FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
hevc_filter.c
Go to the documentation of this file.
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 #include "bit_depth_template.c"
32 
33 #define LUMA 0
34 #define CB 1
35 #define CR 2
36 
37 static const uint8_t tctable[54] = {
38  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, // QP 0...18
39  1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, // QP 19...37
40  5, 5, 6, 6, 7, 8, 9,10,11,13,14,16,18,20,22,24 // QP 38...53
41 };
42 
43 static const uint8_t betatable[52] = {
44  0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6, 7, 8, // QP 0...18
45  9,10,11,12,13,14,15,16,17,18,20,22,24,26,28,30,32,34,36, // QP 19...37
46  38,40,42,44,46,48,50,52,54,56,58,60,62,64 // QP 38...51
47 };
48 
49 static int chroma_tc(HEVCContext *s, int qp_y, int c_idx, int tc_offset)
50 {
51  static const int qp_c[] = { 29, 30, 31, 32, 33, 33, 34, 34, 35, 35, 36, 36, 37, 37 };
52  int qp_i, offset;
53  int qp;
54  int idxt;
55 
56  // slice qp offset is not used for deblocking
57  if (c_idx == 1)
58  offset = s->pps->cb_qp_offset;
59  else
60  offset = s->pps->cr_qp_offset;
61 
62  qp_i = av_clip_c(qp_y + offset, 0, 57);
63  if (qp_i < 30)
64  qp = qp_i;
65  else if (qp_i > 43)
66  qp = qp_i - 6;
67  else
68  qp = qp_c[qp_i - 30];
69 
70  idxt = av_clip_c(qp + DEFAULT_INTRA_TC_OFFSET + tc_offset, 0, 53);
71  return tctable[idxt];
72 }
73 
74 static int get_qPy_pred(HEVCContext *s, int xC, int yC, int xBase, int yBase, int log2_cb_size)
75 {
76  HEVCLocalContext *lc = s->HEVClc;
77  int ctb_size_mask = (1 << s->sps->log2_ctb_size) - 1;
78  int MinCuQpDeltaSizeMask = (1 << (s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth)) - 1;
79  int xQgBase = xBase - ( xBase & MinCuQpDeltaSizeMask );
80  int yQgBase = yBase - ( yBase & MinCuQpDeltaSizeMask );
81  int min_cb_width = s->sps->min_cb_width;
82  int min_cb_height = s->sps->min_cb_height;
83  int x_cb = xQgBase >> s->sps->log2_min_cb_size;
84  int y_cb = yQgBase >> s->sps->log2_min_cb_size;
85  int availableA = (xBase & ctb_size_mask) && (xQgBase & ctb_size_mask);
86  int availableB = (yBase & ctb_size_mask) && (yQgBase & ctb_size_mask);
87  int qPy_pred;
88  int qPy_a;
89  int qPy_b;
90 
91  // qPy_pred
92  if (lc->first_qp_group) {
93  lc->first_qp_group = !lc->tu.is_cu_qp_delta_coded;
94  qPy_pred = s->sh.slice_qp;
95  } else {
96  qPy_pred = lc->qp_y;
97  if (log2_cb_size < s->sps->log2_ctb_size - s->pps->diff_cu_qp_delta_depth) {
98  static const int offsetX[8][8] = {
99  {-1, 1, 3, 1, 7, 1, 3, 1},
100  { 0, 0, 0, 0, 0, 0, 0, 0},
101  { 1, 3, 1, 3, 1, 3, 1, 3},
102  { 2, 2, 2, 2, 2, 2, 2, 2},
103  { 3, 5, 7, 5, 3, 5, 7, 5},
104  { 4, 4, 4, 4, 4, 4, 4, 4},
105  { 5, 7, 5, 7, 5, 7, 5, 7},
106  { 6, 6, 6, 6, 6, 6, 6, 6}
107  };
108  static const int offsetY[8][8] = {
109  { 7, 0, 1, 2, 3, 4, 5, 6},
110  { 0, 1, 2, 3, 4, 5, 6, 7},
111  { 1, 0, 3, 2, 5, 4, 7, 6},
112  { 0, 1, 2, 3, 4, 5, 6, 7},
113  { 3, 0, 1, 2, 7, 4, 5, 6},
114  { 0, 1, 2, 3, 4, 5, 6, 7},
115  { 1, 0, 3, 2, 5, 4, 7, 6},
116  { 0, 1, 2, 3, 4, 5, 6, 7}
117  };
118  int xC0b = (xC - (xC & ctb_size_mask)) >> s->sps->log2_min_cb_size;
119  int yC0b = (yC - (yC & ctb_size_mask)) >> s->sps->log2_min_cb_size;
120  int idxX = (xQgBase & ctb_size_mask) >> s->sps->log2_min_cb_size;
121  int idxY = (yQgBase & ctb_size_mask) >> s->sps->log2_min_cb_size;
122  int idx_mask = ctb_size_mask >> s->sps->log2_min_cb_size;
123  int x, y;
124 
125  x = FFMIN(xC0b + offsetX[idxX][idxY], min_cb_width - 1);
126  y = FFMIN(yC0b + (offsetY[idxX][idxY] & idx_mask), min_cb_height - 1);
127 
128  if (xC0b == (lc->start_of_tiles_x >> s->sps->log2_min_cb_size) &&
129  offsetX[idxX][idxY] == -1) {
130  x = (lc->end_of_tiles_x >> s->sps->log2_min_cb_size) - 1;
131  y = yC0b - 1;
132  }
133  qPy_pred = s->qp_y_tab[y * min_cb_width + x];
134  }
135  }
136 
137  // qPy_a
138  if (availableA == 0)
139  qPy_a = qPy_pred;
140  else
141  qPy_a = s->qp_y_tab[(x_cb - 1) + y_cb * min_cb_width];
142 
143  // qPy_b
144  if (availableB == 0)
145  qPy_b = qPy_pred;
146  else
147  qPy_b = s->qp_y_tab[x_cb + (y_cb - 1) * min_cb_width];
148 
149  return (qPy_a + qPy_b + 1) >> 1;
150 }
151 
152 void ff_hevc_set_qPy(HEVCContext *s, int xC, int yC, int xBase, int yBase, int log2_cb_size)
153 {
154  int qp_y = get_qPy_pred(s, xC, yC, xBase, yBase, log2_cb_size);
155 
156  if (s->HEVClc->tu.cu_qp_delta != 0) {
157  int off = s->sps->qp_bd_offset;
158  s->HEVClc->qp_y = FFUMOD(qp_y + s->HEVClc->tu.cu_qp_delta + 52 + 2 * off,
159  52 + off) - off;
160  } else
161  s->HEVClc->qp_y = qp_y;
162 }
163 
164 static int get_qPy(HEVCContext *s, int xC, int yC)
165 {
166  int log2_min_cb_size = s->sps->log2_min_cb_size;
167  int x = xC >> log2_min_cb_size;
168  int y = yC >> log2_min_cb_size;
169  return s->qp_y_tab[x + y * s->sps->min_cb_width];
170 }
171 
172 static void copy_CTB(uint8_t *dst, uint8_t *src, int width, int height, int stride)
173 {
174  int i;
175 
176  for(i=0; i< height; i++){
177  memcpy(dst, src, width);
178  dst += stride;
179  src += stride;
180  }
181 }
182 
183 #define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
184 
185 static void sao_filter_CTB(HEVCContext *s, int x, int y)
186 {
187  // TODO: This should be easily parallelizable
188  // TODO: skip CBs when (cu_transquant_bypass_flag || (pcm_loop_filter_disable_flag && pcm_flag))
189  int c_idx = 0;
190  int class = 1, class_index;
191  int edges[4]; // 0 left 1 top 2 right 3 bottom
192  SAOParams *sao[4];
193  int classes[4];
194  int x_shift = 0, y_shift = 0;
195  int x_ctb = x>>s->sps->log2_ctb_size;
196  int y_ctb = y>>s->sps->log2_ctb_size;
197  int ctb_addr_rs = y_ctb * s->sps->ctb_width + x_ctb;
198  int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
199 
200  // flags indicating unfilterable edges
201  uint8_t vert_edge[] = {0,0,0,0};
202  uint8_t horiz_edge[] = {0,0,0,0};
203  uint8_t diag_edge[] = {0,0,0,0};
204  uint8_t lfase[3]; // current, above, left
205  uint8_t no_tile_filter = s->pps->tiles_enabled_flag && !s->pps->loop_filter_across_tiles_enabled_flag;
206  uint8_t left_tile_edge = 0;
207  uint8_t up_tile_edge = 0;
208 
209  sao[0] = &CTB(s->sao, x_ctb, y_ctb);
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  lfase[0] = CTB(s->filter_slice_edges, x_ctb, y_ctb);
215  classes[0] = 0;
216 
217  if (!edges[0]) {
218  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]];
219  sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb);
220  vert_edge[0] = (!lfase[0] && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb)) || left_tile_edge;
221  vert_edge[2] = vert_edge[0];
222  lfase[2] = CTB(s->filter_slice_edges, x_ctb - 1, y_ctb);
223  classes[class] = 2;
224  class++;
225  x_shift = 8;
226  }
227 
228  if (!edges[1]) {
229  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]];
230  sao[class] = &CTB(s->sao, x_ctb, y_ctb - 1);
231  horiz_edge[0] = (!lfase[0] && CTB(s->tab_slice_address, x_ctb, y_ctb) != CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) || up_tile_edge;
232  horiz_edge[1] = horiz_edge[0];
233  lfase[1] = CTB(s->filter_slice_edges, x_ctb, y_ctb - 1);
234  classes[class] = 1;
235  class++;
236  y_shift = 4;
237 
238  if (!edges[0]) {
239  classes[class] = 3;
240  sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb - 1);
241  class++;
242 
243  // Tile check here is done current CTB row/col, not above/left like you'd expect,
244  //but that is because the tile boundary always extends through the whole pic
245  vert_edge[1] = (!lfase[1] && CTB(s->tab_slice_address, x_ctb, y_ctb - 1) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || left_tile_edge;
246  vert_edge[3] = vert_edge[1];
247  horiz_edge[2] = (!lfase[2] && CTB(s->tab_slice_address, x_ctb - 1, y_ctb) != CTB(s->tab_slice_address, x_ctb - 1, y_ctb - 1)) || up_tile_edge;
248  horiz_edge[3] = horiz_edge[2];
249  diag_edge[0] = (!lfase[0] && 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;
250  diag_edge[3] = diag_edge[0];
251 
252  // Does left CTB comes after above CTB?
253  if(CTB(s->tab_slice_address, x_ctb - 1, y_ctb) > CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
254  diag_edge[2] = !lfase[2] || left_tile_edge || up_tile_edge;
255  diag_edge[1] = diag_edge[2];
256  } else if(CTB(s->tab_slice_address, x_ctb - 1, y_ctb) < CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
257  diag_edge[1] = !lfase[1] || left_tile_edge || up_tile_edge;
258  diag_edge[2] = diag_edge[1];
259  } else {
260  // Same slice, only consider tiles
261  diag_edge[2] = left_tile_edge || up_tile_edge;
262  diag_edge[1] = diag_edge[2];
263  }
264  }
265  }
266 
267  for (c_idx = 0; c_idx < 3; c_idx++) {
268  int chroma = c_idx ? 1 : 0;
269  int x0 = x >> chroma;
270  int y0 = y >> chroma;
271  int stride = s->frame->linesize[c_idx];
272  int ctb_size = (1 << (s->sps->log2_ctb_size)) >> s->sps->hshift[c_idx];
273  int width = FFMIN(ctb_size,
274  (s->sps->width >> s->sps->hshift[c_idx]) - x0);
275  int height = FFMIN(ctb_size,
276  (s->sps->height >> s->sps->vshift[c_idx]) - y0);
277 
278  uint8_t *src = &s->frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)];
279  uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)];
280  int offset = (y_shift >> chroma) * stride + ((x_shift >> chroma) << s->sps->pixel_shift);
281 
282  copy_CTB(dst - offset, src - offset,
283  (edges[2] ? width + (x_shift >> chroma) : width) << s->sps->pixel_shift,
284  (edges[3] ? height + (y_shift >> chroma) : height), stride);
285 
286  for (class_index = 0; class_index < class; class_index++) {
287 
288  switch (sao[class_index]->type_idx[c_idx]) {
289  case SAO_BAND:
290  s->hevcdsp.sao_band_filter[classes[class_index]](dst, src, stride, sao[class_index], edges, width, height, c_idx);
291  break;
292  case SAO_EDGE:
293  s->hevcdsp.sao_edge_filter[classes[class_index]](dst, src, stride, sao[class_index], edges, width, height, c_idx, vert_edge[classes[class_index]], horiz_edge[classes[class_index]], diag_edge[classes[class_index]]);
294  break;
295  }
296  }
297  }
298 }
299 
300 static int get_pcm(HEVCContext *s, int x, int y)
301 {
302  int log2_min_pu_size = s->sps->log2_min_pu_size;
303  int x_pu = x >> log2_min_pu_size;
304  int y_pu = y >> log2_min_pu_size;
305 
306  if (x < 0 || x_pu >= s->sps->min_pu_width ||
307  y < 0 || y_pu >= s->sps->min_pu_height)
308  return 2;
309  return s->is_pcm[y_pu * s->sps->min_pu_width + x_pu];
310 }
311 
312 #define TC_CALC(qp, bs) tctable[av_clip((qp) + DEFAULT_INTRA_TC_OFFSET * ((bs) - 1) + ((tc_offset >> 1) << 1), 0, MAX_QP + DEFAULT_INTRA_TC_OFFSET)]
313 
314 static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
315 {
316  uint8_t *src;
317  int x, y;
318  int chroma;
319  int c_tc[2];
320  int beta[2];
321  int tc[2];
322  uint8_t no_p[2] = {0};
323  uint8_t no_q[2] = {0};
324 
325  int log2_ctb_size = s->sps->log2_ctb_size;
326  int x_end, y_end;
327  int ctb_size = 1<<log2_ctb_size;
328  int ctb = (x0 >> log2_ctb_size) + (y0 >> log2_ctb_size) * s->sps->ctb_width;
329  int cur_tc_offset = s->deblock[ctb].tc_offset;
330  int cur_beta_offset = s->deblock[ctb].beta_offset;
331  int left_tc_offset, left_beta_offset;
332  int tc_offset, beta_offset;
333  int pcmf = (s->sps->pcm_enabled_flag && s->sps->pcm.loop_filter_disable_flag) ||
334  s->pps->transquant_bypass_enable_flag;
335 
336  if (x0) {
337  left_tc_offset = s->deblock[ctb-1].tc_offset;
338  left_beta_offset = s->deblock[ctb-1].beta_offset;
339  }
340 
341  x_end = x0+ctb_size;
342  if (x_end > s->sps->width)
343  x_end = s->sps->width;
344  y_end = y0+ctb_size;
345  if (y_end > s->sps->height)
346  y_end = s->sps->height;
347 
348  tc_offset = cur_tc_offset;
349  beta_offset = cur_beta_offset;
350 
351  // vertical filtering luma
352  for (y = y0; y < y_end; y += 8) {
353  for (x = x0 ? x0 : 8; x < x_end; x += 8) {
354  const int bs0 = s->vertical_bs[(x >> 3) + (y >> 2) * s->bs_width];
355  const int bs1 = s->vertical_bs[(x >> 3) + ((y + 4) >> 2) * s->bs_width];
356  if (bs0 || bs1) {
357  const int qp0 = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
358  const int qp1 = (get_qPy(s, x - 1, y + 4) + get_qPy(s, x, y + 4) + 1) >> 1;
359 
360  beta[0] = betatable[av_clip(qp0 + ((beta_offset >> 1) << 1), 0, MAX_QP)];
361  beta[1] = betatable[av_clip(qp1 + ((beta_offset >> 1) << 1), 0, MAX_QP)];
362  tc[0] = bs0 ? TC_CALC(qp0, bs0) : 0;
363  tc[1] = bs1 ? TC_CALC(qp1, bs1) : 0;
364  src = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->sps->pixel_shift)];
365  if (pcmf) {
366  no_p[0] = get_pcm(s, x - 1, y);
367  no_p[1] = get_pcm(s, x - 1, y + 4);
368  no_q[0] = get_pcm(s, x, y);
369  no_q[1] = get_pcm(s, x, y + 4);
370  s->hevcdsp.hevc_v_loop_filter_luma_c(src, s->frame->linesize[LUMA], beta, tc, no_p, no_q);
371  } else
372  s->hevcdsp.hevc_v_loop_filter_luma(src, s->frame->linesize[LUMA], beta, tc, no_p, no_q);
373  }
374  }
375  }
376 
377  // vertical filtering chroma
378  for (chroma = 1; chroma <= 2; chroma++) {
379  for (y = y0; y < y_end; y += 16) {
380  for (x = x0 ? x0:16; x < x_end; x += 16) {
381  const int bs0 = s->vertical_bs[(x >> 3) + (y >> 2) * s->bs_width];
382  const int bs1 = s->vertical_bs[(x >> 3) + ((y + 8) >> 2) * s->bs_width];
383  if ((bs0 == 2) || (bs1 == 2)) {
384  const int qp0 = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
385  const int qp1 = (get_qPy(s, x - 1, y + 8) + get_qPy(s, x, y + 8) + 1) >> 1;
386 
387  c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
388  c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
389  src = &s->frame->data[chroma][(y / 2) * s->frame->linesize[chroma] + ((x / 2) << s->sps->pixel_shift)];
390  if (pcmf) {
391  no_p[0] = get_pcm(s, x - 1, y);
392  no_p[1] = get_pcm(s, x - 1, y + 8);
393  no_q[0] = get_pcm(s, x, y);
394  no_q[1] = get_pcm(s, x, y + 8);
395  s->hevcdsp.hevc_v_loop_filter_chroma_c(src, s->frame->linesize[chroma], c_tc, no_p, no_q);
396  } else
397  s->hevcdsp.hevc_v_loop_filter_chroma(src, s->frame->linesize[chroma], c_tc, no_p, no_q);
398  }
399  }
400  }
401  }
402 
403  // horizontal filtering luma
404  if (x_end != s->sps->width)
405  x_end -= 8;
406  for (y = y0 ? y0 : 8; y < y_end; y += 8) {
407  for (x = x0 ? x0 - 8 : 0; x < x_end; x += 8) {
408  const int bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
409  const int bs1 = s->horizontal_bs[(x + 4 + y * s->bs_width) >> 2];
410  if (bs0 || bs1) {
411  const int qp0 = (get_qPy(s, x, y - 1) + get_qPy(s, x, y) + 1) >> 1;
412  const int qp1 = (get_qPy(s, x + 4, y - 1) + get_qPy(s, x + 4, y) + 1) >> 1;
413 
414  tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
415  beta_offset = x >= x0 ? cur_beta_offset : left_beta_offset;
416 
417  beta[0] = betatable[av_clip(qp0 + ((beta_offset >> 1) << 1), 0, MAX_QP)];
418  beta[1] = betatable[av_clip(qp1 + ((beta_offset >> 1) << 1), 0, MAX_QP)];
419  tc[0] = bs0 ? TC_CALC(qp0, bs0) : 0;
420  tc[1] = bs1 ? TC_CALC(qp1, bs1) : 0;
421  src = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->sps->pixel_shift)];
422  if (pcmf) {
423  no_p[0] = get_pcm(s, x, y - 1);
424  no_p[1] = get_pcm(s, x + 4, y - 1);
425  no_q[0] = get_pcm(s, x, y);
426  no_q[1] = get_pcm(s, x + 4, y);
427  s->hevcdsp.hevc_h_loop_filter_luma_c(src, s->frame->linesize[LUMA], beta, tc, no_p, no_q);
428  } else
429  s->hevcdsp.hevc_h_loop_filter_luma(src, s->frame->linesize[LUMA], beta, tc, no_p, no_q);
430  }
431  }
432  }
433 
434  // horizontal filtering chroma
435  for (chroma = 1; chroma <= 2; chroma++) {
436  for (y = y0 ? y0 : 16; y < y_end; y += 16) {
437  for (x = x0 - 8; x < x_end; x += 16) {
438  int bs0, bs1;
439  // to make sure no memory access over boundary when x = -8
440  // TODO: simplify with row based deblocking
441  if (x < 0) {
442  bs0 = 0;
443  bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
444  } else if (x >= x_end - 8) {
445  bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
446  bs1 = 0;
447  } else {
448  bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
449  bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
450  }
451 
452  if ((bs0 == 2) || (bs1 == 2)) {
453  const int qp0 = (bs0 == 2) ? ((get_qPy(s, x, y - 1) + get_qPy(s, x, y) + 1) >> 1) : 0;
454  const int qp1 = (bs1 == 2) ? ((get_qPy(s, x + 8, y - 1) + get_qPy(s, x + 8, y) + 1) >> 1) : 0;
455 
456  tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
457  c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
458  c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, cur_tc_offset) : 0;
459  src = &s->frame->data[chroma][(y / 2) * s->frame->linesize[chroma] + ((x / 2) << s->sps->pixel_shift)];
460  if (pcmf) {
461  no_p[0] = get_pcm(s, x, y - 1);
462  no_p[1] = get_pcm(s, x + 8, y - 1);
463  no_q[0] = get_pcm(s, x, y);
464  no_q[1] = get_pcm(s, x + 8, y);
465  s->hevcdsp.hevc_h_loop_filter_chroma_c(src, s->frame->linesize[chroma], c_tc, no_p, no_q);
466  } else
467  s->hevcdsp.hevc_h_loop_filter_chroma(src, s->frame->linesize[chroma], c_tc, no_p, no_q);
468  }
469  }
470  }
471  }
472 }
473 
474 static int boundary_strength(HEVCContext *s, MvField *curr,
475  uint8_t curr_cbf_luma, MvField *neigh,
476  uint8_t neigh_cbf_luma, RefPicList *neigh_refPicList,
477  int tu_border)
478 {
479  int mvs = curr->pred_flag[0] + curr->pred_flag[1];
480 
481  if (tu_border) {
482  if (curr->is_intra || neigh->is_intra)
483  return 2;
484  if (curr_cbf_luma || neigh_cbf_luma)
485  return 1;
486  }
487 
488  if (mvs == neigh->pred_flag[0] + neigh->pred_flag[1]) {
489  if (mvs == 2) {
490  // same L0 and L1
491  if (s->ref->refPicList[0].list[curr->ref_idx[0]] == neigh_refPicList[0].list[neigh->ref_idx[0]] &&
492  s->ref->refPicList[0].list[curr->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]] &&
493  neigh_refPicList[0].list[neigh->ref_idx[0]] == neigh_refPicList[1].list[neigh->ref_idx[1]]) {
494  if ((abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
495  abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4) &&
496  (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
497  abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4))
498  return 1;
499  else
500  return 0;
501  } else if (neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
502  neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
503  if (abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
504  abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4)
505  return 1;
506  else
507  return 0;
508  } else if (neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
509  neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
510  if (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
511  abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4)
512  return 1;
513  else
514  return 0;
515  } else {
516  return 1;
517  }
518  } else { // 1 MV
519  Mv A, B;
520  int ref_A;
521  int ref_B;
522 
523  if (curr->pred_flag[0]) {
524  A = curr->mv[0];
525  ref_A = s->ref->refPicList[0].list[curr->ref_idx[0]];
526  } else {
527  A = curr->mv[1];
528  ref_A = s->ref->refPicList[1].list[curr->ref_idx[1]];
529  }
530 
531  if (neigh->pred_flag[0]) {
532  B = neigh->mv[0];
533  ref_B = neigh_refPicList[0].list[neigh->ref_idx[0]];
534  } else {
535  B = neigh->mv[1];
536  ref_B = neigh_refPicList[1].list[neigh->ref_idx[1]];
537  }
538 
539  if (ref_A == ref_B) {
540  if (abs(A.x - B.x) >= 4 || abs(A.y - B.y) >= 4)
541  return 1;
542  else
543  return 0;
544  } else
545  return 1;
546  }
547  }
548 
549  return 1;
550 }
551 
552 void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0, int log2_trafo_size,
553  int slice_or_tiles_up_boundary, int slice_or_tiles_left_boundary)
554 {
555  MvField *tab_mvf = s->ref->tab_mvf;
556  int log2_min_pu_size = s->sps->log2_min_pu_size;
557  int log2_min_tu_size = s->sps->log2_min_tb_size;
558  int min_pu_width = s->sps->min_pu_width;
559  int min_tu_width = s->sps->min_tb_width;
560  int is_intra = tab_mvf[(y0 >> log2_min_pu_size) * min_pu_width + (x0 >> log2_min_pu_size)].is_intra;
561 
562  int i, j;
563  int bs;
564 
565  if (y0 > 0 && (y0 & 7) == 0) {
566  int yp_pu = (y0 - 1) >> log2_min_pu_size;
567  int yq_pu = y0 >> log2_min_pu_size;
568  int yp_tu = (y0 - 1) >> log2_min_tu_size;
569  int yq_tu = y0 >> log2_min_tu_size;
570 
571  for (i = 0; i < (1 << log2_trafo_size); i += 4) {
572  int x_pu = (x0 + i) >> log2_min_pu_size;
573  int x_tu = (x0 + i) >> log2_min_tu_size;
574  MvField *top = &tab_mvf[yp_pu * min_pu_width + x_pu];
575  MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
576  uint8_t top_cbf_luma = s->cbf_luma[yp_tu * min_tu_width + x_tu];
577  uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
578  RefPicList* top_refPicList = ff_hevc_get_ref_list(s, s->ref, x0 + i, y0 - 1);
579 
580  bs = boundary_strength(s, curr, curr_cbf_luma, top, top_cbf_luma, top_refPicList, 1);
581  if (!s->sh.slice_loop_filter_across_slices_enabled_flag && (slice_or_tiles_up_boundary & 1) && (y0 % (1 << s->sps->log2_ctb_size)) == 0)
582  bs = 0;
583  else if (!s->pps->loop_filter_across_tiles_enabled_flag && (slice_or_tiles_up_boundary & 2) && (y0 % (1 << s->sps->log2_ctb_size)) == 0)
584  bs = 0;
585  if (y0 == 0 || s->sh.disable_deblocking_filter_flag == 1)
586  bs = 0;
587  if (bs)
588  s->horizontal_bs[((x0 + i) + y0 * s->bs_width) >> 2] = bs;
589  }
590  }
591 
592  // bs for TU internal horizontal PU boundaries
593  if (log2_trafo_size > s->sps->log2_min_pu_size && !is_intra)
594  for (j = 8; j < (1 << log2_trafo_size); j += 8) {
595  int yp_pu = (y0 + j - 1) >> log2_min_pu_size;
596  int yq_pu = (y0 + j) >> log2_min_pu_size;
597  int yp_tu = (y0 + j - 1) >> log2_min_tu_size;
598  int yq_tu = (y0 + j) >> log2_min_tu_size;
599 
600 
601  for (i = 0; i < (1<<log2_trafo_size); i += 4) {
602  int x_pu = (x0 + i) >> log2_min_pu_size;
603  int x_tu = (x0 + i) >> log2_min_tu_size;
604  MvField *top = &tab_mvf[yp_pu * min_pu_width + x_pu];
605  MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
606  uint8_t top_cbf_luma = s->cbf_luma[yp_tu * min_tu_width + x_tu];
607  uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
608  RefPicList* top_refPicList = ff_hevc_get_ref_list(s, s->ref, x0 + i, y0 + j - 1);
609 
610  bs = boundary_strength(s, curr, curr_cbf_luma, top, top_cbf_luma, top_refPicList, 0);
611  if (s->sh.disable_deblocking_filter_flag == 1)
612  bs = 0;
613  if (bs)
614  s->horizontal_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
615  }
616  }
617 
618  // bs for vertical TU boundaries
619  if (x0 > 0 && (x0 & 7) == 0) {
620  int xp_pu = (x0 - 1) >> log2_min_pu_size;
621  int xq_pu = x0 >> log2_min_pu_size;
622  int xp_tu = (x0 - 1) >> log2_min_tu_size;
623  int xq_tu = x0 >> log2_min_tu_size;
624 
625  for (i = 0; i < (1 << log2_trafo_size); i += 4) {
626  int y_pu = (y0 + i) >> log2_min_pu_size;
627  int y_tu = (y0 + i) >> log2_min_tu_size;
628  MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
629  MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
630 
631  uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
632  uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
633  RefPicList* left_refPicList = ff_hevc_get_ref_list(s, s->ref, x0 - 1, y0 + i);
634 
635  bs = boundary_strength(s, curr, curr_cbf_luma, left, left_cbf_luma, left_refPicList, 1);
636  if (!s->sh.slice_loop_filter_across_slices_enabled_flag && (slice_or_tiles_left_boundary & 1) && (x0 % (1 << s->sps->log2_ctb_size)) == 0)
637  bs = 0;
638  else if (!s->pps->loop_filter_across_tiles_enabled_flag && (slice_or_tiles_left_boundary & 2) && (x0 % (1 << s->sps->log2_ctb_size)) == 0)
639  bs = 0;
640  if (x0 == 0 || s->sh.disable_deblocking_filter_flag == 1)
641  bs = 0;
642  if (bs)
643  s->vertical_bs[(x0 >> 3) + ((y0 + i) >> 2) * s->bs_width] = bs;
644  }
645  }
646 
647  // bs for TU internal vertical PU boundaries
648  if (log2_trafo_size > log2_min_pu_size && !is_intra)
649  for (j = 0; j < (1 << log2_trafo_size); j += 4) {
650  int y_pu = (y0 + j) >> log2_min_pu_size;
651  int y_tu = (y0 + j) >> log2_min_tu_size;
652 
653  for (i = 8; i < (1 << log2_trafo_size); i += 8) {
654  int xp_pu = (x0 + i - 1) >> log2_min_pu_size;
655  int xq_pu = (x0 + i) >> log2_min_pu_size;
656  int xp_tu = (x0 + i - 1) >> log2_min_tu_size;
657  int xq_tu = (x0 + i) >> log2_min_tu_size;
658  MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
659  MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
660  uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
661  uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
662  RefPicList* left_refPicList = ff_hevc_get_ref_list(s, s->ref, x0 + i - 1, y0 + j);
663 
664  bs = boundary_strength(s, curr, curr_cbf_luma, left, left_cbf_luma, left_refPicList, 0);
665  if (s->sh.disable_deblocking_filter_flag == 1)
666  bs = 0;
667  if (bs)
668  s->vertical_bs[((x0 + i) >> 3) + ((y0 + j) >> 2) * s->bs_width] = bs;
669  }
670  }
671 }
672 #undef LUMA
673 #undef CB
674 #undef CR
675 
676 void ff_hevc_hls_filter(HEVCContext *s, int x, int y)
677 {
678  deblocking_filter_CTB(s, x, y);
679  if (s->sps->sao_enabled)
680  sao_filter_CTB(s, x, y);
681 }
682 
683 void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size)
684 {
685  if (y_ctb && x_ctb)
686  ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size);
687  if (y_ctb && x_ctb >= s->sps->width - ctb_size) {
688  ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size);
689  if (s->threads_type == FF_THREAD_FRAME )
690  ff_thread_report_progress(&s->ref->tf, y_ctb - ctb_size, 0);
691  }
692  if (x_ctb && y_ctb >= s->sps->height - ctb_size)
693  ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb);
694 }
Generated on Sat Jan 25 2014 19:51:48 for FFmpeg by   doxygen 1.8.2