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 Libav.
   6  *
   7  * Libav 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  * Libav 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 Libav; 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
  28 #include "libavutil/common.h"
  29 #include "h264chroma.h"
  30 #include "vc1dsp.h"
  31
  32
  33 /** Apply overlap transform to horizontal edge
  34 */
  35 static void vc1_v_overlap_c(uint8_t* src, int stride)
  36 {
  37     int i;
  38     int a, b, c, d;
  39     int d1, d2;
  40     int rnd = 1;
  41     for(i = 0; i < 8; i++) {
  42         a = src[-2*stride];
  43         b = src[-stride];
  44         c = src[0];
  45         d = src[stride];
  46         d1 = (a - d + 3 + rnd) >> 3;
  47         d2 = (a - d + b - c + 4 - rnd) >> 3;
  48
  49         src[-2*stride] = a - d1;
  50         src[-stride] = av_clip_uint8(b - d2);
  51         src[0] = av_clip_uint8(c + d2);
  52         src[stride] = d + d1;
  53         src++;
  54         rnd = !rnd;
  55     }
  56 }
  57
  58 /** Apply overlap transform to vertical edge
  59 */
  60 static void vc1_h_overlap_c(uint8_t* src, int stride)
  61 {
  62     int i;
  63     int a, b, c, d;
  64     int d1, d2;
  65     int rnd = 1;
  66     for(i = 0; i < 8; i++) {
  67         a = src[-2];
  68         b = src[-1];
  69         c = src[0];
  70         d = src[1];
  71         d1 = (a - d + 3 + rnd) >> 3;
  72         d2 = (a - d + b - c + 4 - rnd) >> 3;
  73
  74         src[-2] = a - d1;
  75         src[-1] = av_clip_uint8(b - d2);
  76         src[0] = av_clip_uint8(c + d2);
  77         src[1] = d + d1;
  78         src += stride;
  79         rnd = !rnd;
  80     }
  81 }
  82
  83 static void vc1_v_s_overlap_c(int16_t *top,  int16_t *bottom)
  84 {
  85     int i;
  86     int a, b, c, d;
  87     int d1, d2;
  88     int rnd1 = 4, rnd2 = 3;
  89     for(i = 0; i < 8; i++) {
  90         a = top[48];
  91         b = top[56];
  92         c = bottom[0];
  93         d = bottom[8];
  94         d1 = a - d;
  95         d2 = a - d + b - c;
  96
  97         top[48]   = ((a << 3) - d1 + rnd1) >> 3;
  98         top[56]   = ((b << 3) - d2 + rnd2) >> 3;
  99         bottom[0] = ((c << 3) + d2 + rnd1) >> 3;
 100         bottom[8] = ((d << 3) + d1 + rnd2) >> 3;
 101
 102         bottom++;
 103         top++;
 104         rnd2 = 7 - rnd2;
 105         rnd1 = 7 - rnd1;
 106     }
 107 }
 108
 109 static void vc1_h_s_overlap_c(int16_t *left, int16_t *right)
 110 {
 111     int i;
 112     int a, b, c, d;
 113     int d1, d2;
 114     int rnd1 = 4, rnd2 = 3;
 115     for(i = 0; i < 8; i++) {
 116         a = left[6];
 117         b = left[7];
 118         c = right[0];
 119         d = right[1];
 120         d1 = a - d;
 121         d2 = a - d + b - c;
 122
 123         left[6]  = ((a << 3) - d1 + rnd1) >> 3;
 124         left[7]  = ((b << 3) - d2 + rnd2) >> 3;
 125         right[0] = ((c << 3) + d2 + rnd1) >> 3;
 126         right[1] = ((d << 3) + d1 + rnd2) >> 3;
 127
 128         right += 8;
 129         left += 8;
 130         rnd2 = 7 - rnd2;
 131         rnd1 = 7 - rnd1;
 132     }
 133 }
 134
 135 /**
 136  * VC-1 in-loop deblocking filter for one line
 137  * @param src source block type
 138  * @param stride block stride
 139  * @param pq block quantizer
 140  * @return whether other 3 pairs should be filtered or not
 141  * @see 8.6
 142  */
 143 static av_always_inline int vc1_filter_line(uint8_t* src, int stride, int pq){
 144     int a0 = (2*(src[-2*stride] - src[ 1*stride]) - 5*(src[-1*stride] - src[ 0*stride]) + 4) >> 3;
 145     int a0_sign = a0 >> 31;        /* Store sign */
 146     a0 = (a0 ^ a0_sign) - a0_sign; /* a0 = FFABS(a0); */
 147     if(a0 < pq){
 148         int a1 = FFABS((2*(src[-4*stride] - src[-1*stride]) - 5*(src[-3*stride] - src[-2*stride]) + 4) >> 3);
 149         int a2 = FFABS((2*(src[ 0*stride] - src[ 3*stride]) - 5*(src[ 1*stride] - src[ 2*stride]) + 4) >> 3);
 150         if(a1 < a0 || a2 < a0){
 151             int clip = src[-1*stride] - src[ 0*stride];
 152             int clip_sign = clip >> 31;
 153             clip = ((clip ^ clip_sign) - clip_sign)>>1;
 154             if(clip){
 155                 int a3 = FFMIN(a1, a2);
 156                 int d = 5 * (a3 - a0);
 157                 int d_sign = (d >> 31);
 158                 d = ((d ^ d_sign) - d_sign) >> 3;
 159                 d_sign ^= a0_sign;
 160
 161                 if( d_sign ^ clip_sign )
 162                     d = 0;
 163                 else{
 164                     d = FFMIN(d, clip);
 165                     d = (d ^ d_sign) - d_sign;          /* Restore sign */
 166                     src[-1*stride] = av_clip_uint8(src[-1*stride] - d);
 167                     src[ 0*stride] = av_clip_uint8(src[ 0*stride] + d);
 168                 }
 169                 return 1;
 170             }
 171         }
 172     }
 173     return 0;
 174 }
 175
 176 /**
 177  * VC-1 in-loop deblocking filter
 178  * @param src source block type
 179  * @param step distance between horizontally adjacent elements
 180  * @param stride distance between vertically adjacent elements
 181  * @param len edge length to filter (4 or 8 pixels)
 182  * @param pq block quantizer
 183  * @see 8.6
 184  */
 185 static inline void vc1_loop_filter(uint8_t* src, int step, int stride, int len, int pq)
 186 {
 187     int i;
 188     int filt3;
 189
 190     for(i = 0; i < len; i += 4){
 191         filt3 = vc1_filter_line(src + 2*step, stride, pq);
 192         if(filt3){
 193             vc1_filter_line(src + 0*step, stride, pq);
 194             vc1_filter_line(src + 1*step, stride, pq);
 195             vc1_filter_line(src + 3*step, stride, pq);
 196         }
 197         src += step * 4;
 198     }
 199 }
 200
 201 static void vc1_v_loop_filter4_c(uint8_t *src, int stride, int pq)
 202 {
 203     vc1_loop_filter(src, 1, stride, 4, pq);
 204 }
 205
 206 static void vc1_h_loop_filter4_c(uint8_t *src, int stride, int pq)
 207 {
 208     vc1_loop_filter(src, stride, 1, 4, pq);
 209 }
 210
 211 static void vc1_v_loop_filter8_c(uint8_t *src, int stride, int pq)
 212 {
 213     vc1_loop_filter(src, 1, stride, 8, pq);
 214 }
 215
 216 static void vc1_h_loop_filter8_c(uint8_t *src, int stride, int pq)
 217 {
 218     vc1_loop_filter(src, stride, 1, 8, pq);
 219 }
 220
 221 static void vc1_v_loop_filter16_c(uint8_t *src, int stride, int pq)
 222 {
 223     vc1_loop_filter(src, 1, stride, 16, pq);
 224 }
 225
 226 static void vc1_h_loop_filter16_c(uint8_t *src, int stride, int pq)
 227 {
 228     vc1_loop_filter(src, stride, 1, 16, pq);
 229 }
 230
 231 /** Do inverse transform on 8x8 block
 232 */
 233 static void vc1_inv_trans_8x8_dc_c(uint8_t *dest, int linesize, int16_t *block)
 234 {
 235     int i;
 236     int dc = block[0];
 237     dc = (3 * dc +  1) >> 1;
 238     dc = (3 * dc + 16) >> 5;
 239     for(i = 0; i < 8; i++){
 240         dest[0] = av_clip_uint8(dest[0] + dc);
 241         dest[1] = av_clip_uint8(dest[1] + dc);
 242         dest[2] = av_clip_uint8(dest[2] + dc);
 243         dest[3] = av_clip_uint8(dest[3] + dc);
 244         dest[4] = av_clip_uint8(dest[4] + dc);
 245         dest[5] = av_clip_uint8(dest[5] + dc);
 246         dest[6] = av_clip_uint8(dest[6] + dc);
 247         dest[7] = av_clip_uint8(dest[7] + dc);
 248         dest += linesize;
 249     }
 250 }
 251
 252 static void vc1_inv_trans_8x8_c(int16_t block[64])
 253 {
 254     int i;
 255     register int t1,t2,t3,t4,t5,t6,t7,t8;
 256     int16_t *src, *dst, temp[64];
 257
 258     src = block;
 259     dst = temp;
 260     for(i = 0; i < 8; i++){
 261         t1 = 12 * (src[ 0] + src[32]) + 4;
 262         t2 = 12 * (src[ 0] - src[32]) + 4;
 263         t3 = 16 * src[16] +  6 * src[48];
 264         t4 =  6 * src[16] - 16 * src[48];
 265
 266         t5 = t1 + t3;
 267         t6 = t2 + t4;
 268         t7 = t2 - t4;
 269         t8 = t1 - t3;
 270
 271         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];
 272         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];
 273         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];
 274         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];
 275
 276         dst[0] = (t5 + t1) >> 3;
 277         dst[1] = (t6 + t2) >> 3;
 278         dst[2] = (t7 + t3) >> 3;
 279         dst[3] = (t8 + t4) >> 3;
 280         dst[4] = (t8 - t4) >> 3;
 281         dst[5] = (t7 - t3) >> 3;
 282         dst[6] = (t6 - t2) >> 3;
 283         dst[7] = (t5 - t1) >> 3;
 284
 285         src += 1;
 286         dst += 8;
 287     }
 288
 289     src = temp;
 290     dst = block;
 291     for(i = 0; i < 8; i++){
 292         t1 = 12 * (src[ 0] + src[32]) + 64;
 293         t2 = 12 * (src[ 0] - src[32]) + 64;
 294         t3 = 16 * src[16] +  6 * src[48];
 295         t4 =  6 * src[16] - 16 * src[48];
 296
 297         t5 = t1 + t3;
 298         t6 = t2 + t4;
 299         t7 = t2 - t4;
 300         t8 = t1 - t3;
 301
 302         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];
 303         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];
 304         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];
 305         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];
 306
 307         dst[ 0] = (t5 + t1) >> 7;
 308         dst[ 8] = (t6 + t2) >> 7;
 309         dst[16] = (t7 + t3) >> 7;
 310         dst[24] = (t8 + t4) >> 7;
 311         dst[32] = (t8 - t4 + 1) >> 7;
 312         dst[40] = (t7 - t3 + 1) >> 7;
 313         dst[48] = (t6 - t2 + 1) >> 7;
 314         dst[56] = (t5 - t1 + 1) >> 7;
 315
 316         src++;
 317         dst++;
 318     }
 319 }
 320
 321 /** Do inverse transform on 8x4 part of block
 322 */
 323 static void vc1_inv_trans_8x4_dc_c(uint8_t *dest, int linesize, int16_t *block)
 324 {
 325     int i;
 326     int dc = block[0];
 327     dc = ( 3 * dc +  1) >> 1;
 328     dc = (17 * dc + 64) >> 7;
 329     for(i = 0; i < 4; i++){
 330         dest[0] = av_clip_uint8(dest[0] + dc);
 331         dest[1] = av_clip_uint8(dest[1] + dc);
 332         dest[2] = av_clip_uint8(dest[2] + dc);
 333         dest[3] = av_clip_uint8(dest[3] + dc);
 334         dest[4] = av_clip_uint8(dest[4] + dc);
 335         dest[5] = av_clip_uint8(dest[5] + dc);
 336         dest[6] = av_clip_uint8(dest[6] + dc);
 337         dest[7] = av_clip_uint8(dest[7] + dc);
 338         dest += linesize;
 339     }
 340 }
 341
 342 static void vc1_inv_trans_8x4_c(uint8_t *dest, int linesize, int16_t *block)
 343 {
 344     int i;
 345     register int t1,t2,t3,t4,t5,t6,t7,t8;
 346     int16_t *src, *dst;
 347
 348     src = block;
 349     dst = block;
 350     for(i = 0; i < 4; i++){
 351         t1 = 12 * (src[0] + src[4]) + 4;
 352         t2 = 12 * (src[0] - src[4]) + 4;
 353         t3 = 16 * src[2] +  6 * src[6];
 354         t4 =  6 * src[2] - 16 * src[6];
 355
 356         t5 = t1 + t3;
 357         t6 = t2 + t4;
 358         t7 = t2 - t4;
 359         t8 = t1 - t3;
 360
 361         t1 = 16 * src[1] + 15 * src[3] +  9 * src[5] +  4 * src[7];
 362         t2 = 15 * src[1] -  4 * src[3] - 16 * src[5] -  9 * src[7];
 363         t3 =  9 * src[1] - 16 * src[3] +  4 * src[5] + 15 * src[7];
 364         t4 =  4 * src[1] -  9 * src[3] + 15 * src[5] - 16 * src[7];
 365
 366         dst[0] = (t5 + t1) >> 3;
 367         dst[1] = (t6 + t2) >> 3;
 368         dst[2] = (t7 + t3) >> 3;
 369         dst[3] = (t8 + t4) >> 3;
 370         dst[4] = (t8 - t4) >> 3;
 371         dst[5] = (t7 - t3) >> 3;
 372         dst[6] = (t6 - t2) >> 3;
 373         dst[7] = (t5 - t1) >> 3;
 374
 375         src += 8;
 376         dst += 8;
 377     }
 378
 379     src = block;
 380     for(i = 0; i < 8; i++){
 381         t1 = 17 * (src[ 0] + src[16]) + 64;
 382         t2 = 17 * (src[ 0] - src[16]) + 64;
 383         t3 = 22 * src[ 8] + 10 * src[24];
 384         t4 = 22 * src[24] - 10 * src[ 8];
 385
 386         dest[0*linesize] = av_clip_uint8(dest[0*linesize] + ((t1 + t3) >> 7));
 387         dest[1*linesize] = av_clip_uint8(dest[1*linesize] + ((t2 - t4) >> 7));
 388         dest[2*linesize] = av_clip_uint8(dest[2*linesize] + ((t2 + t4) >> 7));
 389         dest[3*linesize] = av_clip_uint8(dest[3*linesize] + ((t1 - t3) >> 7));
 390
 391         src ++;
 392         dest++;
 393     }
 394 }
 395
 396 /** Do inverse transform on 4x8 parts of block
 397 */
 398 static void vc1_inv_trans_4x8_dc_c(uint8_t *dest, int linesize, int16_t *block)
 399 {
 400     int i;
 401     int dc = block[0];
 402     dc = (17 * dc +  4) >> 3;
 403     dc = (12 * dc + 64) >> 7;
 404     for(i = 0; i < 8; i++){
 405         dest[0] = av_clip_uint8(dest[0] + dc);
 406         dest[1] = av_clip_uint8(dest[1] + dc);
 407         dest[2] = av_clip_uint8(dest[2] + dc);
 408         dest[3] = av_clip_uint8(dest[3] + dc);
 409         dest += linesize;
 410     }
 411 }
 412
 413 static void vc1_inv_trans_4x8_c(uint8_t *dest, int linesize, int16_t *block)
 414 {
 415     int i;
 416     register int t1,t2,t3,t4,t5,t6,t7,t8;
 417     int16_t *src, *dst;
 418
 419     src = block;
 420     dst = block;
 421     for(i = 0; i < 8; i++){
 422         t1 = 17 * (src[0] + src[2]) + 4;
 423         t2 = 17 * (src[0] - src[2]) + 4;
 424         t3 = 22 * src[1] + 10 * src[3];
 425         t4 = 22 * src[3] - 10 * src[1];
 426
 427         dst[0] = (t1 + t3) >> 3;
 428         dst[1] = (t2 - t4) >> 3;
 429         dst[2] = (t2 + t4) >> 3;
 430         dst[3] = (t1 - t3) >> 3;
 431
 432         src += 8;
 433         dst += 8;
 434     }
 435
 436     src = block;
 437     for(i = 0; i < 4; i++){
 438         t1 = 12 * (src[ 0] + src[32]) + 64;
 439         t2 = 12 * (src[ 0] - src[32]) + 64;
 440         t3 = 16 * src[16] +  6 * src[48];
 441         t4 =  6 * src[16] - 16 * src[48];
 442
 443         t5 = t1 + t3;
 444         t6 = t2 + t4;
 445         t7 = t2 - t4;
 446         t8 = t1 - t3;
 447
 448         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];
 449         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];
 450         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];
 451         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];
 452
 453         dest[0*linesize] = av_clip_uint8(dest[0*linesize] + ((t5 + t1) >> 7));
 454         dest[1*linesize] = av_clip_uint8(dest[1*linesize] + ((t6 + t2) >> 7));
 455         dest[2*linesize] = av_clip_uint8(dest[2*linesize] + ((t7 + t3) >> 7));
 456         dest[3*linesize] = av_clip_uint8(dest[3*linesize] + ((t8 + t4) >> 7));
 457         dest[4*linesize] = av_clip_uint8(dest[4*linesize] + ((t8 - t4 + 1) >> 7));
 458         dest[5*linesize] = av_clip_uint8(dest[5*linesize] + ((t7 - t3 + 1) >> 7));
 459         dest[6*linesize] = av_clip_uint8(dest[6*linesize] + ((t6 - t2 + 1) >> 7));
 460         dest[7*linesize] = av_clip_uint8(dest[7*linesize] + ((t5 - t1 + 1) >> 7));
 461
 462         src ++;
 463         dest++;
 464     }
 465 }
 466
 467 /** Do inverse transform on 4x4 part of block
 468 */
 469 static void vc1_inv_trans_4x4_dc_c(uint8_t *dest, int linesize, int16_t *block)
 470 {
 471     int i;
 472     int dc = block[0];
 473     dc = (17 * dc +  4) >> 3;
 474     dc = (17 * dc + 64) >> 7;
 475     for(i = 0; i < 4; i++){
 476         dest[0] = av_clip_uint8(dest[0] + dc);
 477         dest[1] = av_clip_uint8(dest[1] + dc);
 478         dest[2] = av_clip_uint8(dest[2] + dc);
 479         dest[3] = av_clip_uint8(dest[3] + dc);
 480         dest += linesize;
 481     }
 482 }
 483
 484 static void vc1_inv_trans_4x4_c(uint8_t *dest, int linesize, int16_t *block)
 485 {
 486     int i;
 487     register int t1,t2,t3,t4;
 488     int16_t *src, *dst;
 489
 490     src = block;
 491     dst = block;
 492     for(i = 0; i < 4; i++){
 493         t1 = 17 * (src[0] + src[2]) + 4;
 494         t2 = 17 * (src[0] - src[2]) + 4;
 495         t3 = 22 * src[1] + 10 * src[3];
 496         t4 = 22 * src[3] - 10 * src[1];
 497
 498         dst[0] = (t1 + t3) >> 3;
 499         dst[1] = (t2 - t4) >> 3;
 500         dst[2] = (t2 + t4) >> 3;
 501         dst[3] = (t1 - t3) >> 3;
 502
 503         src += 8;
 504         dst += 8;
 505     }
 506
 507     src = block;
 508     for(i = 0; i < 4; i++){
 509         t1 = 17 * (src[ 0] + src[16]) + 64;
 510         t2 = 17 * (src[ 0] - src[16]) + 64;
 511         t3 = 22 * src[ 8] + 10 * src[24];
 512         t4 = 22 * src[24] - 10 * src[ 8];
 513
 514         dest[0*linesize] = av_clip_uint8(dest[0*linesize] + ((t1 + t3) >> 7));
 515         dest[1*linesize] = av_clip_uint8(dest[1*linesize] + ((t2 - t4) >> 7));
 516         dest[2*linesize] = av_clip_uint8(dest[2*linesize] + ((t2 + t4) >> 7));
 517         dest[3*linesize] = av_clip_uint8(dest[3*linesize] + ((t1 - t3) >> 7));
 518
 519         src ++;
 520         dest++;
 521     }
 522 }
 523
 524 /* motion compensation functions */
 525 /** Filter in case of 2 filters */
 526 #define VC1_MSPEL_FILTER_16B(DIR, TYPE)                                 \
 527 static av_always_inline int vc1_mspel_ ## DIR ## _filter_16bits(const TYPE *src, int stride, int mode) \
 528 {                                                                       \
 529     switch(mode){                                                       \
 530     case 0: /* no shift - should not occur */                           \
 531         return 0;                                                       \
 532     case 1: /* 1/4 shift */                                             \
 533         return -4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2]; \
 534     case 2: /* 1/2 shift */                                             \
 535         return -src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2]; \
 536     case 3: /* 3/4 shift */                                             \
 537         return -3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2]; \
 538     }                                                                   \
 539     return 0; /* should not occur */                                    \
 540 }
 541
 542 VC1_MSPEL_FILTER_16B(ver, uint8_t)
 543 VC1_MSPEL_FILTER_16B(hor, int16_t)
 544
 545
 546 /** Filter used to interpolate fractional pel values
 547  */
 548 static av_always_inline int vc1_mspel_filter(const uint8_t *src, int stride, int mode, int r)
 549 {
 550     switch(mode){
 551     case 0: //no shift
 552         return src[0];
 553     case 1: // 1/4 shift
 554         return (-4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2] + 32 - r) >> 6;
 555     case 2: // 1/2 shift
 556         return (-src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2] + 8 - r) >> 4;
 557     case 3: // 3/4 shift
 558         return (-3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2] + 32 - r) >> 6;
 559     }
 560     return 0; //should not occur
 561 }
 562
 563 /** Function used to do motion compensation with bicubic interpolation
 564  */
 565 #define VC1_MSPEL_MC(OP, OPNAME)\
 566 static av_always_inline void OPNAME ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride, int hmode, int vmode, int rnd)\
 567 {\
 568     int     i, j;\
 569 \
 570     if (vmode) { /* Horizontal filter to apply */\
 571         int r;\
 572 \
 573         if (hmode) { /* Vertical filter to apply, output to tmp */\
 574             static const int shift_value[] = { 0, 5, 1, 5 };\
 575             int              shift = (shift_value[hmode]+shift_value[vmode])>>1;\
 576             int16_t          tmp[11*8], *tptr = tmp;\
 577 \
 578             r = (1<<(shift-1)) + rnd-1;\
 579 \
 580             src -= 1;\
 581             for(j = 0; j < 8; j++) {\
 582                 for(i = 0; i < 11; i++)\
 583                     tptr[i] = (vc1_mspel_ver_filter_16bits(src + i, stride, vmode)+r)>>shift;\
 584                 src += stride;\
 585                 tptr += 11;\
 586             }\
 587 \
 588             r = 64-rnd;\
 589             tptr = tmp+1;\
 590             for(j = 0; j < 8; j++) {\
 591                 for(i = 0; i < 8; i++)\
 592                     OP(dst[i], (vc1_mspel_hor_filter_16bits(tptr + i, 1, hmode)+r)>>7);\
 593                 dst += stride;\
 594                 tptr += 11;\
 595             }\
 596 \
 597             return;\
 598         }\
 599         else { /* No horizontal filter, output 8 lines to dst */\
 600             r = 1-rnd;\
 601 \
 602             for(j = 0; j < 8; j++) {\
 603                 for(i = 0; i < 8; i++)\
 604                     OP(dst[i], vc1_mspel_filter(src + i, stride, vmode, r));\
 605                 src += stride;\
 606                 dst += stride;\
 607             }\
 608             return;\
 609         }\
 610     }\
 611 \
 612     /* Horizontal mode with no vertical mode */\
 613     for(j = 0; j < 8; j++) {\
 614         for(i = 0; i < 8; i++)\
 615             OP(dst[i], vc1_mspel_filter(src + i, 1, hmode, rnd));\
 616         dst += stride;\
 617         src += stride;\
 618     }\
 619 }
 620
 621 #define op_put(a, b) a = av_clip_uint8(b)
 622 #define op_avg(a, b) a = (a + av_clip_uint8(b) + 1) >> 1
 623
 624 VC1_MSPEL_MC(op_put, put_)
 625 VC1_MSPEL_MC(op_avg, avg_)
 626
 627 /* pixel functions - really are entry points to vc1_mspel_mc */
 628
 629 #define PUT_VC1_MSPEL(a, b)\
 630 static void put_vc1_mspel_mc ## a ## b ##_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
 631      put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                         \
 632 }\
 633 static void avg_vc1_mspel_mc ## a ## b ##_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \
 634      avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                         \
 635 }
 636
 637 PUT_VC1_MSPEL(1, 0)
 638 PUT_VC1_MSPEL(2, 0)
 639 PUT_VC1_MSPEL(3, 0)
 640
 641 PUT_VC1_MSPEL(0, 1)
 642 PUT_VC1_MSPEL(1, 1)
 643 PUT_VC1_MSPEL(2, 1)
 644 PUT_VC1_MSPEL(3, 1)
 645
 646 PUT_VC1_MSPEL(0, 2)
 647 PUT_VC1_MSPEL(1, 2)
 648 PUT_VC1_MSPEL(2, 2)
 649 PUT_VC1_MSPEL(3, 2)
 650
 651 PUT_VC1_MSPEL(0, 3)
 652 PUT_VC1_MSPEL(1, 3)
 653 PUT_VC1_MSPEL(2, 3)
 654 PUT_VC1_MSPEL(3, 3)
 655
 656 static void put_no_rnd_vc1_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
 657     const int A=(8-x)*(8-y);
 658     const int B=(  x)*(8-y);
 659     const int C=(8-x)*(  y);
 660     const int D=(  x)*(  y);
 661     int i;
 662
 663     assert(x<8 && y<8 && x>=0 && y>=0);
 664
 665     for(i=0; i<h; i++)
 666     {
 667         dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6;
 668         dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6;
 669         dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6;
 670         dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6;
 671         dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6;
 672         dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6;
 673         dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6;
 674         dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6;
 675         dst+= stride;
 676         src+= stride;
 677     }
 678 }
 679
 680 static void put_no_rnd_vc1_chroma_mc4_c(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y){
 681     const int A=(8-x)*(8-y);
 682     const int B=(  x)*(8-y);
 683     const int C=(8-x)*(  y);
 684     const int D=(  x)*(  y);
 685     int i;
 686
 687     assert(x<8 && y<8 && x>=0 && y>=0);
 688
 689     for(i=0; i<h; i++)
 690     {
 691         dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6;
 692         dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6;
 693         dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6;
 694         dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6;
 695         dst+= stride;
 696         src+= stride;
 697     }
 698 }
 699
 700 #define avg2(a,b) ((a+b+1)>>1)
 701 static void avg_no_rnd_vc1_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){
 702     const int A=(8-x)*(8-y);
 703     const int B=(  x)*(8-y);
 704     const int C=(8-x)*(  y);
 705     const int D=(  x)*(  y);
 706     int i;
 707
 708     assert(x<8 && y<8 && x>=0 && y>=0);
 709
 710     for(i=0; i<h; i++)
 711     {
 712         dst[0] = avg2(dst[0], ((A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6));
 713         dst[1] = avg2(dst[1], ((A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6));
 714         dst[2] = avg2(dst[2], ((A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6));
 715         dst[3] = avg2(dst[3], ((A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6));
 716         dst[4] = avg2(dst[4], ((A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6));
 717         dst[5] = avg2(dst[5], ((A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6));
 718         dst[6] = avg2(dst[6], ((A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6));
 719         dst[7] = avg2(dst[7], ((A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6));
 720         dst+= stride;
 721         src+= stride;
 722     }
 723 }
 724
 725 #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER
 726
 727 static void sprite_h_c(uint8_t *dst, const uint8_t *src, int offset, int advance, int count)
 728 {
 729     while (count--) {
 730         int a = src[(offset >> 16)    ];
 731         int b = src[(offset >> 16) + 1];
 732         *dst++ = a + ((b - a) * (offset&0xFFFF) >> 16);
 733         offset += advance;
 734     }
 735 }
 736
 737 static av_always_inline void sprite_v_template(uint8_t *dst, const uint8_t *src1a, const uint8_t *src1b, int offset1,
 738                                             int two_sprites, const uint8_t *src2a, const uint8_t *src2b, int offset2,
 739                                             int alpha, int scaled, int width)
 740 {
 741     int a1, b1, a2, b2;
 742     while (width--) {
 743         a1 = *src1a++;
 744         if (scaled) {
 745             b1 = *src1b++;
 746             a1 = a1 + ((b1 - a1) * offset1 >> 16);
 747         }
 748         if (two_sprites) {
 749             a2 = *src2a++;
 750             if (scaled > 1) {
 751                 b2 = *src2b++;
 752                 a2 = a2 + ((b2 - a2) * offset2 >> 16);
 753             }
 754             a1 = a1 + ((a2 - a1) * alpha >> 16);
 755         }
 756         *dst++ = a1;
 757     }
 758 }
 759
 760 static void sprite_v_single_c(uint8_t *dst, const uint8_t *src1a, const uint8_t *src1b, int offset, int width)
 761 {
 762     sprite_v_template(dst, src1a, src1b, offset, 0, NULL, NULL, 0, 0, 1, width);
 763 }
 764
 765 static void sprite_v_double_noscale_c(uint8_t *dst, const uint8_t *src1a, const uint8_t *src2a, int alpha, int width)
 766 {
 767     sprite_v_template(dst, src1a, NULL, 0, 1, src2a, NULL, 0, alpha, 0, width);
 768 }
 769
 770 static void sprite_v_double_onescale_c(uint8_t *dst, const uint8_t *src1a, const uint8_t *src1b, int offset1,
 771                                                      const uint8_t *src2a, int alpha, int width)
 772 {
 773     sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, NULL, 0, alpha, 1, width);
 774 }
 775
 776 static void sprite_v_double_twoscale_c(uint8_t *dst, const uint8_t *src1a, const uint8_t *src1b, int offset1,
 777                                                      const uint8_t *src2a, const uint8_t *src2b, int offset2,
 778                                        int alpha, int width)
 779 {
 780     sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, src2b, offset2, alpha, 2, width);
 781 }
 782
 783 #endif
 784
 785 av_cold void ff_vc1dsp_init(VC1DSPContext* dsp) {
 786     dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;
 787     dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;
 788     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;
 789     dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;
 790     dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;
 791     dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;
 792     dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;
 793     dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;
 794     dsp->vc1_h_overlap = vc1_h_overlap_c;
 795     dsp->vc1_v_overlap = vc1_v_overlap_c;
 796     dsp->vc1_h_s_overlap = vc1_h_s_overlap_c;
 797     dsp->vc1_v_s_overlap = vc1_v_s_overlap_c;
 798     dsp->vc1_v_loop_filter4 = vc1_v_loop_filter4_c;
 799     dsp->vc1_h_loop_filter4 = vc1_h_loop_filter4_c;
 800     dsp->vc1_v_loop_filter8 = vc1_v_loop_filter8_c;
 801     dsp->vc1_h_loop_filter8 = vc1_h_loop_filter8_c;
 802     dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_c;
 803     dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_c;
 804
 805     dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_pixels8x8_c;
 806     dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_c;
 807     dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_c;
 808     dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_c;
 809     dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_c;
 810     dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_c;
 811     dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_c;
 812     dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_c;
 813     dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_c;
 814     dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_c;
 815     dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_c;
 816     dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_c;
 817     dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_c;
 818     dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_c;
 819     dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_c;
 820     dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_c;
 821
 822     dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_pixels8x8_c;
 823     dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_c;
 824     dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_c;
 825     dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_c;
 826     dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_c;
 827     dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_c;
 828     dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_c;
 829     dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_c;
 830     dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_c;
 831     dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_c;
 832     dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_c;
 833     dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_c;
 834     dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_c;
 835     dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_c;
 836     dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_c;
 837     dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_c;
 838
 839     dsp->put_no_rnd_vc1_chroma_pixels_tab[0]= put_no_rnd_vc1_chroma_mc8_c;
 840     dsp->avg_no_rnd_vc1_chroma_pixels_tab[0]= avg_no_rnd_vc1_chroma_mc8_c;
 841     dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = put_no_rnd_vc1_chroma_mc4_c;
 842
 843 #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER
 844     dsp->sprite_h = sprite_h_c;
 845     dsp->sprite_v_single = sprite_v_single_c;
 846     dsp->sprite_v_double_noscale = sprite_v_double_noscale_c;
 847     dsp->sprite_v_double_onescale = sprite_v_double_onescale_c;
 848     dsp->sprite_v_double_twoscale = sprite_v_double_twoscale_c;
 849 #endif
 850
 851     if (HAVE_ALTIVEC)
 852         ff_vc1dsp_init_altivec(dsp);
 853     if (ARCH_X86)
 854         ff_vc1dsp_init_x86(dsp);
 855 }