git.sesse.net Git - ffmpeg/blob - libavcodec/vp8dsp.c

   1 /*
   2  * Copyright (C) 2010 David Conrad
   3  * Copyright (C) 2010 Ronald S. Bultje
   4  * Copyright (C) 2014 Peter Ross
   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  * VP8 compatible video decoder
  26  */
  27
  28 #include "libavutil/common.h"
  29 #include "libavutil/intreadwrite.h"
  30
  31 #include "mathops.h"
  32 #include "vp8dsp.h"
  33
  34 #define MK_IDCT_DC_ADD4_C(name)                                               \
  35 static void name ## _idct_dc_add4uv_c(uint8_t *dst, int16_t block[4][16],     \
  36                                       ptrdiff_t stride)                       \
  37 {                                                                             \
  38     name ## _idct_dc_add_c(dst + stride * 0 + 0, block[0], stride);           \
  39     name ## _idct_dc_add_c(dst + stride * 0 + 4, block[1], stride);           \
  40     name ## _idct_dc_add_c(dst + stride * 4 + 0, block[2], stride);           \
  41     name ## _idct_dc_add_c(dst + stride * 4 + 4, block[3], stride);           \
  42 }                                                                             \
  43                                                                               \
  44 static void name ## _idct_dc_add4y_c(uint8_t *dst, int16_t block[4][16],      \
  45                                      ptrdiff_t stride)                        \
  46 {                                                                             \
  47     name ## _idct_dc_add_c(dst +  0, block[0], stride);                       \
  48     name ## _idct_dc_add_c(dst +  4, block[1], stride);                       \
  49     name ## _idct_dc_add_c(dst +  8, block[2], stride);                       \
  50     name ## _idct_dc_add_c(dst + 12, block[3], stride);                       \
  51 }
  52
  53 #if CONFIG_VP7_DECODER
  54 static void vp7_luma_dc_wht_c(int16_t block[4][4][16], int16_t dc[16])
  55 {
  56     int i;
  57     unsigned a1, b1, c1, d1;
  58     int16_t tmp[16];
  59
  60     for (i = 0; i < 4; i++) {
  61         a1 = (dc[i * 4 + 0] + dc[i * 4 + 2]) * 23170;
  62         b1 = (dc[i * 4 + 0] - dc[i * 4 + 2]) * 23170;
  63         c1 = dc[i * 4 + 1] * 12540 - dc[i * 4 + 3] * 30274;
  64         d1 = dc[i * 4 + 1] * 30274 + dc[i * 4 + 3] * 12540;
  65         tmp[i * 4 + 0] = (int)(a1 + d1) >> 14;
  66         tmp[i * 4 + 3] = (int)(a1 - d1) >> 14;
  67         tmp[i * 4 + 1] = (int)(b1 + c1) >> 14;
  68         tmp[i * 4 + 2] = (int)(b1 - c1) >> 14;
  69     }
  70
  71     for (i = 0; i < 4; i++) {
  72         a1 = (tmp[i + 0] + tmp[i + 8]) * 23170;
  73         b1 = (tmp[i + 0] - tmp[i + 8]) * 23170;
  74         c1 = tmp[i + 4] * 12540 - tmp[i + 12] * 30274;
  75         d1 = tmp[i + 4] * 30274 + tmp[i + 12] * 12540;
  76         AV_ZERO64(dc + i * 4);
  77         block[0][i][0] = (int)(a1 + d1 + 0x20000) >> 18;
  78         block[3][i][0] = (int)(a1 - d1 + 0x20000) >> 18;
  79         block[1][i][0] = (int)(b1 + c1 + 0x20000) >> 18;
  80         block[2][i][0] = (int)(b1 - c1 + 0x20000) >> 18;
  81     }
  82 }
  83
  84 static void vp7_luma_dc_wht_dc_c(int16_t block[4][4][16], int16_t dc[16])
  85 {
  86     int i, val = (23170 * (23170 * dc[0] >> 14) + 0x20000) >> 18;
  87     dc[0] = 0;
  88
  89     for (i = 0; i < 4; i++) {
  90         block[i][0][0] = val;
  91         block[i][1][0] = val;
  92         block[i][2][0] = val;
  93         block[i][3][0] = val;
  94     }
  95 }
  96
  97 static void vp7_idct_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
  98 {
  99     int i;
 100     unsigned a1, b1, c1, d1;
 101     int16_t tmp[16];
 102
 103     for (i = 0; i < 4; i++) {
 104         a1 = (block[i * 4 + 0] + block[i * 4 + 2]) * 23170;
 105         b1 = (block[i * 4 + 0] - block[i * 4 + 2]) * 23170;
 106         c1 = block[i * 4 + 1] * 12540 - block[i * 4 + 3] * 30274;
 107         d1 = block[i * 4 + 1] * 30274 + block[i * 4 + 3] * 12540;
 108         AV_ZERO64(block + i * 4);
 109         tmp[i * 4 + 0] = (int)(a1 + d1) >> 14;
 110         tmp[i * 4 + 3] = (int)(a1 - d1) >> 14;
 111         tmp[i * 4 + 1] = (int)(b1 + c1) >> 14;
 112         tmp[i * 4 + 2] = (int)(b1 - c1) >> 14;
 113     }
 114
 115     for (i = 0; i < 4; i++) {
 116         a1 = (tmp[i + 0] + tmp[i + 8]) * 23170;
 117         b1 = (tmp[i + 0] - tmp[i + 8]) * 23170;
 118         c1 = tmp[i + 4] * 12540 - tmp[i + 12] * 30274;
 119         d1 = tmp[i + 4] * 30274 + tmp[i + 12] * 12540;
 120         dst[0 * stride + i] = av_clip_uint8(dst[0 * stride + i] +
 121                                             ((int)(a1 + d1 + 0x20000) >> 18));
 122         dst[3 * stride + i] = av_clip_uint8(dst[3 * stride + i] +
 123                                             ((int)(a1 - d1 + 0x20000) >> 18));
 124         dst[1 * stride + i] = av_clip_uint8(dst[1 * stride + i] +
 125                                             ((int)(b1 + c1 + 0x20000) >> 18));
 126         dst[2 * stride + i] = av_clip_uint8(dst[2 * stride + i] +
 127                                             ((int)(b1 - c1 + 0x20000) >> 18));
 128     }
 129 }
 130
 131 static void vp7_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
 132 {
 133     int i, dc = (23170 * (23170 * block[0] >> 14) + 0x20000) >> 18;
 134     block[0] = 0;
 135
 136     for (i = 0; i < 4; i++) {
 137         dst[0] = av_clip_uint8(dst[0] + dc);
 138         dst[1] = av_clip_uint8(dst[1] + dc);
 139         dst[2] = av_clip_uint8(dst[2] + dc);
 140         dst[3] = av_clip_uint8(dst[3] + dc);
 141         dst   += stride;
 142     }
 143 }
 144
 145 MK_IDCT_DC_ADD4_C(vp7)
 146 #endif /* CONFIG_VP7_DECODER */
 147
 148 // TODO: Maybe add dequant
 149 #if CONFIG_VP8_DECODER
 150 static void vp8_luma_dc_wht_c(int16_t block[4][4][16], int16_t dc[16])
 151 {
 152     int i, t0, t1, t2, t3;
 153
 154     for (i = 0; i < 4; i++) {
 155         t0 = dc[0 * 4 + i] + dc[3 * 4 + i];
 156         t1 = dc[1 * 4 + i] + dc[2 * 4 + i];
 157         t2 = dc[1 * 4 + i] - dc[2 * 4 + i];
 158         t3 = dc[0 * 4 + i] - dc[3 * 4 + i];
 159
 160         dc[0 * 4 + i] = t0 + t1;
 161         dc[1 * 4 + i] = t3 + t2;
 162         dc[2 * 4 + i] = t0 - t1;
 163         dc[3 * 4 + i] = t3 - t2;
 164     }
 165
 166     for (i = 0; i < 4; i++) {
 167         t0 = dc[i * 4 + 0] + dc[i * 4 + 3] + 3; // rounding
 168         t1 = dc[i * 4 + 1] + dc[i * 4 + 2];
 169         t2 = dc[i * 4 + 1] - dc[i * 4 + 2];
 170         t3 = dc[i * 4 + 0] - dc[i * 4 + 3] + 3; // rounding
 171         AV_ZERO64(dc + i * 4);
 172
 173         block[i][0][0] = (t0 + t1) >> 3;
 174         block[i][1][0] = (t3 + t2) >> 3;
 175         block[i][2][0] = (t0 - t1) >> 3;
 176         block[i][3][0] = (t3 - t2) >> 3;
 177     }
 178 }
 179
 180 static void vp8_luma_dc_wht_dc_c(int16_t block[4][4][16], int16_t dc[16])
 181 {
 182     int i, val = (dc[0] + 3) >> 3;
 183     dc[0] = 0;
 184
 185     for (i = 0; i < 4; i++) {
 186         block[i][0][0] = val;
 187         block[i][1][0] = val;
 188         block[i][2][0] = val;
 189         block[i][3][0] = val;
 190     }
 191 }
 192
 193 #define MUL_20091(a) ((((a) * 20091) >> 16) + (a))
 194 #define MUL_35468(a)  (((a) * 35468) >> 16)
 195
 196 static void vp8_idct_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
 197 {
 198     int i, t0, t1, t2, t3;
 199     int16_t tmp[16];
 200
 201     for (i = 0; i < 4; i++) {
 202         t0 = block[0 * 4 + i] + block[2 * 4 + i];
 203         t1 = block[0 * 4 + i] - block[2 * 4 + i];
 204         t2 = MUL_35468(block[1 * 4 + i]) - MUL_20091(block[3 * 4 + i]);
 205         t3 = MUL_20091(block[1 * 4 + i]) + MUL_35468(block[3 * 4 + i]);
 206         block[0 * 4 + i] = 0;
 207         block[1 * 4 + i] = 0;
 208         block[2 * 4 + i] = 0;
 209         block[3 * 4 + i] = 0;
 210
 211         tmp[i * 4 + 0] = t0 + t3;
 212         tmp[i * 4 + 1] = t1 + t2;
 213         tmp[i * 4 + 2] = t1 - t2;
 214         tmp[i * 4 + 3] = t0 - t3;
 215     }
 216
 217     for (i = 0; i < 4; i++) {
 218         t0 = tmp[0 * 4 + i] + tmp[2 * 4 + i];
 219         t1 = tmp[0 * 4 + i] - tmp[2 * 4 + i];
 220         t2 = MUL_35468(tmp[1 * 4 + i]) - MUL_20091(tmp[3 * 4 + i]);
 221         t3 = MUL_20091(tmp[1 * 4 + i]) + MUL_35468(tmp[3 * 4 + i]);
 222
 223         dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
 224         dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
 225         dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
 226         dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
 227         dst   += stride;
 228     }
 229 }
 230
 231 static void vp8_idct_dc_add_c(uint8_t *dst, int16_t block[16], ptrdiff_t stride)
 232 {
 233     int i, dc = (block[0] + 4) >> 3;
 234     block[0] = 0;
 235
 236     for (i = 0; i < 4; i++) {
 237         dst[0] = av_clip_uint8(dst[0] + dc);
 238         dst[1] = av_clip_uint8(dst[1] + dc);
 239         dst[2] = av_clip_uint8(dst[2] + dc);
 240         dst[3] = av_clip_uint8(dst[3] + dc);
 241         dst   += stride;
 242     }
 243 }
 244
 245 MK_IDCT_DC_ADD4_C(vp8)
 246 #endif /* CONFIG_VP8_DECODER */
 247
 248 // because I like only having two parameters to pass functions...
 249 #define LOAD_PIXELS                                                           \
 250     int av_unused p3 = p[-4 * stride];                                        \
 251     int av_unused p2 = p[-3 * stride];                                        \
 252     int av_unused p1 = p[-2 * stride];                                        \
 253     int av_unused p0 = p[-1 * stride];                                        \
 254     int av_unused q0 = p[ 0 * stride];                                        \
 255     int av_unused q1 = p[ 1 * stride];                                        \
 256     int av_unused q2 = p[ 2 * stride];                                        \
 257     int av_unused q3 = p[ 3 * stride];
 258
 259 #define clip_int8(n) (cm[(n) + 0x80] - 0x80)
 260
 261 static av_always_inline void filter_common(uint8_t *p, ptrdiff_t stride,
 262                                            int is4tap, int is_vp7)
 263 {
 264     LOAD_PIXELS
 265     int a, f1, f2;
 266     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
 267
 268     a = 3 * (q0 - p0);
 269
 270     if (is4tap)
 271         a += clip_int8(p1 - q1);
 272
 273     a = clip_int8(a);
 274
 275     // We deviate from the spec here with c(a+3) >> 3
 276     // since that's what libvpx does.
 277     f1 = FFMIN(a + 4, 127) >> 3;
 278
 279     if (is_vp7)
 280         f2 = f1 - ((a & 7) == 4);
 281     else
 282         f2 = FFMIN(a + 3, 127) >> 3;
 283
 284     // Despite what the spec says, we do need to clamp here to
 285     // be bitexact with libvpx.
 286     p[-1 * stride] = cm[p0 + f2];
 287     p[ 0 * stride] = cm[q0 - f1];
 288
 289     // only used for _inner on blocks without high edge variance
 290     if (!is4tap) {
 291         a              = (f1 + 1) >> 1;
 292         p[-2 * stride] = cm[p1 + a];
 293         p[ 1 * stride] = cm[q1 - a];
 294     }
 295 }
 296
 297 static av_always_inline void vp7_filter_common(uint8_t *p, ptrdiff_t stride,
 298                                                int is4tap)
 299 {
 300     filter_common(p, stride, is4tap, IS_VP7);
 301 }
 302
 303 static av_always_inline void vp8_filter_common(uint8_t *p, ptrdiff_t stride,
 304                                                int is4tap)
 305 {
 306     filter_common(p, stride, is4tap, IS_VP8);
 307 }
 308
 309 static av_always_inline int vp7_simple_limit(uint8_t *p, ptrdiff_t stride,
 310                                              int flim)
 311 {
 312     LOAD_PIXELS
 313     return FFABS(p0 - q0) <= flim;
 314 }
 315
 316 static av_always_inline int vp8_simple_limit(uint8_t *p, ptrdiff_t stride,
 317                                              int flim)
 318 {
 319     LOAD_PIXELS
 320     return 2 * FFABS(p0 - q0) + (FFABS(p1 - q1) >> 1) <= flim;
 321 }
 322
 323 /**
 324  * E - limit at the macroblock edge
 325  * I - limit for interior difference
 326  */
 327 #define NORMAL_LIMIT(vpn)                                                     \
 328 static av_always_inline int vp ## vpn ## _normal_limit(uint8_t *p,            \
 329                                                        ptrdiff_t stride,      \
 330                                                        int E, int I)          \
 331 {                                                                             \
 332     LOAD_PIXELS                                                               \
 333     return vp ## vpn ## _simple_limit(p, stride, E) &&                        \
 334            FFABS(p3 - p2) <= I && FFABS(p2 - p1) <= I &&                      \
 335            FFABS(p1 - p0) <= I && FFABS(q3 - q2) <= I &&                      \
 336            FFABS(q2 - q1) <= I && FFABS(q1 - q0) <= I;                        \
 337 }
 338
 339 NORMAL_LIMIT(7)
 340 NORMAL_LIMIT(8)
 341
 342 // high edge variance
 343 static av_always_inline int hev(uint8_t *p, ptrdiff_t stride, int thresh)
 344 {
 345     LOAD_PIXELS
 346     return FFABS(p1 - p0) > thresh || FFABS(q1 - q0) > thresh;
 347 }
 348
 349 static av_always_inline void filter_mbedge(uint8_t *p, ptrdiff_t stride)
 350 {
 351     int a0, a1, a2, w;
 352     const uint8_t *cm = ff_crop_tab + MAX_NEG_CROP;
 353
 354     LOAD_PIXELS
 355
 356     w = clip_int8(p1 - q1);
 357     w = clip_int8(w + 3 * (q0 - p0));
 358
 359     a0 = (27 * w + 63) >> 7;
 360     a1 = (18 * w + 63) >> 7;
 361     a2 =  (9 * w + 63) >> 7;
 362
 363     p[-3 * stride] = cm[p2 + a2];
 364     p[-2 * stride] = cm[p1 + a1];
 365     p[-1 * stride] = cm[p0 + a0];
 366     p[ 0 * stride] = cm[q0 - a0];
 367     p[ 1 * stride] = cm[q1 - a1];
 368     p[ 2 * stride] = cm[q2 - a2];
 369 }
 370
 371 #define LOOP_FILTER(vpn, dir, size, stridea, strideb, maybe_inline)           \
 372 static maybe_inline                                                           \
 373 void vpn ## _ ## dir ## _loop_filter ## size ## _c(uint8_t *dst,              \
 374                                                    ptrdiff_t stride,          \
 375                                                    int flim_E, int flim_I,    \
 376                                                    int hev_thresh)            \
 377 {                                                                             \
 378     int i;                                                                    \
 379     for (i = 0; i < size; i++)                                                \
 380         if (vpn ## _normal_limit(dst + i * stridea, strideb,                  \
 381                                  flim_E, flim_I)) {                           \
 382             if (hev(dst + i * stridea, strideb, hev_thresh))                  \
 383                 vpn ## _filter_common(dst + i * stridea, strideb, 1);         \
 384             else                                                              \
 385                 filter_mbedge(dst + i * stridea, strideb);                    \
 386         }                                                                     \
 387 }                                                                             \
 388                                                                               \
 389 static maybe_inline                                                           \
 390 void vpn ## _ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst,        \
 391                                                          ptrdiff_t stride,    \
 392                                                          int flim_E,          \
 393                                                          int flim_I,          \
 394                                                          int hev_thresh)      \
 395 {                                                                             \
 396     int i;                                                                    \
 397     for (i = 0; i < size; i++)                                                \
 398         if (vpn ## _normal_limit(dst + i * stridea, strideb,                  \
 399                                  flim_E, flim_I)) {                           \
 400             int hv = hev(dst + i * stridea, strideb, hev_thresh);             \
 401             if (hv)                                                           \
 402                 vpn ## _filter_common(dst + i * stridea, strideb, 1);         \
 403             else                                                              \
 404                 vpn ## _filter_common(dst + i * stridea, strideb, 0);         \
 405         }                                                                     \
 406 }
 407
 408 #define UV_LOOP_FILTER(vpn, dir, stridea, strideb)                            \
 409 LOOP_FILTER(vpn, dir, 8, stridea, strideb, av_always_inline)                  \
 410 static void vpn ## _ ## dir ## _loop_filter8uv_c(uint8_t *dstU,               \
 411                                                  uint8_t *dstV,               \
 412                                                  ptrdiff_t stride, int fE,    \
 413                                                  int fI, int hev_thresh)      \
 414 {                                                                             \
 415     vpn ## _ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);     \
 416     vpn ## _ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);     \
 417 }                                                                             \
 418                                                                               \
 419 static void vpn ## _ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU,         \
 420                                                        uint8_t *dstV,         \
 421                                                        ptrdiff_t stride,      \
 422                                                        int fE, int fI,        \
 423                                                        int hev_thresh)        \
 424 {                                                                             \
 425     vpn ## _ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI,            \
 426                                              hev_thresh);                     \
 427     vpn ## _ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI,            \
 428                                              hev_thresh);                     \
 429 }
 430
 431 #define LOOP_FILTER_SIMPLE(vpn)                                               \
 432 static void vpn ## _v_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride,    \
 433                                            int flim)                          \
 434 {                                                                             \
 435     int i;                                                                    \
 436     for (i = 0; i < 16; i++)                                                  \
 437         if (vpn ## _simple_limit(dst + i, stride, flim))                      \
 438             vpn ## _filter_common(dst + i, stride, 1);                        \
 439 }                                                                             \
 440                                                                               \
 441 static void vpn ## _h_loop_filter_simple_c(uint8_t *dst, ptrdiff_t stride,    \
 442                                            int flim)                          \
 443 {                                                                             \
 444     int i;                                                                    \
 445     for (i = 0; i < 16; i++)                                                  \
 446         if (vpn ## _simple_limit(dst + i * stride, 1, flim))                  \
 447             vpn ## _filter_common(dst + i * stride, 1, 1);                    \
 448 }
 449
 450 #define LOOP_FILTERS(vpn)                \
 451     LOOP_FILTER(vpn, v, 16, 1, stride, ) \
 452     LOOP_FILTER(vpn, h, 16, stride, 1, ) \
 453     UV_LOOP_FILTER(vpn, v, 1, stride)    \
 454     UV_LOOP_FILTER(vpn, h, stride, 1)    \
 455     LOOP_FILTER_SIMPLE(vpn)              \
 456
 457 static const uint8_t subpel_filters[7][6] = {
 458     { 0,  6, 123,  12,  1, 0 },
 459     { 2, 11, 108,  36,  8, 1 },
 460     { 0,  9,  93,  50,  6, 0 },
 461     { 3, 16,  77,  77, 16, 3 },
 462     { 0,  6,  50,  93,  9, 0 },
 463     { 1,  8,  36, 108, 11, 2 },
 464     { 0,  1,  12, 123,  6, 0 },
 465 };
 466
 467 #define PUT_PIXELS(WIDTH)                                                     \
 468 static void put_vp8_pixels ## WIDTH ## _c(uint8_t *dst, ptrdiff_t dststride,  \
 469                                           uint8_t *src, ptrdiff_t srcstride,  \
 470                                           int h, int x, int y)                \
 471 {                                                                             \
 472     int i;                                                                    \
 473     for (i = 0; i < h; i++, dst += dststride, src += srcstride)               \
 474         memcpy(dst, src, WIDTH);                                              \
 475 }
 476
 477 PUT_PIXELS(16)
 478 PUT_PIXELS(8)
 479 PUT_PIXELS(4)
 480
 481 #define FILTER_6TAP(src, F, stride)                                           \
 482     cm[(F[2] * src[x + 0 * stride] - F[1] * src[x - 1 * stride] +             \
 483         F[0] * src[x - 2 * stride] + F[3] * src[x + 1 * stride] -             \
 484         F[4] * src[x + 2 * stride] + F[5] * src[x + 3 * stride] + 64) >> 7]
 485
 486 #define FILTER_4TAP(src, F, stride)                                           \
 487     cm[(F[2] * src[x + 0 * stride] - F[1] * src[x - 1 * stride] +             \
 488         F[3] * src[x + 1 * stride] - F[4] * src[x + 2 * stride] + 64) >> 7]
 489
 490 #define VP8_EPEL_H(SIZE, TAPS)                                                \
 491 static void put_vp8_epel ## SIZE ## _h ## TAPS ## _c(uint8_t *dst,            \
 492                                                      ptrdiff_t dststride,     \
 493                                                      uint8_t *src,            \
 494                                                      ptrdiff_t srcstride,     \
 495                                                      int h, int mx, int my)   \
 496 {                                                                             \
 497     const uint8_t *filter = subpel_filters[mx - 1];                           \
 498     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
 499     int x, y;                                                                 \
 500     for (y = 0; y < h; y++) {                                                 \
 501         for (x = 0; x < SIZE; x++)                                            \
 502             dst[x] = FILTER_ ## TAPS ## TAP(src, filter, 1);                  \
 503         dst += dststride;                                                     \
 504         src += srcstride;                                                     \
 505     }                                                                         \
 506 }
 507
 508 #define VP8_EPEL_V(SIZE, TAPS)                                                \
 509 static void put_vp8_epel ## SIZE ## _v ## TAPS ## _c(uint8_t *dst,            \
 510                                                      ptrdiff_t dststride,     \
 511                                                      uint8_t *src,            \
 512                                                      ptrdiff_t srcstride,     \
 513                                                      int h, int mx, int my)   \
 514 {                                                                             \
 515     const uint8_t *filter = subpel_filters[my - 1];                           \
 516     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
 517     int x, y;                                                                 \
 518     for (y = 0; y < h; y++) {                                                 \
 519         for (x = 0; x < SIZE; x++)                                            \
 520             dst[x] = FILTER_ ## TAPS ## TAP(src, filter, srcstride);          \
 521         dst += dststride;                                                     \
 522         src += srcstride;                                                     \
 523     }                                                                         \
 524 }
 525
 526 #define VP8_EPEL_HV(SIZE, HTAPS, VTAPS)                                       \
 527 static void                                                                   \
 528 put_vp8_epel ## SIZE ## _h ## HTAPS ## v ## VTAPS ## _c(uint8_t *dst,         \
 529                                                         ptrdiff_t dststride,  \
 530                                                         uint8_t *src,         \
 531                                                         ptrdiff_t srcstride,  \
 532                                                         int h, int mx,        \
 533                                                         int my)               \
 534 {                                                                             \
 535     const uint8_t *filter = subpel_filters[mx - 1];                           \
 536     const uint8_t *cm     = ff_crop_tab + MAX_NEG_CROP;                       \
 537     int x, y;                                                                 \
 538     uint8_t tmp_array[(2 * SIZE + VTAPS - 1) * SIZE];                         \
 539     uint8_t *tmp = tmp_array;                                                 \
 540     src -= (2 - (VTAPS == 4)) * srcstride;                                    \
 541                                                                               \
 542     for (y = 0; y < h + VTAPS - 1; y++) {                                     \
 543         for (x = 0; x < SIZE; x++)                                            \
 544             tmp[x] = FILTER_ ## HTAPS ## TAP(src, filter, 1);                 \
 545         tmp += SIZE;                                                          \
 546         src += srcstride;                                                     \
 547     }                                                                         \
 548     tmp    = tmp_array + (2 - (VTAPS == 4)) * SIZE;                           \
 549     filter = subpel_filters[my - 1];                                          \
 550                                                                               \
 551     for (y = 0; y < h; y++) {                                                 \
 552         for (x = 0; x < SIZE; x++)                                            \
 553             dst[x] = FILTER_ ## VTAPS ## TAP(tmp, filter, SIZE);              \
 554         dst += dststride;                                                     \
 555         tmp += SIZE;                                                          \
 556     }                                                                         \
 557 }
 558
 559 VP8_EPEL_H(16, 4)
 560 VP8_EPEL_H(8,  4)
 561 VP8_EPEL_H(4,  4)
 562 VP8_EPEL_H(16, 6)
 563 VP8_EPEL_H(8,  6)
 564 VP8_EPEL_H(4,  6)
 565 VP8_EPEL_V(16, 4)
 566 VP8_EPEL_V(8,  4)
 567 VP8_EPEL_V(4,  4)
 568 VP8_EPEL_V(16, 6)
 569 VP8_EPEL_V(8,  6)
 570 VP8_EPEL_V(4,  6)
 571
 572 VP8_EPEL_HV(16, 4, 4)
 573 VP8_EPEL_HV(8,  4, 4)
 574 VP8_EPEL_HV(4,  4, 4)
 575 VP8_EPEL_HV(16, 4, 6)
 576 VP8_EPEL_HV(8,  4, 6)
 577 VP8_EPEL_HV(4,  4, 6)
 578 VP8_EPEL_HV(16, 6, 4)
 579 VP8_EPEL_HV(8,  6, 4)
 580 VP8_EPEL_HV(4,  6, 4)
 581 VP8_EPEL_HV(16, 6, 6)
 582 VP8_EPEL_HV(8,  6, 6)
 583 VP8_EPEL_HV(4,  6, 6)
 584
 585 #define VP8_BILINEAR(SIZE)                                                    \
 586 static void put_vp8_bilinear ## SIZE ## _h_c(uint8_t *dst, ptrdiff_t dstride, \
 587                                              uint8_t *src, ptrdiff_t sstride, \
 588                                              int h, int mx, int my)           \
 589 {                                                                             \
 590     int a = 8 - mx, b = mx;                                                   \
 591     int x, y;                                                                 \
 592     for (y = 0; y < h; y++) {                                                 \
 593         for (x = 0; x < SIZE; x++)                                            \
 594             dst[x] = (a * src[x] + b * src[x + 1] + 4) >> 3;                  \
 595         dst += dstride;                                                       \
 596         src += sstride;                                                       \
 597     }                                                                         \
 598 }                                                                             \
 599                                                                               \
 600 static void put_vp8_bilinear ## SIZE ## _v_c(uint8_t *dst, ptrdiff_t dstride, \
 601                                              uint8_t *src, ptrdiff_t sstride, \
 602                                              int h, int mx, int my)           \
 603 {                                                                             \
 604     int c = 8 - my, d = my;                                                   \
 605     int x, y;                                                                 \
 606     for (y = 0; y < h; y++) {                                                 \
 607         for (x = 0; x < SIZE; x++)                                            \
 608             dst[x] = (c * src[x] + d * src[x + sstride] + 4) >> 3;            \
 609         dst += dstride;                                                       \
 610         src += sstride;                                                       \
 611     }                                                                         \
 612 }                                                                             \
 613                                                                               \
 614 static void put_vp8_bilinear ## SIZE ## _hv_c(uint8_t *dst,                   \
 615                                               ptrdiff_t dstride,              \
 616                                               uint8_t *src,                   \
 617                                               ptrdiff_t sstride,              \
 618                                               int h, int mx, int my)          \
 619 {                                                                             \
 620     int a = 8 - mx, b = mx;                                                   \
 621     int c = 8 - my, d = my;                                                   \
 622     int x, y;                                                                 \
 623     uint8_t tmp_array[(2 * SIZE + 1) * SIZE];                                 \
 624     uint8_t *tmp = tmp_array;                                                 \
 625     for (y = 0; y < h + 1; y++) {                                             \
 626         for (x = 0; x < SIZE; x++)                                            \
 627             tmp[x] = (a * src[x] + b * src[x + 1] + 4) >> 3;                  \
 628         tmp += SIZE;                                                          \
 629         src += sstride;                                                       \
 630     }                                                                         \
 631     tmp = tmp_array;                                                          \
 632     for (y = 0; y < h; y++) {                                                 \
 633         for (x = 0; x < SIZE; x++)                                            \
 634             dst[x] = (c * tmp[x] + d * tmp[x + SIZE] + 4) >> 3;               \
 635         dst += dstride;                                                       \
 636         tmp += SIZE;                                                          \
 637     }                                                                         \
 638 }
 639
 640 VP8_BILINEAR(16)
 641 VP8_BILINEAR(8)
 642 VP8_BILINEAR(4)
 643
 644 #define VP78_MC_FUNC(IDX, SIZE)                                               \
 645     dsp->put_vp8_epel_pixels_tab[IDX][0][0] = put_vp8_pixels ## SIZE ## _c;   \
 646     dsp->put_vp8_epel_pixels_tab[IDX][0][1] = put_vp8_epel ## SIZE ## _h4_c;  \
 647     dsp->put_vp8_epel_pixels_tab[IDX][0][2] = put_vp8_epel ## SIZE ## _h6_c;  \
 648     dsp->put_vp8_epel_pixels_tab[IDX][1][0] = put_vp8_epel ## SIZE ## _v4_c;  \
 649     dsp->put_vp8_epel_pixels_tab[IDX][1][1] = put_vp8_epel ## SIZE ## _h4v4_c; \
 650     dsp->put_vp8_epel_pixels_tab[IDX][1][2] = put_vp8_epel ## SIZE ## _h6v4_c; \
 651     dsp->put_vp8_epel_pixels_tab[IDX][2][0] = put_vp8_epel ## SIZE ## _v6_c;  \
 652     dsp->put_vp8_epel_pixels_tab[IDX][2][1] = put_vp8_epel ## SIZE ## _h4v6_c; \
 653     dsp->put_vp8_epel_pixels_tab[IDX][2][2] = put_vp8_epel ## SIZE ## _h6v6_c
 654
 655 #define VP78_BILINEAR_MC_FUNC(IDX, SIZE)                                      \
 656     dsp->put_vp8_bilinear_pixels_tab[IDX][0][0] = put_vp8_pixels   ## SIZE ## _c; \
 657     dsp->put_vp8_bilinear_pixels_tab[IDX][0][1] = put_vp8_bilinear ## SIZE ## _h_c; \
 658     dsp->put_vp8_bilinear_pixels_tab[IDX][0][2] = put_vp8_bilinear ## SIZE ## _h_c; \
 659     dsp->put_vp8_bilinear_pixels_tab[IDX][1][0] = put_vp8_bilinear ## SIZE ## _v_c; \
 660     dsp->put_vp8_bilinear_pixels_tab[IDX][1][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
 661     dsp->put_vp8_bilinear_pixels_tab[IDX][1][2] = put_vp8_bilinear ## SIZE ## _hv_c; \
 662     dsp->put_vp8_bilinear_pixels_tab[IDX][2][0] = put_vp8_bilinear ## SIZE ## _v_c; \
 663     dsp->put_vp8_bilinear_pixels_tab[IDX][2][1] = put_vp8_bilinear ## SIZE ## _hv_c; \
 664     dsp->put_vp8_bilinear_pixels_tab[IDX][2][2] = put_vp8_bilinear ## SIZE ## _hv_c
 665
 666 av_cold void ff_vp78dsp_init(VP8DSPContext *dsp)
 667 {
 668     VP78_MC_FUNC(0, 16);
 669     VP78_MC_FUNC(1, 8);
 670     VP78_MC_FUNC(2, 4);
 671
 672     VP78_BILINEAR_MC_FUNC(0, 16);
 673     VP78_BILINEAR_MC_FUNC(1, 8);
 674     VP78_BILINEAR_MC_FUNC(2, 4);
 675
 676     if (ARCH_AARCH64)
 677         ff_vp78dsp_init_aarch64(dsp);
 678     if (ARCH_ARM)
 679         ff_vp78dsp_init_arm(dsp);
 680     if (ARCH_PPC)
 681         ff_vp78dsp_init_ppc(dsp);
 682     if (ARCH_X86)
 683         ff_vp78dsp_init_x86(dsp);
 684 }
 685
 686 #if CONFIG_VP7_DECODER
 687 LOOP_FILTERS(vp7)
 688
 689 av_cold void ff_vp7dsp_init(VP8DSPContext *dsp)
 690 {
 691     dsp->vp8_luma_dc_wht    = vp7_luma_dc_wht_c;
 692     dsp->vp8_luma_dc_wht_dc = vp7_luma_dc_wht_dc_c;
 693     dsp->vp8_idct_add       = vp7_idct_add_c;
 694     dsp->vp8_idct_dc_add    = vp7_idct_dc_add_c;
 695     dsp->vp8_idct_dc_add4y  = vp7_idct_dc_add4y_c;
 696     dsp->vp8_idct_dc_add4uv = vp7_idct_dc_add4uv_c;
 697
 698     dsp->vp8_v_loop_filter16y = vp7_v_loop_filter16_c;
 699     dsp->vp8_h_loop_filter16y = vp7_h_loop_filter16_c;
 700     dsp->vp8_v_loop_filter8uv = vp7_v_loop_filter8uv_c;
 701     dsp->vp8_h_loop_filter8uv = vp7_h_loop_filter8uv_c;
 702
 703     dsp->vp8_v_loop_filter16y_inner = vp7_v_loop_filter16_inner_c;
 704     dsp->vp8_h_loop_filter16y_inner = vp7_h_loop_filter16_inner_c;
 705     dsp->vp8_v_loop_filter8uv_inner = vp7_v_loop_filter8uv_inner_c;
 706     dsp->vp8_h_loop_filter8uv_inner = vp7_h_loop_filter8uv_inner_c;
 707
 708     dsp->vp8_v_loop_filter_simple = vp7_v_loop_filter_simple_c;
 709     dsp->vp8_h_loop_filter_simple = vp7_h_loop_filter_simple_c;
 710 }
 711 #endif /* CONFIG_VP7_DECODER */
 712
 713 #if CONFIG_VP8_DECODER
 714 LOOP_FILTERS(vp8)
 715
 716 av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
 717 {
 718     dsp->vp8_luma_dc_wht    = vp8_luma_dc_wht_c;
 719     dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c;
 720     dsp->vp8_idct_add       = vp8_idct_add_c;
 721     dsp->vp8_idct_dc_add    = vp8_idct_dc_add_c;
 722     dsp->vp8_idct_dc_add4y  = vp8_idct_dc_add4y_c;
 723     dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c;
 724
 725     dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
 726     dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
 727     dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
 728     dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
 729
 730     dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
 731     dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
 732     dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
 733     dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
 734
 735     dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
 736     dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
 737
 738     if (ARCH_AARCH64)
 739         ff_vp8dsp_init_aarch64(dsp);
 740     if (ARCH_ARM)
 741         ff_vp8dsp_init_arm(dsp);
 742     if (ARCH_X86)
 743         ff_vp8dsp_init_x86(dsp);
 744     if (ARCH_MIPS)
 745         ff_vp8dsp_init_mips(dsp);
 746 }
 747 #endif /* CONFIG_VP8_DECODER */