git.sesse.net Git - ffmpeg/blob - libavcodec/vc1dsp.c

   1 /*
   2  * VC-1 and WMV3 decoder - DSP functions
   3  * Copyright (c) 2006 Konstantin Shishkov
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; either
  10  * version 2.1 of the License, or (at your option) any later version.
  11  *
  12  * FFmpeg is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 /**
  23  * @file
  24  * VC-1 and WMV3 decoder
  25  */
  26
  27 #include "libavutil/avassert.h"
  28 #include "libavutil/common.h"
  29 #include "libavutil/intreadwrite.h"
  30 #include "h264chroma.h"
  31 #include "qpeldsp.h"
  32 #include "rnd_avg.h"
  33 #include "vc1dsp.h"
  34 #include "startcode.h"
  35
  36 /* Apply overlap transform to horizontal edge */
  37 static void vc1_v_overlap_c(uint8_t *src, int stride)
  38 {
  39     int i;
  40     int a, b, c, d;
  41     int d1, d2;
  42     int rnd = 1;
  43     for (i = 0; i < 8; i++) {
  44         a  = src[-2 * stride];
  45         b  = src[-stride];
  46         c  = src[0];
  47         d  = src[stride];
  48         d1 = (a - d + 3 + rnd) >> 3;
  49         d2 = (a - d + b - c + 4 - rnd) >> 3;
  50
  51         src[-2 * stride] = a - d1;
  52         src[-stride]     = av_clip_uint8(b - d2);
  53         src[0]           = av_clip_uint8(c + d2);
  54         src[stride]      = d + d1;
  55         src++;
  56         rnd = !rnd;
  57     }
  58 }
  59
  60 /* Apply overlap transform to vertical edge */
  61 static void vc1_h_overlap_c(uint8_t *src, int stride)
  62 {
  63     int i;
  64     int a, b, c, d;
  65     int d1, d2;
  66     int rnd = 1;
  67     for (i = 0; i < 8; i++) {
  68         a  = src[-2];
  69         b  = src[-1];
  70         c  = src[0];
  71         d  = src[1];
  72         d1 = (a - d + 3 + rnd) >> 3;
  73         d2 = (a - d + b - c + 4 - rnd) >> 3;
  74
  75         src[-2] = a - d1;
  76         src[-1] = av_clip_uint8(b - d2);
  77         src[0]  = av_clip_uint8(c + d2);
  78         src[1]  = d + d1;
  79         src    += stride;
  80         rnd     = !rnd;
  81     }
  82 }
  83
  84 static void vc1_v_s_overlap_c(int16_t *top, int16_t *bottom)
  85 {
  86     int i;
  87     int a, b, c, d;
  88     int d1, d2;
  89     int rnd1 = 4, rnd2 = 3;
  90     for (i = 0; i < 8; i++) {
  91         a  = top[48];
  92         b  = top[56];
  93         c  = bottom[0];
  94         d  = bottom[8];
  95         d1 = a - d;
  96         d2 = a - d + b - c;
  97
  98         top[48]   = ((a << 3) - d1 + rnd1) >> 3;
  99         top[56]   = ((b << 3) - d2 + rnd2) >> 3;
 100         bottom[0] = ((c << 3) + d2 + rnd1) >> 3;
 101         bottom[8] = ((d << 3) + d1 + rnd2) >> 3;
 102
 103         bottom++;
 104         top++;
 105         rnd2 = 7 - rnd2;
 106         rnd1 = 7 - rnd1;
 107     }
 108 }
 109
 110 static void vc1_h_s_overlap_c(int16_t *left, int16_t *right)
 111 {
 112     int i;
 113     int a, b, c, d;
 114     int d1, d2;
 115     int rnd1 = 4, rnd2 = 3;
 116     for (i = 0; i < 8; i++) {
 117         a  = left[6];
 118         b  = left[7];
 119         c  = right[0];
 120         d  = right[1];
 121         d1 = a - d;
 122         d2 = a - d + b - c;
 123
 124         left[6]  = ((a << 3) - d1 + rnd1) >> 3;
 125         left[7]  = ((b << 3) - d2 + rnd2) >> 3;
 126         right[0] = ((c << 3) + d2 + rnd1) >> 3;
 127         right[1] = ((d << 3) + d1 + rnd2) >> 3;
 128
 129         right += 8;
 130         left  += 8;
 131         rnd2   = 7 - rnd2;
 132         rnd1   = 7 - rnd1;
 133     }
 134 }
 135
 136 /**
 137  * VC-1 in-loop deblocking filter for one line
 138  * @param src source block type
 139  * @param stride block stride
 140  * @param pq block quantizer
 141  * @return whether other 3 pairs should be filtered or not
 142  * @see 8.6
 143  */
 144 static av_always_inline int vc1_filter_line(uint8_t *src, int stride, int pq)
 145 {
 146     int a0 = (2 * (src[-2 * stride] - src[1 * stride]) -
 147               5 * (src[-1 * stride] - src[0 * stride]) + 4) >> 3;
 148     int a0_sign = a0 >> 31;        /* Store sign */
 149
 150     a0 = (a0 ^ a0_sign) - a0_sign; /* a0 = FFABS(a0); */
 151     if (a0 < pq) {
 152         int a1 = FFABS((2 * (src[-4 * stride] - src[-1 * stride]) -
 153                         5 * (src[-3 * stride] - src[-2 * stride]) + 4) >> 3);
 154         int a2 = FFABS((2 * (src[ 0 * stride] - src[ 3 * stride]) -
 155                         5 * (src[ 1 * stride] - src[ 2 * stride]) + 4) >> 3);
 156         if (a1 < a0 || a2 < a0) {
 157             int clip      = src[-1 * stride] - src[0 * stride];
 158             int clip_sign = clip >> 31;
 159
 160             clip = ((clip ^ clip_sign) - clip_sign) >> 1;
 161             if (clip) {
 162                 int a3     = FFMIN(a1, a2);
 163                 int d      = 5 * (a3 - a0);
 164                 int d_sign = (d >> 31);
 165
 166                 d       = ((d ^ d_sign) - d_sign) >> 3;
 167                 d_sign ^= a0_sign;
 168
 169                 if (d_sign ^ clip_sign)
 170                     d = 0;
 171                 else {
 172                     d = FFMIN(d, clip);
 173                     d = (d ^ d_sign) - d_sign; /* Restore sign */
 174                     src[-1 * stride] = av_clip_uint8(src[-1 * stride] - d);
 175                     src[ 0 * stride] = av_clip_uint8(src[ 0 * stride] + d);
 176                 }
 177                 return 1;
 178             }
 179         }
 180     }
 181     return 0;
 182 }
 183
 184 /**
 185  * VC-1 in-loop deblocking filter
 186  * @param src source block type
 187  * @param step distance between horizontally adjacent elements
 188  * @param stride distance between vertically adjacent elements
 189  * @param len edge length to filter (4 or 8 pixels)
 190  * @param pq block quantizer
 191  * @see 8.6
 192  */
 193 static inline void vc1_loop_filter(uint8_t *src, int step, int stride,
 194                                    int len, int pq)
 195 {
 196     int i;
 197     int filt3;
 198
 199     for (i = 0; i < len; i += 4) {
 200         filt3 = vc1_filter_line(src + 2 * step, stride, pq);
 201         if (filt3) {
 202             vc1_filter_line(src + 0 * step, stride, pq);
 203             vc1_filter_line(src + 1 * step, stride, pq);
 204             vc1_filter_line(src + 3 * step, stride, pq);
 205         }
 206         src += step * 4;
 207     }
 208 }
 209
 210 static void vc1_v_loop_filter4_c(uint8_t *src, int stride, int pq)
 211 {
 212     vc1_loop_filter(src, 1, stride, 4, pq);
 213 }
 214
 215 static void vc1_h_loop_filter4_c(uint8_t *src, int stride, int pq)
 216 {
 217     vc1_loop_filter(src, stride, 1, 4, pq);
 218 }
 219
 220 static void vc1_v_loop_filter8_c(uint8_t *src, int stride, int pq)
 221 {
 222     vc1_loop_filter(src, 1, stride, 8, pq);
 223 }
 224
 225 static void vc1_h_loop_filter8_c(uint8_t *src, int stride, int pq)
 226 {
 227     vc1_loop_filter(src, stride, 1, 8, pq);
 228 }
 229
 230 static void vc1_v_loop_filter16_c(uint8_t *src, int stride, int pq)
 231 {
 232     vc1_loop_filter(src, 1, stride, 16, pq);
 233 }
 234
 235 static void vc1_h_loop_filter16_c(uint8_t *src, int stride, int pq)
 236 {
 237     vc1_loop_filter(src, stride, 1, 16, pq);
 238 }
 239
 240 /* Do inverse transform on 8x8 block */
 241 static void vc1_inv_trans_8x8_dc_c(uint8_t *dest, int linesize, int16_t *block)
 242 {
 243     int i;
 244     int dc = block[0];
 245
 246     dc = (3 * dc +  1) >> 1;
 247     dc = (3 * dc + 16) >> 5;
 248
 249     for (i = 0; i < 8; i++) {
 250         dest[0] = av_clip_uint8(dest[0] + dc);
 251         dest[1] = av_clip_uint8(dest[1] + dc);
 252         dest[2] = av_clip_uint8(dest[2] + dc);
 253         dest[3] = av_clip_uint8(dest[3] + dc);
 254         dest[4] = av_clip_uint8(dest[4] + dc);
 255         dest[5] = av_clip_uint8(dest[5] + dc);
 256         dest[6] = av_clip_uint8(dest[6] + dc);
 257         dest[7] = av_clip_uint8(dest[7] + dc);
 258         dest += linesize;
 259     }
 260 }
 261
 262 static void vc1_inv_trans_8x8_c(int16_t block[64])
 263 {
 264     int i;
 265     register int t1, t2, t3, t4, t5, t6, t7, t8;
 266     int16_t *src, *dst, temp[64];
 267
 268     src = block;
 269     dst = temp;
 270     for (i = 0; i < 8; i++) {
 271         t1 = 12 * (src[ 0] + src[32]) + 4;
 272         t2 = 12 * (src[ 0] - src[32]) + 4;
 273         t3 = 16 * src[16] +  6 * src[48];
 274         t4 =  6 * src[16] - 16 * src[48];
 275
 276         t5 = t1 + t3;
 277         t6 = t2 + t4;
 278         t7 = t2 - t4;
 279         t8 = t1 - t3;
 280
 281         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];
 282         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];
 283         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];
 284         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];
 285
 286         dst[0] = (t5 + t1) >> 3;
 287         dst[1] = (t6 + t2) >> 3;
 288         dst[2] = (t7 + t3) >> 3;
 289         dst[3] = (t8 + t4) >> 3;
 290         dst[4] = (t8 - t4) >> 3;
 291         dst[5] = (t7 - t3) >> 3;
 292         dst[6] = (t6 - t2) >> 3;
 293         dst[7] = (t5 - t1) >> 3;
 294
 295         src += 1;
 296         dst += 8;
 297     }
 298
 299     src = temp;
 300     dst = block;
 301     for (i = 0; i < 8; i++) {
 302         t1 = 12 * (src[ 0] + src[32]) + 64;
 303         t2 = 12 * (src[ 0] - src[32]) + 64;
 304         t3 = 16 * src[16] +  6 * src[48];
 305         t4 =  6 * src[16] - 16 * src[48];
 306
 307         t5 = t1 + t3;
 308         t6 = t2 + t4;
 309         t7 = t2 - t4;
 310         t8 = t1 - t3;
 311
 312         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];
 313         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];
 314         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];
 315         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];
 316
 317         dst[ 0] = (t5 + t1) >> 7;
 318         dst[ 8] = (t6 + t2) >> 7;
 319         dst[16] = (t7 + t3) >> 7;
 320         dst[24] = (t8 + t4) >> 7;
 321         dst[32] = (t8 - t4 + 1) >> 7;
 322         dst[40] = (t7 - t3 + 1) >> 7;
 323         dst[48] = (t6 - t2 + 1) >> 7;
 324         dst[56] = (t5 - t1 + 1) >> 7;
 325
 326         src++;
 327         dst++;
 328     }
 329 }
 330
 331 /* Do inverse transform on 8x4 part of block */
 332 static void vc1_inv_trans_8x4_dc_c(uint8_t *dest, int linesize, int16_t *block)
 333 {
 334     int i;
 335     int dc = block[0];
 336
 337     dc =  (3 * dc +  1) >> 1;
 338     dc = (17 * dc + 64) >> 7;
 339
 340     for (i = 0; i < 4; i++) {
 341         dest[0] = av_clip_uint8(dest[0] + dc);
 342         dest[1] = av_clip_uint8(dest[1] + dc);
 343         dest[2] = av_clip_uint8(dest[2] + dc);
 344         dest[3] = av_clip_uint8(dest[3] + dc);
 345         dest[4] = av_clip_uint8(dest[4] + dc);
 346         dest[5] = av_clip_uint8(dest[5] + dc);
 347         dest[6] = av_clip_uint8(dest[6] + dc);
 348         dest[7] = av_clip_uint8(dest[7] + dc);
 349         dest += linesize;
 350     }
 351 }
 352
 353 static void vc1_inv_trans_8x4_c(uint8_t *dest, int linesize, int16_t *block)
 354 {
 355     int i;
 356     register int t1, t2, t3, t4, t5, t6, t7, t8;
 357     int16_t *src, *dst;
 358
 359     src = block;
 360     dst = block;
 361
 362     for (i = 0; i < 4; i++) {
 363         t1 = 12 * (src[0] + src[4]) + 4;
 364         t2 = 12 * (src[0] - src[4]) + 4;
 365         t3 = 16 * src[2] +  6 * src[6];
 366         t4 =  6 * src[2] - 16 * src[6];
 367
 368         t5 = t1 + t3;
 369         t6 = t2 + t4;
 370         t7 = t2 - t4;
 371         t8 = t1 - t3;
 372
 373         t1 = 16 * src[1] + 15 * src[3] +  9 * src[5] +  4 * src[7];
 374         t2 = 15 * src[1] -  4 * src[3] - 16 * src[5] -  9 * src[7];
 375         t3 =  9 * src[1] - 16 * src[3] +  4 * src[5] + 15 * src[7];
 376         t4 =  4 * src[1] -  9 * src[3] + 15 * src[5] - 16 * src[7];
 377
 378         dst[0] = (t5 + t1) >> 3;
 379         dst[1] = (t6 + t2) >> 3;
 380         dst[2] = (t7 + t3) >> 3;
 381         dst[3] = (t8 + t4) >> 3;
 382         dst[4] = (t8 - t4) >> 3;
 383         dst[5] = (t7 - t3) >> 3;
 384         dst[6] = (t6 - t2) >> 3;
 385         dst[7] = (t5 - t1) >> 3;
 386
 387         src += 8;
 388         dst += 8;
 389     }
 390
 391     src = block;
 392     for (i = 0; i < 8; i++) {
 393         t1 = 17 * (src[ 0] + src[16]) + 64;
 394         t2 = 17 * (src[ 0] - src[16]) + 64;
 395         t3 = 22 * src[ 8] + 10 * src[24];
 396         t4 = 22 * src[24] - 10 * src[ 8];
 397
 398         dest[0 * linesize] = av_clip_uint8(dest[0 * linesize] + ((t1 + t3) >> 7));
 399         dest[1 * linesize] = av_clip_uint8(dest[1 * linesize] + ((t2 - t4) >> 7));
 400         dest[2 * linesize] = av_clip_uint8(dest[2 * linesize] + ((t2 + t4) >> 7));
 401         dest[3 * linesize] = av_clip_uint8(dest[3 * linesize] + ((t1 - t3) >> 7));
 402
 403         src++;
 404         dest++;
 405     }
 406 }
 407
 408 /* Do inverse transform on 4x8 parts of block */
 409 static void vc1_inv_trans_4x8_dc_c(uint8_t *dest, int linesize, int16_t *block)
 410 {
 411     int i;
 412     int dc = block[0];
 413
 414     dc = (17 * dc +  4) >> 3;
 415     dc = (12 * dc + 64) >> 7;
 416
 417     for (i = 0; i < 8; i++) {
 418         dest[0] = av_clip_uint8(dest[0] + dc);
 419         dest[1] = av_clip_uint8(dest[1] + dc);
 420         dest[2] = av_clip_uint8(dest[2] + dc);
 421         dest[3] = av_clip_uint8(dest[3] + dc);
 422         dest += linesize;
 423     }
 424 }
 425
 426 static void vc1_inv_trans_4x8_c(uint8_t *dest, int linesize, int16_t *block)
 427 {
 428     int i;
 429     register int t1, t2, t3, t4, t5, t6, t7, t8;
 430     int16_t *src, *dst;
 431
 432     src = block;
 433     dst = block;
 434
 435     for (i = 0; i < 8; i++) {
 436         t1 = 17 * (src[0] + src[2]) + 4;
 437         t2 = 17 * (src[0] - src[2]) + 4;
 438         t3 = 22 * src[1] + 10 * src[3];
 439         t4 = 22 * src[3] - 10 * src[1];
 440
 441         dst[0] = (t1 + t3) >> 3;
 442         dst[1] = (t2 - t4) >> 3;
 443         dst[2] = (t2 + t4) >> 3;
 444         dst[3] = (t1 - t3) >> 3;
 445
 446         src += 8;
 447         dst += 8;
 448     }
 449
 450     src = block;
 451     for (i = 0; i < 4; i++) {
 452         t1 = 12 * (src[ 0] + src[32]) + 64;
 453         t2 = 12 * (src[ 0] - src[32]) + 64;
 454         t3 = 16 * src[16] +  6 * src[48];
 455         t4 =  6 * src[16] - 16 * src[48];
 456
 457         t5 = t1 + t3;
 458         t6 = t2 + t4;
 459         t7 = t2 - t4;
 460         t8 = t1 - t3;
 461
 462         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];
 463         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];
 464         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];
 465         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];
 466
 467         dest[0 * linesize] = av_clip_uint8(dest[0 * linesize] + ((t5 + t1)     >> 7));
 468         dest[1 * linesize] = av_clip_uint8(dest[1 * linesize] + ((t6 + t2)     >> 7));
 469         dest[2 * linesize] = av_clip_uint8(dest[2 * linesize] + ((t7 + t3)     >> 7));
 470         dest[3 * linesize] = av_clip_uint8(dest[3 * linesize] + ((t8 + t4)     >> 7));
 471         dest[4 * linesize] = av_clip_uint8(dest[4 * linesize] + ((t8 - t4 + 1) >> 7));
 472         dest[5 * linesize] = av_clip_uint8(dest[5 * linesize] + ((t7 - t3 + 1) >> 7));
 473         dest[6 * linesize] = av_clip_uint8(dest[6 * linesize] + ((t6 - t2 + 1) >> 7));
 474         dest[7 * linesize] = av_clip_uint8(dest[7 * linesize] + ((t5 - t1 + 1) >> 7));
 475
 476         src++;
 477         dest++;
 478     }
 479 }
 480
 481 /* Do inverse transform on 4x4 part of block */
 482 static void vc1_inv_trans_4x4_dc_c(uint8_t *dest, int linesize, int16_t *block)
 483 {
 484     int i;
 485     int dc = block[0];
 486
 487     dc = (17 * dc +  4) >> 3;
 488     dc = (17 * dc + 64) >> 7;
 489
 490     for (i = 0; i < 4; i++) {
 491         dest[0] = av_clip_uint8(dest[0] + dc);
 492         dest[1] = av_clip_uint8(dest[1] + dc);
 493         dest[2] = av_clip_uint8(dest[2] + dc);
 494         dest[3] = av_clip_uint8(dest[3] + dc);
 495         dest += linesize;
 496     }
 497 }
 498
 499 static void vc1_inv_trans_4x4_c(uint8_t *dest, int linesize, int16_t *block)
 500 {
 501     int i;
 502     register int t1, t2, t3, t4;
 503     int16_t *src, *dst;
 504
 505     src = block;
 506     dst = block;
 507     for (i = 0; i < 4; i++) {
 508         t1 = 17 * (src[0] + src[2]) + 4;
 509         t2 = 17 * (src[0] - src[2]) + 4;
 510         t3 = 22 * src[1] + 10 * src[3];
 511         t4 = 22 * src[3] - 10 * src[1];
 512
 513         dst[0] = (t1 + t3) >> 3;
 514         dst[1] = (t2 - t4) >> 3;
 515         dst[2] = (t2 + t4) >> 3;
 516         dst[3] = (t1 - t3) >> 3;
 517
 518         src += 8;
 519         dst += 8;
 520     }
 521
 522     src = block;
 523     for (i = 0; i < 4; i++) {
 524         t1 = 17 * (src[0] + src[16]) + 64;
 525         t2 = 17 * (src[0] - src[16]) + 64;
 526         t3 = 22 * src[8] + 10 * src[24];
 527         t4 = 22 * src[24] - 10 * src[8];
 528
 529         dest[0 * linesize] = av_clip_uint8(dest[0 * linesize] + ((t1 + t3) >> 7));
 530         dest[1 * linesize] = av_clip_uint8(dest[1 * linesize] + ((t2 - t4) >> 7));
 531         dest[2 * linesize] = av_clip_uint8(dest[2 * linesize] + ((t2 + t4) >> 7));
 532         dest[3 * linesize] = av_clip_uint8(dest[3 * linesize] + ((t1 - t3) >> 7));
 533
 534         src++;
 535         dest++;
 536     }
 537 }
 538
 539 /* motion compensation functions */
 540
 541 /* Filter in case of 2 filters */
 542 #define VC1_MSPEL_FILTER_16B(DIR, TYPE)                                       \
 543 static av_always_inline int vc1_mspel_ ## DIR ## _filter_16bits(const TYPE *src, \
 544                                                                 int stride,   \
 545                                                                 int mode)     \
 546 {                                                                             \
 547     switch(mode) {                                                            \
 548     case 0: /* no shift - should not occur */                                 \
 549         return 0;                                                             \
 550     case 1: /* 1/4 shift */                                                   \
 551         return -4 * src[-stride] + 53 * src[0] +                              \
 552                18 * src[stride]  -  3 * src[stride * 2];                      \
 553     case 2: /* 1/2 shift */                                                   \
 554         return -1 * src[-stride] +  9 * src[0] +                              \
 555                 9 * src[stride]  -  1 * src[stride * 2];                      \
 556     case 3: /* 3/4 shift */                                                   \
 557         return -3 * src[-stride] + 18 * src[0] +                              \
 558                53 * src[stride]  -  4 * src[stride * 2];                      \
 559     }                                                                         \
 560     return 0; /* should not occur */                                          \
 561 }
 562
 563 VC1_MSPEL_FILTER_16B(ver, uint8_t)
 564 VC1_MSPEL_FILTER_16B(hor, int16_t)
 565
 566 /* Filter used to interpolate fractional pel values */
 567 static av_always_inline int vc1_mspel_filter(const uint8_t *src, int stride,
 568                                              int mode, int r)
 569 {
 570     switch (mode) {
 571     case 0: // no shift
 572         return src[0];
 573     case 1: // 1/4 shift
 574         return (-4 * src[-stride] + 53 * src[0] +
 575                 18 * src[stride]  -  3 * src[stride * 2] + 32 - r) >> 6;
 576     case 2: // 1/2 shift
 577         return (-1 * src[-stride] +  9 * src[0] +
 578                  9 * src[stride]  -  1 * src[stride * 2] + 8 - r) >> 4;
 579     case 3: // 3/4 shift
 580         return (-3 * src[-stride] + 18 * src[0] +
 581                 53 * src[stride]  -  4 * src[stride * 2] + 32 - r) >> 6;
 582     }
 583     return 0; // should not occur
 584 }
 585
 586 /* Function used to do motion compensation with bicubic interpolation */
 587 #define VC1_MSPEL_MC(OP, OP4, OPNAME)                                         \
 588 static av_always_inline void OPNAME ## vc1_mspel_mc(uint8_t *dst,             \
 589                                                     const uint8_t *src,       \
 590                                                     ptrdiff_t stride,         \
 591                                                     int hmode,                \
 592                                                     int vmode,                \
 593                                                     int rnd)                  \
 594 {                                                                             \
 595     int i, j;                                                                 \
 596                                                                               \
 597     if (vmode) { /* Horizontal filter to apply */                             \
 598         int r;                                                                \
 599                                                                               \
 600         if (hmode) { /* Vertical filter to apply, output to tmp */            \
 601             static const int shift_value[] = { 0, 5, 1, 5 };                  \
 602             int shift = (shift_value[hmode] + shift_value[vmode]) >> 1;       \
 603             int16_t tmp[11 * 8], *tptr = tmp;                                 \
 604                                                                               \
 605             r = (1 << (shift - 1)) + rnd - 1;                                 \
 606                                                                               \
 607             src -= 1;                                                         \
 608             for (j = 0; j < 8; j++) {                                         \
 609                 for (i = 0; i < 11; i++)                                      \
 610                     tptr[i] = (vc1_mspel_ver_filter_16bits(src + i, stride, vmode) + r) >> shift; \
 611                 src  += stride;                                               \
 612                 tptr += 11;                                                   \
 613             }                                                                 \
 614                                                                               \
 615             r    = 64 - rnd;                                                  \
 616             tptr = tmp + 1;                                                   \
 617             for (j = 0; j < 8; j++) {                                         \
 618                 for (i = 0; i < 8; i++)                                       \
 619                     OP(dst[i], (vc1_mspel_hor_filter_16bits(tptr + i, 1, hmode) + r) >> 7); \
 620                 dst  += stride;                                               \
 621                 tptr += 11;                                                   \
 622             }                                                                 \
 623                                                                               \
 624             return;                                                           \
 625         } else { /* No horizontal filter, output 8 lines to dst */            \
 626             r = 1 - rnd;                                                      \
 627                                                                               \
 628             for (j = 0; j < 8; j++) {                                         \
 629                 for (i = 0; i < 8; i++)                                       \
 630                     OP(dst[i], vc1_mspel_filter(src + i, stride, vmode, r));  \
 631                 src += stride;                                                \
 632                 dst += stride;                                                \
 633             }                                                                 \
 634             return;                                                           \
 635         }                                                                     \
 636     }                                                                         \
 637                                                                               \
 638     /* Horizontal mode with no vertical mode */                               \
 639     for (j = 0; j < 8; j++) {                                                 \
 640         for (i = 0; i < 8; i++)                                               \
 641             OP(dst[i], vc1_mspel_filter(src + i, 1, hmode, rnd));             \
 642         dst += stride;                                                        \
 643         src += stride;                                                        \
 644     }                                                                         \
 645 }\
 646 static av_always_inline void OPNAME ## vc1_mspel_mc_16(uint8_t *dst,          \
 647                                                        const uint8_t *src,    \
 648                                                        ptrdiff_t stride,      \
 649                                                        int hmode,             \
 650                                                        int vmode,             \
 651                                                        int rnd)               \
 652 {                                                                             \
 653     int i, j;                                                                 \
 654                                                                               \
 655     if (vmode) { /* Horizontal filter to apply */                             \
 656         int r;                                                                \
 657                                                                               \
 658         if (hmode) { /* Vertical filter to apply, output to tmp */            \
 659             static const int shift_value[] = { 0, 5, 1, 5 };                  \
 660             int shift = (shift_value[hmode] + shift_value[vmode]) >> 1;       \
 661             int16_t tmp[19 * 16], *tptr = tmp;                                \
 662                                                                               \
 663             r = (1 << (shift - 1)) + rnd - 1;                                 \
 664                                                                               \
 665             src -= 1;                                                         \
 666             for (j = 0; j < 16; j++) {                                        \
 667                 for (i = 0; i < 19; i++)                                      \
 668                     tptr[i] = (vc1_mspel_ver_filter_16bits(src + i, stride, vmode) + r) >> shift; \
 669                 src  += stride;                                               \
 670                 tptr += 19;                                                   \
 671             }                                                                 \
 672                                                                               \
 673             r    = 64 - rnd;                                                  \
 674             tptr = tmp + 1;                                                   \
 675             for (j = 0; j < 16; j++) {                                        \
 676                 for (i = 0; i < 16; i++)                                      \
 677                     OP(dst[i], (vc1_mspel_hor_filter_16bits(tptr + i, 1, hmode) + r) >> 7); \
 678                 dst  += stride;                                               \
 679                 tptr += 19;                                                   \
 680             }                                                                 \
 681                                                                               \
 682             return;                                                           \
 683         } else { /* No horizontal filter, output 8 lines to dst */            \
 684             r = 1 - rnd;                                                      \
 685                                                                               \
 686             for (j = 0; j < 16; j++) {                                        \
 687                 for (i = 0; i < 16; i++)                                      \
 688                     OP(dst[i], vc1_mspel_filter(src + i, stride, vmode, r));  \
 689                 src += stride;                                                \
 690                 dst += stride;                                                \
 691             }                                                                 \
 692             return;                                                           \
 693         }                                                                     \
 694     }                                                                         \
 695                                                                               \
 696     /* Horizontal mode with no vertical mode */                               \
 697     for (j = 0; j < 16; j++) {                                                \
 698         for (i = 0; i < 16; i++)                                              \
 699             OP(dst[i], vc1_mspel_filter(src + i, 1, hmode, rnd));             \
 700         dst += stride;                                                        \
 701         src += stride;                                                        \
 702     }                                                                         \
 703 }\
 704 static void OPNAME ## pixels8x8_c(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int rnd){\
 705     int i;\
 706     for(i=0; i<8; i++){\
 707         OP4(*(uint32_t*)(block  ), AV_RN32(pixels  ));\
 708         OP4(*(uint32_t*)(block+4), AV_RN32(pixels+4));\
 709         pixels+=line_size;\
 710         block +=line_size;\
 711     }\
 712 }\
 713 static void OPNAME ## pixels16x16_c(uint8_t *block, const uint8_t *pixels, ptrdiff_t line_size, int rnd){\
 714     int i;\
 715     for(i=0; i<16; i++){\
 716         OP4(*(uint32_t*)(block   ), AV_RN32(pixels   ));\
 717         OP4(*(uint32_t*)(block+ 4), AV_RN32(pixels+ 4));\
 718         OP4(*(uint32_t*)(block+ 8), AV_RN32(pixels+ 8));\
 719         OP4(*(uint32_t*)(block+12), AV_RN32(pixels+12));\
 720         pixels+=line_size;\
 721         block +=line_size;\
 722     }\
 723 }
 724
 725 #define op_put(a, b) (a) = av_clip_uint8(b)
 726 #define op_avg(a, b) (a) = ((a) + av_clip_uint8(b) + 1) >> 1
 727 #define op4_avg(a, b) (a) = rnd_avg32(a, b)
 728 #define op4_put(a, b) (a) = (b)
 729
 730 VC1_MSPEL_MC(op_put, op4_put, put_)
 731 VC1_MSPEL_MC(op_avg, op4_avg, avg_)
 732
 733 /* pixel functions - really are entry points to vc1_mspel_mc */
 734
 735 #define PUT_VC1_MSPEL(a, b)                                                   \
 736 static void put_vc1_mspel_mc ## a ## b ## _c(uint8_t *dst,                    \
 737                                              const uint8_t *src,              \
 738                                              ptrdiff_t stride, int rnd)       \
 739 {                                                                             \
 740     put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                            \
 741 }                                                                             \
 742 static void avg_vc1_mspel_mc ## a ## b ## _c(uint8_t *dst,                    \
 743                                              const uint8_t *src,              \
 744                                              ptrdiff_t stride, int rnd)       \
 745 {                                                                             \
 746     avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                            \
 747 }                                                                             \
 748 static void put_vc1_mspel_mc ## a ## b ## _16_c(uint8_t *dst,                 \
 749                                                 const uint8_t *src,           \
 750                                                 ptrdiff_t stride, int rnd)    \
 751 {                                                                             \
 752     put_vc1_mspel_mc_16(dst, src, stride, a, b, rnd);                         \
 753 }                                                                             \
 754 static void avg_vc1_mspel_mc ## a ## b ## _16_c(uint8_t *dst,                 \
 755                                                 const uint8_t *src,           \
 756                                                 ptrdiff_t stride, int rnd)    \
 757 {                                                                             \
 758     avg_vc1_mspel_mc_16(dst, src, stride, a, b, rnd);                         \
 759 }
 760
 761 PUT_VC1_MSPEL(1, 0)
 762 PUT_VC1_MSPEL(2, 0)
 763 PUT_VC1_MSPEL(3, 0)
 764
 765 PUT_VC1_MSPEL(0, 1)
 766 PUT_VC1_MSPEL(1, 1)
 767 PUT_VC1_MSPEL(2, 1)
 768 PUT_VC1_MSPEL(3, 1)
 769
 770 PUT_VC1_MSPEL(0, 2)
 771 PUT_VC1_MSPEL(1, 2)
 772 PUT_VC1_MSPEL(2, 2)
 773 PUT_VC1_MSPEL(3, 2)
 774
 775 PUT_VC1_MSPEL(0, 3)
 776 PUT_VC1_MSPEL(1, 3)
 777 PUT_VC1_MSPEL(2, 3)
 778 PUT_VC1_MSPEL(3, 3)
 779
 780 #define chroma_mc(a) \
 781     ((A * src[a] + B * src[a + 1] + \
 782       C * src[stride + a] + D * src[stride + a + 1] + 32 - 4) >> 6)
 783 static void put_no_rnd_vc1_chroma_mc8_c(uint8_t *dst /* align 8 */,
 784                                         uint8_t *src /* align 1 */,
 785                                         int stride, int h, int x, int y)
 786 {
 787     const int A = (8 - x) * (8 - y);
 788     const int B =     (x) * (8 - y);
 789     const int C = (8 - x) *     (y);
 790     const int D =     (x) *     (y);
 791     int i;
 792
 793     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
 794
 795     for (i = 0; i < h; i++) {
 796         dst[0] = chroma_mc(0);
 797         dst[1] = chroma_mc(1);
 798         dst[2] = chroma_mc(2);
 799         dst[3] = chroma_mc(3);
 800         dst[4] = chroma_mc(4);
 801         dst[5] = chroma_mc(5);
 802         dst[6] = chroma_mc(6);
 803         dst[7] = chroma_mc(7);
 804         dst += stride;
 805         src += stride;
 806     }
 807 }
 808
 809 static void put_no_rnd_vc1_chroma_mc4_c(uint8_t *dst, uint8_t *src,
 810                                         int stride, int h, int x, int y)
 811 {
 812     const int A = (8 - x) * (8 - y);
 813     const int B =     (x) * (8 - y);
 814     const int C = (8 - x) *     (y);
 815     const int D =     (x) *     (y);
 816     int i;
 817
 818     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
 819
 820     for (i = 0; i < h; i++) {
 821         dst[0] = chroma_mc(0);
 822         dst[1] = chroma_mc(1);
 823         dst[2] = chroma_mc(2);
 824         dst[3] = chroma_mc(3);
 825         dst += stride;
 826         src += stride;
 827     }
 828 }
 829
 830 #define avg2(a, b) (((a) + (b) + 1) >> 1)
 831 static void avg_no_rnd_vc1_chroma_mc8_c(uint8_t *dst /* align 8 */,
 832                                         uint8_t *src /* align 1 */,
 833                                         int stride, int h, int x, int y)
 834 {
 835     const int A = (8 - x) * (8 - y);
 836     const int B =     (x) * (8 - y);
 837     const int C = (8 - x) *     (y);
 838     const int D =     (x) *     (y);
 839     int i;
 840
 841     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
 842
 843     for (i = 0; i < h; i++) {
 844         dst[0] = avg2(dst[0], chroma_mc(0));
 845         dst[1] = avg2(dst[1], chroma_mc(1));
 846         dst[2] = avg2(dst[2], chroma_mc(2));
 847         dst[3] = avg2(dst[3], chroma_mc(3));
 848         dst[4] = avg2(dst[4], chroma_mc(4));
 849         dst[5] = avg2(dst[5], chroma_mc(5));
 850         dst[6] = avg2(dst[6], chroma_mc(6));
 851         dst[7] = avg2(dst[7], chroma_mc(7));
 852         dst += stride;
 853         src += stride;
 854     }
 855 }
 856
 857 static void avg_no_rnd_vc1_chroma_mc4_c(uint8_t *dst /* align 8 */,
 858                                         uint8_t *src /* align 1 */,
 859                                         int stride, int h, int x, int y)
 860 {
 861     const int A = (8 - x) * (8 - y);
 862     const int B = (    x) * (8 - y);
 863     const int C = (8 - x) * (    y);
 864     const int D = (    x) * (    y);
 865     int i;
 866
 867     av_assert2(x < 8 && y < 8 && x >= 0 && y >= 0);
 868
 869     for (i = 0; i < h; i++) {
 870         dst[0] = avg2(dst[0], chroma_mc(0));
 871         dst[1] = avg2(dst[1], chroma_mc(1));
 872         dst[2] = avg2(dst[2], chroma_mc(2));
 873         dst[3] = avg2(dst[3], chroma_mc(3));
 874         dst += stride;
 875         src += stride;
 876     }
 877 }
 878
 879 #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER
 880
 881 static void sprite_h_c(uint8_t *dst, const uint8_t *src, int offset,
 882                        int advance, int count)
 883 {
 884     while (count--) {
 885         int a = src[(offset >> 16)];
 886         int b = src[(offset >> 16) + 1];
 887         *dst++  = a + ((b - a) * (offset & 0xFFFF) >> 16);
 888         offset += advance;
 889     }
 890 }
 891
 892 static av_always_inline void sprite_v_template(uint8_t *dst,
 893                                                const uint8_t *src1a,
 894                                                const uint8_t *src1b,
 895                                                int offset1,
 896                                                int two_sprites,
 897                                                const uint8_t *src2a,
 898                                                const uint8_t *src2b,
 899                                                int offset2,
 900                                                int alpha, int scaled,
 901                                                int width)
 902 {
 903     int a1, b1, a2, b2;
 904     while (width--) {
 905         a1 = *src1a++;
 906         if (scaled) {
 907             b1 = *src1b++;
 908             a1 = a1 + ((b1 - a1) * offset1 >> 16);
 909         }
 910         if (two_sprites) {
 911             a2 = *src2a++;
 912             if (scaled > 1) {
 913                 b2 = *src2b++;
 914                 a2 = a2 + ((b2 - a2) * offset2 >> 16);
 915             }
 916             a1 = a1 + ((a2 - a1) * alpha >> 16);
 917         }
 918         *dst++ = a1;
 919     }
 920 }
 921
 922 static void sprite_v_single_c(uint8_t *dst, const uint8_t *src1a,
 923                               const uint8_t *src1b,
 924                               int offset, int width)
 925 {
 926     sprite_v_template(dst, src1a, src1b, offset, 0, NULL, NULL, 0, 0, 1, width);
 927 }
 928
 929 static void sprite_v_double_noscale_c(uint8_t *dst, const uint8_t *src1a,
 930                                       const uint8_t *src2a,
 931                                       int alpha, int width)
 932 {
 933     sprite_v_template(dst, src1a, NULL, 0, 1, src2a, NULL, 0, alpha, 0, width);
 934 }
 935
 936 static void sprite_v_double_onescale_c(uint8_t *dst,
 937                                        const uint8_t *src1a,
 938                                        const uint8_t *src1b,
 939                                        int offset1,
 940                                        const uint8_t *src2a,
 941                                        int alpha, int width)
 942 {
 943     sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, NULL, 0, alpha, 1,
 944                       width);
 945 }
 946
 947 static void sprite_v_double_twoscale_c(uint8_t *dst,
 948                                        const uint8_t *src1a,
 949                                        const uint8_t *src1b,
 950                                        int offset1,
 951                                        const uint8_t *src2a,
 952                                        const uint8_t *src2b,
 953                                        int offset2,
 954                                        int alpha,
 955                                        int width)
 956 {
 957     sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, src2b, offset2,
 958                       alpha, 2, width);
 959 }
 960
 961 #endif /* CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER */
 962 #define FN_ASSIGN(X, Y) \
 963     dsp->put_vc1_mspel_pixels_tab[1][X+4*Y] = put_vc1_mspel_mc##X##Y##_c; \
 964     dsp->put_vc1_mspel_pixels_tab[0][X+4*Y] = put_vc1_mspel_mc##X##Y##_16_c; \
 965     dsp->avg_vc1_mspel_pixels_tab[1][X+4*Y] = avg_vc1_mspel_mc##X##Y##_c; \
 966     dsp->avg_vc1_mspel_pixels_tab[0][X+4*Y] = avg_vc1_mspel_mc##X##Y##_16_c
 967
 968 av_cold void ff_vc1dsp_init(VC1DSPContext *dsp)
 969 {
 970     dsp->vc1_inv_trans_8x8    = vc1_inv_trans_8x8_c;
 971     dsp->vc1_inv_trans_4x8    = vc1_inv_trans_4x8_c;
 972     dsp->vc1_inv_trans_8x4    = vc1_inv_trans_8x4_c;
 973     dsp->vc1_inv_trans_4x4    = vc1_inv_trans_4x4_c;
 974     dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;
 975     dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;
 976     dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;
 977     dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;
 978
 979     dsp->vc1_h_overlap        = vc1_h_overlap_c;
 980     dsp->vc1_v_overlap        = vc1_v_overlap_c;
 981     dsp->vc1_h_s_overlap      = vc1_h_s_overlap_c;
 982     dsp->vc1_v_s_overlap      = vc1_v_s_overlap_c;
 983
 984     dsp->vc1_v_loop_filter4   = vc1_v_loop_filter4_c;
 985     dsp->vc1_h_loop_filter4   = vc1_h_loop_filter4_c;
 986     dsp->vc1_v_loop_filter8   = vc1_v_loop_filter8_c;
 987     dsp->vc1_h_loop_filter8   = vc1_h_loop_filter8_c;
 988     dsp->vc1_v_loop_filter16  = vc1_v_loop_filter16_c;
 989     dsp->vc1_h_loop_filter16  = vc1_h_loop_filter16_c;
 990
 991     dsp->put_vc1_mspel_pixels_tab[0][0] = put_pixels16x16_c;
 992     dsp->avg_vc1_mspel_pixels_tab[0][0] = avg_pixels16x16_c;
 993     dsp->put_vc1_mspel_pixels_tab[1][0] = put_pixels8x8_c;
 994     dsp->avg_vc1_mspel_pixels_tab[1][0] = avg_pixels8x8_c;
 995     FN_ASSIGN(0, 1);
 996     FN_ASSIGN(0, 2);
 997     FN_ASSIGN(0, 3);
 998
 999     FN_ASSIGN(1, 0);
1000     FN_ASSIGN(1, 1);
1001     FN_ASSIGN(1, 2);
1002     FN_ASSIGN(1, 3);
1003
1004     FN_ASSIGN(2, 0);
1005     FN_ASSIGN(2, 1);
1006     FN_ASSIGN(2, 2);
1007     FN_ASSIGN(2, 3);
1008
1009     FN_ASSIGN(3, 0);
1010     FN_ASSIGN(3, 1);
1011     FN_ASSIGN(3, 2);
1012     FN_ASSIGN(3, 3);
1013
1014     dsp->put_no_rnd_vc1_chroma_pixels_tab[0] = put_no_rnd_vc1_chroma_mc8_c;
1015     dsp->avg_no_rnd_vc1_chroma_pixels_tab[0] = avg_no_rnd_vc1_chroma_mc8_c;
1016     dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = put_no_rnd_vc1_chroma_mc4_c;
1017     dsp->avg_no_rnd_vc1_chroma_pixels_tab[1] = avg_no_rnd_vc1_chroma_mc4_c;
1018
1019 #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER
1020     dsp->sprite_h                 = sprite_h_c;
1021     dsp->sprite_v_single          = sprite_v_single_c;
1022     dsp->sprite_v_double_noscale  = sprite_v_double_noscale_c;
1023     dsp->sprite_v_double_onescale = sprite_v_double_onescale_c;
1024     dsp->sprite_v_double_twoscale = sprite_v_double_twoscale_c;
1025 #endif /* CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER */
1026
1027     dsp->startcode_find_candidate = ff_startcode_find_candidate_c;
1028
1029     if (ARCH_AARCH64)
1030         ff_vc1dsp_init_aarch64(dsp);
1031     if (ARCH_ARM)
1032         ff_vc1dsp_init_arm(dsp);
1033     if (ARCH_PPC)
1034         ff_vc1dsp_init_ppc(dsp);
1035     if (ARCH_X86)
1036         ff_vc1dsp_init_x86(dsp);
1037 }