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

   1 /*
   2  * HEVC video decoder
   3  *
   4  * Copyright (C) 2012 - 2013 Guillaume Martres
   5  *
   6  * This file is part of Libav.
   7  *
   8  * Libav 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  * Libav 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 Libav; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 #include "get_bits.h"
  24 #include "hevcdec.h"
  25
  26 #include "bit_depth_template.c"
  27
  28 static void FUNC(put_pcm)(uint8_t *_dst, ptrdiff_t stride, int size,
  29                           GetBitContext *gb, int pcm_bit_depth)
  30 {
  31     int x, y;
  32     pixel *dst = (pixel *)_dst;
  33
  34     stride /= sizeof(pixel);
  35
  36     for (y = 0; y < size; y++) {
  37         for (x = 0; x < size; x++)
  38             dst[x] = get_bits(gb, pcm_bit_depth) << (BIT_DEPTH - pcm_bit_depth);
  39         dst += stride;
  40     }
  41 }
  42
  43 static av_always_inline void FUNC(add_residual)(uint8_t *_dst, int16_t *res,
  44                                                 ptrdiff_t stride, int size)
  45 {
  46     int x, y;
  47     pixel *dst = (pixel *)_dst;
  48
  49     stride /= sizeof(pixel);
  50
  51     for (y = 0; y < size; y++) {
  52         for (x = 0; x < size; x++) {
  53             dst[x] = av_clip_pixel(dst[x] + *res);
  54             res++;
  55         }
  56         dst += stride;
  57     }
  58 }
  59
  60 static void FUNC(add_residual4x4)(uint8_t *_dst, int16_t *res,
  61                                   ptrdiff_t stride)
  62 {
  63     FUNC(add_residual)(_dst, res, stride, 4);
  64 }
  65
  66 static void FUNC(add_residual8x8)(uint8_t *_dst, int16_t *res,
  67                                   ptrdiff_t stride)
  68 {
  69     FUNC(add_residual)(_dst, res, stride, 8);
  70 }
  71
  72 static void FUNC(add_residual16x16)(uint8_t *_dst, int16_t *res,
  73                                     ptrdiff_t stride)
  74 {
  75     FUNC(add_residual)(_dst, res, stride, 16);
  76 }
  77
  78 static void FUNC(add_residual32x32)(uint8_t *_dst, int16_t *res,
  79                                     ptrdiff_t stride)
  80 {
  81     FUNC(add_residual)(_dst, res, stride, 32);
  82 }
  83
  84 static void FUNC(dequant)(int16_t *coeffs)
  85 {
  86     int shift  = 13 - BIT_DEPTH;
  87 #if BIT_DEPTH <= 13
  88     int offset = 1 << (shift - 1);
  89 #else
  90     int offset = 0;
  91 #endif
  92     int x, y;
  93
  94     for (y = 0; y < 4 * 4; y += 4) {
  95         for (x = 0; x < 4; x++)
  96             coeffs[y + x] = (coeffs[y + x] + offset) >> shift;
  97     }
  98 }
  99
 100 #define SET(dst, x)   (dst) = (x)
 101 #define SCALE(dst, x) (dst) = av_clip_int16(((x) + add) >> shift)
 102
 103 #define TR_4x4_LUMA(dst, src, step, assign)                             \
 104     do {                                                                \
 105         int c0 = src[0 * step] + src[2 * step];                         \
 106         int c1 = src[2 * step] + src[3 * step];                         \
 107         int c2 = src[0 * step] - src[3 * step];                         \
 108         int c3 = 74 * src[1 * step];                                    \
 109                                                                         \
 110         assign(dst[2 * step], 74 * (src[0 * step] -                     \
 111                                     src[2 * step] +                     \
 112                                     src[3 * step]));                    \
 113         assign(dst[0 * step], 29 * c0 + 55 * c1 + c3);                  \
 114         assign(dst[1 * step], 55 * c2 - 29 * c1 + c3);                  \
 115         assign(dst[3 * step], 55 * c0 + 29 * c2 - c3);                  \
 116     } while (0)
 117
 118 static void FUNC(transform_4x4_luma)(int16_t *coeffs)
 119 {
 120     int i;
 121     int shift    = 7;
 122     int add      = 1 << (shift - 1);
 123     int16_t *src = coeffs;
 124
 125     for (i = 0; i < 4; i++) {
 126         TR_4x4_LUMA(src, src, 4, SCALE);
 127         src++;
 128     }
 129
 130     shift = 20 - BIT_DEPTH;
 131     add   = 1 << (shift - 1);
 132     for (i = 0; i < 4; i++) {
 133         TR_4x4_LUMA(coeffs, coeffs, 1, SCALE);
 134         coeffs += 4;
 135     }
 136 }
 137
 138 #undef TR_4x4_LUMA
 139
 140 #define TR_4(dst, src, dstep, sstep, assign, end)                       \
 141     do {                                                                \
 142         const int e0 = transform[8 * 0][0] * src[0 * sstep] +           \
 143                        transform[8 * 2][0] * src[2 * sstep];            \
 144         const int e1 = transform[8 * 0][1] * src[0 * sstep] +           \
 145                        transform[8 * 2][1] * src[2 * sstep];            \
 146         const int o0 = transform[8 * 1][0] * src[1 * sstep] +           \
 147                        transform[8 * 3][0] * src[3 * sstep];            \
 148         const int o1 = transform[8 * 1][1] * src[1 * sstep] +           \
 149                        transform[8 * 3][1] * src[3 * sstep];            \
 150                                                                         \
 151         assign(dst[0 * dstep], e0 + o0);                                \
 152         assign(dst[1 * dstep], e1 + o1);                                \
 153         assign(dst[2 * dstep], e1 - o1);                                \
 154         assign(dst[3 * dstep], e0 - o0);                                \
 155     } while (0)
 156
 157 #define TR_8(dst, src, dstep, sstep, assign, end)                 \
 158     do {                                                          \
 159         int i, j;                                                 \
 160         int e_8[4];                                               \
 161         int o_8[4] = { 0 };                                       \
 162         for (i = 0; i < 4; i++)                                   \
 163             for (j = 1; j < end; j += 2)                          \
 164                 o_8[i] += transform[4 * j][i] * src[j * sstep];   \
 165         TR_4(e_8, src, 1, 2 * sstep, SET, 4);                     \
 166                                                                   \
 167         for (i = 0; i < 4; i++) {                                 \
 168             assign(dst[i * dstep], e_8[i] + o_8[i]);              \
 169             assign(dst[(7 - i) * dstep], e_8[i] - o_8[i]);        \
 170         }                                                         \
 171     } while (0)
 172
 173 #define TR_16(dst, src, dstep, sstep, assign, end)                \
 174     do {                                                          \
 175         int i, j;                                                 \
 176         int e_16[8];                                              \
 177         int o_16[8] = { 0 };                                      \
 178         for (i = 0; i < 8; i++)                                   \
 179             for (j = 1; j < end; j += 2)                          \
 180                 o_16[i] += transform[2 * j][i] * src[j * sstep];  \
 181         TR_8(e_16, src, 1, 2 * sstep, SET, 8);                    \
 182                                                                   \
 183         for (i = 0; i < 8; i++) {                                 \
 184             assign(dst[i * dstep], e_16[i] + o_16[i]);            \
 185             assign(dst[(15 - i) * dstep], e_16[i] - o_16[i]);     \
 186         }                                                         \
 187     } while (0)
 188
 189 #define TR_32(dst, src, dstep, sstep, assign, end)                \
 190     do {                                                          \
 191         int i, j;                                                 \
 192         int e_32[16];                                             \
 193         int o_32[16] = { 0 };                                     \
 194         for (i = 0; i < 16; i++)                                  \
 195             for (j = 1; j < end; j += 2)                          \
 196                 o_32[i] += transform[j][i] * src[j * sstep];      \
 197         TR_16(e_32, src, 1, 2 * sstep, SET, end / 2);             \
 198                                                                   \
 199         for (i = 0; i < 16; i++) {                                \
 200             assign(dst[i * dstep], e_32[i] + o_32[i]);            \
 201             assign(dst[(31 - i) * dstep], e_32[i] - o_32[i]);     \
 202         }                                                         \
 203     } while (0)
 204
 205 #define IDCT_VAR4(H)                                              \
 206     int limit2 = FFMIN(col_limit + 4, H)
 207 #define IDCT_VAR8(H)                                              \
 208     int limit  = FFMIN(col_limit, H);                             \
 209     int limit2 = FFMIN(col_limit + 4, H)
 210 #define IDCT_VAR16(H)   IDCT_VAR8(H)
 211 #define IDCT_VAR32(H)   IDCT_VAR8(H)
 212
 213 #define IDCT(H)                                                   \
 214 static void FUNC(idct_ ## H ## x ## H )(int16_t *coeffs,          \
 215                                         int col_limit)            \
 216 {                                                                 \
 217     int i;                                                        \
 218     int      shift = 7;                                           \
 219     int      add   = 1 << (shift - 1);                            \
 220     int16_t *src   = coeffs;                                      \
 221     IDCT_VAR ## H(H);                                             \
 222                                                                   \
 223     for (i = 0; i < H; i++) {                                     \
 224         TR_ ## H(src, src, H, H, SCALE, limit2);                  \
 225         if (limit2 < H && i%4 == 0 && !!i)                        \
 226             limit2 -= 4;                                          \
 227         src++;                                                    \
 228     }                                                             \
 229                                                                   \
 230     shift = 20 - BIT_DEPTH;                                       \
 231     add   = 1 << (shift - 1);                                     \
 232     for (i = 0; i < H; i++) {                                     \
 233         TR_ ## H(coeffs, coeffs, 1, 1, SCALE, limit);             \
 234         coeffs += H;                                              \
 235     }                                                             \
 236 }
 237
 238 #define IDCT_DC(H)                                                \
 239 static void FUNC(idct_ ## H ## x ## H ## _dc)(int16_t *coeffs)    \
 240 {                                                                 \
 241     int i, j;                                                     \
 242     int shift = 14 - BIT_DEPTH;                                   \
 243     int add   = 1 << (shift - 1);                                 \
 244     int coeff = (((coeffs[0] + 1) >> 1) + add) >> shift;          \
 245                                                                   \
 246     for (j = 0; j < H; j++) {                                     \
 247         for (i = 0; i < H; i++) {                                 \
 248             coeffs[i + j * H] = coeff;                            \
 249         }                                                         \
 250     }                                                             \
 251 }
 252
 253 IDCT( 4)
 254 IDCT( 8)
 255 IDCT(16)
 256 IDCT(32)
 257 IDCT_DC( 4)
 258 IDCT_DC( 8)
 259 IDCT_DC(16)
 260 IDCT_DC(32)
 261 #undef TR_4
 262 #undef TR_8
 263 #undef TR_16
 264 #undef TR_32
 265
 266 #undef SET
 267 #undef SCALE
 268 #undef ADD_AND_SCALE
 269
 270 static void FUNC(sao_band_filter)(uint8_t *_dst, uint8_t *_src,
 271                                   ptrdiff_t stride, SAOParams *sao,
 272                                   int *borders, int width, int height,
 273                                   int c_idx, int class)
 274 {
 275     pixel *dst = (pixel *)_dst;
 276     pixel *src = (pixel *)_src;
 277     int offset_table[32] = { 0 };
 278     int k, y, x;
 279     int chroma = !!c_idx;
 280     int shift  = BIT_DEPTH - 5;
 281     int *sao_offset_val = sao->offset_val[c_idx];
 282     int sao_left_class  = sao->band_position[c_idx];
 283     int init_y = 0, init_x = 0;
 284
 285     stride /= sizeof(pixel);
 286
 287     switch (class) {
 288     case 0:
 289         if (!borders[2])
 290             width -= (8 >> chroma) + 2;
 291         if (!borders[3])
 292             height -= (4 >> chroma) + 2;
 293         break;
 294     case 1:
 295         init_y = -(4 >> chroma) - 2;
 296         if (!borders[2])
 297             width -= (8 >> chroma) + 2;
 298         height = (4 >> chroma) + 2;
 299         break;
 300     case 2:
 301         init_x = -(8 >> chroma) - 2;
 302         width  =  (8 >> chroma) + 2;
 303         if (!borders[3])
 304             height -= (4 >> chroma) + 2;
 305         break;
 306     case 3:
 307         init_y = -(4 >> chroma) - 2;
 308         init_x = -(8 >> chroma) - 2;
 309         width  =  (8 >> chroma) + 2;
 310         height =  (4 >> chroma) + 2;
 311         break;
 312     }
 313
 314     dst = dst + (init_y * stride + init_x);
 315     src = src + (init_y * stride + init_x);
 316     for (k = 0; k < 4; k++)
 317         offset_table[(k + sao_left_class) & 31] = sao_offset_val[k + 1];
 318     for (y = 0; y < height; y++) {
 319         for (x = 0; x < width; x++)
 320             dst[x] = av_clip_pixel(src[x] + offset_table[src[x] >> shift]);
 321         dst += stride;
 322         src += stride;
 323     }
 324 }
 325
 326 static void FUNC(sao_band_filter_0)(uint8_t *dst, uint8_t *src,
 327                                     ptrdiff_t stride, SAOParams *sao,
 328                                     int *borders, int width, int height,
 329                                     int c_idx)
 330 {
 331     FUNC(sao_band_filter)(dst, src, stride, sao, borders,
 332                           width, height, c_idx, 0);
 333 }
 334
 335 static void FUNC(sao_band_filter_1)(uint8_t *dst, uint8_t *src,
 336                                     ptrdiff_t stride, SAOParams *sao,
 337                                     int *borders, int width, int height,
 338                                     int c_idx)
 339 {
 340     FUNC(sao_band_filter)(dst, src, stride, sao, borders,
 341                           width, height, c_idx, 1);
 342 }
 343
 344 static void FUNC(sao_band_filter_2)(uint8_t *dst, uint8_t *src,
 345                                     ptrdiff_t stride, SAOParams *sao,
 346                                     int *borders, int width, int height,
 347                                     int c_idx)
 348 {
 349     FUNC(sao_band_filter)(dst, src, stride, sao, borders,
 350                           width, height, c_idx, 2);
 351 }
 352
 353 static void FUNC(sao_band_filter_3)(uint8_t *_dst, uint8_t *_src,
 354                                     ptrdiff_t stride, SAOParams *sao,
 355                                     int *borders, int width, int height,
 356                                     int c_idx)
 357 {
 358     FUNC(sao_band_filter)(_dst, _src, stride, sao, borders,
 359                           width, height, c_idx, 3);
 360 }
 361
 362 static void FUNC(sao_edge_filter_0)(uint8_t *_dst, uint8_t *_src,
 363                                     ptrdiff_t stride, SAOParams *sao,
 364                                     int *borders, int _width, int _height,
 365                                     int c_idx, uint8_t vert_edge,
 366                                     uint8_t horiz_edge, uint8_t diag_edge)
 367 {
 368     int x, y;
 369     pixel *dst = (pixel *)_dst;
 370     pixel *src = (pixel *)_src;
 371     int chroma = !!c_idx;
 372     int *sao_offset_val = sao->offset_val[c_idx];
 373     int sao_eo_class    = sao->eo_class[c_idx];
 374     int init_x = 0, init_y = 0, width = _width, height = _height;
 375
 376     static const int8_t pos[4][2][2] = {
 377         { { -1,  0 }, {  1, 0 } }, // horizontal
 378         { {  0, -1 }, {  0, 1 } }, // vertical
 379         { { -1, -1 }, {  1, 1 } }, // 45 degree
 380         { {  1, -1 }, { -1, 1 } }, // 135 degree
 381     };
 382     static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
 383
 384 #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
 385
 386     stride /= sizeof(pixel);
 387
 388     if (!borders[2])
 389         width -= (8 >> chroma) + 2;
 390     if (!borders[3])
 391         height -= (4 >> chroma) + 2;
 392
 393     dst = dst + (init_y * stride + init_x);
 394     src = src + (init_y * stride + init_x);
 395     init_y = init_x = 0;
 396     if (sao_eo_class != SAO_EO_VERT) {
 397         if (borders[0]) {
 398             int offset_val = sao_offset_val[0];
 399             ptrdiff_t y_stride   = 0;
 400             for (y = 0; y < height; y++) {
 401                 dst[y_stride] = av_clip_pixel(src[y_stride] + offset_val);
 402                 y_stride     += stride;
 403             }
 404             init_x = 1;
 405         }
 406         if (borders[2]) {
 407             int offset_val = sao_offset_val[0];
 408             ptrdiff_t x_stride = width - 1;
 409             for (x = 0; x < height; x++) {
 410                 dst[x_stride] = av_clip_pixel(src[x_stride] + offset_val);
 411                 x_stride     += stride;
 412             }
 413             width--;
 414         }
 415     }
 416     if (sao_eo_class != SAO_EO_HORIZ) {
 417         if (borders[1]) {
 418             int offset_val = sao_offset_val[0];
 419             for (x = init_x; x < width; x++)
 420                 dst[x] = av_clip_pixel(src[x] + offset_val);
 421             init_y = 1;
 422         }
 423         if (borders[3]) {
 424             int offset_val = sao_offset_val[0];
 425             ptrdiff_t y_stride = stride * (height - 1);
 426             for (x = init_x; x < width; x++)
 427                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + offset_val);
 428             height--;
 429         }
 430     }
 431     {
 432         ptrdiff_t y_stride = init_y * stride;
 433         int pos_0_0  = pos[sao_eo_class][0][0];
 434         int pos_0_1  = pos[sao_eo_class][0][1];
 435         int pos_1_0  = pos[sao_eo_class][1][0];
 436         int pos_1_1  = pos[sao_eo_class][1][1];
 437
 438         ptrdiff_t y_stride_0_1 = (init_y + pos_0_1) * stride;
 439         ptrdiff_t y_stride_1_1 = (init_y + pos_1_1) * stride;
 440         for (y = init_y; y < height; y++) {
 441             for (x = init_x; x < width; x++) {
 442                 int diff0         = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
 443                 int diff1         = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
 444                 int offset_val    = edge_idx[2 + diff0 + diff1];
 445                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
 446             }
 447             y_stride     += stride;
 448             y_stride_0_1 += stride;
 449             y_stride_1_1 += stride;
 450         }
 451     }
 452
 453     {
 454         // Restore pixels that can't be modified
 455         int save_upper_left = !diag_edge && sao_eo_class == SAO_EO_135D && !borders[0] && !borders[1];
 456         if (vert_edge && sao_eo_class != SAO_EO_VERT)
 457             for (y = init_y+save_upper_left; y< height; y++)
 458                 dst[y*stride] = src[y*stride];
 459         if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
 460             for(x = init_x+save_upper_left; x<width; x++)
 461                 dst[x] = src[x];
 462         if(diag_edge && sao_eo_class == SAO_EO_135D)
 463             dst[0] = src[0];
 464     }
 465
 466 #undef CMP
 467 }
 468
 469 static void FUNC(sao_edge_filter_1)(uint8_t *_dst, uint8_t *_src,
 470                                     ptrdiff_t stride, SAOParams *sao,
 471                                     int *borders, int _width, int _height,
 472                                     int c_idx, uint8_t vert_edge,
 473                                     uint8_t horiz_edge, uint8_t diag_edge)
 474 {
 475     int x, y;
 476     pixel *dst = (pixel *)_dst;
 477     pixel *src = (pixel *)_src;
 478     int chroma = !!c_idx;
 479     int *sao_offset_val = sao->offset_val[c_idx];
 480     int sao_eo_class    = sao->eo_class[c_idx];
 481     int init_x = 0, init_y = 0, width = _width, height = _height;
 482
 483     static const int8_t pos[4][2][2] = {
 484         { { -1, 0  }, { 1,  0 } }, // horizontal
 485         { { 0,  -1 }, { 0,  1 } }, // vertical
 486         { { -1, -1 }, { 1,  1 } }, // 45 degree
 487         { { 1,  -1 }, { -1, 1 } }, // 135 degree
 488     };
 489     static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
 490
 491 #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
 492
 493     stride /= sizeof(pixel);
 494
 495     init_y = -(4 >> chroma) - 2;
 496     if (!borders[2])
 497         width -= (8 >> chroma) + 2;
 498     height = (4 >> chroma) + 2;
 499
 500     dst = dst + (init_y * stride + init_x);
 501     src = src + (init_y * stride + init_x);
 502     init_y = init_x = 0;
 503     if (sao_eo_class != SAO_EO_VERT) {
 504         if (borders[0]) {
 505             int offset_val = sao_offset_val[0];
 506             ptrdiff_t y_stride = 0;
 507             for (y = 0; y < height; y++) {
 508                 dst[y_stride] = av_clip_pixel(src[y_stride] + offset_val);
 509                 y_stride     += stride;
 510             }
 511             init_x = 1;
 512         }
 513         if (borders[2]) {
 514             int offset_val = sao_offset_val[0];
 515             ptrdiff_t x_stride = width - 1;
 516             for (x = 0; x < height; x++) {
 517                 dst[x_stride] = av_clip_pixel(src[x_stride] + offset_val);
 518                 x_stride     += stride;
 519             }
 520             width--;
 521         }
 522     }
 523     {
 524         ptrdiff_t y_stride = init_y * stride;
 525         int pos_0_0  = pos[sao_eo_class][0][0];
 526         int pos_0_1  = pos[sao_eo_class][0][1];
 527         int pos_1_0  = pos[sao_eo_class][1][0];
 528         int pos_1_1  = pos[sao_eo_class][1][1];
 529
 530         ptrdiff_t y_stride_0_1 = (init_y + pos_0_1) * stride;
 531         ptrdiff_t y_stride_1_1 = (init_y + pos_1_1) * stride;
 532         for (y = init_y; y < height; y++) {
 533             for (x = init_x; x < width; x++) {
 534                 int diff0         = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
 535                 int diff1         = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
 536                 int offset_val    = edge_idx[2 + diff0 + diff1];
 537                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
 538             }
 539             y_stride     += stride;
 540             y_stride_0_1 += stride;
 541             y_stride_1_1 += stride;
 542         }
 543     }
 544
 545     {
 546         // Restore pixels that can't be modified
 547         int save_lower_left = !diag_edge && sao_eo_class == SAO_EO_45D && !borders[0];
 548         if(vert_edge && sao_eo_class != SAO_EO_VERT)
 549             for(y = init_y; y< height-save_lower_left; y++)
 550                 dst[y*stride] = src[y*stride];
 551         if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
 552             for(x = init_x+save_lower_left; x<width; x++)
 553                 dst[(height-1)*stride+x] = src[(height-1)*stride+x];
 554         if(diag_edge && sao_eo_class == SAO_EO_45D)
 555             dst[stride*(height-1)] = src[stride*(height-1)];
 556     }
 557
 558 #undef CMP
 559 }
 560
 561 static void FUNC(sao_edge_filter_2)(uint8_t *_dst, uint8_t *_src,
 562                                     ptrdiff_t stride, SAOParams *sao,
 563                                     int *borders, int _width, int _height,
 564                                     int c_idx, uint8_t vert_edge,
 565                                     uint8_t horiz_edge, uint8_t diag_edge)
 566 {
 567     int x, y;
 568     pixel *dst = (pixel *)_dst;
 569     pixel *src = (pixel *)_src;
 570     int chroma = !!c_idx;
 571     int *sao_offset_val = sao->offset_val[c_idx];
 572     int sao_eo_class    = sao->eo_class[c_idx];
 573     int init_x = 0, init_y = 0, width = _width, height = _height;
 574
 575     static const int8_t pos[4][2][2] = {
 576         { { -1,  0 }, {  1, 0 } }, // horizontal
 577         { {  0, -1 }, {  0, 1 } }, // vertical
 578         { { -1, -1 }, {  1, 1 } }, // 45 degree
 579         { {  1, -1 }, { -1, 1 } }, // 135 degree
 580     };
 581     static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
 582
 583 #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
 584
 585     stride /= sizeof(pixel);
 586
 587     init_x = -(8 >> chroma) - 2;
 588     width  =  (8 >> chroma) + 2;
 589     if (!borders[3])
 590         height -= (4 >> chroma) + 2;
 591
 592     dst = dst + (init_y * stride + init_x);
 593     src = src + (init_y * stride + init_x);
 594     init_y = init_x = 0;
 595     if (sao_eo_class != SAO_EO_HORIZ) {
 596         if (borders[1]) {
 597             int offset_val = sao_offset_val[0];
 598             for (x = init_x; x < width; x++)
 599                 dst[x] = av_clip_pixel(src[x] + offset_val);
 600             init_y = 1;
 601         }
 602         if (borders[3]) {
 603             int offset_val = sao_offset_val[0];
 604             ptrdiff_t y_stride = stride * (height - 1);
 605             for (x = init_x; x < width; x++)
 606                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + offset_val);
 607             height--;
 608         }
 609     }
 610     {
 611         ptrdiff_t y_stride = init_y * stride;
 612         int pos_0_0  = pos[sao_eo_class][0][0];
 613         int pos_0_1  = pos[sao_eo_class][0][1];
 614         int pos_1_0  = pos[sao_eo_class][1][0];
 615         int pos_1_1  = pos[sao_eo_class][1][1];
 616
 617         ptrdiff_t y_stride_0_1 = (init_y + pos_0_1) * stride;
 618         ptrdiff_t y_stride_1_1 = (init_y + pos_1_1) * stride;
 619         for (y = init_y; y < height; y++) {
 620             for (x = init_x; x < width; x++) {
 621                 int diff0         = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
 622                 int diff1         = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
 623                 int offset_val    = edge_idx[2 + diff0 + diff1];
 624                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
 625             }
 626             y_stride     += stride;
 627             y_stride_0_1 += stride;
 628             y_stride_1_1 += stride;
 629         }
 630     }
 631
 632     {
 633         // Restore pixels that can't be modified
 634         int save_upper_right = !diag_edge && sao_eo_class == SAO_EO_45D && !borders[1];
 635         if(vert_edge && sao_eo_class != SAO_EO_VERT)
 636             for(y = init_y+save_upper_right; y< height; y++)
 637                 dst[y*stride+width-1] = src[y*stride+width-1];
 638         if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
 639             for(x = init_x; x<width-save_upper_right; x++)
 640                 dst[x] = src[x];
 641         if(diag_edge && sao_eo_class == SAO_EO_45D)
 642             dst[width-1] = src[width-1];
 643     }
 644 #undef CMP
 645 }
 646
 647 static void FUNC(sao_edge_filter_3)(uint8_t *_dst, uint8_t *_src,
 648                                     ptrdiff_t stride, SAOParams *sao,
 649                                     int *borders, int _width, int _height,
 650                                     int c_idx, uint8_t vert_edge,
 651                                     uint8_t horiz_edge, uint8_t diag_edge)
 652 {
 653     int x, y;
 654     pixel *dst = (pixel *)_dst;
 655     pixel *src = (pixel *)_src;
 656     int chroma = !!c_idx;
 657     int *sao_offset_val = sao->offset_val[c_idx];
 658     int sao_eo_class    = sao->eo_class[c_idx];
 659     int init_x = 0, init_y = 0, width = _width, height = _height;
 660
 661     static const int8_t pos[4][2][2] = {
 662         { { -1,  0 }, {  1, 0 } }, // horizontal
 663         { {  0, -1 }, {  0, 1 } }, // vertical
 664         { { -1, -1 }, {  1, 1 } }, // 45 degree
 665         { {  1, -1 }, { -1, 1 } }, // 135 degree
 666     };
 667     static const uint8_t edge_idx[] = { 1, 2, 0, 3, 4 };
 668
 669 #define CMP(a, b) ((a) > (b) ? 1 : ((a) == (b) ? 0 : -1))
 670
 671     stride /= sizeof(pixel);
 672
 673     init_y = -(4 >> chroma) - 2;
 674     init_x = -(8 >> chroma) - 2;
 675     width  =  (8 >> chroma) + 2;
 676     height =  (4 >> chroma) + 2;
 677
 678
 679     dst    = dst + (init_y * stride + init_x);
 680     src    = src + (init_y * stride + init_x);
 681     init_y = init_x = 0;
 682
 683     {
 684         ptrdiff_t y_stride = init_y * stride;
 685         int pos_0_0  = pos[sao_eo_class][0][0];
 686         int pos_0_1  = pos[sao_eo_class][0][1];
 687         int pos_1_0  = pos[sao_eo_class][1][0];
 688         int pos_1_1  = pos[sao_eo_class][1][1];
 689
 690         ptrdiff_t y_stride_0_1 = (init_y + pos_0_1) * stride;
 691         ptrdiff_t y_stride_1_1 = (init_y + pos_1_1) * stride;
 692
 693         for (y = init_y; y < height; y++) {
 694             for (x = init_x; x < width; x++) {
 695                 int diff0         = CMP(src[x + y_stride], src[x + pos_0_0 + y_stride_0_1]);
 696                 int diff1         = CMP(src[x + y_stride], src[x + pos_1_0 + y_stride_1_1]);
 697                 int offset_val    = edge_idx[2 + diff0 + diff1];
 698                 dst[x + y_stride] = av_clip_pixel(src[x + y_stride] + sao_offset_val[offset_val]);
 699             }
 700             y_stride     += stride;
 701             y_stride_0_1 += stride;
 702             y_stride_1_1 += stride;
 703         }
 704     }
 705
 706     {
 707         // Restore pixels that can't be modified
 708         int save_lower_right = !diag_edge && sao_eo_class == SAO_EO_135D;
 709         if(vert_edge && sao_eo_class != SAO_EO_VERT)
 710             for(y = init_y; y< height-save_lower_right; y++)
 711                 dst[y*stride+width-1] = src[y*stride+width-1];
 712         if(horiz_edge && sao_eo_class != SAO_EO_HORIZ)
 713             for(x = init_x; x<width-save_lower_right; x++)
 714                 dst[(height-1)*stride+x] = src[(height-1)*stride+x];
 715         if(diag_edge && sao_eo_class == SAO_EO_135D)
 716             dst[stride*(height-1)+width-1] = src[stride*(height-1)+width-1];
 717     }
 718 #undef CMP
 719 }
 720
 721 #undef SET
 722 #undef SCALE
 723 #undef TR_4
 724 #undef TR_8
 725 #undef TR_16
 726 #undef TR_32
 727
 728 static av_always_inline void
 729 FUNC(put_hevc_qpel_pixels)(int16_t *dst, ptrdiff_t dststride,
 730                            uint8_t *_src, ptrdiff_t _srcstride,
 731                            int width, int height, int mx, int my,
 732                            int16_t* mcbuffer)
 733 {
 734     int x, y;
 735     pixel *src          = (pixel *)_src;
 736     ptrdiff_t srcstride = _srcstride / sizeof(pixel);
 737
 738     dststride /= sizeof(*dst);
 739     for (y = 0; y < height; y++) {
 740         for (x = 0; x < width; x++)
 741             dst[x] = src[x] << (14 - BIT_DEPTH);
 742         src += srcstride;
 743         dst += dststride;
 744     }
 745 }
 746
 747 #define QPEL_FILTER_1(src, stride)      \
 748     (1 * -src[x - 3 * stride] +         \
 749      4 *  src[x - 2 * stride] -         \
 750     10 *  src[x -     stride] +         \
 751     58 *  src[x]              +         \
 752     17 *  src[x +     stride] -         \
 753      5 *  src[x + 2 * stride] +         \
 754      1 *  src[x + 3 * stride])
 755
 756 #define QPEL_FILTER_2(src, stride)      \
 757     (1  * -src[x - 3 * stride] +        \
 758      4  *  src[x - 2 * stride] -        \
 759     11  *  src[x -     stride] +        \
 760     40  *  src[x]              +        \
 761     40  *  src[x +     stride] -        \
 762     11  *  src[x + 2 * stride] +        \
 763      4  *  src[x + 3 * stride] -        \
 764      1  *  src[x + 4 * stride])
 765
 766 #define QPEL_FILTER_3(src, stride)      \
 767     (1  * src[x - 2 * stride] -         \
 768      5  * src[x -     stride] +         \
 769     17  * src[x]              +         \
 770     58  * src[x + stride]     -         \
 771     10  * src[x + 2 * stride] +         \
 772      4  * src[x + 3 * stride] -         \
 773      1  * src[x + 4 * stride])
 774
 775
 776 #define PUT_HEVC_QPEL_H(H)                                                     \
 777 static void FUNC(put_hevc_qpel_h ## H)(int16_t *dst,  ptrdiff_t dststride,     \
 778                                        uint8_t *_src, ptrdiff_t _srcstride,    \
 779                                        int width, int height,                  \
 780                                        int16_t* mcbuffer)                      \
 781 {                                                                              \
 782     int x, y;                                                                  \
 783     pixel *src = (pixel*)_src;                                                 \
 784     ptrdiff_t srcstride = _srcstride / sizeof(pixel);                          \
 785                                                                                \
 786     dststride /= sizeof(*dst);                                                 \
 787     for (y = 0; y < height; y++) {                                             \
 788         for (x = 0; x < width; x++)                                            \
 789             dst[x] = QPEL_FILTER_ ## H(src, 1) >> (BIT_DEPTH - 8);             \
 790         src += srcstride;                                                      \
 791         dst += dststride;                                                      \
 792     }                                                                          \
 793 }
 794
 795 #define PUT_HEVC_QPEL_V(V)                                                     \
 796 static void FUNC(put_hevc_qpel_v ## V)(int16_t *dst,  ptrdiff_t dststride,     \
 797                                        uint8_t *_src, ptrdiff_t _srcstride,    \
 798                                        int width, int height,                  \
 799                                        int16_t* mcbuffer)                      \
 800 {                                                                              \
 801     int x, y;                                                                  \
 802     pixel *src = (pixel*)_src;                                                 \
 803     ptrdiff_t srcstride = _srcstride / sizeof(pixel);                          \
 804                                                                                \
 805     dststride /= sizeof(*dst);                                                 \
 806     for (y = 0; y < height; y++)  {                                            \
 807         for (x = 0; x < width; x++)                                            \
 808             dst[x] = QPEL_FILTER_ ## V(src, srcstride) >> (BIT_DEPTH - 8);     \
 809         src += srcstride;                                                      \
 810         dst += dststride;                                                      \
 811     }                                                                          \
 812 }
 813
 814 #define PUT_HEVC_QPEL_HV(H, V)                                                 \
 815 static void FUNC(put_hevc_qpel_h ## H ## v ## V)(int16_t *dst,                 \
 816                                                  ptrdiff_t dststride,          \
 817                                                  uint8_t *_src,                \
 818                                                  ptrdiff_t _srcstride,         \
 819                                                  int width, int height,        \
 820                                                  int16_t* mcbuffer)            \
 821 {                                                                              \
 822     int x, y;                                                                  \
 823     pixel *src = (pixel*)_src;                                                 \
 824     ptrdiff_t srcstride = _srcstride / sizeof(pixel);                          \
 825                                                                                \
 826     int16_t tmp_array[(MAX_PB_SIZE + 7) * MAX_PB_SIZE];                        \
 827     int16_t *tmp = tmp_array;                                                  \
 828                                                                                \
 829     dststride /= sizeof(*dst);                                                 \
 830     src -= ff_hevc_qpel_extra_before[V] * srcstride;                           \
 831                                                                                \
 832     for (y = 0; y < height + ff_hevc_qpel_extra[V]; y++) {                     \
 833         for (x = 0; x < width; x++)                                            \
 834             tmp[x] = QPEL_FILTER_ ## H(src, 1) >> (BIT_DEPTH - 8);             \
 835         src += srcstride;                                                      \
 836         tmp += MAX_PB_SIZE;                                                    \
 837     }                                                                          \
 838                                                                                \
 839     tmp = tmp_array + ff_hevc_qpel_extra_before[V] * MAX_PB_SIZE;              \
 840                                                                                \
 841     for (y = 0; y < height; y++) {                                             \
 842         for (x = 0; x < width; x++)                                            \
 843             dst[x] = QPEL_FILTER_ ## V(tmp, MAX_PB_SIZE) >> 6;                 \
 844         tmp += MAX_PB_SIZE;                                                    \
 845         dst += dststride;                                                      \
 846     }                                                                          \
 847 }
 848
 849 PUT_HEVC_QPEL_H(1)
 850 PUT_HEVC_QPEL_H(2)
 851 PUT_HEVC_QPEL_H(3)
 852 PUT_HEVC_QPEL_V(1)
 853 PUT_HEVC_QPEL_V(2)
 854 PUT_HEVC_QPEL_V(3)
 855 PUT_HEVC_QPEL_HV(1, 1)
 856 PUT_HEVC_QPEL_HV(1, 2)
 857 PUT_HEVC_QPEL_HV(1, 3)
 858 PUT_HEVC_QPEL_HV(2, 1)
 859 PUT_HEVC_QPEL_HV(2, 2)
 860 PUT_HEVC_QPEL_HV(2, 3)
 861 PUT_HEVC_QPEL_HV(3, 1)
 862 PUT_HEVC_QPEL_HV(3, 2)
 863 PUT_HEVC_QPEL_HV(3, 3)
 864
 865 #define QPEL(W)                                                                             \
 866 static void FUNC(put_hevc_qpel_pixels_ ## W)(int16_t *dst, ptrdiff_t dststride,             \
 867                                              uint8_t *src, ptrdiff_t srcstride,             \
 868                                              int height, int mx, int my,                    \
 869                                              int16_t *mcbuffer)                             \
 870 {                                                                                           \
 871     FUNC(put_hevc_qpel_pixels)(dst, dststride, src, srcstride, W, height,                   \
 872                                mx, my, mcbuffer);                                           \
 873 }                                                                                           \
 874                                                                                             \
 875 static void FUNC(put_hevc_qpel_h_ ## W)(int16_t *dst, ptrdiff_t dststride,                  \
 876                                         uint8_t *src, ptrdiff_t srcstride,                  \
 877                                         int height, int mx, int my,                         \
 878                                         int16_t *mcbuffer)                                  \
 879 {                                                                                           \
 880     if (mx == 1)                                                                            \
 881         FUNC(put_hevc_qpel_h1)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
 882     else if (mx == 2)                                                                       \
 883         FUNC(put_hevc_qpel_h2)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
 884     else                                                                                    \
 885         FUNC(put_hevc_qpel_h3)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
 886 }                                                                                           \
 887                                                                                             \
 888 static void FUNC(put_hevc_qpel_v_ ## W)(int16_t *dst, ptrdiff_t dststride,                  \
 889                                              uint8_t *src, ptrdiff_t srcstride,             \
 890                                              int height, int mx, int my,                    \
 891                                              int16_t *mcbuffer)                             \
 892 {                                                                                           \
 893     if (my == 1)                                                                            \
 894         FUNC(put_hevc_qpel_v1)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
 895     else if (my == 2)                                                                       \
 896         FUNC(put_hevc_qpel_v2)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
 897     else                                                                                    \
 898         FUNC(put_hevc_qpel_v3)(dst, dststride, src, srcstride, W, height, mcbuffer);        \
 899 }                                                                                           \
 900                                                                                             \
 901 static void FUNC(put_hevc_qpel_hv_ ## W)(int16_t *dst, ptrdiff_t dststride,                 \
 902                                              uint8_t *src, ptrdiff_t srcstride,             \
 903                                              int height, int mx, int my,                    \
 904                                              int16_t *mcbuffer)                             \
 905 {                                                                                           \
 906     if (my == 1) {                                                                          \
 907         if (mx == 1)                                                                        \
 908             FUNC(put_hevc_qpel_h1v1)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 909         else if (mx == 2)                                                                   \
 910             FUNC(put_hevc_qpel_h2v1)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 911         else                                                                                \
 912             FUNC(put_hevc_qpel_h3v1)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 913     } else if (my == 2) {                                                                   \
 914         if (mx == 1)                                                                        \
 915             FUNC(put_hevc_qpel_h1v2)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 916         else if (mx == 2)                                                                   \
 917             FUNC(put_hevc_qpel_h2v2)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 918         else                                                                                \
 919             FUNC(put_hevc_qpel_h3v2)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 920     } else {                                                                                \
 921         if (mx == 1)                                                                        \
 922             FUNC(put_hevc_qpel_h1v3)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 923         else if (mx == 2)                                                                   \
 924             FUNC(put_hevc_qpel_h2v3)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 925         else                                                                                \
 926             FUNC(put_hevc_qpel_h3v3)(dst, dststride, src, srcstride, W, height, mcbuffer);  \
 927     }                                                                                       \
 928 }
 929
 930 QPEL(64)
 931 QPEL(48)
 932 QPEL(32)
 933 QPEL(24)
 934 QPEL(16)
 935 QPEL(12)
 936 QPEL(8)
 937 QPEL(4)
 938
 939 static inline void FUNC(put_hevc_epel_pixels)(int16_t *dst, ptrdiff_t dststride,
 940                                               uint8_t *_src, ptrdiff_t _srcstride,
 941                                               int width, int height, int mx, int my,
 942                                               int16_t* mcbuffer)
 943 {
 944     int x, y;
 945     pixel *src          = (pixel *)_src;
 946     ptrdiff_t srcstride = _srcstride / sizeof(pixel);
 947
 948     dststride /= sizeof(*dst);
 949     for (y = 0; y < height; y++) {
 950         for (x = 0; x < width; x++)
 951             dst[x] = src[x] << (14 - BIT_DEPTH);
 952         src += srcstride;
 953         dst += dststride;
 954     }
 955 }
 956
 957 #define EPEL_FILTER(src, stride)                \
 958     (filter_0 * src[x - stride] +               \
 959      filter_1 * src[x]          +               \
 960      filter_2 * src[x + stride] +               \
 961      filter_3 * src[x + 2 * stride])
 962
 963 static inline void FUNC(put_hevc_epel_h)(int16_t *dst, ptrdiff_t dststride,
 964                                          uint8_t *_src, ptrdiff_t _srcstride,
 965                                          int width, int height, int mx, int my,
 966                                          int16_t* mcbuffer)
 967 {
 968     int x, y;
 969     pixel *src = (pixel *)_src;
 970     ptrdiff_t srcstride  = _srcstride / sizeof(pixel);
 971     const int16_t *filter = ff_hevc_epel_coeffs[mx - 1];
 972     int8_t filter_0 = filter[0];
 973     int8_t filter_1 = filter[1];
 974     int8_t filter_2 = filter[2];
 975     int8_t filter_3 = filter[3];
 976     dststride /= sizeof(*dst);
 977     for (y = 0; y < height; y++) {
 978         for (x = 0; x < width; x++)
 979             dst[x] = EPEL_FILTER(src, 1) >> (BIT_DEPTH - 8);
 980         src += srcstride;
 981         dst += dststride;
 982     }
 983 }
 984
 985 static inline void FUNC(put_hevc_epel_v)(int16_t *dst, ptrdiff_t dststride,
 986                                          uint8_t *_src, ptrdiff_t _srcstride,
 987                                          int width, int height, int mx, int my,
 988                                          int16_t* mcbuffer)
 989 {
 990     int x, y;
 991     pixel *src = (pixel *)_src;
 992     ptrdiff_t srcstride = _srcstride / sizeof(pixel);
 993     const int16_t *filter = ff_hevc_epel_coeffs[my - 1];
 994     int8_t filter_0 = filter[0];
 995     int8_t filter_1 = filter[1];
 996     int8_t filter_2 = filter[2];
 997     int8_t filter_3 = filter[3];
 998
 999     dststride /= sizeof(*dst);
1000     for (y = 0; y < height; y++) {
1001         for (x = 0; x < width; x++)
1002             dst[x] = EPEL_FILTER(src, srcstride) >> (BIT_DEPTH - 8);
1003         src += srcstride;
1004         dst += dststride;
1005     }
1006 }
1007
1008 static inline void FUNC(put_hevc_epel_hv)(int16_t *dst, ptrdiff_t dststride,
1009                                           uint8_t *_src, ptrdiff_t _srcstride,
1010                                           int width, int height, int mx, int my,
1011                                           int16_t* mcbuffer)
1012 {
1013     int x, y;
1014     pixel *src = (pixel *)_src;
1015     ptrdiff_t srcstride = _srcstride / sizeof(pixel);
1016     const int16_t *filter_h = ff_hevc_epel_coeffs[mx - 1];
1017     const int16_t *filter_v = ff_hevc_epel_coeffs[my - 1];
1018     int8_t filter_0 = filter_h[0];
1019     int8_t filter_1 = filter_h[1];
1020     int8_t filter_2 = filter_h[2];
1021     int8_t filter_3 = filter_h[3];
1022     int16_t tmp_array[(MAX_PB_SIZE + 3) * MAX_PB_SIZE];
1023     int16_t *tmp = tmp_array;
1024
1025     dststride /= sizeof(*dst);
1026     src -= EPEL_EXTRA_BEFORE * srcstride;
1027
1028     for (y = 0; y < height + EPEL_EXTRA; y++) {
1029         for (x = 0; x < width; x++)
1030             tmp[x] = EPEL_FILTER(src, 1) >> (BIT_DEPTH - 8);
1031         src += srcstride;
1032         tmp += MAX_PB_SIZE;
1033     }
1034
1035     tmp      = tmp_array + EPEL_EXTRA_BEFORE * MAX_PB_SIZE;
1036     filter_0 = filter_v[0];
1037     filter_1 = filter_v[1];
1038     filter_2 = filter_v[2];
1039     filter_3 = filter_v[3];
1040     for (y = 0; y < height; y++) {
1041         for (x = 0; x < width; x++)
1042             dst[x] = EPEL_FILTER(tmp, MAX_PB_SIZE) >> 6;
1043         tmp += MAX_PB_SIZE;
1044         dst += dststride;
1045     }
1046 }
1047
1048 #define EPEL(W)                                                                 \
1049 static void FUNC(put_hevc_epel_pixels_ ## W)(int16_t *dst, ptrdiff_t dststride, \
1050                                              uint8_t *src, ptrdiff_t srcstride, \
1051                                              int height, int mx, int my,        \
1052                                              int16_t *mcbuffer)                 \
1053 {                                                                               \
1054     FUNC(put_hevc_epel_pixels)(dst, dststride, src, srcstride,                  \
1055                                W, height, mx, my, mcbuffer);                    \
1056 }                                                                               \
1057 static void FUNC(put_hevc_epel_h_ ## W)(int16_t *dst, ptrdiff_t dststride,      \
1058                                         uint8_t *src, ptrdiff_t srcstride,      \
1059                                         int height, int mx, int my,             \
1060                                         int16_t *mcbuffer)                      \
1061 {                                                                               \
1062     FUNC(put_hevc_epel_h)(dst, dststride, src, srcstride,                       \
1063                           W, height, mx, my, mcbuffer);                         \
1064 }                                                                               \
1065 static void FUNC(put_hevc_epel_v_ ## W)(int16_t *dst, ptrdiff_t dststride,      \
1066                                         uint8_t *src, ptrdiff_t srcstride,      \
1067                                         int height, int mx, int my,             \
1068                                         int16_t *mcbuffer)                      \
1069 {                                                                               \
1070     FUNC(put_hevc_epel_v)(dst, dststride, src, srcstride,                       \
1071                           W, height, mx, my, mcbuffer);                         \
1072 }                                                                               \
1073 static void FUNC(put_hevc_epel_hv_ ## W)(int16_t *dst, ptrdiff_t dststride,     \
1074                                          uint8_t *src, ptrdiff_t srcstride,     \
1075                                          int height, int mx, int my,            \
1076                                          int16_t *mcbuffer)                     \
1077 {                                                                               \
1078     FUNC(put_hevc_epel_hv)(dst, dststride, src, srcstride,                      \
1079                            W, height, mx, my, mcbuffer);                        \
1080 }
1081
1082 EPEL(32)
1083 EPEL(24)
1084 EPEL(16)
1085 EPEL(12)
1086 EPEL(8)
1087 EPEL(6)
1088 EPEL(4)
1089 EPEL(2)
1090
1091 static av_always_inline void
1092 FUNC(put_unweighted_pred)(uint8_t *_dst, ptrdiff_t _dststride,
1093                           int16_t *src, ptrdiff_t srcstride,
1094                           int width, int height)
1095 {
1096     int x, y;
1097     pixel *dst          = (pixel *)_dst;
1098     ptrdiff_t dststride = _dststride / sizeof(pixel);
1099
1100     int shift = 14 - BIT_DEPTH;
1101 #if BIT_DEPTH < 14
1102     int offset = 1 << (shift - 1);
1103 #else
1104     int offset = 0;
1105 #endif
1106     srcstride /= sizeof(*src);
1107     for (y = 0; y < height; y++) {
1108         for (x = 0; x < width; x++)
1109             dst[x] = av_clip_pixel((src[x] + offset) >> shift);
1110         dst += dststride;
1111         src += srcstride;
1112     }
1113 }
1114
1115 static av_always_inline void
1116 FUNC(put_unweighted_pred_avg)(uint8_t *_dst, ptrdiff_t _dststride,
1117                               int16_t *src1, int16_t *src2,
1118                               ptrdiff_t srcstride,
1119                               int width, int height)
1120 {
1121     int x, y;
1122     pixel *dst          = (pixel *)_dst;
1123     ptrdiff_t dststride = _dststride / sizeof(pixel);
1124
1125     int shift = 14 + 1 - BIT_DEPTH;
1126 #if BIT_DEPTH < 14
1127     int offset = 1 << (shift - 1);
1128 #else
1129     int offset = 0;
1130 #endif
1131
1132     srcstride /= sizeof(*src1);
1133     for (y = 0; y < height; y++) {
1134         for (x = 0; x < width; x++)
1135             dst[x] = av_clip_pixel((src1[x] + src2[x] + offset) >> shift);
1136         dst  += dststride;
1137         src1 += srcstride;
1138         src2 += srcstride;
1139     }
1140 }
1141
1142 static av_always_inline void
1143 FUNC(weighted_pred)(uint8_t denom, int16_t wlxFlag, int16_t olxFlag,
1144                     uint8_t *_dst, ptrdiff_t _dststride,
1145                     int16_t *src, ptrdiff_t srcstride,
1146                     int width, int height)
1147 {
1148     int shift, log2Wd, wx, ox, x, y, offset;
1149     pixel *dst          = (pixel *)_dst;
1150     ptrdiff_t dststride = _dststride / sizeof(pixel);
1151
1152     shift  = 14 - BIT_DEPTH;
1153     log2Wd = denom + shift;
1154     offset = 1 << (log2Wd - 1);
1155     wx     = wlxFlag;
1156     ox     = olxFlag * (1 << (BIT_DEPTH - 8));
1157
1158     srcstride /= sizeof(*src);
1159     for (y = 0; y < height; y++) {
1160         for (x = 0; x < width; x++) {
1161             if (log2Wd >= 1) {
1162                 dst[x] = av_clip_pixel(((src[x] * wx + offset) >> log2Wd) + ox);
1163             } else {
1164                 dst[x] = av_clip_pixel(src[x] * wx + ox);
1165             }
1166         }
1167         dst += dststride;
1168         src += srcstride;
1169     }
1170 }
1171
1172 static av_always_inline void
1173 FUNC(weighted_pred_avg)(uint8_t denom,
1174                         int16_t wl0Flag, int16_t wl1Flag,
1175                         int16_t ol0Flag, int16_t ol1Flag,
1176                         uint8_t *_dst, ptrdiff_t _dststride,
1177                         int16_t *src1, int16_t *src2,
1178                         ptrdiff_t srcstride,
1179                         int width, int height)
1180 {
1181     int shift, log2Wd, w0, w1, o0, o1, x, y;
1182     pixel *dst = (pixel *)_dst;
1183     ptrdiff_t dststride = _dststride / sizeof(pixel);
1184
1185     shift  = 14 - BIT_DEPTH;
1186     log2Wd = denom + shift;
1187     w0     = wl0Flag;
1188     w1     = wl1Flag;
1189     o0     = ol0Flag * (1 << (BIT_DEPTH - 8));
1190     o1     = ol1Flag * (1 << (BIT_DEPTH - 8));
1191
1192     srcstride /= sizeof(*src1);
1193     for (y = 0; y < height; y++) {
1194         for (x = 0; x < width; x++)
1195             dst[x] = av_clip_pixel((src1[x] * w0 + src2[x] * w1 +
1196                                     ((o0 + o1 + 1) << log2Wd)) >> (log2Wd + 1));
1197         dst  += dststride;
1198         src1 += srcstride;
1199         src2 += srcstride;
1200     }
1201 }
1202
1203 #define PUT_PRED(w)                                                                            \
1204 static void FUNC(put_unweighted_pred_ ## w)(uint8_t *dst, ptrdiff_t dststride,                 \
1205                                             int16_t *src, ptrdiff_t srcstride,                 \
1206                                             int height)                                        \
1207 {                                                                                              \
1208     FUNC(put_unweighted_pred)(dst, dststride, src, srcstride, w, height);                      \
1209 }                                                                                              \
1210 static void FUNC(put_unweighted_pred_avg_ ## w)(uint8_t *dst, ptrdiff_t dststride,             \
1211                                                 int16_t *src1, int16_t *src2,                  \
1212                                                 ptrdiff_t srcstride, int height)               \
1213 {                                                                                              \
1214     FUNC(put_unweighted_pred_avg)(dst, dststride, src1, src2, srcstride, w, height);           \
1215 }                                                                                              \
1216 static void FUNC(put_weighted_pred_ ## w)(uint8_t denom, int16_t weight, int16_t offset,       \
1217                                           uint8_t *dst, ptrdiff_t dststride,                   \
1218                                           int16_t *src, ptrdiff_t srcstride, int height)       \
1219 {                                                                                              \
1220     FUNC(weighted_pred)(denom, weight, offset,                                                 \
1221                         dst, dststride, src, srcstride, w, height);                            \
1222 }                                                                                              \
1223 static void FUNC(put_weighted_pred_avg_ ## w)(uint8_t denom, int16_t weight0, int16_t weight1, \
1224                                               int16_t offset0, int16_t offset1,                \
1225                                               uint8_t *dst, ptrdiff_t dststride,               \
1226                                               int16_t *src1, int16_t *src2,                    \
1227                                               ptrdiff_t srcstride, int height)                 \
1228 {                                                                                              \
1229     FUNC(weighted_pred_avg)(denom, weight0, weight1, offset0, offset1,                         \
1230                             dst, dststride, src1, src2, srcstride, w, height);                 \
1231 }
1232
1233 PUT_PRED(64)
1234 PUT_PRED(48)
1235 PUT_PRED(32)
1236 PUT_PRED(24)
1237 PUT_PRED(16)
1238 PUT_PRED(12)
1239 PUT_PRED(8)
1240 PUT_PRED(6)
1241 PUT_PRED(4)
1242 PUT_PRED(2)
1243
1244 // line zero
1245 #define P3 pix[-4 * xstride]
1246 #define P2 pix[-3 * xstride]
1247 #define P1 pix[-2 * xstride]
1248 #define P0 pix[-1 * xstride]
1249 #define Q0 pix[0 * xstride]
1250 #define Q1 pix[1 * xstride]
1251 #define Q2 pix[2 * xstride]
1252 #define Q3 pix[3 * xstride]
1253
1254 // line three. used only for deblocking decision
1255 #define TP3 pix[-4 * xstride + 3 * ystride]
1256 #define TP2 pix[-3 * xstride + 3 * ystride]
1257 #define TP1 pix[-2 * xstride + 3 * ystride]
1258 #define TP0 pix[-1 * xstride + 3 * ystride]
1259 #define TQ0 pix[0  * xstride + 3 * ystride]
1260 #define TQ1 pix[1  * xstride + 3 * ystride]
1261 #define TQ2 pix[2  * xstride + 3 * ystride]
1262 #define TQ3 pix[3  * xstride + 3 * ystride]
1263
1264 static void FUNC(hevc_loop_filter_luma)(uint8_t *_pix,
1265                                         ptrdiff_t _xstride, ptrdiff_t _ystride,
1266                                         int beta, int *_tc,
1267                                         uint8_t *_no_p, uint8_t *_no_q)
1268 {
1269     int d, j;
1270     pixel *pix        = (pixel *)_pix;
1271     ptrdiff_t xstride = _xstride / sizeof(pixel);
1272     ptrdiff_t ystride = _ystride / sizeof(pixel);
1273
1274     beta <<= BIT_DEPTH - 8;
1275
1276     for (j = 0; j < 2; j++) {
1277         const int dp0  = abs(P2  - 2 * P1  + P0);
1278         const int dq0  = abs(Q2  - 2 * Q1  + Q0);
1279         const int dp3  = abs(TP2 - 2 * TP1 + TP0);
1280         const int dq3  = abs(TQ2 - 2 * TQ1 + TQ0);
1281         const int d0   = dp0 + dq0;
1282         const int d3   = dp3 + dq3;
1283         const int tc   = _tc[j]   << (BIT_DEPTH - 8);
1284         const int no_p = _no_p[j];
1285         const int no_q = _no_q[j];
1286
1287         if (d0 + d3 >= beta) {
1288             pix += 4 * ystride;
1289             continue;
1290         } else {
1291             const int beta_3 = beta >> 3;
1292             const int beta_2 = beta >> 2;
1293             const int tc25   = ((tc * 5 + 1) >> 1);
1294
1295             if (abs(P3  -  P0) + abs(Q3  -  Q0) < beta_3 && abs(P0  -  Q0) < tc25 &&
1296                 abs(TP3 - TP0) + abs(TQ3 - TQ0) < beta_3 && abs(TP0 - TQ0) < tc25 &&
1297                                       (d0 << 1) < beta_2 &&      (d3 << 1) < beta_2) {
1298                 // strong filtering
1299                 const int tc2 = tc << 1;
1300                 for (d = 0; d < 4; d++) {
1301                     const int p3 = P3;
1302                     const int p2 = P2;
1303                     const int p1 = P1;
1304                     const int p0 = P0;
1305                     const int q0 = Q0;
1306                     const int q1 = Q1;
1307                     const int q2 = Q2;
1308                     const int q3 = Q3;
1309                     if (!no_p) {
1310                         P0 = p0 + av_clip(((p2 + 2 * p1 + 2 * p0 + 2 * q0 + q1 + 4) >> 3) - p0, -tc2, tc2);
1311                         P1 = p1 + av_clip(((p2 + p1 + p0 + q0 + 2) >> 2) - p1, -tc2, tc2);
1312                         P2 = p2 + av_clip(((2 * p3 + 3 * p2 + p1 + p0 + q0 + 4) >> 3) - p2, -tc2, tc2);
1313                     }
1314                     if (!no_q) {
1315                         Q0 = q0 + av_clip(((p1 + 2 * p0 + 2 * q0 + 2 * q1 + q2 + 4) >> 3) - q0, -tc2, tc2);
1316                         Q1 = q1 + av_clip(((p0 + q0 + q1 + q2 + 2) >> 2) - q1, -tc2, tc2);
1317                         Q2 = q2 + av_clip(((2 * q3 + 3 * q2 + q1 + q0 + p0 + 4) >> 3) - q2, -tc2, tc2);
1318                     }
1319                     pix += ystride;
1320                 }
1321             } else { // normal filtering
1322                 int nd_p = 1;
1323                 int nd_q = 1;
1324                 const int tc_2 = tc >> 1;
1325                 if (dp0 + dp3 < ((beta + (beta >> 1)) >> 3))
1326                     nd_p = 2;
1327                 if (dq0 + dq3 < ((beta + (beta >> 1)) >> 3))
1328                     nd_q = 2;
1329
1330                 for (d = 0; d < 4; d++) {
1331                     const int p2 = P2;
1332                     const int p1 = P1;
1333                     const int p0 = P0;
1334                     const int q0 = Q0;
1335                     const int q1 = Q1;
1336                     const int q2 = Q2;
1337                     int delta0   = (9 * (q0 - p0) - 3 * (q1 - p1) + 8) >> 4;
1338                     if (abs(delta0) < 10 * tc) {
1339                         delta0 = av_clip(delta0, -tc, tc);
1340                         if (!no_p)
1341                             P0 = av_clip_pixel(p0 + delta0);
1342                         if (!no_q)
1343                             Q0 = av_clip_pixel(q0 - delta0);
1344                         if (!no_p && nd_p > 1) {
1345                             const int deltap1 = av_clip((((p2 + p0 + 1) >> 1) - p1 + delta0) >> 1, -tc_2, tc_2);
1346                             P1 = av_clip_pixel(p1 + deltap1);
1347                         }
1348                         if (!no_q && nd_q > 1) {
1349                             const int deltaq1 = av_clip((((q2 + q0 + 1) >> 1) - q1 - delta0) >> 1, -tc_2, tc_2);
1350                             Q1 = av_clip_pixel(q1 + deltaq1);
1351                         }
1352                     }
1353                     pix += ystride;
1354                 }
1355             }
1356         }
1357     }
1358 }
1359
1360 static void FUNC(hevc_loop_filter_chroma)(uint8_t *_pix, ptrdiff_t _xstride,
1361                                           ptrdiff_t _ystride, int *_tc,
1362                                           uint8_t *_no_p, uint8_t *_no_q)
1363 {
1364     int d, j, no_p, no_q;
1365     pixel *pix        = (pixel *)_pix;
1366     ptrdiff_t xstride = _xstride / sizeof(pixel);
1367     ptrdiff_t ystride = _ystride / sizeof(pixel);
1368
1369     for (j = 0; j < 2; j++) {
1370         const int tc = _tc[j] << (BIT_DEPTH - 8);
1371         if (tc <= 0) {
1372             pix += 4 * ystride;
1373             continue;
1374         }
1375         no_p = _no_p[j];
1376         no_q = _no_q[j];
1377
1378         for (d = 0; d < 4; d++) {
1379             int delta0;
1380             const int p1 = P1;
1381             const int p0 = P0;
1382             const int q0 = Q0;
1383             const int q1 = Q1;
1384             delta0 = av_clip((((q0 - p0) * 4) + p1 - q1 + 4) >> 3, -tc, tc);
1385             if (!no_p)
1386                 P0 = av_clip_pixel(p0 + delta0);
1387             if (!no_q)
1388                 Q0 = av_clip_pixel(q0 - delta0);
1389             pix += ystride;
1390         }
1391     }
1392 }
1393
1394 static void FUNC(hevc_h_loop_filter_chroma)(uint8_t *pix, ptrdiff_t stride,
1395                                             int *tc, uint8_t *no_p,
1396                                             uint8_t *no_q)
1397 {
1398     FUNC(hevc_loop_filter_chroma)(pix, stride, sizeof(pixel), tc, no_p, no_q);
1399 }
1400
1401 static void FUNC(hevc_v_loop_filter_chroma)(uint8_t *pix, ptrdiff_t stride,
1402                                             int *tc, uint8_t *no_p,
1403                                             uint8_t *no_q)
1404 {
1405     FUNC(hevc_loop_filter_chroma)(pix, sizeof(pixel), stride, tc, no_p, no_q);
1406 }
1407
1408 static void FUNC(hevc_h_loop_filter_luma)(uint8_t *pix, ptrdiff_t stride,
1409                                           int beta, int *tc, uint8_t *no_p,
1410                                           uint8_t *no_q)
1411 {
1412     FUNC(hevc_loop_filter_luma)(pix, stride, sizeof(pixel),
1413                                 beta, tc, no_p, no_q);
1414 }
1415
1416 static void FUNC(hevc_v_loop_filter_luma)(uint8_t *pix, ptrdiff_t stride,
1417                                           int beta, int *tc, uint8_t *no_p,
1418                                           uint8_t *no_q)
1419 {
1420     FUNC(hevc_loop_filter_luma)(pix, sizeof(pixel), stride,
1421                                 beta, tc, no_p, no_q);
1422 }
1423
1424 #undef P3
1425 #undef P2
1426 #undef P1
1427 #undef P0
1428 #undef Q0
1429 #undef Q1
1430 #undef Q2
1431 #undef Q3
1432
1433 #undef TP3
1434 #undef TP2
1435 #undef TP1
1436 #undef TP0
1437 #undef TQ0
1438 #undef TQ1
1439 #undef TQ2
1440 #undef TQ3