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 463     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 464     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 465 }\
 466 static void x264_pixel_satd_x4_##size##cpu( pixel *fenc, pixel *pix0, pixel *pix1, pixel *pix2, pixel *pix3, int i_stride, int scores[4] )\
 467 {\
 468     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 469     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 470     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 471     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 472 }
 473 #define SATD_X_DECL6( cpu )\
 474 SATD_X( 16x16, cpu )\
 475 SATD_X( 16x8, cpu )\
 476 SATD_X( 8x16, cpu )\
 477 SATD_X( 8x8, cpu )\
 478 SATD_X( 8x4, cpu )\
 479 SATD_X( 4x8, cpu )
 480 #define SATD_X_DECL7( cpu )\
 481 SATD_X_DECL6( cpu )\
 482 SATD_X( 4x4, cpu )
 483
 484 SATD_X_DECL7()
 485 #if HAVE_MMX
 486 SATD_X_DECL7( _mmxext )
 487 #if !HIGH_BIT_DEPTH
 488 SATD_X_DECL6( _sse2 )
 489 SATD_X_DECL7( _ssse3 )
 490 SATD_X_DECL7( _sse4 )
 491 SATD_X_DECL7( _avx )
 492 #endif // !HIGH_BIT_DEPTH
 493 #endif
 494
 495 #if !HIGH_BIT_DEPTH
 496 #if HAVE_ARMV6
 497 SATD_X_DECL7( _neon )
 498 #endif
 499 #endif // !HIGH_BIT_DEPTH
 500
 501 #define INTRA_MBCMP_8x8( mbcmp, cpu )\
 502 void x264_intra_##mbcmp##_x3_8x8##cpu( pixel *fenc, pixel edge[33], int res[3] )\
 503 {\
 504     ALIGNED_ARRAY_16( pixel, pix, [8*FDEC_STRIDE] );\
 505     x264_predict_8x8_v_c( pix, edge );\
 506     res[0] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 507     x264_predict_8x8_h_c( pix, edge );\
 508     res[1] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 509     x264_predict_8x8_dc_c( pix, edge );\
 510     res[2] = x264_pixel_##mbcmp##_8x8##cpu( pix, FDEC_STRIDE, fenc, FENC_STRIDE );\
 511 }
 512
 513 INTRA_MBCMP_8x8( sad, )
 514 INTRA_MBCMP_8x8(sa8d, )
 515 #if HIGH_BIT_DEPTH && HAVE_MMX
 516 INTRA_MBCMP_8x8( sad, _mmxext)
 517 INTRA_MBCMP_8x8( sad, _sse2  )
 518 INTRA_MBCMP_8x8( sad, _ssse3 )
 519 INTRA_MBCMP_8x8(sa8d, _sse2  )
 520 #endif
 521
 522 #define INTRA_MBCMP( mbcmp, size, pred1, pred2, pred3, chroma, cpu )\
 523 void x264_intra_##mbcmp##_x3_##size##x##size##chroma##cpu( pixel *fenc, pixel *fdec, int res[3] )\
 524 {\
 525     x264_predict_##size##x##size##chroma##_##pred1##_c( fdec );\
 526     res[0] = x264_pixel_##mbcmp##_##size##x##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 527     x264_predict_##size##x##size##chroma##_##pred2##_c( fdec );\
 528     res[1] = x264_pixel_##mbcmp##_##size##x##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 529     x264_predict_##size##x##size##chroma##_##pred3##_c( fdec );\
 530     res[2] = x264_pixel_##mbcmp##_##size##x##size##cpu( fdec, FDEC_STRIDE, fenc, FENC_STRIDE );\
 531 }
 532
 533 INTRA_MBCMP( sad,  4,  v, h, dc,  , )
 534 INTRA_MBCMP(satd,  4,  v, h, dc,  , )
 535 INTRA_MBCMP( sad,  8, dc, h,  v, c, )
 536 INTRA_MBCMP(satd,  8, dc, h,  v, c, )
 537 INTRA_MBCMP( sad, 16,  v, h, dc,  , )
 538 INTRA_MBCMP(satd, 16,  v, h, dc,  , )
 539
 540 #if HIGH_BIT_DEPTH && HAVE_MMX
 541 INTRA_MBCMP( sad,  4,  v, h, dc,  , _mmxext)
 542 INTRA_MBCMP(satd,  4,  v, h, dc,  , _mmxext)
 543 INTRA_MBCMP( sad,  8, dc, h,  v, c, _mmxext)
 544 INTRA_MBCMP(satd,  8, dc, h,  v, c, _mmxext)
 545 INTRA_MBCMP( sad, 16,  v, h, dc,  , _mmxext)
 546 INTRA_MBCMP(satd, 16,  v, h, dc,  , _mmxext)
 547 INTRA_MBCMP( sad,  8, dc, h,  v, c, _sse2  )
 548 INTRA_MBCMP( sad, 16,  v, h, dc,  , _sse2  )
 549 INTRA_MBCMP( sad,  8, dc, h,  v, c, _ssse3 )
 550 INTRA_MBCMP( sad, 16,  v, h, dc,  , _ssse3 )
 551 #endif
 552
 553 /****************************************************************************
 554  * structural similarity metric
 555  ****************************************************************************/
 556 static void ssim_4x4x2_core( const pixel *pix1, int stride1,
 557                              const pixel *pix2, int stride2,
 558                              int sums[2][4])
 559 {
 560     for( int z = 0; z < 2; z++ )
 561     {
 562         uint32_t s1 = 0, s2 = 0, ss = 0, s12 = 0;
 563         for( int y = 0; y < 4; y++ )
 564             for( int x = 0; x < 4; x++ )
 565             {
 566                 int a = pix1[x+y*stride1];
 567                 int b = pix2[x+y*stride2];
 568                 s1  += a;
 569                 s2  += b;
 570                 ss  += a*a;
 571                 ss  += b*b;
 572                 s12 += a*b;
 573             }
 574         sums[z][0] = s1;
 575         sums[z][1] = s2;
 576         sums[z][2] = ss;
 577         sums[z][3] = s12;
 578         pix1 += 4;
 579         pix2 += 4;
 580     }
 581 }
 582
 583 static float ssim_end1( int s1, int s2, int ss, int s12 )
 584 {
 585 /* Maximum value for 10-bit is: ss*64 = (2^10-1)^2*16*4*64 = 4286582784, which will overflow in some cases.
 586  * s1*s1, s2*s2, and s1*s2 also obtain this value for edge cases: ((2^10-1)*16*4)^2 = 4286582784.
 587  * Maximum value for 9-bit is: ss*64 = (2^9-1)^2*16*4*64 = 1069551616, which will not overflow. */
 588 #if BIT_DEPTH > 9
 589 #define type float
 590     static const float ssim_c1 = .01*.01*PIXEL_MAX*PIXEL_MAX*64;
 591     static const float ssim_c2 = .03*.03*PIXEL_MAX*PIXEL_MAX*64*63;
 592 #else
 593 #define type int
 594     static const int ssim_c1 = (int)(.01*.01*PIXEL_MAX*PIXEL_MAX*64 + .5);
 595     static const int ssim_c2 = (int)(.03*.03*PIXEL_MAX*PIXEL_MAX*64*63 + .5);
 596 #endif
 597     type fs1 = s1;
 598     type fs2 = s2;
 599     type fss = ss;
 600     type fs12 = s12;
 601     type vars = fss*64 - fs1*fs1 - fs2*fs2;
 602     type covar = fs12*64 - fs1*fs2;
 603     return (float)(2*fs1*fs2 + ssim_c1) * (float)(2*covar + ssim_c2)
 604          / ((float)(fs1*fs1 + fs2*fs2 + ssim_c1) * (float)(vars + ssim_c2));
 605 #undef type
 606 }
 607
 608 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 609 {
 610     float ssim = 0.0;
 611     for( int i = 0; i < width; i++ )
 612         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 613                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 614                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 615                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 616     return ssim;
 617 }
 618
 619 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 620                            pixel *pix1, int stride1,
 621                            pixel *pix2, int stride2,
 622                            int width, int height, void *buf )
 623 {
 624     int z = 0;
 625     float ssim = 0.0;
 626     int (*sum0)[4] = buf;
 627     int (*sum1)[4] = sum0 + (width >> 2) + 3;
 628     width >>= 2;
 629     height >>= 2;
 630     for( int y = 1; y < height; y++ )
 631     {
 632         for( ; z <= y; z++ )
 633         {
 634             XCHG( void*, sum0, sum1 );
 635             for( int x = 0; x < width; x+=2 )
 636                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 637         }
 638         for( int x = 0; x < width-1; x += 4 )
 639             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 640     }
 641     return ssim;
 642 }
 643
 644 int pixel_vsad( pixel *src, int stride, int height )
 645 {
 646     int score = 0;
 647     for( int i = 1; i < height; i++, src += stride )
 648         for( int j = 0; j < 16; j++ )
 649             score += abs(src[j] - src[j+stride]);
 650     return score;
 651 }
 652
 653 int x264_field_vsad( x264_t *h, int mb_x, int mb_y )
 654 {
 655     int score_field, score_frame;
 656     int stride = h->fenc->i_stride[0];
 657     int mb_stride = h->mb.i_mb_stride;
 658     pixel *fenc = h->fenc->plane[0] + 16 * (mb_x + mb_y * stride);
 659     int mb_xy = mb_x + mb_y*mb_stride;
 660
 661     /* We don't want to analyze pixels outside the frame, as it gives inaccurate results. */
 662     int mbpair_height = X264_MIN( h->param.i_height - mb_y * 16, 32 );
 663     score_frame  = h->pixf.vsad( fenc,          stride, mbpair_height );
 664     score_field  = h->pixf.vsad( fenc,        stride*2, mbpair_height >> 1 );
 665     score_field += h->pixf.vsad( fenc+stride, stride*2, mbpair_height >> 1 );
 666
 667     if( mb_x > 0 )
 668         score_field += 512 - h->mb.field[mb_xy        -1]*1024;
 669     if( mb_y > 0 )
 670         score_field += 512 - h->mb.field[mb_xy-mb_stride]*1024;
 671
 672     return (score_field < score_frame);
 673 }
 674
 675 /****************************************************************************
 676  * successive elimination
 677  ****************************************************************************/
 678 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 679                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 680 {
 681     int nmv = 0;
 682     for( int i = 0; i < width; i++, sums++ )
 683     {
 684         int ads = abs( enc_dc[0] - sums[0] )
 685                 + abs( enc_dc[1] - sums[8] )
 686                 + abs( enc_dc[2] - sums[delta] )
 687                 + abs( enc_dc[3] - sums[delta+8] )
 688                 + cost_mvx[i];
 689         if( ads < thresh )
 690             mvs[nmv++] = i;
 691     }
 692     return nmv;
 693 }
 694
 695 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 696                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 697 {
 698     int nmv = 0;
 699     for( int i = 0; i < width; i++, sums++ )
 700     {
 701         int ads = abs( enc_dc[0] - sums[0] )
 702                 + abs( enc_dc[1] - sums[delta] )
 703                 + cost_mvx[i];
 704         if( ads < thresh )
 705             mvs[nmv++] = i;
 706     }
 707     return nmv;
 708 }
 709
 710 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 711                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 712 {
 713     int nmv = 0;
 714     for( int i = 0; i<width; i++, sums++ )
 715     {
 716         int ads = abs( enc_dc[0] - sums[0] )
 717                 + cost_mvx[i];
 718         if( ads < thresh )
 719             mvs[nmv++] = i;
 720     }
 721     return nmv;
 722 }
 723
 724
 725 /****************************************************************************
 726  * x264_pixel_init:
 727  ****************************************************************************/
 728 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 729 {
 730     memset( pixf, 0, sizeof(*pixf) );
 731
 732 #define INIT2_NAME( name1, name2, cpu ) \
 733     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 734     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 735 #define INIT4_NAME( name1, name2, cpu ) \
 736     INIT2_NAME( name1, name2, cpu ) \
 737     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 738     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 739 #define INIT5_NAME( name1, name2, cpu ) \
 740     INIT4_NAME( name1, name2, cpu ) \
 741     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 742 #define INIT6_NAME( name1, name2, cpu ) \
 743     INIT5_NAME( name1, name2, cpu ) \
 744     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 745 #define INIT7_NAME( name1, name2, cpu ) \
 746     INIT6_NAME( name1, name2, cpu ) \
 747     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 748 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 749 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 750 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 751 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 752 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 753
 754 #define INIT_ADS( cpu ) \
 755     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 756     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 757     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 758
 759     INIT7( sad, );
 760     INIT7_NAME( sad_aligned, sad, );
 761     INIT7( sad_x3, );
 762     INIT7( sad_x4, );
 763     INIT7( ssd, );
 764     INIT7( satd, );
 765     INIT7( satd_x3, );
 766     INIT7( satd_x4, );
 767     INIT4( hadamard_ac, );
 768     INIT_ADS( );
 769
 770     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 771     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 772     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 773     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 774
 775     pixf->ssd_nv12_core = pixel_ssd_nv12_core;
 776     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 777     pixf->ssim_end4 = ssim_end4;
 778     pixf->var2_8x8 = pixel_var2_8x8;
 779     pixf->vsad = pixel_vsad;
 780
 781     pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4;
 782     pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4;
 783     pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8;
 784     pixf->intra_sa8d_x3_8x8   = x264_intra_sa8d_x3_8x8;
 785     pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c;
 786     pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c;
 787     pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16;
 788     pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16;
 789
 790 #if HIGH_BIT_DEPTH
 791 #if HAVE_MMX
 792     if( cpu&X264_CPU_MMXEXT )
 793     {
 794         INIT7( sad, _mmxext );
 795         INIT7( sad_x3, _mmxext );
 796         INIT7( sad_x4, _mmxext );
 797         INIT7( satd, _mmxext );
 798         INIT7( satd_x3, _mmxext );
 799         INIT7( satd_x4, _mmxext );
 800         INIT4( hadamard_ac, _mmxext );
 801         INIT7( ssd, _mmxext );
 802         INIT_ADS( _mmxext );
 803
 804         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_mmxext;
 805         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
 806         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
 807         pixf->var2_8x8 = x264_pixel_var2_8x8_mmxext;
 808
 809         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmxext;
 810         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
 811         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmxext;
 812         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmxext;
 813         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
 814         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmxext;
 815         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
 816     }
 817     if( cpu&X264_CPU_SSE2 )
 818     {
 819         INIT4_NAME( sad_aligned, sad, _sse2_aligned );
 820         INIT5( ssd, _sse2 );
 821
 822         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 823         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 824 #if ARCH_X86_64
 825         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 826 #endif
 827         pixf->ssd_nv12_core = x264_pixel_ssd_nv12_core_sse2;
 828         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 829         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 830         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 831         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
 832         pixf->var2_8x8 = x264_pixel_var2_8x8_sse2;
 833     }
 834     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 835     {
 836         INIT5( sad, _sse2 );
 837         INIT2( sad_x3, _sse2 );
 838         INIT2( sad_x4, _sse2 );
 839         INIT_ADS( _sse2 );
 840
 841         if( !(cpu&X264_CPU_STACK_MOD4) )
 842         {
 843             INIT4( hadamard_ac, _sse2 );
 844         }
 845
 846         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_sse2;
 847         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_sse2;
 848         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_sse2;
 849     }
 850     if( cpu&X264_CPU_SSE2_IS_FAST )
 851     {
 852         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 853         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 854         pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_sse2;
 855         pixf->sad_x3[PIXEL_8x4]  = x264_pixel_sad_x3_8x4_sse2;
 856         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 857         pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_sse2;
 858         pixf->sad_x4[PIXEL_8x4]  = x264_pixel_sad_x4_8x4_sse2;
 859     }
 860     if( cpu&X264_CPU_SSSE3 )
 861     {
 862         INIT4_NAME( sad_aligned, sad, _ssse3_aligned );
 863         INIT7( sad, _ssse3 );
 864         INIT7( sad_x3, _ssse3 );
 865         INIT7( sad_x4, _ssse3 );
 866         INIT_ADS( _ssse3 );
 867
 868         if( !(cpu&X264_CPU_STACK_MOD4) )
 869         {
 870             INIT4( hadamard_ac, _ssse3 );
 871         }
 872
 873         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 874         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 875         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_ssse3;
 876         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 877         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 878     }
 879 #endif // HAVE_MMX
 880 #else // !HIGH_BIT_DEPTH
 881 #if HAVE_MMX
 882     if( cpu&X264_CPU_MMX )
 883     {
 884         INIT7( ssd, _mmx );
 885     }
 886
 887     if( cpu&X264_CPU_MMXEXT )
 888     {
 889         INIT7( sad, _mmxext );
 890         INIT7_NAME( sad_aligned, sad, _mmxext );
 891         INIT7( sad_x3, _mmxext );
 892         INIT7( sad_x4, _mmxext );
 893         INIT7( satd, _mmxext );
 894         INIT7( satd_x3, _mmxext );
 895         INIT7( satd_x4, _mmxext );
 896         INIT4( hadamard_ac, _mmxext );
 897         INIT_ADS( _mmxext );
 898         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
 899         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
 900         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_mmxext;
 901 #if ARCH_X86
 902         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
 903         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmxext;
 904         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
 905         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_mmxext;
 906         pixf->var2_8x8 = x264_pixel_var2_8x8_mmxext;
 907         pixf->vsad = x264_pixel_vsad_mmxext;
 908
 909         if( cpu&X264_CPU_CACHELINE_32 )
 910         {
 911             INIT5( sad, _cache32_mmxext );
 912             INIT4( sad_x3, _cache32_mmxext );
 913             INIT4( sad_x4, _cache32_mmxext );
 914         }
 915         else if( cpu&X264_CPU_CACHELINE_64 )
 916         {
 917             INIT5( sad, _cache64_mmxext );
 918             INIT4( sad_x3, _cache64_mmxext );
 919             INIT4( sad_x4, _cache64_mmxext );
 920         }
 921 #else
 922         if( cpu&X264_CPU_CACHELINE_64 )
 923         {
 924             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
 925             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmxext;
 926             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmxext;
 927             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
 928             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmxext;
 929             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
 930             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmxext;
 931         }
 932 #endif
 933         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
 934         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmxext;
 935         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
 936         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmxext;
 937         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmxext;
 938         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
 939         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmxext;
 940     }
 941
 942     if( cpu&X264_CPU_SSE2 )
 943     {
 944         INIT5( ssd, _sse2slow );
 945         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
 946         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 947         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_sse2;
 948         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 949         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 950         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 951         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 952 #if ARCH_X86_64
 953         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 954 #endif
 955         pixf->var2_8x8 = x264_pixel_var2_8x8_sse2;
 956         pixf->vsad = x264_pixel_vsad_sse2;
 957     }
 958
 959     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 960     {
 961         INIT2( sad, _sse2 );
 962         INIT2( sad_x3, _sse2 );
 963         INIT2( sad_x4, _sse2 );
 964         INIT6( satd, _sse2 );
 965         INIT6( satd_x3, _sse2 );
 966         INIT6( satd_x4, _sse2 );
 967         if( !(cpu&X264_CPU_STACK_MOD4) )
 968         {
 969             INIT4( hadamard_ac, _sse2 );
 970         }
 971         INIT_ADS( _sse2 );
 972         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
 973         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
 974         if( cpu&X264_CPU_CACHELINE_64 )
 975         {
 976             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
 977 #if ARCH_X86
 978             INIT2( sad, _cache64_sse2 );
 979             INIT2( sad_x3, _cache64_sse2 );
 980             INIT2( sad_x4, _cache64_sse2 );
 981 #endif
 982            if( cpu&X264_CPU_SSE2_IS_FAST )
 983            {
 984                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
 985                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
 986            }
 987         }
 988
 989         if( cpu&X264_CPU_SSE_MISALIGN )
 990         {
 991             INIT2( sad_x3, _sse2_misalign );
 992             INIT2( sad_x4, _sse2_misalign );
 993         }
 994     }
 995
 996     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
 997     {
 998         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 999         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
1000         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
1001         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
1002         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
1003         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
1004         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
1005         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
1006     }
1007
1008     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
1009     {
1010         INIT2( sad, _sse3 );
1011         INIT2( sad_x3, _sse3 );
1012         INIT2( sad_x4, _sse3 );
1013     }
1014
1015     if( cpu&X264_CPU_SSSE3 )
1016     {
1017         if( !(cpu&X264_CPU_STACK_MOD4) )
1018         {
1019             INIT4( hadamard_ac, _ssse3 );
1020         }
1021         INIT_ADS( _ssse3 );
1022         if( !(cpu&X264_CPU_SLOW_ATOM) )
1023         {
1024             INIT7( ssd, _ssse3 );
1025             pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
1026             pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
1027             INIT7( satd, _ssse3 );
1028             INIT7( satd_x3, _ssse3 );
1029             INIT7( satd_x4, _ssse3 );
1030         }
1031         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
1032         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
1033         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
1034         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
1035         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
1036 #if ARCH_X86_64
1037         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
1038 #endif
1039         pixf->var2_8x8 = x264_pixel_var2_8x8_ssse3;
1040         if( cpu&X264_CPU_CACHELINE_64 )
1041         {
1042             INIT2( sad, _cache64_ssse3 );
1043             INIT2( sad_x3, _cache64_ssse3 );
1044             INIT2( sad_x4, _cache64_ssse3 );
1045         }
1046         if( cpu&X264_CPU_SLOW_ATOM || !(cpu&X264_CPU_SHUFFLE_IS_FAST) )
1047         {
1048             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
1049         }
1050     }
1051
1052     if( cpu&X264_CPU_SSE4 )
1053     {
1054         INIT7( satd, _sse4 );
1055         INIT7( satd_x3, _sse4 );
1056         INIT7( satd_x4, _sse4 );
1057         if( !(cpu&X264_CPU_STACK_MOD4) )
1058         {
1059             INIT4( hadamard_ac, _sse4 );
1060         }
1061         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
1062         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
1063         pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_sse4;
1064         /* Slower on Conroe, so only enable under SSE4 */
1065         pixf->intra_sad_x3_8x8  = x264_intra_sad_x3_8x8_ssse3;
1066     }
1067
1068     if( cpu&X264_CPU_AVX )
1069     {
1070         INIT7( satd, _avx );
1071         INIT7( satd_x3, _avx );
1072         INIT7( satd_x4, _avx );
1073         pixf->ads[PIXEL_16x16] = x264_pixel_ads4_avx;
1074         pixf->ads[PIXEL_16x8]  = x264_pixel_ads2_avx;
1075         if( !(cpu&X264_CPU_STACK_MOD4) )
1076         {
1077             INIT4( hadamard_ac, _avx );
1078         }
1079         INIT5( ssd, _avx );
1080 #if ARCH_X86_64
1081         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_avx;
1082         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_avx;
1083         pixf->intra_sa8d_x3_8x8= x264_intra_sa8d_x3_8x8_avx;
1084 #endif
1085         pixf->ssd_nv12_core    = x264_pixel_ssd_nv12_core_avx;
1086         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_avx;
1087         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_avx;
1088         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_avx;
1089         pixf->ssim_end4        = x264_pixel_ssim_end4_avx;
1090         pixf->intra_sad_x3_4x4 = x264_intra_sad_x3_4x4_avx;
1091         pixf->intra_sad_x3_8x8 = x264_intra_sad_x3_8x8_avx;
1092     }
1093 #endif //HAVE_MMX
1094
1095 #if HAVE_ARMV6
1096     if( cpu&X264_CPU_ARMV6 )
1097     {
1098         pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1099         pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1100         pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_4x8_armv6;
1101         pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_4x4_armv6;
1102     }
1103     if( cpu&X264_CPU_NEON )
1104     {
1105         INIT5( sad, _neon );
1106         INIT5( sad_aligned, _neon );
1107         INIT7( sad_x3, _neon );
1108         INIT7( sad_x4, _neon );
1109         INIT7( ssd, _neon );
1110         INIT7( satd, _neon );
1111         INIT7( satd_x3, _neon );
1112         INIT7( satd_x4, _neon );
1113         INIT4( hadamard_ac, _neon );
1114         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_neon;
1115         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_neon;
1116         pixf->var[PIXEL_8x8]    = x264_pixel_var_8x8_neon;
1117         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
1118         pixf->var2_8x8          = x264_pixel_var2_8x8_neon;
1119
1120         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
1121         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
1122
1123         if( cpu&X264_CPU_FAST_NEON_MRC )
1124         {
1125             pixf->sad[PIXEL_4x8] = x264_pixel_sad_4x8_neon;
1126             pixf->sad[PIXEL_4x4] = x264_pixel_sad_4x4_neon;
1127             pixf->sad_aligned[PIXEL_4x8] = x264_pixel_sad_aligned_4x8_neon;
1128             pixf->sad_aligned[PIXEL_4x4] = x264_pixel_sad_aligned_4x4_neon;
1129         }
1130         else    // really just scheduled for dual issue / A8
1131         {
1132             INIT5( sad_aligned, _neon_dual );
1133         }
1134     }
1135 #endif
1136 #endif // HIGH_BIT_DEPTH
1137 #if HAVE_ALTIVEC
1138     if( cpu&X264_CPU_ALTIVEC )
1139     {
1140         x264_pixel_altivec_init( pixf );
1141     }
1142 #endif
1143 #if !HIGH_BIT_DEPTH
1144 #if ARCH_UltraSPARC
1145     INIT4( sad, _vis );
1146     INIT4( sad_x3, _vis );
1147     INIT4( sad_x4, _vis );
1148 #endif
1149 #endif // !HIGH_BIT_DEPTH
1150
1151     pixf->ads[PIXEL_8x16] =
1152     pixf->ads[PIXEL_8x4] =
1153     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
1154     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
1155 }
1156