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

   1 /*
   2  * HEVC video decoder
   3  *
   4  * Copyright (C) 2012 - 2013 Guillaume Martres
   5  * Copyright (C) 2013 Seppo Tomperi
   6  * Copyright (C) 2013 Wassim Hamidouche
   7  *
   8  * This file is part of FFmpeg.
   9  *
  10  * FFmpeg is free software; you can redistribute it and/or
  11  * modify it under the terms of the GNU Lesser General Public
  12  * License as published by the Free Software Foundation; either
  13  * version 2.1 of the License, or (at your option) any later version.
  14  *
  15  * FFmpeg is distributed in the hope that it will be useful,
  16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  18  * Lesser General Public License for more details.
  19  *
  20  * You should have received a copy of the GNU Lesser General Public
  21  * License along with FFmpeg; if not, write to the Free Software
  22  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  23  */
  24
  25 #include "libavutil/common.h"
  26 #include "libavutil/internal.h"
  27
  28 #include "cabac_functions.h"
  29 #include "golomb.h"
  30 #include "hevc.h"
  31
  32 #include "bit_depth_template.c"
  33
  34 #define LUMA 0
  35 #define CB 1
  36 #define CR 2
  37
  38 static const uint8_t tctable[54] = {
  39     0, 0, 0, 0, 0, 0, 0,  0,  0,  0,  0,  0,  0,  0,  0,  0, 0, 0, 1, // QP  0...18
  40     1, 1, 1, 1, 1, 1, 1,  1,  2,  2,  2,  2,  3,  3,  3,  3, 4, 4, 4, // QP 19...37
  41     5, 5, 6, 6, 7, 8, 9, 10, 11, 13, 14, 16, 18, 20, 22, 24           // QP 38...53
  42 };
  43
  44 static const uint8_t betatable[52] = {
  45      0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  6,  7,  8, // QP 0...18
  46      9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 20, 22, 24, 26, 28, 30, 32, 34, 36, // QP 19...37
  47     38, 40, 42, 44, 46, 48, 50, 52, 54, 56, 58, 60, 62, 64                      // QP 38...51
  48 };
  49
  50 static int chroma_tc(HEVCContext *s, int qp_y, int c_idx, int tc_offset)
  51 {
  52     static const int qp_c[] = {
  53         29, 30, 31, 32, 33, 33, 34, 34, 35, 35, 36, 36, 37, 37
  54     };
  55     int qp, qp_i, offset, idxt;
  56
  57     // slice qp offset is not used for deblocking
  58     if (c_idx == 1)
  59         offset = s->pps->cb_qp_offset;
  60     else
  61         offset = s->pps->cr_qp_offset;
  62
  63     qp_i = av_clip(qp_y + offset, 0, 57);
  64     if (s->sps->chroma_format_idc == 1) {
  65         if (qp_i < 30)
  66             qp = qp_i;
  67         else if (qp_i > 43)
  68             qp = qp_i - 6;
  69         else
  70             qp = qp_c[qp_i - 30];
  71     } else {
  72         qp = av_clip(qp_i, 0, 51);
  73     }
  74
  75     idxt = av_clip(qp + DEFAULT_INTRA_TC_OFFSET + tc_offset, 0, 53);
  76     return tctable[idxt];
  77 }
  78
  79 static int get_qPy_pred(HEVCContext *s, int xBase, int yBase, int log2_cb_size)
  80 {
  81     HEVCLocalContext *lc     = s->HEVClc;
  82     int ctb_size_mask        = (1 << s->sps->log2_ctb_size) - 1;
  83     int MinCuQpDeltaSizeMask = (1 << (s->sps->log2_ctb_size -
  84                                       s->pps->diff_cu_qp_delta_depth)) - 1;
  85     int xQgBase              = xBase - (xBase & MinCuQpDeltaSizeMask);
  86     int yQgBase              = yBase - (yBase & MinCuQpDeltaSizeMask);
  87     int min_cb_width         = s->sps->min_cb_width;
  88     int x_cb                 = xQgBase >> s->sps->log2_min_cb_size;
  89     int y_cb                 = yQgBase >> s->sps->log2_min_cb_size;
  90     int availableA           = (xBase   & ctb_size_mask) &&
  91                                (xQgBase & ctb_size_mask);
  92     int availableB           = (yBase   & ctb_size_mask) &&
  93                                (yQgBase & ctb_size_mask);
  94     int qPy_pred, qPy_a, qPy_b;
  95
  96     // qPy_pred
  97     if (lc->first_qp_group || (!xQgBase && !yQgBase)) {
  98         lc->first_qp_group = !lc->tu.is_cu_qp_delta_coded;
  99         qPy_pred = s->sh.slice_qp;
 100     } else {
 101         qPy_pred = lc->qPy_pred;
 102     }
 103
 104     // qPy_a
 105     if (availableA == 0)
 106         qPy_a = qPy_pred;
 107     else
 108         qPy_a = s->qp_y_tab[(x_cb - 1) + y_cb * min_cb_width];
 109
 110     // qPy_b
 111     if (availableB == 0)
 112         qPy_b = qPy_pred;
 113     else
 114         qPy_b = s->qp_y_tab[x_cb + (y_cb - 1) * min_cb_width];
 115
 116     av_assert2(qPy_a >= -s->sps->qp_bd_offset && qPy_a < 52);
 117     av_assert2(qPy_b >= -s->sps->qp_bd_offset && qPy_b < 52);
 118
 119     return (qPy_a + qPy_b + 1) >> 1;
 120 }
 121
 122 void ff_hevc_set_qPy(HEVCContext *s, int xBase, int yBase, int log2_cb_size)
 123 {
 124     int qp_y = get_qPy_pred(s, xBase, yBase, log2_cb_size);
 125
 126     if (s->HEVClc->tu.cu_qp_delta != 0) {
 127         int off = s->sps->qp_bd_offset;
 128         s->HEVClc->qp_y = FFUMOD(qp_y + s->HEVClc->tu.cu_qp_delta + 52 + 2 * off,
 129                                  52 + off) - off;
 130     } else
 131         s->HEVClc->qp_y = qp_y;
 132 }
 133
 134 static int get_qPy(HEVCContext *s, int xC, int yC)
 135 {
 136     int log2_min_cb_size  = s->sps->log2_min_cb_size;
 137     int x                 = xC >> log2_min_cb_size;
 138     int y                 = yC >> log2_min_cb_size;
 139     return s->qp_y_tab[x + y * s->sps->min_cb_width];
 140 }
 141
 142 static void copy_CTB(uint8_t *dst, uint8_t *src,
 143                      int width, int height, int stride_dst, int stride_src)
 144 {
 145     int i;
 146
 147     for (i = 0; i < height; i++) {
 148         memcpy(dst, src, width);
 149         dst += stride_dst;
 150         src += stride_src;
 151     }
 152 }
 153
 154 static void restore_tqb_pixels(HEVCContext *s, int x0, int y0, int width, int height, int c_idx)
 155 {
 156     if ( s->pps->transquant_bypass_enable_flag ||
 157             (s->sps->pcm.loop_filter_disable_flag && s->sps->pcm_enabled_flag)) {
 158         int x, y;
 159         ptrdiff_t stride_dst = s->sao_frame->linesize[c_idx];
 160         ptrdiff_t stride_src = s->frame->linesize[c_idx];
 161         int min_pu_size  = 1 << s->sps->log2_min_pu_size;
 162         int hshift       = s->sps->hshift[c_idx];
 163         int vshift       = s->sps->vshift[c_idx];
 164         int x_min        = ((x0         ) >> s->sps->log2_min_pu_size);
 165         int y_min        = ((y0         ) >> s->sps->log2_min_pu_size);
 166         int x_max        = ((x0 + width ) >> s->sps->log2_min_pu_size);
 167         int y_max        = ((y0 + height) >> s->sps->log2_min_pu_size);
 168         int len          = min_pu_size >> hshift;
 169         for (y = y_min; y < y_max; y++) {
 170             for (x = x_min; x < x_max; x++) {
 171                 if (s->is_pcm[y * s->sps->min_pu_width + x]) {
 172                     int n;
 173                     uint8_t *src = &s->frame->data[c_idx][    ((y << s->sps->log2_min_pu_size) >> vshift) * stride_src + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
 174                     uint8_t *dst = &s->sao_frame->data[c_idx][((y << s->sps->log2_min_pu_size) >> vshift) * stride_dst + (((x << s->sps->log2_min_pu_size) >> hshift) << s->sps->pixel_shift)];
 175                     for (n = 0; n < (min_pu_size >> vshift); n++) {
 176                         memcpy(src, dst, len);
 177                         src += stride_src;
 178                         dst += stride_dst;
 179                     }
 180                 }
 181             }
 182         }
 183     }
 184 }
 185
 186 #define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
 187
 188 static void sao_filter_CTB(HEVCContext *s, int x, int y)
 189 {
 190     int c_idx;
 191     int edges[4];  // 0 left 1 top 2 right 3 bottom
 192     int x_ctb                = x >> s->sps->log2_ctb_size;
 193     int y_ctb                = y >> s->sps->log2_ctb_size;
 194     int ctb_addr_rs          = y_ctb * s->sps->ctb_width + x_ctb;
 195     int ctb_addr_ts          = s->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
 196     SAOParams *sao           = &CTB(s->sao, x_ctb, y_ctb);
 197     // flags indicating unfilterable edges
 198     uint8_t vert_edge[]      = { 0, 0 };
 199     uint8_t horiz_edge[]     = { 0, 0 };
 200     uint8_t diag_edge[]      = { 0, 0, 0, 0 };
 201     uint8_t lfase            = CTB(s->filter_slice_edges, x_ctb, y_ctb);
 202     uint8_t no_tile_filter   = s->pps->tiles_enabled_flag &&
 203                                !s->pps->loop_filter_across_tiles_enabled_flag;
 204     uint8_t restore          = no_tile_filter || !lfase;
 205     uint8_t left_tile_edge   = 0;
 206     uint8_t right_tile_edge  = 0;
 207     uint8_t up_tile_edge     = 0;
 208     uint8_t bottom_tile_edge = 0;
 209
 210     edges[0]   = x_ctb == 0;
 211     edges[1]   = y_ctb == 0;
 212     edges[2]   = x_ctb == s->sps->ctb_width  - 1;
 213     edges[3]   = y_ctb == s->sps->ctb_height - 1;
 214
 215     if (restore) {
 216         if (!edges[0]) {
 217             left_tile_edge  = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
 218             vert_edge[0]    = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb)) || left_tile_edge;
 219         }
 220         if (!edges[2]) {
 221             right_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs+1]];
 222             vert_edge[1]    = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb)) || right_tile_edge;
 223         }
 224         if (!edges[1]) {
 225             up_tile_edge     = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->sps->ctb_width]];
 226             horiz_edge[0]    = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) || up_tile_edge;
 227         }
 228         if (!edges[3]) {
 229             bottom_tile_edge = no_tile_filter && s->pps->tile_id[ctb_addr_ts] != s->pps->tile_id[s->pps->ctb_addr_rs_to_ts[ctb_addr_rs + s->sps->ctb_width]];
 230             horiz_edge[1]    = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb + 1)) || bottom_tile_edge;
 231         }
 232         if (!edges[0] && !edges[1]) {
 233             diag_edge[0] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || left_tile_edge || up_tile_edge;
 234         }
 235         if (!edges[1] && !edges[2]) {
 236             diag_edge[1] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb - 1)) || right_tile_edge || up_tile_edge;
 237         }
 238         if (!edges[2] && !edges[3]) {
 239             diag_edge[2] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb + 1, y_ctb + 1)) || right_tile_edge || bottom_tile_edge;
 240         }
 241         if (!edges[0] && !edges[3]) {
 242             diag_edge[3] = (!lfase && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb + 1)) || left_tile_edge || bottom_tile_edge;
 243         }
 244     }
 245
 246     for (c_idx = 0; c_idx < 3; c_idx++) {
 247         int x0       = x >> s->sps->hshift[c_idx];
 248         int y0       = y >> s->sps->vshift[c_idx];
 249         int stride_src = s->frame->linesize[c_idx];
 250         int stride_dst = s->sao_frame->linesize[c_idx];
 251         int ctb_size_h = (1 << (s->sps->log2_ctb_size)) >> s->sps->hshift[c_idx];
 252         int ctb_size_v = (1 << (s->sps->log2_ctb_size)) >> s->sps->vshift[c_idx];
 253         int width    = FFMIN(ctb_size_h, (s->sps->width  >> s->sps->hshift[c_idx]) - x0);
 254         int height   = FFMIN(ctb_size_v, (s->sps->height >> s->sps->vshift[c_idx]) - y0);
 255         uint8_t *src = &s->frame->data[c_idx][y0 * stride_src + (x0 << s->sps->pixel_shift)];
 256         uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride_dst + (x0 << s->sps->pixel_shift)];
 257
 258         switch (sao->type_idx[c_idx]) {
 259         case SAO_BAND:
 260             copy_CTB(dst, src, width << s->sps->pixel_shift, height, stride_dst, stride_src);
 261             s->hevcdsp.sao_band_filter(src, dst,
 262                                        stride_src, stride_dst,
 263                                        sao,
 264                                        edges, width,
 265                                        height, c_idx);
 266             restore_tqb_pixels(s, x, y, width, height, c_idx);
 267             sao->type_idx[c_idx] = SAO_APPLIED;
 268             break;
 269         case SAO_EDGE:
 270         {
 271             uint8_t left_pixels = !edges[0] && (CTB(s->sao, x_ctb-1, y_ctb).type_idx[c_idx] != SAO_APPLIED);
 272             if (!edges[1]) {
 273                 uint8_t top_left  = !edges[0] && (CTB(s->sao, x_ctb-1, y_ctb-1).type_idx[c_idx] != SAO_APPLIED);
 274                 uint8_t top_right = !edges[2] && (CTB(s->sao, x_ctb+1, y_ctb-1).type_idx[c_idx] != SAO_APPLIED);
 275                 if (CTB(s->sao, x_ctb  , y_ctb-1).type_idx[c_idx] == 0)
 276                     memcpy( dst - stride_dst - (top_left << s->sps->pixel_shift),
 277                             src - stride_src - (top_left << s->sps->pixel_shift),
 278                             (top_left + width + top_right) << s->sps->pixel_shift);
 279                 else {
 280                     if (top_left)
 281                         memcpy( dst - stride_dst - (1 << s->sps->pixel_shift),
 282                                 src - stride_src - (1 << s->sps->pixel_shift),
 283                                 1 << s->sps->pixel_shift);
 284                     if(top_right)
 285                         memcpy( dst - stride_dst + (width << s->sps->pixel_shift),
 286                                 src - stride_src + (width << s->sps->pixel_shift),
 287                                 1 << s->sps->pixel_shift);
 288                 }
 289             }
 290             if (!edges[3]) {                                                                // bottom and bottom right
 291                 uint8_t bottom_left = !edges[0] && (CTB(s->sao, x_ctb-1, y_ctb+1).type_idx[c_idx] != SAO_APPLIED);
 292                 memcpy( dst + height * stride_dst - (bottom_left << s->sps->pixel_shift),
 293                         src + height * stride_src - (bottom_left << s->sps->pixel_shift),
 294                         (width + 1 + bottom_left) << s->sps->pixel_shift);
 295             }
 296             copy_CTB(dst - (left_pixels << s->sps->pixel_shift),
 297                      src - (left_pixels << s->sps->pixel_shift),
 298                      (width + 1 + left_pixels) << s->sps->pixel_shift, height, stride_dst, stride_src);
 299             s->hevcdsp.sao_edge_filter[restore](src, dst,
 300                                                 stride_src, stride_dst,
 301                                                 sao,
 302                                                 edges, width,
 303                                                 height, c_idx,
 304                                                 vert_edge,
 305                                                 horiz_edge,
 306                                                 diag_edge);
 307             restore_tqb_pixels(s, x, y, width, height, c_idx);
 308             sao->type_idx[c_idx] = SAO_APPLIED;
 309             break;
 310         }
 311         }
 312     }
 313 }
 314
 315 static int get_pcm(HEVCContext *s, int x, int y)
 316 {
 317     int log2_min_pu_size = s->sps->log2_min_pu_size;
 318     int x_pu, y_pu;
 319
 320     if (x < 0 || y < 0)
 321         return 2;
 322
 323     x_pu = x >> log2_min_pu_size;
 324     y_pu = y >> log2_min_pu_size;
 325
 326     if (x_pu >= s->sps->min_pu_width || y_pu >= s->sps->min_pu_height)
 327         return 2;
 328     return s->is_pcm[y_pu * s->sps->min_pu_width + x_pu];
 329 }
 330
 331 #define TC_CALC(qp, bs)                                                 \
 332     tctable[av_clip((qp) + DEFAULT_INTRA_TC_OFFSET * ((bs) - 1) +       \
 333                     (tc_offset >> 1 << 1),                              \
 334                     0, MAX_QP + DEFAULT_INTRA_TC_OFFSET)]
 335
 336 static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
 337 {
 338     uint8_t *src;
 339     int x, y;
 340     int chroma, beta;
 341     int32_t c_tc[2], tc[2];
 342     uint8_t no_p[2] = { 0 };
 343     uint8_t no_q[2] = { 0 };
 344
 345     int log2_ctb_size = s->sps->log2_ctb_size;
 346     int x_end, x_end2, y_end;
 347     int ctb_size        = 1 << log2_ctb_size;
 348     int ctb             = (x0 >> log2_ctb_size) +
 349                           (y0 >> log2_ctb_size) * s->sps->ctb_width;
 350     int cur_tc_offset   = s->deblock[ctb].tc_offset;
 351     int cur_beta_offset = s->deblock[ctb].beta_offset;
 352     int left_tc_offset, left_beta_offset;
 353     int tc_offset, beta_offset;
 354     int pcmf = (s->sps->pcm_enabled_flag &&
 355                 s->sps->pcm.loop_filter_disable_flag) ||
 356                s->pps->transquant_bypass_enable_flag;
 357
 358     if (x0) {
 359         left_tc_offset   = s->deblock[ctb - 1].tc_offset;
 360         left_beta_offset = s->deblock[ctb - 1].beta_offset;
 361     } else {
 362         left_tc_offset   = 0;
 363         left_beta_offset = 0;
 364     }
 365
 366     x_end = x0 + ctb_size;
 367     if (x_end > s->sps->width)
 368         x_end = s->sps->width;
 369     y_end = y0 + ctb_size;
 370     if (y_end > s->sps->height)
 371         y_end = s->sps->height;
 372
 373     tc_offset   = cur_tc_offset;
 374     beta_offset = cur_beta_offset;
 375
 376     x_end2 = x_end;
 377     if (x_end2 != s->sps->width)
 378         x_end2 -= 8;
 379     for (y = y0; y < y_end; y += 8) {
 380         // vertical filtering luma
 381         for (x = x0 ? x0 : 8; x < x_end; x += 8) {
 382             const int bs0 = s->vertical_bs[(x +  y      * s->bs_width) >> 2];
 383             const int bs1 = s->vertical_bs[(x + (y + 4) * s->bs_width) >> 2];
 384             if (bs0 || bs1) {
 385                 const int qp = (get_qPy(s, x - 1, y)     + get_qPy(s, x, y)     + 1) >> 1;
 386
 387                 beta = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
 388
 389                 tc[0]   = bs0 ? TC_CALC(qp, bs0) : 0;
 390                 tc[1]   = bs1 ? TC_CALC(qp, bs1) : 0;
 391                 src     = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->sps->pixel_shift)];
 392                 if (pcmf) {
 393                     no_p[0] = get_pcm(s, x - 1, y);
 394                     no_p[1] = get_pcm(s, x - 1, y + 4);
 395                     no_q[0] = get_pcm(s, x, y);
 396                     no_q[1] = get_pcm(s, x, y + 4);
 397                     s->hevcdsp.hevc_v_loop_filter_luma_c(src,
 398                                                          s->frame->linesize[LUMA],
 399                                                          beta, tc, no_p, no_q);
 400                 } else
 401                     s->hevcdsp.hevc_v_loop_filter_luma(src,
 402                                                        s->frame->linesize[LUMA],
 403                                                        beta, tc, no_p, no_q);
 404             }
 405         }
 406
 407         if(!y)
 408              continue;
 409
 410         // horizontal filtering luma
 411         for (x = x0 ? x0 - 8 : 0; x < x_end2; x += 8) {
 412             const int bs0 = s->horizontal_bs[( x      + y * s->bs_width) >> 2];
 413             const int bs1 = s->horizontal_bs[((x + 4) + y * s->bs_width) >> 2];
 414             if (bs0 || bs1) {
 415                 const int qp = (get_qPy(s, x, y - 1)     + get_qPy(s, x, y)     + 1) >> 1;
 416
 417                 tc_offset   = x >= x0 ? cur_tc_offset : left_tc_offset;
 418                 beta_offset = x >= x0 ? cur_beta_offset : left_beta_offset;
 419
 420                 beta = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
 421                 tc[0]   = bs0 ? TC_CALC(qp, bs0) : 0;
 422                 tc[1]   = bs1 ? TC_CALC(qp, bs1) : 0;
 423                 src     = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->sps->pixel_shift)];
 424                 if (pcmf) {
 425                     no_p[0] = get_pcm(s, x, y - 1);
 426                     no_p[1] = get_pcm(s, x + 4, y - 1);
 427                     no_q[0] = get_pcm(s, x, y);
 428                     no_q[1] = get_pcm(s, x + 4, y);
 429                     s->hevcdsp.hevc_h_loop_filter_luma_c(src,
 430                                                          s->frame->linesize[LUMA],
 431                                                          beta, tc, no_p, no_q);
 432                 } else
 433                     s->hevcdsp.hevc_h_loop_filter_luma(src,
 434                                                        s->frame->linesize[LUMA],
 435                                                        beta, tc, no_p, no_q);
 436             }
 437         }
 438     }
 439
 440     for (chroma = 1; chroma <= 2; chroma++) {
 441         int h = 1 << s->sps->hshift[chroma];
 442         int v = 1 << s->sps->vshift[chroma];
 443
 444         // vertical filtering chroma
 445         for (y = y0; y < y_end; y += (8 * v)) {
 446             for (x = x0 ? x0 : 8 * h; x < x_end; x += (8 * h)) {
 447                 const int bs0 = s->vertical_bs[(x +  y            * s->bs_width) >> 2];
 448                 const int bs1 = s->vertical_bs[(x + (y + (4 * v)) * s->bs_width) >> 2];
 449
 450                 if ((bs0 == 2) || (bs1 == 2)) {
 451                     const int qp0 = (get_qPy(s, x - 1, y)           + get_qPy(s, x, y)           + 1) >> 1;
 452                     const int qp1 = (get_qPy(s, x - 1, y + (4 * v)) + get_qPy(s, x, y + (4 * v)) + 1) >> 1;
 453
 454                     c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
 455                     c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
 456                     src       = &s->frame->data[chroma][(y >> s->sps->vshift[chroma]) * s->frame->linesize[chroma] + ((x >> s->sps->hshift[chroma]) << s->sps->pixel_shift)];
 457                     if (pcmf) {
 458                         no_p[0] = get_pcm(s, x - 1, y);
 459                         no_p[1] = get_pcm(s, x - 1, y + (4 * v));
 460                         no_q[0] = get_pcm(s, x, y);
 461                         no_q[1] = get_pcm(s, x, y + (4 * v));
 462                         s->hevcdsp.hevc_v_loop_filter_chroma_c(src,
 463                                                                s->frame->linesize[chroma],
 464                                                                c_tc, no_p, no_q);
 465                     } else
 466                         s->hevcdsp.hevc_v_loop_filter_chroma(src,
 467                                                              s->frame->linesize[chroma],
 468                                                              c_tc, no_p, no_q);
 469                 }
 470             }
 471
 472             if(!y)
 473                  continue;
 474
 475             // horizontal filtering chroma
 476             tc_offset = x0 ? left_tc_offset : cur_tc_offset;
 477             x_end2 = x_end;
 478             if (x_end != s->sps->width)
 479                 x_end2 = x_end - 8 * h;
 480             for (x = x0 ? x0 - 8 * h : 0; x < x_end2; x += (8 * h)) {
 481                 const int bs0 = s->horizontal_bs[( x          + y * s->bs_width) >> 2];
 482                 const int bs1 = s->horizontal_bs[((x + 4 * h) + y * s->bs_width) >> 2];
 483                 if ((bs0 == 2) || (bs1 == 2)) {
 484                     const int qp0 = bs0 == 2 ? (get_qPy(s, x,           y - 1) + get_qPy(s, x,           y) + 1) >> 1 : 0;
 485                     const int qp1 = bs1 == 2 ? (get_qPy(s, x + (4 * h), y - 1) + get_qPy(s, x + (4 * h), y) + 1) >> 1 : 0;
 486
 487                     c_tc[0]   = bs0 == 2 ? chroma_tc(s, qp0, chroma, tc_offset)     : 0;
 488                     c_tc[1]   = bs1 == 2 ? chroma_tc(s, qp1, chroma, cur_tc_offset) : 0;
 489                     src       = &s->frame->data[chroma][(y >> s->sps->vshift[1]) * s->frame->linesize[chroma] + ((x >> s->sps->hshift[1]) << s->sps->pixel_shift)];
 490                     if (pcmf) {
 491                         no_p[0] = get_pcm(s, x,           y - 1);
 492                         no_p[1] = get_pcm(s, x + (4 * h), y - 1);
 493                         no_q[0] = get_pcm(s, x,           y);
 494                         no_q[1] = get_pcm(s, x + (4 * h), y);
 495                         s->hevcdsp.hevc_h_loop_filter_chroma_c(src,
 496                                                                s->frame->linesize[chroma],
 497                                                                c_tc, no_p, no_q);
 498                     } else
 499                         s->hevcdsp.hevc_h_loop_filter_chroma(src,
 500                                                              s->frame->linesize[chroma],
 501                                                              c_tc, no_p, no_q);
 502                 }
 503             }
 504         }
 505     }
 506 }
 507
 508 static int boundary_strength(HEVCContext *s, MvField *curr, MvField *neigh,
 509                              RefPicList *neigh_refPicList)
 510 {
 511     if (curr->pred_flag == PF_BI &&  neigh->pred_flag == PF_BI) {
 512         // same L0 and L1
 513         if (s->ref->refPicList[0].list[curr->ref_idx[0]] == neigh_refPicList[0].list[neigh->ref_idx[0]]  &&
 514             s->ref->refPicList[0].list[curr->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]] &&
 515             neigh_refPicList[0].list[neigh->ref_idx[0]] == neigh_refPicList[1].list[neigh->ref_idx[1]]) {
 516             if ((FFABS(neigh->mv[0].x - curr->mv[0].x) >= 4 || FFABS(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
 517                  FFABS(neigh->mv[1].x - curr->mv[1].x) >= 4 || FFABS(neigh->mv[1].y - curr->mv[1].y) >= 4) &&
 518                 (FFABS(neigh->mv[1].x - curr->mv[0].x) >= 4 || FFABS(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
 519                  FFABS(neigh->mv[0].x - curr->mv[1].x) >= 4 || FFABS(neigh->mv[0].y - curr->mv[1].y) >= 4))
 520                 return 1;
 521             else
 522                 return 0;
 523         } else if (neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
 524                    neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
 525             if (FFABS(neigh->mv[0].x - curr->mv[0].x) >= 4 || FFABS(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
 526                 FFABS(neigh->mv[1].x - curr->mv[1].x) >= 4 || FFABS(neigh->mv[1].y - curr->mv[1].y) >= 4)
 527                 return 1;
 528             else
 529                 return 0;
 530         } else if (neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
 531                    neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
 532             if (FFABS(neigh->mv[1].x - curr->mv[0].x) >= 4 || FFABS(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
 533                 FFABS(neigh->mv[0].x - curr->mv[1].x) >= 4 || FFABS(neigh->mv[0].y - curr->mv[1].y) >= 4)
 534                 return 1;
 535             else
 536                 return 0;
 537         } else {
 538             return 1;
 539         }
 540     } else if ((curr->pred_flag != PF_BI) && (neigh->pred_flag != PF_BI)){ // 1 MV
 541         Mv A, B;
 542         int ref_A, ref_B;
 543
 544         if (curr->pred_flag & 1) {
 545             A     = curr->mv[0];
 546             ref_A = s->ref->refPicList[0].list[curr->ref_idx[0]];
 547         } else {
 548             A     = curr->mv[1];
 549             ref_A = s->ref->refPicList[1].list[curr->ref_idx[1]];
 550         }
 551
 552         if (neigh->pred_flag & 1) {
 553             B     = neigh->mv[0];
 554             ref_B = neigh_refPicList[0].list[neigh->ref_idx[0]];
 555         } else {
 556             B     = neigh->mv[1];
 557             ref_B = neigh_refPicList[1].list[neigh->ref_idx[1]];
 558         }
 559
 560         if (ref_A == ref_B) {
 561             if (FFABS(A.x - B.x) >= 4 || FFABS(A.y - B.y) >= 4)
 562                 return 1;
 563             else
 564                 return 0;
 565         } else
 566             return 1;
 567     }
 568
 569     return 1;
 570 }
 571
 572 void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
 573                                            int log2_trafo_size)
 574 {
 575     HEVCLocalContext *lc = s->HEVClc;
 576     MvField *tab_mvf     = s->ref->tab_mvf;
 577     int log2_min_pu_size = s->sps->log2_min_pu_size;
 578     int log2_min_tu_size = s->sps->log2_min_tb_size;
 579     int min_pu_width     = s->sps->min_pu_width;
 580     int min_tu_width     = s->sps->min_tb_width;
 581     int is_intra = tab_mvf[(y0 >> log2_min_pu_size) * min_pu_width +
 582                            (x0 >> log2_min_pu_size)].pred_flag == PF_INTRA;
 583     int boundary_upper, boundary_left;
 584     int i, j, bs;
 585
 586     boundary_upper = y0 > 0 && !(y0 & 7);
 587     if (boundary_upper &&
 588         ((!s->sh.slice_loop_filter_across_slices_enabled_flag &&
 589           lc->boundary_flags & BOUNDARY_UPPER_SLICE &&
 590           (y0 % (1 << s->sps->log2_ctb_size)) == 0) ||
 591          (!s->pps->loop_filter_across_tiles_enabled_flag &&
 592           lc->boundary_flags & BOUNDARY_UPPER_TILE &&
 593           (y0 % (1 << s->sps->log2_ctb_size)) == 0)))
 594         boundary_upper = 0;
 595
 596     if (boundary_upper) {
 597         RefPicList *rpl_top = (lc->boundary_flags & BOUNDARY_UPPER_SLICE) ?
 598                               ff_hevc_get_ref_list(s, s->ref, x0, y0 - 1) :
 599                               s->ref->refPicList;
 600         int yp_pu = (y0 - 1) >> log2_min_pu_size;
 601         int yq_pu =  y0      >> log2_min_pu_size;
 602         int yp_tu = (y0 - 1) >> log2_min_tu_size;
 603         int yq_tu =  y0      >> log2_min_tu_size;
 604
 605             for (i = 0; i < (1 << log2_trafo_size); i += 4) {
 606                 int x_pu = (x0 + i) >> log2_min_pu_size;
 607                 int x_tu = (x0 + i) >> log2_min_tu_size;
 608                 MvField *top  = &tab_mvf[yp_pu * min_pu_width + x_pu];
 609                 MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
 610                 uint8_t top_cbf_luma  = s->cbf_luma[yp_tu * min_tu_width + x_tu];
 611                 uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
 612
 613                 if (curr->pred_flag == PF_INTRA || top->pred_flag == PF_INTRA)
 614                     bs = 2;
 615                 else if (curr_cbf_luma || top_cbf_luma)
 616                     bs = 1;
 617                 else
 618                     bs = boundary_strength(s, curr, top, rpl_top);
 619                 s->horizontal_bs[((x0 + i) + y0 * s->bs_width) >> 2] = bs;
 620             }
 621     }
 622
 623     // bs for vertical TU boundaries
 624     boundary_left = x0 > 0 && !(x0 & 7);
 625     if (boundary_left &&
 626         ((!s->sh.slice_loop_filter_across_slices_enabled_flag &&
 627           lc->boundary_flags & BOUNDARY_LEFT_SLICE &&
 628           (x0 % (1 << s->sps->log2_ctb_size)) == 0) ||
 629          (!s->pps->loop_filter_across_tiles_enabled_flag &&
 630           lc->boundary_flags & BOUNDARY_LEFT_TILE &&
 631           (x0 % (1 << s->sps->log2_ctb_size)) == 0)))
 632         boundary_left = 0;
 633
 634     if (boundary_left) {
 635         RefPicList *rpl_left = (lc->boundary_flags & BOUNDARY_LEFT_SLICE) ?
 636                                ff_hevc_get_ref_list(s, s->ref, x0 - 1, y0) :
 637                                s->ref->refPicList;
 638         int xp_pu = (x0 - 1) >> log2_min_pu_size;
 639         int xq_pu =  x0      >> log2_min_pu_size;
 640         int xp_tu = (x0 - 1) >> log2_min_tu_size;
 641         int xq_tu =  x0      >> log2_min_tu_size;
 642
 643             for (i = 0; i < (1 << log2_trafo_size); i += 4) {
 644                 int y_pu      = (y0 + i) >> log2_min_pu_size;
 645                 int y_tu      = (y0 + i) >> log2_min_tu_size;
 646                 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
 647                 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
 648                 uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
 649                 uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
 650
 651                 if (curr->pred_flag == PF_INTRA || left->pred_flag == PF_INTRA)
 652                     bs = 2;
 653                 else if (curr_cbf_luma || left_cbf_luma)
 654                     bs = 1;
 655                 else
 656                     bs = boundary_strength(s, curr, left, rpl_left);
 657                 s->vertical_bs[(x0 + (y0 + i) * s->bs_width) >> 2] = bs;
 658             }
 659     }
 660
 661     if (log2_trafo_size > log2_min_pu_size && !is_intra) {
 662         RefPicList *rpl = s->ref->refPicList;
 663
 664         // bs for TU internal horizontal PU boundaries
 665         for (j = 8; j < (1 << log2_trafo_size); j += 8) {
 666             int yp_pu = (y0 + j - 1) >> log2_min_pu_size;
 667             int yq_pu = (y0 + j)     >> log2_min_pu_size;
 668
 669             for (i = 0; i < (1 << log2_trafo_size); i += 4) {
 670                 int x_pu = (x0 + i) >> log2_min_pu_size;
 671                 MvField *top  = &tab_mvf[yp_pu * min_pu_width + x_pu];
 672                 MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
 673
 674                 bs = boundary_strength(s, curr, top, rpl);
 675                 s->horizontal_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
 676             }
 677         }
 678
 679         // bs for TU internal vertical PU boundaries
 680         for (j = 0; j < (1 << log2_trafo_size); j += 4) {
 681             int y_pu = (y0 + j) >> log2_min_pu_size;
 682
 683             for (i = 8; i < (1 << log2_trafo_size); i += 8) {
 684                 int xp_pu = (x0 + i - 1) >> log2_min_pu_size;
 685                 int xq_pu = (x0 + i)     >> log2_min_pu_size;
 686                 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
 687                 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
 688
 689                 bs = boundary_strength(s, curr, left, rpl);
 690                 s->vertical_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
 691             }
 692         }
 693     }
 694 }
 695
 696 #undef LUMA
 697 #undef CB
 698 #undef CR
 699
 700 void ff_hevc_hls_filter(HEVCContext *s, int x, int y, int ctb_size)
 701 {
 702     int x_end = x >= s->sps->width  - ctb_size;
 703     deblocking_filter_CTB(s, x, y);
 704     if (s->sps->sao_enabled) {
 705         int y_end = y >= s->sps->height - ctb_size;
 706         if (y && x)
 707             sao_filter_CTB(s, x - ctb_size, y - ctb_size);
 708         if (x && y_end)
 709             sao_filter_CTB(s, x - ctb_size, y);
 710         if (y && x_end) {
 711             sao_filter_CTB(s, x, y - ctb_size);
 712             if (s->threads_type & FF_THREAD_FRAME )
 713                 ff_thread_report_progress(&s->ref->tf, y, 0);
 714         }
 715         if (x_end && y_end) {
 716             sao_filter_CTB(s, x , y);
 717             if (s->threads_type & FF_THREAD_FRAME )
 718                 ff_thread_report_progress(&s->ref->tf, y + ctb_size, 0);
 719         }
 720     } else if (s->threads_type & FF_THREAD_FRAME && x_end)
 721         ff_thread_report_progress(&s->ref->tf, y + ctb_size - 4, 0);
 722 }
 723
 724 void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size)
 725 {
 726     int x_end = x_ctb >= s->sps->width  - ctb_size;
 727     int y_end = y_ctb >= s->sps->height - ctb_size;
 728     if (y_ctb && x_ctb)
 729         ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size, ctb_size);
 730     if (y_ctb && x_end)
 731         ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size, ctb_size);
 732     if (x_ctb && y_end)
 733         ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb, ctb_size);
 734 }