git.sesse.net Git - ffmpeg/blob - libavcodec/dsputil.c

   1 /*
   2  * DSP utils
   3  * Copyright (c) 2000, 2001 Fabrice Bellard
   4  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
   5  *
   6  * This file is part of FFmpeg.
   7  *
   8  * FFmpeg is free software; you can redistribute it and/or
   9  * modify it under the terms of the GNU Lesser General Public
  10  * License as published by the Free Software Foundation; either
  11  * version 2.1 of the License, or (at your option) any later version.
  12  *
  13  * FFmpeg is distributed in the hope that it will be useful,
  14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16  * Lesser General Public License for more details.
  17  *
  18  * You should have received a copy of the GNU Lesser General Public
  19  * License along with FFmpeg; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 /**
  24  * @file
  25  * DSP utils
  26  */
  27
  28 #include "libavutil/attributes.h"
  29 #include "libavutil/internal.h"
  30 #include "avcodec.h"
  31 #include "copy_block.h"
  32 #include "dsputil.h"
  33 #include "simple_idct.h"
  34 #include "mpegvideo.h"
  35 #include "config.h"
  36
  37 uint32_t ff_square_tab[512] = { 0, };
  38
  39 static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
  40                   int line_size, int h)
  41 {
  42     int s = 0, i;
  43     uint32_t *sq = ff_square_tab + 256;
  44
  45     for (i = 0; i < h; i++) {
  46         s    += sq[pix1[0] - pix2[0]];
  47         s    += sq[pix1[1] - pix2[1]];
  48         s    += sq[pix1[2] - pix2[2]];
  49         s    += sq[pix1[3] - pix2[3]];
  50         pix1 += line_size;
  51         pix2 += line_size;
  52     }
  53     return s;
  54 }
  55
  56 static int sse8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
  57                   int line_size, int h)
  58 {
  59     int s = 0, i;
  60     uint32_t *sq = ff_square_tab + 256;
  61
  62     for (i = 0; i < h; i++) {
  63         s    += sq[pix1[0] - pix2[0]];
  64         s    += sq[pix1[1] - pix2[1]];
  65         s    += sq[pix1[2] - pix2[2]];
  66         s    += sq[pix1[3] - pix2[3]];
  67         s    += sq[pix1[4] - pix2[4]];
  68         s    += sq[pix1[5] - pix2[5]];
  69         s    += sq[pix1[6] - pix2[6]];
  70         s    += sq[pix1[7] - pix2[7]];
  71         pix1 += line_size;
  72         pix2 += line_size;
  73     }
  74     return s;
  75 }
  76
  77 static int sse16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
  78                    int line_size, int h)
  79 {
  80     int s = 0, i;
  81     uint32_t *sq = ff_square_tab + 256;
  82
  83     for (i = 0; i < h; i++) {
  84         s += sq[pix1[0]  - pix2[0]];
  85         s += sq[pix1[1]  - pix2[1]];
  86         s += sq[pix1[2]  - pix2[2]];
  87         s += sq[pix1[3]  - pix2[3]];
  88         s += sq[pix1[4]  - pix2[4]];
  89         s += sq[pix1[5]  - pix2[5]];
  90         s += sq[pix1[6]  - pix2[6]];
  91         s += sq[pix1[7]  - pix2[7]];
  92         s += sq[pix1[8]  - pix2[8]];
  93         s += sq[pix1[9]  - pix2[9]];
  94         s += sq[pix1[10] - pix2[10]];
  95         s += sq[pix1[11] - pix2[11]];
  96         s += sq[pix1[12] - pix2[12]];
  97         s += sq[pix1[13] - pix2[13]];
  98         s += sq[pix1[14] - pix2[14]];
  99         s += sq[pix1[15] - pix2[15]];
 100
 101         pix1 += line_size;
 102         pix2 += line_size;
 103     }
 104     return s;
 105 }
 106
 107 static int sum_abs_dctelem_c(int16_t *block)
 108 {
 109     int sum = 0, i;
 110
 111     for (i = 0; i < 64; i++)
 112         sum += FFABS(block[i]);
 113     return sum;
 114 }
 115
 116 #define avg2(a, b) ((a + b + 1) >> 1)
 117 #define avg4(a, b, c, d) ((a + b + c + d + 2) >> 2)
 118
 119 static inline int pix_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 120                               int line_size, int h)
 121 {
 122     int s = 0, i;
 123
 124     for (i = 0; i < h; i++) {
 125         s    += abs(pix1[0]  - pix2[0]);
 126         s    += abs(pix1[1]  - pix2[1]);
 127         s    += abs(pix1[2]  - pix2[2]);
 128         s    += abs(pix1[3]  - pix2[3]);
 129         s    += abs(pix1[4]  - pix2[4]);
 130         s    += abs(pix1[5]  - pix2[5]);
 131         s    += abs(pix1[6]  - pix2[6]);
 132         s    += abs(pix1[7]  - pix2[7]);
 133         s    += abs(pix1[8]  - pix2[8]);
 134         s    += abs(pix1[9]  - pix2[9]);
 135         s    += abs(pix1[10] - pix2[10]);
 136         s    += abs(pix1[11] - pix2[11]);
 137         s    += abs(pix1[12] - pix2[12]);
 138         s    += abs(pix1[13] - pix2[13]);
 139         s    += abs(pix1[14] - pix2[14]);
 140         s    += abs(pix1[15] - pix2[15]);
 141         pix1 += line_size;
 142         pix2 += line_size;
 143     }
 144     return s;
 145 }
 146
 147 static int pix_abs16_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 148                           int line_size, int h)
 149 {
 150     int s = 0, i;
 151
 152     for (i = 0; i < h; i++) {
 153         s    += abs(pix1[0]  - avg2(pix2[0],  pix2[1]));
 154         s    += abs(pix1[1]  - avg2(pix2[1],  pix2[2]));
 155         s    += abs(pix1[2]  - avg2(pix2[2],  pix2[3]));
 156         s    += abs(pix1[3]  - avg2(pix2[3],  pix2[4]));
 157         s    += abs(pix1[4]  - avg2(pix2[4],  pix2[5]));
 158         s    += abs(pix1[5]  - avg2(pix2[5],  pix2[6]));
 159         s    += abs(pix1[6]  - avg2(pix2[6],  pix2[7]));
 160         s    += abs(pix1[7]  - avg2(pix2[7],  pix2[8]));
 161         s    += abs(pix1[8]  - avg2(pix2[8],  pix2[9]));
 162         s    += abs(pix1[9]  - avg2(pix2[9],  pix2[10]));
 163         s    += abs(pix1[10] - avg2(pix2[10], pix2[11]));
 164         s    += abs(pix1[11] - avg2(pix2[11], pix2[12]));
 165         s    += abs(pix1[12] - avg2(pix2[12], pix2[13]));
 166         s    += abs(pix1[13] - avg2(pix2[13], pix2[14]));
 167         s    += abs(pix1[14] - avg2(pix2[14], pix2[15]));
 168         s    += abs(pix1[15] - avg2(pix2[15], pix2[16]));
 169         pix1 += line_size;
 170         pix2 += line_size;
 171     }
 172     return s;
 173 }
 174
 175 static int pix_abs16_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 176                           int line_size, int h)
 177 {
 178     int s = 0, i;
 179     uint8_t *pix3 = pix2 + line_size;
 180
 181     for (i = 0; i < h; i++) {
 182         s    += abs(pix1[0]  - avg2(pix2[0],  pix3[0]));
 183         s    += abs(pix1[1]  - avg2(pix2[1],  pix3[1]));
 184         s    += abs(pix1[2]  - avg2(pix2[2],  pix3[2]));
 185         s    += abs(pix1[3]  - avg2(pix2[3],  pix3[3]));
 186         s    += abs(pix1[4]  - avg2(pix2[4],  pix3[4]));
 187         s    += abs(pix1[5]  - avg2(pix2[5],  pix3[5]));
 188         s    += abs(pix1[6]  - avg2(pix2[6],  pix3[6]));
 189         s    += abs(pix1[7]  - avg2(pix2[7],  pix3[7]));
 190         s    += abs(pix1[8]  - avg2(pix2[8],  pix3[8]));
 191         s    += abs(pix1[9]  - avg2(pix2[9],  pix3[9]));
 192         s    += abs(pix1[10] - avg2(pix2[10], pix3[10]));
 193         s    += abs(pix1[11] - avg2(pix2[11], pix3[11]));
 194         s    += abs(pix1[12] - avg2(pix2[12], pix3[12]));
 195         s    += abs(pix1[13] - avg2(pix2[13], pix3[13]));
 196         s    += abs(pix1[14] - avg2(pix2[14], pix3[14]));
 197         s    += abs(pix1[15] - avg2(pix2[15], pix3[15]));
 198         pix1 += line_size;
 199         pix2 += line_size;
 200         pix3 += line_size;
 201     }
 202     return s;
 203 }
 204
 205 static int pix_abs16_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 206                            int line_size, int h)
 207 {
 208     int s = 0, i;
 209     uint8_t *pix3 = pix2 + line_size;
 210
 211     for (i = 0; i < h; i++) {
 212         s    += abs(pix1[0]  - avg4(pix2[0],  pix2[1],  pix3[0],  pix3[1]));
 213         s    += abs(pix1[1]  - avg4(pix2[1],  pix2[2],  pix3[1],  pix3[2]));
 214         s    += abs(pix1[2]  - avg4(pix2[2],  pix2[3],  pix3[2],  pix3[3]));
 215         s    += abs(pix1[3]  - avg4(pix2[3],  pix2[4],  pix3[3],  pix3[4]));
 216         s    += abs(pix1[4]  - avg4(pix2[4],  pix2[5],  pix3[4],  pix3[5]));
 217         s    += abs(pix1[5]  - avg4(pix2[5],  pix2[6],  pix3[5],  pix3[6]));
 218         s    += abs(pix1[6]  - avg4(pix2[6],  pix2[7],  pix3[6],  pix3[7]));
 219         s    += abs(pix1[7]  - avg4(pix2[7],  pix2[8],  pix3[7],  pix3[8]));
 220         s    += abs(pix1[8]  - avg4(pix2[8],  pix2[9],  pix3[8],  pix3[9]));
 221         s    += abs(pix1[9]  - avg4(pix2[9],  pix2[10], pix3[9],  pix3[10]));
 222         s    += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));
 223         s    += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));
 224         s    += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));
 225         s    += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));
 226         s    += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));
 227         s    += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));
 228         pix1 += line_size;
 229         pix2 += line_size;
 230         pix3 += line_size;
 231     }
 232     return s;
 233 }
 234
 235 static inline int pix_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 236                              int line_size, int h)
 237 {
 238     int s = 0, i;
 239
 240     for (i = 0; i < h; i++) {
 241         s    += abs(pix1[0] - pix2[0]);
 242         s    += abs(pix1[1] - pix2[1]);
 243         s    += abs(pix1[2] - pix2[2]);
 244         s    += abs(pix1[3] - pix2[3]);
 245         s    += abs(pix1[4] - pix2[4]);
 246         s    += abs(pix1[5] - pix2[5]);
 247         s    += abs(pix1[6] - pix2[6]);
 248         s    += abs(pix1[7] - pix2[7]);
 249         pix1 += line_size;
 250         pix2 += line_size;
 251     }
 252     return s;
 253 }
 254
 255 static int pix_abs8_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 256                          int line_size, int h)
 257 {
 258     int s = 0, i;
 259
 260     for (i = 0; i < h; i++) {
 261         s    += abs(pix1[0] - avg2(pix2[0], pix2[1]));
 262         s    += abs(pix1[1] - avg2(pix2[1], pix2[2]));
 263         s    += abs(pix1[2] - avg2(pix2[2], pix2[3]));
 264         s    += abs(pix1[3] - avg2(pix2[3], pix2[4]));
 265         s    += abs(pix1[4] - avg2(pix2[4], pix2[5]));
 266         s    += abs(pix1[5] - avg2(pix2[5], pix2[6]));
 267         s    += abs(pix1[6] - avg2(pix2[6], pix2[7]));
 268         s    += abs(pix1[7] - avg2(pix2[7], pix2[8]));
 269         pix1 += line_size;
 270         pix2 += line_size;
 271     }
 272     return s;
 273 }
 274
 275 static int pix_abs8_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 276                          int line_size, int h)
 277 {
 278     int s = 0, i;
 279     uint8_t *pix3 = pix2 + line_size;
 280
 281     for (i = 0; i < h; i++) {
 282         s    += abs(pix1[0] - avg2(pix2[0], pix3[0]));
 283         s    += abs(pix1[1] - avg2(pix2[1], pix3[1]));
 284         s    += abs(pix1[2] - avg2(pix2[2], pix3[2]));
 285         s    += abs(pix1[3] - avg2(pix2[3], pix3[3]));
 286         s    += abs(pix1[4] - avg2(pix2[4], pix3[4]));
 287         s    += abs(pix1[5] - avg2(pix2[5], pix3[5]));
 288         s    += abs(pix1[6] - avg2(pix2[6], pix3[6]));
 289         s    += abs(pix1[7] - avg2(pix2[7], pix3[7]));
 290         pix1 += line_size;
 291         pix2 += line_size;
 292         pix3 += line_size;
 293     }
 294     return s;
 295 }
 296
 297 static int pix_abs8_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,
 298                           int line_size, int h)
 299 {
 300     int s = 0, i;
 301     uint8_t *pix3 = pix2 + line_size;
 302
 303     for (i = 0; i < h; i++) {
 304         s    += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));
 305         s    += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));
 306         s    += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));
 307         s    += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));
 308         s    += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));
 309         s    += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));
 310         s    += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));
 311         s    += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));
 312         pix1 += line_size;
 313         pix2 += line_size;
 314         pix3 += line_size;
 315     }
 316     return s;
 317 }
 318
 319 static int nsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)
 320 {
 321     int score1 = 0, score2 = 0, x, y;
 322
 323     for (y = 0; y < h; y++) {
 324         for (x = 0; x < 16; x++)
 325             score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
 326         if (y + 1 < h) {
 327             for (x = 0; x < 15; x++)
 328                 score2 += FFABS(s1[x]     - s1[x + stride] -
 329                                 s1[x + 1] + s1[x + stride + 1]) -
 330                           FFABS(s2[x]     - s2[x + stride] -
 331                                 s2[x + 1] + s2[x + stride + 1]);
 332         }
 333         s1 += stride;
 334         s2 += stride;
 335     }
 336
 337     if (c)
 338         return score1 + FFABS(score2) * c->avctx->nsse_weight;
 339     else
 340         return score1 + FFABS(score2) * 8;
 341 }
 342
 343 static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)
 344 {
 345     int score1 = 0, score2 = 0, x, y;
 346
 347     for (y = 0; y < h; y++) {
 348         for (x = 0; x < 8; x++)
 349             score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);
 350         if (y + 1 < h) {
 351             for (x = 0; x < 7; x++)
 352                 score2 += FFABS(s1[x]     - s1[x + stride] -
 353                                 s1[x + 1] + s1[x + stride + 1]) -
 354                           FFABS(s2[x]     - s2[x + stride] -
 355                                 s2[x + 1] + s2[x + stride + 1]);
 356         }
 357         s1 += stride;
 358         s2 += stride;
 359     }
 360
 361     if (c)
 362         return score1 + FFABS(score2) * c->avctx->nsse_weight;
 363     else
 364         return score1 + FFABS(score2) * 8;
 365 }
 366
 367 static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,
 368                     int stride, int h)
 369 {
 370     return 0;
 371 }
 372
 373 void ff_set_cmp(DSPContext *c, me_cmp_func *cmp, int type)
 374 {
 375     int i;
 376
 377     memset(cmp, 0, sizeof(void *) * 6);
 378
 379     for (i = 0; i < 6; i++) {
 380         switch (type & 0xFF) {
 381         case FF_CMP_SAD:
 382             cmp[i] = c->sad[i];
 383             break;
 384         case FF_CMP_SATD:
 385             cmp[i] = c->hadamard8_diff[i];
 386             break;
 387         case FF_CMP_SSE:
 388             cmp[i] = c->sse[i];
 389             break;
 390         case FF_CMP_DCT:
 391             cmp[i] = c->dct_sad[i];
 392             break;
 393         case FF_CMP_DCT264:
 394             cmp[i] = c->dct264_sad[i];
 395             break;
 396         case FF_CMP_DCTMAX:
 397             cmp[i] = c->dct_max[i];
 398             break;
 399         case FF_CMP_PSNR:
 400             cmp[i] = c->quant_psnr[i];
 401             break;
 402         case FF_CMP_BIT:
 403             cmp[i] = c->bit[i];
 404             break;
 405         case FF_CMP_RD:
 406             cmp[i] = c->rd[i];
 407             break;
 408         case FF_CMP_VSAD:
 409             cmp[i] = c->vsad[i];
 410             break;
 411         case FF_CMP_VSSE:
 412             cmp[i] = c->vsse[i];
 413             break;
 414         case FF_CMP_ZERO:
 415             cmp[i] = zero_cmp;
 416             break;
 417         case FF_CMP_NSSE:
 418             cmp[i] = c->nsse[i];
 419             break;
 420 #if CONFIG_DWT
 421         case FF_CMP_W53:
 422             cmp[i]= c->w53[i];
 423             break;
 424         case FF_CMP_W97:
 425             cmp[i]= c->w97[i];
 426             break;
 427 #endif
 428         default:
 429             av_log(NULL, AV_LOG_ERROR,
 430                    "internal error in cmp function selection\n");
 431         }
 432     }
 433 }
 434
 435 #define BUTTERFLY2(o1, o2, i1, i2)              \
 436     o1 = (i1) + (i2);                           \
 437     o2 = (i1) - (i2);
 438
 439 #define BUTTERFLY1(x, y)                        \
 440     {                                           \
 441         int a, b;                               \
 442         a = x;                                  \
 443         b = y;                                  \
 444         x = a + b;                              \
 445         y = a - b;                              \
 446     }
 447
 448 #define BUTTERFLYA(x, y) (FFABS((x) + (y)) + FFABS((x) - (y)))
 449
 450 static int hadamard8_diff8x8_c(MpegEncContext *s, uint8_t *dst,
 451                                uint8_t *src, int stride, int h)
 452 {
 453     int i, temp[64], sum = 0;
 454
 455     av_assert2(h == 8);
 456
 457     for (i = 0; i < 8; i++) {
 458         // FIXME: try pointer walks
 459         BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
 460                    src[stride * i + 0] - dst[stride * i + 0],
 461                    src[stride * i + 1] - dst[stride * i + 1]);
 462         BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
 463                    src[stride * i + 2] - dst[stride * i + 2],
 464                    src[stride * i + 3] - dst[stride * i + 3]);
 465         BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
 466                    src[stride * i + 4] - dst[stride * i + 4],
 467                    src[stride * i + 5] - dst[stride * i + 5]);
 468         BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
 469                    src[stride * i + 6] - dst[stride * i + 6],
 470                    src[stride * i + 7] - dst[stride * i + 7]);
 471
 472         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
 473         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
 474         BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
 475         BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
 476
 477         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
 478         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
 479         BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
 480         BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
 481     }
 482
 483     for (i = 0; i < 8; i++) {
 484         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
 485         BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
 486         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
 487         BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
 488
 489         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
 490         BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
 491         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
 492         BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
 493
 494         sum += BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i]) +
 495                BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i]) +
 496                BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i]) +
 497                BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
 498     }
 499     return sum;
 500 }
 501
 502 static int hadamard8_intra8x8_c(MpegEncContext *s, uint8_t *src,
 503                                 uint8_t *dummy, int stride, int h)
 504 {
 505     int i, temp[64], sum = 0;
 506
 507     av_assert2(h == 8);
 508
 509     for (i = 0; i < 8; i++) {
 510         // FIXME: try pointer walks
 511         BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],
 512                    src[stride * i + 0], src[stride * i + 1]);
 513         BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],
 514                    src[stride * i + 2], src[stride * i + 3]);
 515         BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],
 516                    src[stride * i + 4], src[stride * i + 5]);
 517         BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],
 518                    src[stride * i + 6], src[stride * i + 7]);
 519
 520         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);
 521         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);
 522         BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);
 523         BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);
 524
 525         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);
 526         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);
 527         BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);
 528         BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);
 529     }
 530
 531     for (i = 0; i < 8; i++) {
 532         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);
 533         BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);
 534         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);
 535         BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);
 536
 537         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);
 538         BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);
 539         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);
 540         BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);
 541
 542         sum +=
 543             BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i])
 544             + BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i])
 545             + BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i])
 546             + BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);
 547     }
 548
 549     sum -= FFABS(temp[8 * 0] + temp[8 * 4]); // -mean
 550
 551     return sum;
 552 }
 553
 554 static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,
 555                         uint8_t *src2, int stride, int h)
 556 {
 557     LOCAL_ALIGNED_16(int16_t, temp, [64]);
 558
 559     av_assert2(h == 8);
 560
 561     s->pdsp.diff_pixels(temp, src1, src2, stride);
 562     s->fdsp.fdct(temp);
 563     return s->dsp.sum_abs_dctelem(temp);
 564 }
 565
 566 #if CONFIG_GPL
 567 #define DCT8_1D                                         \
 568     {                                                   \
 569         const int s07 = SRC(0) + SRC(7);                \
 570         const int s16 = SRC(1) + SRC(6);                \
 571         const int s25 = SRC(2) + SRC(5);                \
 572         const int s34 = SRC(3) + SRC(4);                \
 573         const int a0  = s07 + s34;                      \
 574         const int a1  = s16 + s25;                      \
 575         const int a2  = s07 - s34;                      \
 576         const int a3  = s16 - s25;                      \
 577         const int d07 = SRC(0) - SRC(7);                \
 578         const int d16 = SRC(1) - SRC(6);                \
 579         const int d25 = SRC(2) - SRC(5);                \
 580         const int d34 = SRC(3) - SRC(4);                \
 581         const int a4  = d16 + d25 + (d07 + (d07 >> 1)); \
 582         const int a5  = d07 - d34 - (d25 + (d25 >> 1)); \
 583         const int a6  = d07 + d34 - (d16 + (d16 >> 1)); \
 584         const int a7  = d16 - d25 + (d34 + (d34 >> 1)); \
 585         DST(0, a0 + a1);                                \
 586         DST(1, a4 + (a7 >> 2));                         \
 587         DST(2, a2 + (a3 >> 1));                         \
 588         DST(3, a5 + (a6 >> 2));                         \
 589         DST(4, a0 - a1);                                \
 590         DST(5, a6 - (a5 >> 2));                         \
 591         DST(6, (a2 >> 1) - a3);                         \
 592         DST(7, (a4 >> 2) - a7);                         \
 593     }
 594
 595 static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,
 596                            uint8_t *src2, int stride, int h)
 597 {
 598     int16_t dct[8][8];
 599     int i, sum = 0;
 600
 601     s->pdsp.diff_pixels(dct[0], src1, src2, stride);
 602
 603 #define SRC(x) dct[i][x]
 604 #define DST(x, v) dct[i][x] = v
 605     for (i = 0; i < 8; i++)
 606         DCT8_1D
 607 #undef SRC
 608 #undef DST
 609
 610 #define SRC(x) dct[x][i]
 611 #define DST(x, v) sum += FFABS(v)
 612         for (i = 0; i < 8; i++)
 613             DCT8_1D
 614 #undef SRC
 615 #undef DST
 616             return sum;
 617 }
 618 #endif
 619
 620 static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,
 621                         uint8_t *src2, int stride, int h)
 622 {
 623     LOCAL_ALIGNED_16(int16_t, temp, [64]);
 624     int sum = 0, i;
 625
 626     av_assert2(h == 8);
 627
 628     s->pdsp.diff_pixels(temp, src1, src2, stride);
 629     s->fdsp.fdct(temp);
 630
 631     for (i = 0; i < 64; i++)
 632         sum = FFMAX(sum, FFABS(temp[i]));
 633
 634     return sum;
 635 }
 636
 637 static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,
 638                            uint8_t *src2, int stride, int h)
 639 {
 640     LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);
 641     int16_t *const bak = temp + 64;
 642     int sum = 0, i;
 643
 644     av_assert2(h == 8);
 645     s->mb_intra = 0;
 646
 647     s->pdsp.diff_pixels(temp, src1, src2, stride);
 648
 649     memcpy(bak, temp, 64 * sizeof(int16_t));
 650
 651     s->block_last_index[0 /* FIXME */] =
 652         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
 653     s->dct_unquantize_inter(s, temp, 0, s->qscale);
 654     ff_simple_idct_8(temp); // FIXME
 655
 656     for (i = 0; i < 64; i++)
 657         sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);
 658
 659     return sum;
 660 }
 661
 662 static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
 663                    int stride, int h)
 664 {
 665     const uint8_t *scantable = s->intra_scantable.permutated;
 666     LOCAL_ALIGNED_16(int16_t, temp, [64]);
 667     LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);
 668     LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);
 669     int i, last, run, bits, level, distortion, start_i;
 670     const int esc_length = s->ac_esc_length;
 671     uint8_t *length, *last_length;
 672
 673     av_assert2(h == 8);
 674
 675     copy_block8(lsrc1, src1, 8, stride, 8);
 676     copy_block8(lsrc2, src2, 8, stride, 8);
 677
 678     s->pdsp.diff_pixels(temp, lsrc1, lsrc2, 8);
 679
 680     s->block_last_index[0 /* FIXME */] =
 681     last                               =
 682         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
 683
 684     bits = 0;
 685
 686     if (s->mb_intra) {
 687         start_i     = 1;
 688         length      = s->intra_ac_vlc_length;
 689         last_length = s->intra_ac_vlc_last_length;
 690         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
 691     } else {
 692         start_i     = 0;
 693         length      = s->inter_ac_vlc_length;
 694         last_length = s->inter_ac_vlc_last_length;
 695     }
 696
 697     if (last >= start_i) {
 698         run = 0;
 699         for (i = start_i; i < last; i++) {
 700             int j = scantable[i];
 701             level = temp[j];
 702
 703             if (level) {
 704                 level += 64;
 705                 if ((level & (~127)) == 0)
 706                     bits += length[UNI_AC_ENC_INDEX(run, level)];
 707                 else
 708                     bits += esc_length;
 709                 run = 0;
 710             } else
 711                 run++;
 712         }
 713         i = scantable[last];
 714
 715         level = temp[i] + 64;
 716
 717         av_assert2(level - 64);
 718
 719         if ((level & (~127)) == 0) {
 720             bits += last_length[UNI_AC_ENC_INDEX(run, level)];
 721         } else
 722             bits += esc_length;
 723     }
 724
 725     if (last >= 0) {
 726         if (s->mb_intra)
 727             s->dct_unquantize_intra(s, temp, 0, s->qscale);
 728         else
 729             s->dct_unquantize_inter(s, temp, 0, s->qscale);
 730     }
 731
 732     s->idsp.idct_add(lsrc2, 8, temp);
 733
 734     distortion = s->dsp.sse[1](NULL, lsrc2, lsrc1, 8, 8);
 735
 736     return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);
 737 }
 738
 739 static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,
 740                     int stride, int h)
 741 {
 742     const uint8_t *scantable = s->intra_scantable.permutated;
 743     LOCAL_ALIGNED_16(int16_t, temp, [64]);
 744     int i, last, run, bits, level, start_i;
 745     const int esc_length = s->ac_esc_length;
 746     uint8_t *length, *last_length;
 747
 748     av_assert2(h == 8);
 749
 750     s->pdsp.diff_pixels(temp, src1, src2, stride);
 751
 752     s->block_last_index[0 /* FIXME */] =
 753     last                               =
 754         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);
 755
 756     bits = 0;
 757
 758     if (s->mb_intra) {
 759         start_i     = 1;
 760         length      = s->intra_ac_vlc_length;
 761         last_length = s->intra_ac_vlc_last_length;
 762         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma
 763     } else {
 764         start_i     = 0;
 765         length      = s->inter_ac_vlc_length;
 766         last_length = s->inter_ac_vlc_last_length;
 767     }
 768
 769     if (last >= start_i) {
 770         run = 0;
 771         for (i = start_i; i < last; i++) {
 772             int j = scantable[i];
 773             level = temp[j];
 774
 775             if (level) {
 776                 level += 64;
 777                 if ((level & (~127)) == 0)
 778                     bits += length[UNI_AC_ENC_INDEX(run, level)];
 779                 else
 780                     bits += esc_length;
 781                 run = 0;
 782             } else
 783                 run++;
 784         }
 785         i = scantable[last];
 786
 787         level = temp[i] + 64;
 788
 789         av_assert2(level - 64);
 790
 791         if ((level & (~127)) == 0)
 792             bits += last_length[UNI_AC_ENC_INDEX(run, level)];
 793         else
 794             bits += esc_length;
 795     }
 796
 797     return bits;
 798 }
 799
 800 #define VSAD_INTRA(size)                                                \
 801 static int vsad_intra ## size ## _c(MpegEncContext *c,                  \
 802                                     uint8_t *s, uint8_t *dummy,         \
 803                                     int stride, int h)                  \
 804 {                                                                       \
 805     int score = 0, x, y;                                                \
 806                                                                         \
 807     for (y = 1; y < h; y++) {                                           \
 808         for (x = 0; x < size; x += 4) {                                 \
 809             score += FFABS(s[x]     - s[x + stride])     +              \
 810                      FFABS(s[x + 1] - s[x + stride + 1]) +              \
 811                      FFABS(s[x + 2] - s[x + 2 + stride]) +              \
 812                      FFABS(s[x + 3] - s[x + 3 + stride]);               \
 813         }                                                               \
 814         s += stride;                                                    \
 815     }                                                                   \
 816                                                                         \
 817     return score;                                                       \
 818 }
 819 VSAD_INTRA(8)
 820 VSAD_INTRA(16)
 821
 822 #define VSAD(size)                                                             \
 823 static int vsad ## size ## _c(MpegEncContext *c,                               \
 824                               uint8_t *s1, uint8_t *s2,                        \
 825                               int stride, int h)                               \
 826 {                                                                              \
 827     int score = 0, x, y;                                                       \
 828                                                                                \
 829     for (y = 1; y < h; y++) {                                                  \
 830         for (x = 0; x < size; x++)                                             \
 831             score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);   \
 832         s1 += stride;                                                          \
 833         s2 += stride;                                                          \
 834     }                                                                          \
 835                                                                                \
 836     return score;                                                              \
 837 }
 838 VSAD(8)
 839 VSAD(16)
 840
 841 #define SQ(a) ((a) * (a))
 842 #define VSSE_INTRA(size)                                                \
 843 static int vsse_intra ## size ## _c(MpegEncContext *c,                  \
 844                                     uint8_t *s, uint8_t *dummy,         \
 845                                     int stride, int h)                  \
 846 {                                                                       \
 847     int score = 0, x, y;                                                \
 848                                                                         \
 849     for (y = 1; y < h; y++) {                                           \
 850         for (x = 0; x < size; x += 4) {                                 \
 851             score += SQ(s[x]     - s[x + stride]) +                     \
 852                      SQ(s[x + 1] - s[x + stride + 1]) +                 \
 853                      SQ(s[x + 2] - s[x + stride + 2]) +                 \
 854                      SQ(s[x + 3] - s[x + stride + 3]);                  \
 855         }                                                               \
 856         s += stride;                                                    \
 857     }                                                                   \
 858                                                                         \
 859     return score;                                                       \
 860 }
 861 VSSE_INTRA(8)
 862 VSSE_INTRA(16)
 863
 864 #define VSSE(size)                                                             \
 865 static int vsse ## size ## _c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,     \
 866                     int stride, int h)                                         \
 867 {                                                                              \
 868     int score = 0, x, y;                                                       \
 869                                                                                \
 870     for (y = 1; y < h; y++) {                                                  \
 871         for (x = 0; x < size; x++)                                             \
 872             score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);      \
 873         s1 += stride;                                                          \
 874         s2 += stride;                                                          \
 875     }                                                                          \
 876                                                                                \
 877     return score;                                                              \
 878 }
 879 VSSE(8)
 880 VSSE(16)
 881
 882 #define WRAPPER8_16_SQ(name8, name16)                                   \
 883 static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src,        \
 884                   int stride, int h)                                    \
 885 {                                                                       \
 886     int score = 0;                                                      \
 887                                                                         \
 888     score += name8(s, dst, src, stride, 8);                             \
 889     score += name8(s, dst + 8, src + 8, stride, 8);                     \
 890     if (h == 16) {                                                      \
 891         dst   += 8 * stride;                                            \
 892         src   += 8 * stride;                                            \
 893         score += name8(s, dst, src, stride, 8);                         \
 894         score += name8(s, dst + 8, src + 8, stride, 8);                 \
 895     }                                                                   \
 896     return score;                                                       \
 897 }
 898
 899 WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)
 900 WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)
 901 WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)
 902 #if CONFIG_GPL
 903 WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)
 904 #endif
 905 WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)
 906 WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)
 907 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
 908 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
 909
 910 /* init static data */
 911 av_cold void ff_dsputil_static_init(void)
 912 {
 913     int i;
 914
 915     for (i = 0; i < 512; i++)
 916         ff_square_tab[i] = (i - 256) * (i - 256);
 917 }
 918
 919 int ff_check_alignment(void)
 920 {
 921     static int did_fail = 0;
 922     LOCAL_ALIGNED_16(int, aligned, [4]);
 923
 924     if ((intptr_t)aligned & 15) {
 925         if (!did_fail) {
 926 #if HAVE_MMX || HAVE_ALTIVEC
 927             av_log(NULL, AV_LOG_ERROR,
 928                 "Compiler did not align stack variables. Libavcodec has been miscompiled\n"
 929                 "and may be very slow or crash. This is not a bug in libavcodec,\n"
 930                 "but in the compiler. You may try recompiling using gcc >= 4.2.\n"
 931                 "Do not report crashes to FFmpeg developers.\n");
 932 #endif
 933             did_fail=1;
 934         }
 935         return -1;
 936     }
 937     return 0;
 938 }
 939
 940 av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)
 941 {
 942     ff_check_alignment();
 943
 944     c->sum_abs_dctelem = sum_abs_dctelem_c;
 945
 946     /* TODO [0] 16  [1] 8 */
 947     c->pix_abs[0][0] = pix_abs16_c;
 948     c->pix_abs[0][1] = pix_abs16_x2_c;
 949     c->pix_abs[0][2] = pix_abs16_y2_c;
 950     c->pix_abs[0][3] = pix_abs16_xy2_c;
 951     c->pix_abs[1][0] = pix_abs8_c;
 952     c->pix_abs[1][1] = pix_abs8_x2_c;
 953     c->pix_abs[1][2] = pix_abs8_y2_c;
 954     c->pix_abs[1][3] = pix_abs8_xy2_c;
 955
 956 #define SET_CMP_FUNC(name)                      \
 957     c->name[0] = name ## 16_c;                  \
 958     c->name[1] = name ## 8x8_c;
 959
 960     SET_CMP_FUNC(hadamard8_diff)
 961     c->hadamard8_diff[4] = hadamard8_intra16_c;
 962     c->hadamard8_diff[5] = hadamard8_intra8x8_c;
 963     SET_CMP_FUNC(dct_sad)
 964     SET_CMP_FUNC(dct_max)
 965 #if CONFIG_GPL
 966     SET_CMP_FUNC(dct264_sad)
 967 #endif
 968     c->sad[0] = pix_abs16_c;
 969     c->sad[1] = pix_abs8_c;
 970     c->sse[0] = sse16_c;
 971     c->sse[1] = sse8_c;
 972     c->sse[2] = sse4_c;
 973     SET_CMP_FUNC(quant_psnr)
 974     SET_CMP_FUNC(rd)
 975     SET_CMP_FUNC(bit)
 976     c->vsad[0] = vsad16_c;
 977     c->vsad[1] = vsad8_c;
 978     c->vsad[4] = vsad_intra16_c;
 979     c->vsad[5] = vsad_intra8_c;
 980     c->vsse[0] = vsse16_c;
 981     c->vsse[1] = vsse8_c;
 982     c->vsse[4] = vsse_intra16_c;
 983     c->vsse[5] = vsse_intra8_c;
 984     c->nsse[0] = nsse16_c;
 985     c->nsse[1] = nsse8_c;
 986 #if CONFIG_SNOW_DECODER || CONFIG_SNOW_ENCODER
 987     ff_dsputil_init_dwt(c);
 988 #endif
 989
 990     if (ARCH_ALPHA)
 991         ff_dsputil_init_alpha(c, avctx);
 992     if (ARCH_ARM)
 993         ff_dsputil_init_arm(c, avctx);
 994     if (ARCH_PPC)
 995         ff_dsputil_init_ppc(c, avctx);
 996     if (ARCH_X86)
 997         ff_dsputil_init_x86(c, avctx);
 998 }
 999
1000 av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
1001 {
1002     ff_dsputil_init(c, avctx);
1003 }
1004
1005 av_cold void avpriv_dsputil_init(DSPContext *c, AVCodecContext *avctx)
1006 {
1007     ff_dsputil_init(c, avctx);
1008 }