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 #ifdef HAVE_MMX
  27 #   include "x86/pixel.h"
  28 #endif
  29 #ifdef ARCH_PPC
  30 #   include "ppc/pixel.h"
  31 #endif
  32 #ifdef ARCH_UltraSparc
  33 #   include "sparc/pixel.h"
  34 #endif
  35
  36
  37 /****************************************************************************
  38  * pixel_sad_WxH
  39  ****************************************************************************/
  40 #define PIXEL_SAD_C( name, lx, ly ) \
  41 static int name( uint8_t *pix1, int i_stride_pix1,  \
  42                  uint8_t *pix2, int i_stride_pix2 ) \
  43 {                                                   \
  44     int i_sum = 0;                                  \
  45     int x, y;                                       \
  46     for( y = 0; y < ly; y++ )                       \
  47     {                                               \
  48         for( x = 0; x < lx; x++ )                   \
  49         {                                           \
  50             i_sum += abs( pix1[x] - pix2[x] );      \
  51         }                                           \
  52         pix1 += i_stride_pix1;                      \
  53         pix2 += i_stride_pix2;                      \
  54     }                                               \
  55     return i_sum;                                   \
  56 }
  57
  58
  59 PIXEL_SAD_C( x264_pixel_sad_16x16, 16, 16 )
  60 PIXEL_SAD_C( x264_pixel_sad_16x8,  16,  8 )
  61 PIXEL_SAD_C( x264_pixel_sad_8x16,   8, 16 )
  62 PIXEL_SAD_C( x264_pixel_sad_8x8,    8,  8 )
  63 PIXEL_SAD_C( x264_pixel_sad_8x4,    8,  4 )
  64 PIXEL_SAD_C( x264_pixel_sad_4x8,    4,  8 )
  65 PIXEL_SAD_C( x264_pixel_sad_4x4,    4,  4 )
  66
  67
  68 /****************************************************************************
  69  * pixel_ssd_WxH
  70  ****************************************************************************/
  71 #define PIXEL_SSD_C( name, lx, ly ) \
  72 static int name( uint8_t *pix1, int i_stride_pix1,  \
  73                  uint8_t *pix2, int i_stride_pix2 ) \
  74 {                                                   \
  75     int i_sum = 0;                                  \
  76     int x, y;                                       \
  77     for( y = 0; y < ly; y++ )                       \
  78     {                                               \
  79         for( x = 0; x < lx; x++ )                   \
  80         {                                           \
  81             int d = pix1[x] - pix2[x];              \
  82             i_sum += d*d;                           \
  83         }                                           \
  84         pix1 += i_stride_pix1;                      \
  85         pix2 += i_stride_pix2;                      \
  86     }                                               \
  87     return i_sum;                                   \
  88 }
  89
  90 PIXEL_SSD_C( x264_pixel_ssd_16x16, 16, 16 )
  91 PIXEL_SSD_C( x264_pixel_ssd_16x8,  16,  8 )
  92 PIXEL_SSD_C( x264_pixel_ssd_8x16,   8, 16 )
  93 PIXEL_SSD_C( x264_pixel_ssd_8x8,    8,  8 )
  94 PIXEL_SSD_C( x264_pixel_ssd_8x4,    8,  4 )
  95 PIXEL_SSD_C( x264_pixel_ssd_4x8,    4,  8 )
  96 PIXEL_SSD_C( x264_pixel_ssd_4x4,    4,  4 )
  97
  98 int64_t x264_pixel_ssd_wxh( x264_pixel_function_t *pf, uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2, int i_width, int i_height )
  99 {
 100     int64_t i_ssd = 0;
 101     int x, y;
 102     int align = !(((intptr_t)pix1 | (intptr_t)pix2 | i_pix1 | i_pix2) & 15);
 103
 104 #define SSD(size) i_ssd += pf->ssd[size]( pix1 + y*i_pix1 + x, i_pix1, \
 105                                           pix2 + y*i_pix2 + x, i_pix2 );
 106     for( y = 0; y < i_height-15; y += 16 )
 107     {
 108         x = 0;
 109         if( align )
 110             for( ; x < i_width-15; x += 16 )
 111                 SSD(PIXEL_16x16);
 112         for( ; x < i_width-7; x += 8 )
 113             SSD(PIXEL_8x16);
 114     }
 115     if( y < i_height-7 )
 116         for( x = 0; x < i_width-7; x += 8 )
 117             SSD(PIXEL_8x8);
 118 #undef SSD
 119
 120 #define SSD1 { int d = pix1[y*i_pix1+x] - pix2[y*i_pix2+x]; i_ssd += d*d; }
 121     if( i_width % 8 != 0 )
 122     {
 123         for( y = 0; y < (i_height & ~7); y++ )
 124             for( x = i_width & ~7; x < i_width; x++ )
 125                 SSD1;
 126     }
 127     if( i_height % 8 != 0 )
 128     {
 129         for( y = i_height & ~7; y < i_height; y++ )
 130             for( x = 0; x < i_width; x++ )
 131                 SSD1;
 132     }
 133 #undef SSD1
 134
 135     return i_ssd;
 136 }
 137
 138
 139 /****************************************************************************
 140  * pixel_var_wxh
 141  ****************************************************************************/
 142 #define PIXEL_VAR_C( name, w, shift ) \
 143 static int name( uint8_t *pix, int i_stride ) \
 144 {                                             \
 145     uint32_t var = 0, sum = 0, sqr = 0;       \
 146     int x, y;                                 \
 147     for( y = 0; y < w; y++ )                  \
 148     {                                         \
 149         for( x = 0; x < w; x++ )              \
 150         {                                     \
 151             sum += pix[x];                    \
 152             sqr += pix[x] * pix[x];           \
 153         }                                     \
 154         pix += i_stride;                      \
 155     }                                         \
 156     var = sqr - (sum * sum >> shift);         \
 157     return var;                               \
 158 }
 159
 160 PIXEL_VAR_C( x264_pixel_var_16x16, 16, 8 )
 161 PIXEL_VAR_C( x264_pixel_var_8x8,    8, 6 )
 162
 163 /****************************************************************************
 164  * pixel_var2_wxh
 165  ****************************************************************************/
 166 static int pixel_var2_8x8( uint8_t *pix1, int i_stride1, uint8_t *pix2, int i_stride2, int *ssd )
 167 {
 168     uint32_t var = 0, sum = 0, sqr = 0;
 169     int x, y;
 170     for( y = 0; y < 8; y++ )
 171     {
 172         for( x = 0; x < 8; x++ )
 173         {
 174             int diff = pix1[x] - pix2[x];
 175             sum += diff;
 176             sqr += diff * diff;
 177         }
 178         pix1 += i_stride1;
 179         pix2 += i_stride2;
 180     }
 181     sum = abs(sum);
 182     var = sqr - (sum * sum >> 6);
 183     *ssd = sqr;
 184     return var;
 185 }
 186
 187
 188 #define HADAMARD4(d0,d1,d2,d3,s0,s1,s2,s3) {\
 189     int t0 = s0 + s1;\
 190     int t1 = s0 - s1;\
 191     int t2 = s2 + s3;\
 192     int t3 = s2 - s3;\
 193     d0 = t0 + t2;\
 194     d2 = t0 - t2;\
 195     d1 = t1 + t3;\
 196     d3 = t1 - t3;\
 197 }
 198
 199 // in: a pseudo-simd number of the form x+(y<<16)
 200 // return: abs(x)+(abs(y)<<16)
 201 static ALWAYS_INLINE uint32_t abs2( uint32_t a )
 202 {
 203     uint32_t s = ((a>>15)&0x10001)*0xffff;
 204     return (a+s)^s;
 205 }
 206
 207 /****************************************************************************
 208  * pixel_satd_WxH: sum of 4x4 Hadamard transformed differences
 209  ****************************************************************************/
 210
 211 static NOINLINE int x264_pixel_satd_4x4( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
 212 {
 213     uint32_t tmp[4][2];
 214     uint32_t a0,a1,a2,a3,b0,b1;
 215     int sum=0, i;
 216     for( i=0; i<4; i++, pix1+=i_pix1, pix2+=i_pix2 )
 217     {
 218         a0 = pix1[0] - pix2[0];
 219         a1 = pix1[1] - pix2[1];
 220         b0 = (a0+a1) + ((a0-a1)<<16);
 221         a2 = pix1[2] - pix2[2];
 222         a3 = pix1[3] - pix2[3];
 223         b1 = (a2+a3) + ((a2-a3)<<16);
 224         tmp[i][0] = b0 + b1;
 225         tmp[i][1] = b0 - b1;
 226     }
 227     for( i=0; i<2; i++ )
 228     {
 229         HADAMARD4( a0,a1,a2,a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 230         a0 = abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 231         sum += ((uint16_t)a0) + (a0>>16);
 232     }
 233     return sum >> 1;
 234 }
 235
 236 static NOINLINE int x264_pixel_satd_8x4( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
 237 {
 238     uint32_t tmp[4][4];
 239     uint32_t a0,a1,a2,a3;
 240     int sum=0, i;
 241     for( i=0; i<4; i++, pix1+=i_pix1, pix2+=i_pix2 )
 242     {
 243         a0 = (pix1[0] - pix2[0]) + ((pix1[4] - pix2[4]) << 16);
 244         a1 = (pix1[1] - pix2[1]) + ((pix1[5] - pix2[5]) << 16);
 245         a2 = (pix1[2] - pix2[2]) + ((pix1[6] - pix2[6]) << 16);
 246         a3 = (pix1[3] - pix2[3]) + ((pix1[7] - pix2[7]) << 16);
 247         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], a0,a1,a2,a3 );
 248     }
 249     for( i=0; i<4; i++ )
 250     {
 251         HADAMARD4( a0,a1,a2,a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 252         sum += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 253     }
 254     return (((uint16_t)sum) + ((uint32_t)sum>>16)) >> 1;
 255 }
 256
 257 #define PIXEL_SATD_C( w, h, sub )\
 258 static int x264_pixel_satd_##w##x##h( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )\
 259 {\
 260     int sum = sub( pix1, i_pix1, pix2, i_pix2 )\
 261             + sub( pix1+4*i_pix1, i_pix1, pix2+4*i_pix2, i_pix2 );\
 262     if( w==16 )\
 263         sum+= sub( pix1+8, i_pix1, pix2+8, i_pix2 )\
 264             + sub( pix1+8+4*i_pix1, i_pix1, pix2+8+4*i_pix2, i_pix2 );\
 265     if( h==16 )\
 266         sum+= sub( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )\
 267             + sub( pix1+12*i_pix1, i_pix1, pix2+12*i_pix2, i_pix2 );\
 268     if( w==16 && h==16 )\
 269         sum+= sub( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 )\
 270             + sub( pix1+8+12*i_pix1, i_pix1, pix2+8+12*i_pix2, i_pix2 );\
 271     return sum;\
 272 }
 273 PIXEL_SATD_C( 16, 16, x264_pixel_satd_8x4 )
 274 PIXEL_SATD_C( 16, 8,  x264_pixel_satd_8x4 )
 275 PIXEL_SATD_C( 8,  16, x264_pixel_satd_8x4 )
 276 PIXEL_SATD_C( 8,  8,  x264_pixel_satd_8x4 )
 277 PIXEL_SATD_C( 4,  8,  x264_pixel_satd_4x4 )
 278
 279
 280 static NOINLINE int sa8d_8x8( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
 281 {
 282     uint32_t tmp[8][4];
 283     uint32_t a0,a1,a2,a3,a4,a5,a6,a7,b0,b1,b2,b3;
 284     int sum=0, i;
 285     for( i=0; i<8; i++, pix1+=i_pix1, pix2+=i_pix2 )
 286     {
 287         a0 = pix1[0] - pix2[0];
 288         a1 = pix1[1] - pix2[1];
 289         b0 = (a0+a1) + ((a0-a1)<<16);
 290         a2 = pix1[2] - pix2[2];
 291         a3 = pix1[3] - pix2[3];
 292         b1 = (a2+a3) + ((a2-a3)<<16);
 293         a4 = pix1[4] - pix2[4];
 294         a5 = pix1[5] - pix2[5];
 295         b2 = (a4+a5) + ((a4-a5)<<16);
 296         a6 = pix1[6] - pix2[6];
 297         a7 = pix1[7] - pix2[7];
 298         b3 = (a6+a7) + ((a6-a7)<<16);
 299         HADAMARD4( tmp[i][0], tmp[i][1], tmp[i][2], tmp[i][3], b0,b1,b2,b3 );
 300     }
 301     for( i=0; i<4; i++ )
 302     {
 303         HADAMARD4( a0,a1,a2,a3, tmp[0][i], tmp[1][i], tmp[2][i], tmp[3][i] );
 304         HADAMARD4( a4,a5,a6,a7, tmp[4][i], tmp[5][i], tmp[6][i], tmp[7][i] );
 305         b0  = abs2(a0+a4) + abs2(a0-a4);
 306         b0 += abs2(a1+a5) + abs2(a1-a5);
 307         b0 += abs2(a2+a6) + abs2(a2-a6);
 308         b0 += abs2(a3+a7) + abs2(a3-a7);
 309         sum += (uint16_t)b0 + (b0>>16);
 310     }
 311     return sum;
 312 }
 313
 314 static int x264_pixel_sa8d_8x8( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
 315 {
 316     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 );
 317     return (sum+2)>>2;
 318 }
 319
 320 static int x264_pixel_sa8d_16x16( uint8_t *pix1, int i_pix1, uint8_t *pix2, int i_pix2 )
 321 {
 322     int sum = sa8d_8x8( pix1, i_pix1, pix2, i_pix2 )
 323             + sa8d_8x8( pix1+8, i_pix1, pix2+8, i_pix2 )
 324             + sa8d_8x8( pix1+8*i_pix1, i_pix1, pix2+8*i_pix2, i_pix2 )
 325             + sa8d_8x8( pix1+8+8*i_pix1, i_pix1, pix2+8+8*i_pix2, i_pix2 );
 326     return (sum+2)>>2;
 327 }
 328
 329
 330 static NOINLINE uint64_t pixel_hadamard_ac( uint8_t *pix, int stride )
 331 {
 332     uint32_t tmp[32];
 333     uint32_t a0,a1,a2,a3,dc;
 334     int sum4=0, sum8=0, i;
 335     for( i=0; i<8; i++, pix+=stride )
 336     {
 337         uint32_t *t = tmp + (i&3) + (i&4)*4;
 338         a0 = (pix[0]+pix[1]) + ((pix[0]-pix[1])<<16);
 339         a1 = (pix[2]+pix[3]) + ((pix[2]-pix[3])<<16);
 340         t[0] = a0 + a1;
 341         t[4] = a0 - a1;
 342         a2 = (pix[4]+pix[5]) + ((pix[4]-pix[5])<<16);
 343         a3 = (pix[6]+pix[7]) + ((pix[6]-pix[7])<<16);
 344         t[8] = a2 + a3;
 345         t[12] = a2 - a3;
 346     }
 347     for( i=0; i<8; i++ )
 348     {
 349         HADAMARD4( a0,a1,a2,a3, tmp[i*4+0], tmp[i*4+1], tmp[i*4+2], tmp[i*4+3] );
 350         tmp[i*4+0] = a0;
 351         tmp[i*4+1] = a1;
 352         tmp[i*4+2] = a2;
 353         tmp[i*4+3] = a3;
 354         sum4 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 355     }
 356     for( i=0; i<8; i++ )
 357     {
 358         HADAMARD4( a0,a1,a2,a3, tmp[i], tmp[8+i], tmp[16+i], tmp[24+i] );
 359         sum8 += abs2(a0) + abs2(a1) + abs2(a2) + abs2(a3);
 360     }
 361     dc = (uint16_t)(tmp[0] + tmp[8] + tmp[16] + tmp[24]);
 362     sum4 = (uint16_t)sum4 + ((uint32_t)sum4>>16) - dc;
 363     sum8 = (uint16_t)sum8 + ((uint32_t)sum8>>16) - dc;
 364     return ((uint64_t)sum8<<32) + sum4;
 365 }
 366
 367 #define HADAMARD_AC(w,h) \
 368 static uint64_t x264_pixel_hadamard_ac_##w##x##h( uint8_t *pix, int stride )\
 369 {\
 370     uint64_t sum = pixel_hadamard_ac( pix, stride );\
 371     if( w==16 )\
 372         sum += pixel_hadamard_ac( pix+8, stride );\
 373     if( h==16 )\
 374         sum += pixel_hadamard_ac( pix+8*stride, stride );\
 375     if( w==16 && h==16 )\
 376         sum += pixel_hadamard_ac( pix+8*stride+8, stride );\
 377     return ((sum>>34)<<32) + ((uint32_t)sum>>1);\
 378 }
 379 HADAMARD_AC( 16, 16 )
 380 HADAMARD_AC( 16, 8 )
 381 HADAMARD_AC( 8, 16 )
 382 HADAMARD_AC( 8, 8 )
 383
 384
 385 /****************************************************************************
 386  * pixel_sad_x4
 387  ****************************************************************************/
 388 #define SAD_X( size ) \
 389 static void x264_pixel_sad_x3_##size( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, int i_stride, int scores[3] )\
 390 {\
 391     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 392     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 393     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 394 }\
 395 static void x264_pixel_sad_x4_##size( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, uint8_t *pix3, int i_stride, int scores[4] )\
 396 {\
 397     scores[0] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix0, i_stride );\
 398     scores[1] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix1, i_stride );\
 399     scores[2] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix2, i_stride );\
 400     scores[3] = x264_pixel_sad_##size( fenc, FENC_STRIDE, pix3, i_stride );\
 401 }
 402
 403 SAD_X( 16x16 )
 404 SAD_X( 16x8 )
 405 SAD_X( 8x16 )
 406 SAD_X( 8x8 )
 407 SAD_X( 8x4 )
 408 SAD_X( 4x8 )
 409 SAD_X( 4x4 )
 410
 411 #ifdef ARCH_UltraSparc
 412 SAD_X( 16x16_vis )
 413 SAD_X( 16x8_vis )
 414 SAD_X( 8x16_vis )
 415 SAD_X( 8x8_vis )
 416 #endif
 417
 418 /****************************************************************************
 419  * pixel_satd_x4
 420  * no faster than single satd, but needed for satd to be a drop-in replacement for sad
 421  ****************************************************************************/
 422
 423 #define SATD_X( size, cpu ) \
 424 static void x264_pixel_satd_x3_##size##cpu( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, int i_stride, int scores[3] )\
 425 {\
 426     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 427     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 428     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 429 }\
 430 static void x264_pixel_satd_x4_##size##cpu( uint8_t *fenc, uint8_t *pix0, uint8_t *pix1, uint8_t *pix2, uint8_t *pix3, int i_stride, int scores[4] )\
 431 {\
 432     scores[0] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix0, i_stride );\
 433     scores[1] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix1, i_stride );\
 434     scores[2] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix2, i_stride );\
 435     scores[3] = x264_pixel_satd_##size##cpu( fenc, FENC_STRIDE, pix3, i_stride );\
 436 }
 437 #define SATD_X_DECL6( cpu )\
 438 SATD_X( 16x16, cpu )\
 439 SATD_X( 16x8, cpu )\
 440 SATD_X( 8x16, cpu )\
 441 SATD_X( 8x8, cpu )\
 442 SATD_X( 8x4, cpu )\
 443 SATD_X( 4x8, cpu )
 444 #define SATD_X_DECL7( cpu )\
 445 SATD_X_DECL6( cpu )\
 446 SATD_X( 4x4, cpu )
 447
 448 SATD_X_DECL7()
 449 #ifdef HAVE_MMX
 450 SATD_X_DECL7( _mmxext )
 451 SATD_X_DECL6( _sse2 )
 452 SATD_X_DECL7( _ssse3 )
 453 SATD_X_DECL7( _sse4 )
 454 #endif
 455
 456 /****************************************************************************
 457  * structural similarity metric
 458  ****************************************************************************/
 459 static void ssim_4x4x2_core( const uint8_t *pix1, int stride1,
 460                              const uint8_t *pix2, int stride2,
 461                              int sums[2][4])
 462 {
 463     int x, y, z;
 464     for(z=0; z<2; z++)
 465     {
 466         uint32_t s1=0, s2=0, ss=0, s12=0;
 467         for(y=0; y<4; y++)
 468             for(x=0; x<4; x++)
 469             {
 470                 int a = pix1[x+y*stride1];
 471                 int b = pix2[x+y*stride2];
 472                 s1  += a;
 473                 s2  += b;
 474                 ss  += a*a;
 475                 ss  += b*b;
 476                 s12 += a*b;
 477             }
 478         sums[z][0] = s1;
 479         sums[z][1] = s2;
 480         sums[z][2] = ss;
 481         sums[z][3] = s12;
 482         pix1 += 4;
 483         pix2 += 4;
 484     }
 485 }
 486
 487 static float ssim_end1( int s1, int s2, int ss, int s12 )
 488 {
 489     static const int ssim_c1 = (int)(.01*.01*255*255*64 + .5);
 490     static const int ssim_c2 = (int)(.03*.03*255*255*64*63 + .5);
 491     int vars = ss*64 - s1*s1 - s2*s2;
 492     int covar = s12*64 - s1*s2;
 493     return (float)(2*s1*s2 + ssim_c1) * (float)(2*covar + ssim_c2)\
 494            / ((float)(s1*s1 + s2*s2 + ssim_c1) * (float)(vars + ssim_c2));
 495 }
 496
 497 static float ssim_end4( int sum0[5][4], int sum1[5][4], int width )
 498 {
 499     int i;
 500     float ssim = 0.0;
 501     for( i = 0; i < width; i++ )
 502         ssim += ssim_end1( sum0[i][0] + sum0[i+1][0] + sum1[i][0] + sum1[i+1][0],
 503                            sum0[i][1] + sum0[i+1][1] + sum1[i][1] + sum1[i+1][1],
 504                            sum0[i][2] + sum0[i+1][2] + sum1[i][2] + sum1[i+1][2],
 505                            sum0[i][3] + sum0[i+1][3] + sum1[i][3] + sum1[i+1][3] );
 506     return ssim;
 507 }
 508
 509 float x264_pixel_ssim_wxh( x264_pixel_function_t *pf,
 510                            uint8_t *pix1, int stride1,
 511                            uint8_t *pix2, int stride2,
 512                            int width, int height, void *buf )
 513 {
 514     int x, y, z;
 515     float ssim = 0.0;
 516     int (*sum0)[4] = buf;
 517     int (*sum1)[4] = sum0 + width/4+3;
 518     width >>= 2;
 519     height >>= 2;
 520     z = 0;
 521     for( y = 1; y < height; y++ )
 522     {
 523         for( ; z <= y; z++ )
 524         {
 525             XCHG( void*, sum0, sum1 );
 526             for( x = 0; x < width; x+=2 )
 527                 pf->ssim_4x4x2_core( &pix1[4*(x+z*stride1)], stride1, &pix2[4*(x+z*stride2)], stride2, &sum0[x] );
 528         }
 529         for( x = 0; x < width-1; x += 4 )
 530             ssim += pf->ssim_end4( sum0+x, sum1+x, X264_MIN(4,width-x-1) );
 531     }
 532     return ssim;
 533 }
 534
 535
 536 /****************************************************************************
 537  * successive elimination
 538  ****************************************************************************/
 539 static int x264_pixel_ads4( int enc_dc[4], uint16_t *sums, int delta,
 540                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 541 {
 542     int nmv=0, i;
 543     for( i=0; i<width; i++, sums++ )
 544     {
 545         int ads = abs( enc_dc[0] - sums[0] )
 546                 + abs( enc_dc[1] - sums[8] )
 547                 + abs( enc_dc[2] - sums[delta] )
 548                 + abs( enc_dc[3] - sums[delta+8] )
 549                 + cost_mvx[i];
 550         if( ads < thresh )
 551             mvs[nmv++] = i;
 552     }
 553     return nmv;
 554 }
 555
 556 static int x264_pixel_ads2( int enc_dc[2], uint16_t *sums, int delta,
 557                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 558 {
 559     int nmv=0, i;
 560     for( i=0; i<width; i++, sums++ )
 561     {
 562         int ads = abs( enc_dc[0] - sums[0] )
 563                 + abs( enc_dc[1] - sums[delta] )
 564                 + cost_mvx[i];
 565         if( ads < thresh )
 566             mvs[nmv++] = i;
 567     }
 568     return nmv;
 569 }
 570
 571 static int x264_pixel_ads1( int enc_dc[1], uint16_t *sums, int delta,
 572                             uint16_t *cost_mvx, int16_t *mvs, int width, int thresh )
 573 {
 574     int nmv=0, i;
 575     for( i=0; i<width; i++, sums++ )
 576     {
 577         int ads = abs( enc_dc[0] - sums[0] )
 578                 + cost_mvx[i];
 579         if( ads < thresh )
 580             mvs[nmv++] = i;
 581     }
 582     return nmv;
 583 }
 584
 585
 586 /****************************************************************************
 587  * x264_pixel_init:
 588  ****************************************************************************/
 589 void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
 590 {
 591     memset( pixf, 0, sizeof(*pixf) );
 592
 593 #define INIT2_NAME( name1, name2, cpu ) \
 594     pixf->name1[PIXEL_16x16] = x264_pixel_##name2##_16x16##cpu;\
 595     pixf->name1[PIXEL_16x8]  = x264_pixel_##name2##_16x8##cpu;
 596 #define INIT4_NAME( name1, name2, cpu ) \
 597     INIT2_NAME( name1, name2, cpu ) \
 598     pixf->name1[PIXEL_8x16]  = x264_pixel_##name2##_8x16##cpu;\
 599     pixf->name1[PIXEL_8x8]   = x264_pixel_##name2##_8x8##cpu;
 600 #define INIT5_NAME( name1, name2, cpu ) \
 601     INIT4_NAME( name1, name2, cpu ) \
 602     pixf->name1[PIXEL_8x4]   = x264_pixel_##name2##_8x4##cpu;
 603 #define INIT6_NAME( name1, name2, cpu ) \
 604     INIT5_NAME( name1, name2, cpu ) \
 605     pixf->name1[PIXEL_4x8]   = x264_pixel_##name2##_4x8##cpu;
 606 #define INIT7_NAME( name1, name2, cpu ) \
 607     INIT6_NAME( name1, name2, cpu ) \
 608     pixf->name1[PIXEL_4x4]   = x264_pixel_##name2##_4x4##cpu;
 609 #define INIT2( name, cpu ) INIT2_NAME( name, name, cpu )
 610 #define INIT4( name, cpu ) INIT4_NAME( name, name, cpu )
 611 #define INIT5( name, cpu ) INIT5_NAME( name, name, cpu )
 612 #define INIT6( name, cpu ) INIT6_NAME( name, name, cpu )
 613 #define INIT7( name, cpu ) INIT7_NAME( name, name, cpu )
 614
 615 #define INIT_ADS( cpu ) \
 616     pixf->ads[PIXEL_16x16] = x264_pixel_ads4##cpu;\
 617     pixf->ads[PIXEL_16x8] = x264_pixel_ads2##cpu;\
 618     pixf->ads[PIXEL_8x8] = x264_pixel_ads1##cpu;
 619
 620     INIT7( sad, );
 621     INIT7_NAME( sad_aligned, sad, );
 622     INIT7( sad_x3, );
 623     INIT7( sad_x4, );
 624     INIT7( ssd, );
 625     INIT7( satd, );
 626     INIT7( satd_x3, );
 627     INIT7( satd_x4, );
 628     INIT4( hadamard_ac, );
 629     INIT_ADS( );
 630
 631     pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16;
 632     pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8;
 633     pixf->var[PIXEL_16x16] = x264_pixel_var_16x16;
 634     pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8;
 635
 636     pixf->ssim_4x4x2_core = ssim_4x4x2_core;
 637     pixf->ssim_end4 = ssim_end4;
 638     pixf->var2_8x8 = pixel_var2_8x8;
 639
 640 #ifdef HAVE_MMX
 641     if( cpu&X264_CPU_MMX )
 642     {
 643         INIT7( ssd, _mmx );
 644     }
 645
 646     if( cpu&X264_CPU_MMXEXT )
 647     {
 648         INIT7( sad, _mmxext );
 649         INIT7_NAME( sad_aligned, sad, _mmxext );
 650         INIT7( sad_x3, _mmxext );
 651         INIT7( sad_x4, _mmxext );
 652         INIT7( satd, _mmxext );
 653         INIT7( satd_x3, _mmxext );
 654         INIT7( satd_x4, _mmxext );
 655         INIT4( hadamard_ac, _mmxext );
 656         INIT_ADS( _mmxext );
 657         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_mmxext;
 658         pixf->var[PIXEL_8x8]   = x264_pixel_var_8x8_mmxext;
 659 #ifdef ARCH_X86
 660         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_mmxext;
 661         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_mmxext;
 662         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_mmxext;
 663         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_mmxext;
 664         pixf->var2_8x8 = x264_pixel_var2_8x8_mmxext;
 665
 666         if( cpu&X264_CPU_CACHELINE_32 )
 667         {
 668             INIT5( sad, _cache32_mmxext );
 669             INIT4( sad_x3, _cache32_mmxext );
 670             INIT4( sad_x4, _cache32_mmxext );
 671         }
 672         else if( cpu&X264_CPU_CACHELINE_64 )
 673         {
 674             INIT5( sad, _cache64_mmxext );
 675             INIT4( sad_x3, _cache64_mmxext );
 676             INIT4( sad_x4, _cache64_mmxext );
 677         }
 678 #else
 679         if( cpu&X264_CPU_CACHELINE_64 )
 680         {
 681             pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_cache64_mmxext;
 682             pixf->sad[PIXEL_8x8]  = x264_pixel_sad_8x8_cache64_mmxext;
 683             pixf->sad[PIXEL_8x4]  = x264_pixel_sad_8x4_cache64_mmxext;
 684             pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_mmxext;
 685             pixf->sad_x3[PIXEL_8x8]  = x264_pixel_sad_x3_8x8_cache64_mmxext;
 686             pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_mmxext;
 687             pixf->sad_x4[PIXEL_8x8]  = x264_pixel_sad_x4_8x8_cache64_mmxext;
 688         }
 689 #endif
 690         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_mmxext;
 691         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_mmxext;
 692         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_mmxext;
 693         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_mmxext;
 694         pixf->intra_sad_x3_8x8    = x264_intra_sad_x3_8x8_mmxext;
 695         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_mmxext;
 696         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_mmxext;
 697     }
 698
 699     if( cpu&X264_CPU_SSE2 )
 700     {
 701         INIT5( ssd, _sse2slow );
 702         INIT2_NAME( sad_aligned, sad, _sse2_aligned );
 703         pixf->var[PIXEL_16x16] = x264_pixel_var_16x16_sse2;
 704         pixf->ssim_4x4x2_core  = x264_pixel_ssim_4x4x2_core_sse2;
 705         pixf->ssim_end4        = x264_pixel_ssim_end4_sse2;
 706         pixf->sa8d[PIXEL_16x16] = x264_pixel_sa8d_16x16_sse2;
 707         pixf->sa8d[PIXEL_8x8]   = x264_pixel_sa8d_8x8_sse2;
 708 #ifdef ARCH_X86_64
 709         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_sse2;
 710 #endif
 711         pixf->var2_8x8 = x264_pixel_var2_8x8_sse2;
 712     }
 713
 714     if( (cpu&X264_CPU_SSE2) && !(cpu&X264_CPU_SSE2_IS_SLOW) )
 715     {
 716         INIT2( sad, _sse2 );
 717         INIT2( sad_x3, _sse2 );
 718         INIT2( sad_x4, _sse2 );
 719         INIT6( satd, _sse2 );
 720         INIT6( satd_x3, _sse2 );
 721         INIT6( satd_x4, _sse2 );
 722         if( !(cpu&X264_CPU_STACK_MOD4) )
 723         {
 724             INIT4( hadamard_ac, _sse2 );
 725         }
 726         INIT_ADS( _sse2 );
 727         pixf->var[PIXEL_8x8] = x264_pixel_var_8x8_sse2;
 728         pixf->intra_sad_x3_16x16 = x264_intra_sad_x3_16x16_sse2;
 729         if( cpu&X264_CPU_CACHELINE_64 )
 730         {
 731             INIT2( ssd, _sse2); /* faster for width 16 on p4 */
 732 #ifdef ARCH_X86
 733             INIT2( sad, _cache64_sse2 );
 734             INIT2( sad_x3, _cache64_sse2 );
 735             INIT2( sad_x4, _cache64_sse2 );
 736 #endif
 737            if( cpu&X264_CPU_SSE2_IS_FAST )
 738            {
 739                pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_cache64_sse2;
 740                pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_cache64_sse2;
 741            }
 742         }
 743
 744         if( cpu&X264_CPU_SSE_MISALIGN )
 745         {
 746             INIT2( sad_x3, _sse2_misalign );
 747             INIT2( sad_x4, _sse2_misalign );
 748         }
 749     }
 750
 751     if( cpu&X264_CPU_SSE2_IS_FAST && !(cpu&X264_CPU_CACHELINE_64) )
 752     {
 753         pixf->sad_aligned[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 754         pixf->sad[PIXEL_8x16] = x264_pixel_sad_8x16_sse2;
 755         pixf->sad_x3[PIXEL_8x16] = x264_pixel_sad_x3_8x16_sse2;
 756         pixf->sad_x3[PIXEL_8x8] = x264_pixel_sad_x3_8x8_sse2;
 757         pixf->sad_x3[PIXEL_8x4] = x264_pixel_sad_x3_8x4_sse2;
 758         pixf->sad_x4[PIXEL_8x16] = x264_pixel_sad_x4_8x16_sse2;
 759         pixf->sad_x4[PIXEL_8x8] = x264_pixel_sad_x4_8x8_sse2;
 760         pixf->sad_x4[PIXEL_8x4] = x264_pixel_sad_x4_8x4_sse2;
 761     }
 762
 763     if( (cpu&X264_CPU_SSE3) && (cpu&X264_CPU_CACHELINE_64) )
 764     {
 765         INIT2( sad, _sse3 );
 766         INIT2( sad_x3, _sse3 );
 767         INIT2( sad_x4, _sse3 );
 768     }
 769
 770     if( cpu&X264_CPU_SSSE3 )
 771     {
 772         INIT7( ssd, _ssse3 );
 773         INIT7( satd, _ssse3 );
 774         INIT7( satd_x3, _ssse3 );
 775         INIT7( satd_x4, _ssse3 );
 776         if( !(cpu&X264_CPU_STACK_MOD4) )
 777         {
 778             INIT4( hadamard_ac, _ssse3 );
 779         }
 780         INIT_ADS( _ssse3 );
 781         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_ssse3;
 782         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_ssse3;
 783         pixf->intra_satd_x3_16x16 = x264_intra_satd_x3_16x16_ssse3;
 784         pixf->intra_sad_x3_16x16  = x264_intra_sad_x3_16x16_ssse3;
 785         pixf->intra_satd_x3_8x8c  = x264_intra_satd_x3_8x8c_ssse3;
 786         pixf->intra_sad_x3_8x8c   = x264_intra_sad_x3_8x8c_ssse3;
 787         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_ssse3;
 788 #ifdef ARCH_X86_64
 789         pixf->intra_sa8d_x3_8x8 = x264_intra_sa8d_x3_8x8_ssse3;
 790 #endif
 791         pixf->var2_8x8 = x264_pixel_var2_8x8_ssse3;
 792         if( cpu&X264_CPU_CACHELINE_64 )
 793         {
 794             INIT2( sad, _cache64_ssse3 );
 795             INIT2( sad_x3, _cache64_ssse3 );
 796             INIT2( sad_x4, _cache64_ssse3 );
 797         }
 798         if( !(cpu&X264_CPU_SHUFFLE_IS_FAST) )
 799         {
 800             INIT5( ssd, _sse2 ); /* on conroe, sse2 is faster for width8/16 */
 801         }
 802     }
 803
 804     if( cpu&X264_CPU_SSE4 )
 805     {
 806         INIT7( satd, _sse4 );
 807         INIT7( satd_x3, _sse4 );
 808         INIT7( satd_x4, _sse4 );
 809         if( !(cpu&X264_CPU_STACK_MOD4) )
 810         {
 811             INIT4( hadamard_ac, _sse4 );
 812         }
 813         pixf->sa8d[PIXEL_16x16]= x264_pixel_sa8d_16x16_sse4;
 814         pixf->sa8d[PIXEL_8x8]  = x264_pixel_sa8d_8x8_sse4;
 815     }
 816 #endif //HAVE_MMX
 817
 818 #ifdef ARCH_PPC
 819     if( cpu&X264_CPU_ALTIVEC )
 820     {
 821         x264_pixel_altivec_init( pixf );
 822     }
 823 #endif
 824 #ifdef ARCH_UltraSparc
 825     INIT4( sad, _vis );
 826     INIT4( sad_x3, _vis );
 827     INIT4( sad_x4, _vis );
 828 #endif
 829
 830     pixf->ads[PIXEL_8x16] =
 831     pixf->ads[PIXEL_8x4] =
 832     pixf->ads[PIXEL_4x8] = pixf->ads[PIXEL_16x8];
 833     pixf->ads[PIXEL_4x4] = pixf->ads[PIXEL_8x8];
 834 }
 835