4 * Copyright (C) 2012 - 2013 Guillaume Martres
5 * Copyright (C) 2013 Seppo Tomperi
6 * Copyright (C) 2013 Wassim Hamidouche
8 * This file is part of FFmpeg.
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.
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.
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
25 #include "libavutil/common.h"
26 #include "libavutil/internal.h"
28 #include "cabac_functions.h"
32 #include "bit_depth_template.c"
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
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
50 static int chroma_tc(HEVCContext *s, int qp_y, int c_idx, int tc_offset)
52 static const int qp_c[] = {
53 29, 30, 31, 32, 33, 33, 34, 34, 35, 35, 36, 36, 37, 37
55 int qp, qp_i, offset, idxt;
57 // slice qp offset is not used for deblocking
59 offset = s->pps->cb_qp_offset;
61 offset = s->pps->cr_qp_offset;
63 qp_i = av_clip_c(qp_y + offset, 0, 57);
71 idxt = av_clip_c(qp + DEFAULT_INTRA_TC_OFFSET + tc_offset, 0, 53);
75 static int get_qPy_pred(HEVCContext *s, int xC, int yC,
76 int xBase, int yBase, int log2_cb_size)
78 HEVCLocalContext *lc = s->HEVClc;
79 int ctb_size_mask = (1 << s->sps->log2_ctb_size) - 1;
80 int MinCuQpDeltaSizeMask = (1 << (s->sps->log2_ctb_size -
81 s->pps->diff_cu_qp_delta_depth)) - 1;
82 int xQgBase = xBase - (xBase & MinCuQpDeltaSizeMask);
83 int yQgBase = yBase - (yBase & MinCuQpDeltaSizeMask);
84 int min_cb_width = s->sps->min_cb_width;
85 int min_cb_height = s->sps->min_cb_height;
86 int x_cb = xQgBase >> s->sps->log2_min_cb_size;
87 int y_cb = yQgBase >> s->sps->log2_min_cb_size;
88 int availableA = (xBase & ctb_size_mask) &&
89 (xQgBase & ctb_size_mask);
90 int availableB = (yBase & ctb_size_mask) &&
91 (yQgBase & ctb_size_mask);
92 int qPy_pred, qPy_a, qPy_b;
95 if (lc->first_qp_group) {
96 lc->first_qp_group = !lc->tu.is_cu_qp_delta_coded;
97 qPy_pred = s->sh.slice_qp;
100 if (log2_cb_size < s->sps->log2_ctb_size -
101 s->pps->diff_cu_qp_delta_depth) {
102 static const int offsetX[8][8] = {
103 { -1, 1, 3, 1, 7, 1, 3, 1 },
104 { 0, 0, 0, 0, 0, 0, 0, 0 },
105 { 1, 3, 1, 3, 1, 3, 1, 3 },
106 { 2, 2, 2, 2, 2, 2, 2, 2 },
107 { 3, 5, 7, 5, 3, 5, 7, 5 },
108 { 4, 4, 4, 4, 4, 4, 4, 4 },
109 { 5, 7, 5, 7, 5, 7, 5, 7 },
110 { 6, 6, 6, 6, 6, 6, 6, 6 }
112 static const int offsetY[8][8] = {
113 { 7, 0, 1, 2, 3, 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 { 3, 0, 1, 2, 7, 4, 5, 6 },
118 { 0, 1, 2, 3, 4, 5, 6, 7 },
119 { 1, 0, 3, 2, 5, 4, 7, 6 },
120 { 0, 1, 2, 3, 4, 5, 6, 7 }
122 int xC0b = (xC - (xC & ctb_size_mask)) >> s->sps->log2_min_cb_size;
123 int yC0b = (yC - (yC & ctb_size_mask)) >> s->sps->log2_min_cb_size;
124 int idxX = (xQgBase & ctb_size_mask) >> s->sps->log2_min_cb_size;
125 int idxY = (yQgBase & ctb_size_mask) >> s->sps->log2_min_cb_size;
126 int idx_mask = ctb_size_mask >> s->sps->log2_min_cb_size;
129 x = FFMIN(xC0b + offsetX[idxX][idxY], min_cb_width - 1);
130 y = FFMIN(yC0b + (offsetY[idxX][idxY] & idx_mask), min_cb_height - 1);
132 if (xC0b == (lc->start_of_tiles_x >> s->sps->log2_min_cb_size) &&
133 offsetX[idxX][idxY] == -1) {
134 x = (lc->end_of_tiles_x >> s->sps->log2_min_cb_size) - 1;
137 qPy_pred = s->qp_y_tab[y * min_cb_width + x];
145 qPy_a = s->qp_y_tab[(x_cb - 1) + y_cb * min_cb_width];
151 qPy_b = s->qp_y_tab[x_cb + (y_cb - 1) * min_cb_width];
153 return (qPy_a + qPy_b + 1) >> 1;
156 void ff_hevc_set_qPy(HEVCContext *s, int xC, int yC,
157 int xBase, int yBase, int log2_cb_size)
159 int qp_y = get_qPy_pred(s, xC, yC, xBase, yBase, log2_cb_size);
161 if (s->HEVClc->tu.cu_qp_delta != 0) {
162 int off = s->sps->qp_bd_offset;
163 s->HEVClc->qp_y = ((qp_y + s->HEVClc->tu.cu_qp_delta + 52 + 2 * off) %
166 s->HEVClc->qp_y = qp_y;
169 static int get_qPy(HEVCContext *s, int xC, int yC)
171 int log2_min_cb_size = s->sps->log2_min_cb_size;
172 int x = xC >> log2_min_cb_size;
173 int y = yC >> log2_min_cb_size;
174 return s->qp_y_tab[x + y * s->sps->min_cb_width];
177 static void copy_CTB(uint8_t *dst, uint8_t *src,
178 int width, int height, int stride)
182 for (i = 0; i < height; i++) {
183 memcpy(dst, src, width);
189 #define CTB(tab, x, y) ((tab)[(y) * s->sps->ctb_width + (x)])
191 static void sao_filter_CTB(HEVCContext *s, int x, int y)
193 // TODO: This should be easily parallelizable
194 // TODO: skip CBs when (cu_transquant_bypass_flag || (pcm_loop_filter_disable_flag && pcm_flag))
196 int class = 1, class_index;
197 int edges[4]; // 0 left 1 top 2 right 3 bottom
200 int x_shift = 0, y_shift = 0;
201 int x_ctb = x >> s->sps->log2_ctb_size;
202 int y_ctb = y >> s->sps->log2_ctb_size;
203 int ctb_addr_rs = y_ctb * s->sps->ctb_width + x_ctb;
204 int ctb_addr_ts = s->pps->ctb_addr_rs_to_ts[ctb_addr_rs];
206 // flags indicating unfilterable edges
207 uint8_t vert_edge[] = { 0, 0, 0, 0 };
208 uint8_t horiz_edge[] = { 0, 0, 0, 0 };
209 uint8_t diag_edge[] = { 0, 0, 0, 0 };
210 uint8_t lfase[3]; // current, above, left
211 uint8_t no_tile_filter = s->pps->tiles_enabled_flag &&
212 !s->pps->loop_filter_across_tiles_enabled_flag;
213 uint8_t left_tile_edge = 0;
214 uint8_t up_tile_edge = 0;
216 sao[0] = &CTB(s->sao, x_ctb, y_ctb);
217 edges[0] = x_ctb == 0;
218 edges[1] = y_ctb == 0;
219 edges[2] = x_ctb == s->sps->ctb_width - 1;
220 edges[3] = y_ctb == s->sps->ctb_height - 1;
221 lfase[0] = CTB(s->filter_slice_edges, x_ctb, y_ctb);
225 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]];
226 sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb);
227 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;
228 vert_edge[2] = vert_edge[0];
229 lfase[2] = CTB(s->filter_slice_edges, x_ctb - 1, y_ctb);
236 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]];
237 sao[class] = &CTB(s->sao, x_ctb, y_ctb - 1);
238 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;
239 horiz_edge[1] = horiz_edge[0];
240 lfase[1] = CTB(s->filter_slice_edges, x_ctb, y_ctb - 1);
247 sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb - 1);
250 // Tile check here is done current CTB row/col, not above/left like you'd expect,
251 //but that is because the tile boundary always extends through the whole pic
252 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;
253 vert_edge[3] = vert_edge[1];
254 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;
255 horiz_edge[3] = horiz_edge[2];
256 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;
257 diag_edge[3] = diag_edge[0];
259 // Does left CTB comes after above CTB?
260 if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) >
261 CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
262 diag_edge[2] = !lfase[2] || left_tile_edge || up_tile_edge;
263 diag_edge[1] = diag_edge[2];
264 } else if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) <
265 CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
266 diag_edge[1] = !lfase[1] || left_tile_edge || up_tile_edge;
267 diag_edge[2] = diag_edge[1];
269 // Same slice, only consider tiles
270 diag_edge[2] = left_tile_edge || up_tile_edge;
271 diag_edge[1] = diag_edge[2];
276 for (c_idx = 0; c_idx < 3; c_idx++) {
277 int chroma = c_idx ? 1 : 0;
278 int x0 = x >> chroma;
279 int y0 = y >> chroma;
280 int stride = s->frame->linesize[c_idx];
281 int ctb_size = (1 << (s->sps->log2_ctb_size)) >> s->sps->hshift[c_idx];
282 int width = FFMIN(ctb_size,
283 (s->sps->width >> s->sps->hshift[c_idx]) - x0);
284 int height = FFMIN(ctb_size,
285 (s->sps->height >> s->sps->vshift[c_idx]) - y0);
287 uint8_t *src = &s->frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)];
288 uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride + (x0 << s->sps->pixel_shift)];
289 int offset = (y_shift >> chroma) * stride + ((x_shift >> chroma) << s->sps->pixel_shift);
291 copy_CTB(dst - offset, src - offset,
292 (edges[2] ? width + (x_shift >> chroma) : width) << s->sps->pixel_shift,
293 (edges[3] ? height + (y_shift >> chroma) : height), stride);
295 for (class_index = 0; class_index < class; class_index++) {
297 switch (sao[class_index]->type_idx[c_idx]) {
299 s->hevcdsp.sao_band_filter[classes[class_index]](dst, src,
306 s->hevcdsp.sao_edge_filter[classes[class_index]](dst, src,
311 vert_edge[classes[class_index]],
312 horiz_edge[classes[class_index]],
313 diag_edge[classes[class_index]]);
320 static int get_pcm(HEVCContext *s, int x, int y)
322 int log2_min_pu_size = s->sps->log2_min_pu_size;
323 int x_pu = x >> log2_min_pu_size;
324 int y_pu = y >> log2_min_pu_size;
326 if (x < 0 || x_pu >= s->sps->min_pu_width ||
327 y < 0 || y_pu >= s->sps->min_pu_height)
329 return s->is_pcm[y_pu * s->sps->min_pu_width + x_pu];
332 #define TC_CALC(qp, bs) \
333 tctable[av_clip((qp) + DEFAULT_INTRA_TC_OFFSET * ((bs) - 1) + \
334 (tc_offset >> 1 << 1), \
335 0, MAX_QP + DEFAULT_INTRA_TC_OFFSET)]
337 static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
342 int c_tc[2], beta[2], tc[2];
343 uint8_t no_p[2] = { 0 };
344 uint8_t no_q[2] = { 0 };
346 int log2_ctb_size = s->sps->log2_ctb_size;
348 int ctb_size = 1 << log2_ctb_size;
349 int ctb = (x0 >> log2_ctb_size) +
350 (y0 >> log2_ctb_size) * s->sps->ctb_width;
351 int cur_tc_offset = s->deblock[ctb].tc_offset;
352 int cur_beta_offset = s->deblock[ctb].beta_offset;
353 int left_tc_offset, left_beta_offset;
354 int tc_offset, beta_offset;
355 int pcmf = (s->sps->pcm_enabled_flag &&
356 s->sps->pcm.loop_filter_disable_flag) ||
357 s->pps->transquant_bypass_enable_flag;
360 left_tc_offset = s->deblock[ctb - 1].tc_offset;
361 left_beta_offset = s->deblock[ctb - 1].beta_offset;
364 x_end = x0 + ctb_size;
365 if (x_end > s->sps->width)
366 x_end = s->sps->width;
367 y_end = y0 + ctb_size;
368 if (y_end > s->sps->height)
369 y_end = s->sps->height;
371 tc_offset = cur_tc_offset;
372 beta_offset = cur_beta_offset;
374 // vertical filtering luma
375 for (y = y0; y < y_end; y += 8) {
376 for (x = x0 ? x0 : 8; x < x_end; x += 8) {
377 const int bs0 = s->vertical_bs[(x >> 3) + (y >> 2) * s->bs_width];
378 const int bs1 = s->vertical_bs[(x >> 3) + ((y + 4) >> 2) * s->bs_width];
380 const int qp0 = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
381 const int qp1 = (get_qPy(s, x - 1, y + 4) + get_qPy(s, x, y + 4) + 1) >> 1;
383 beta[0] = betatable[av_clip(qp0 + (beta_offset >> 1 << 1), 0, MAX_QP)];
384 beta[1] = betatable[av_clip(qp1 + (beta_offset >> 1 << 1), 0, MAX_QP)];
385 tc[0] = bs0 ? TC_CALC(qp0, bs0) : 0;
386 tc[1] = bs1 ? TC_CALC(qp1, bs1) : 0;
387 src = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->sps->pixel_shift)];
389 no_p[0] = get_pcm(s, x - 1, y);
390 no_p[1] = get_pcm(s, x - 1, y + 4);
391 no_q[0] = get_pcm(s, x, y);
392 no_q[1] = get_pcm(s, x, y + 4);
393 s->hevcdsp.hevc_v_loop_filter_luma_c(src,
394 s->frame->linesize[LUMA],
395 beta, tc, no_p, no_q);
397 s->hevcdsp.hevc_v_loop_filter_luma(src,
398 s->frame->linesize[LUMA],
399 beta, tc, no_p, no_q);
404 // vertical filtering chroma
405 for (chroma = 1; chroma <= 2; chroma++) {
406 for (y = y0; y < y_end; y += 16) {
407 for (x = x0 ? x0 : 16; x < x_end; x += 16) {
408 const int bs0 = s->vertical_bs[(x >> 3) + (y >> 2) * s->bs_width];
409 const int bs1 = s->vertical_bs[(x >> 3) + ((y + 8) >> 2) * s->bs_width];
410 if ((bs0 == 2) || (bs1 == 2)) {
411 const int qp0 = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
412 const int qp1 = (get_qPy(s, x - 1, y + 8) + get_qPy(s, x, y + 8) + 1) >> 1;
414 c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
415 c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
416 src = &s->frame->data[chroma][y / 2 * s->frame->linesize[chroma] + ((x / 2) << s->sps->pixel_shift)];
418 no_p[0] = get_pcm(s, x - 1, y);
419 no_p[1] = get_pcm(s, x - 1, y + 8);
420 no_q[0] = get_pcm(s, x, y);
421 no_q[1] = get_pcm(s, x, y + 8);
422 s->hevcdsp.hevc_v_loop_filter_chroma_c(src,
423 s->frame->linesize[chroma],
426 s->hevcdsp.hevc_v_loop_filter_chroma(src,
427 s->frame->linesize[chroma],
434 // horizontal filtering luma
435 if (x_end != s->sps->width)
437 for (y = y0 ? y0 : 8; y < y_end; y += 8) {
438 for (x = x0 ? x0 - 8 : 0; x < x_end; x += 8) {
439 const int bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
440 const int bs1 = s->horizontal_bs[(x + 4 + y * s->bs_width) >> 2];
442 const int qp0 = (get_qPy(s, x, y - 1) + get_qPy(s, x, y) + 1) >> 1;
443 const int qp1 = (get_qPy(s, x + 4, y - 1) + get_qPy(s, x + 4, y) + 1) >> 1;
445 tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
446 beta_offset = x >= x0 ? cur_beta_offset : left_beta_offset;
448 beta[0] = betatable[av_clip(qp0 + (beta_offset >> 1 << 1), 0, MAX_QP)];
449 beta[1] = betatable[av_clip(qp1 + (beta_offset >> 1 << 1), 0, MAX_QP)];
450 tc[0] = bs0 ? TC_CALC(qp0, bs0) : 0;
451 tc[1] = bs1 ? TC_CALC(qp1, bs1) : 0;
452 src = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->sps->pixel_shift)];
454 no_p[0] = get_pcm(s, x, y - 1);
455 no_p[1] = get_pcm(s, x + 4, y - 1);
456 no_q[0] = get_pcm(s, x, y);
457 no_q[1] = get_pcm(s, x + 4, y);
458 s->hevcdsp.hevc_h_loop_filter_luma_c(src,
459 s->frame->linesize[LUMA],
460 beta, tc, no_p, no_q);
462 s->hevcdsp.hevc_h_loop_filter_luma(src,
463 s->frame->linesize[LUMA],
464 beta, tc, no_p, no_q);
469 // horizontal filtering chroma
470 for (chroma = 1; chroma <= 2; chroma++) {
471 for (y = y0 ? y0 : 16; y < y_end; y += 16) {
472 for (x = x0 - 8; x < x_end; x += 16) {
474 // to make sure no memory access over boundary when x = -8
475 // TODO: simplify with row based deblocking
478 bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
479 } else if (x >= x_end - 8) {
480 bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
483 bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
484 bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
487 if ((bs0 == 2) || (bs1 == 2)) {
488 const int qp0 = bs0 == 2 ? (get_qPy(s, x, y - 1) + get_qPy(s, x, y) + 1) >> 1 : 0;
489 const int qp1 = bs1 == 2 ? (get_qPy(s, x + 8, y - 1) + get_qPy(s, x + 8, y) + 1) >> 1 : 0;
491 tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
492 c_tc[0] = bs0 == 2 ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
493 c_tc[1] = bs1 == 2 ? chroma_tc(s, qp1, chroma, cur_tc_offset) : 0;
494 src = &s->frame->data[chroma][y / 2 * s->frame->linesize[chroma] + ((x / 2) << s->sps->pixel_shift)];
496 no_p[0] = get_pcm(s, x, y - 1);
497 no_p[1] = get_pcm(s, x + 8, y - 1);
498 no_q[0] = get_pcm(s, x, y);
499 no_q[1] = get_pcm(s, x + 8, y);
500 s->hevcdsp.hevc_h_loop_filter_chroma_c(src,
501 s->frame->linesize[chroma],
504 s->hevcdsp.hevc_h_loop_filter_chroma(src,
505 s->frame->linesize[chroma],
513 static int boundary_strength(HEVCContext *s, MvField *curr,
514 uint8_t curr_cbf_luma, MvField *neigh,
515 uint8_t neigh_cbf_luma,
516 RefPicList *neigh_refPicList,
519 int mvs = curr->pred_flag[0] + curr->pred_flag[1];
522 if (curr->is_intra || neigh->is_intra)
524 if (curr_cbf_luma || neigh_cbf_luma)
528 if (mvs == neigh->pred_flag[0] + neigh->pred_flag[1]) {
531 if (s->ref->refPicList[0].list[curr->ref_idx[0]] == neigh_refPicList[0].list[neigh->ref_idx[0]] &&
532 s->ref->refPicList[0].list[curr->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]] &&
533 neigh_refPicList[0].list[neigh->ref_idx[0]] == neigh_refPicList[1].list[neigh->ref_idx[1]]) {
534 if ((abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
535 abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4) &&
536 (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
537 abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4))
541 } else if (neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
542 neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
543 if (abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
544 abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4)
548 } else if (neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
549 neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
550 if (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
551 abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4)
562 if (curr->pred_flag[0]) {
564 ref_A = s->ref->refPicList[0].list[curr->ref_idx[0]];
567 ref_A = s->ref->refPicList[1].list[curr->ref_idx[1]];
570 if (neigh->pred_flag[0]) {
572 ref_B = neigh_refPicList[0].list[neigh->ref_idx[0]];
575 ref_B = neigh_refPicList[1].list[neigh->ref_idx[1]];
578 if (ref_A == ref_B) {
579 if (abs(A.x - B.x) >= 4 || abs(A.y - B.y) >= 4)
591 void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
593 int slice_or_tiles_up_boundary,
594 int slice_or_tiles_left_boundary)
596 MvField *tab_mvf = s->ref->tab_mvf;
597 int log2_min_pu_size = s->sps->log2_min_pu_size;
598 int log2_min_tu_size = s->sps->log2_min_tb_size;
599 int min_pu_width = s->sps->min_pu_width;
600 int min_tu_width = s->sps->min_tb_width;
601 int is_intra = tab_mvf[(y0 >> log2_min_pu_size) * min_pu_width +
602 (x0 >> log2_min_pu_size)].is_intra;
605 if (y0 > 0 && (y0 & 7) == 0) {
606 int yp_pu = (y0 - 1) >> log2_min_pu_size;
607 int yq_pu = y0 >> log2_min_pu_size;
608 int yp_tu = (y0 - 1) >> log2_min_tu_size;
609 int yq_tu = y0 >> log2_min_tu_size;
611 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
612 int x_pu = (x0 + i) >> log2_min_pu_size;
613 int x_tu = (x0 + i) >> log2_min_tu_size;
614 MvField *top = &tab_mvf[yp_pu * min_pu_width + x_pu];
615 MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
616 uint8_t top_cbf_luma = s->cbf_luma[yp_tu * min_tu_width + x_tu];
617 uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
618 RefPicList *top_refPicList = ff_hevc_get_ref_list(s, s->ref,
621 bs = boundary_strength(s, curr, curr_cbf_luma,
622 top, top_cbf_luma, top_refPicList, 1);
623 if (!s->sh.slice_loop_filter_across_slices_enabled_flag &&
624 (slice_or_tiles_up_boundary & 1) &&
625 (y0 % (1 << s->sps->log2_ctb_size)) == 0)
627 else if (!s->pps->loop_filter_across_tiles_enabled_flag &&
628 (slice_or_tiles_up_boundary & 2) &&
629 (y0 % (1 << s->sps->log2_ctb_size)) == 0)
631 if (y0 == 0 || s->sh.disable_deblocking_filter_flag == 1)
634 s->horizontal_bs[((x0 + i) + y0 * s->bs_width) >> 2] = bs;
638 // bs for TU internal horizontal PU boundaries
639 if (log2_trafo_size > s->sps->log2_min_pu_size && !is_intra)
640 for (j = 8; j < (1 << log2_trafo_size); j += 8) {
641 int yp_pu = (y0 + j - 1) >> log2_min_pu_size;
642 int yq_pu = (y0 + j) >> log2_min_pu_size;
643 int yp_tu = (y0 + j - 1) >> log2_min_tu_size;
644 int yq_tu = (y0 + j) >> log2_min_tu_size;
646 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
647 int x_pu = (x0 + i) >> log2_min_pu_size;
648 int x_tu = (x0 + i) >> log2_min_tu_size;
649 MvField *top = &tab_mvf[yp_pu * min_pu_width + x_pu];
650 MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
651 uint8_t top_cbf_luma = s->cbf_luma[yp_tu * min_tu_width + x_tu];
652 uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
653 RefPicList *top_refPicList = ff_hevc_get_ref_list(s, s->ref,
657 bs = boundary_strength(s, curr, curr_cbf_luma,
658 top, top_cbf_luma, top_refPicList, 0);
659 if (s->sh.disable_deblocking_filter_flag == 1)
662 s->horizontal_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
666 // bs for vertical TU boundaries
667 if (x0 > 0 && (x0 & 7) == 0) {
668 int xp_pu = (x0 - 1) >> log2_min_pu_size;
669 int xq_pu = x0 >> log2_min_pu_size;
670 int xp_tu = (x0 - 1) >> log2_min_tu_size;
671 int xq_tu = x0 >> log2_min_tu_size;
673 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
674 int y_pu = (y0 + i) >> log2_min_pu_size;
675 int y_tu = (y0 + i) >> log2_min_tu_size;
676 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
677 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
679 uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
680 uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
681 RefPicList *left_refPicList = ff_hevc_get_ref_list(s, s->ref,
684 bs = boundary_strength(s, curr, curr_cbf_luma,
685 left, left_cbf_luma, left_refPicList, 1);
686 if (!s->sh.slice_loop_filter_across_slices_enabled_flag &&
687 (slice_or_tiles_left_boundary & 1) &&
688 (x0 % (1 << s->sps->log2_ctb_size)) == 0)
690 else if (!s->pps->loop_filter_across_tiles_enabled_flag &&
691 (slice_or_tiles_left_boundary & 2) &&
692 (x0 % (1 << s->sps->log2_ctb_size)) == 0)
694 if (x0 == 0 || s->sh.disable_deblocking_filter_flag == 1)
697 s->vertical_bs[(x0 >> 3) + ((y0 + i) >> 2) * s->bs_width] = bs;
701 // bs for TU internal vertical PU boundaries
702 if (log2_trafo_size > log2_min_pu_size && !is_intra)
703 for (j = 0; j < (1 << log2_trafo_size); j += 4) {
704 int y_pu = (y0 + j) >> log2_min_pu_size;
705 int y_tu = (y0 + j) >> log2_min_tu_size;
707 for (i = 8; i < (1 << log2_trafo_size); i += 8) {
708 int xp_pu = (x0 + i - 1) >> log2_min_pu_size;
709 int xq_pu = (x0 + i) >> log2_min_pu_size;
710 int xp_tu = (x0 + i - 1) >> log2_min_tu_size;
711 int xq_tu = (x0 + i) >> log2_min_tu_size;
712 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
713 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
714 uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
715 uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
716 RefPicList *left_refPicList = ff_hevc_get_ref_list(s, s->ref,
720 bs = boundary_strength(s, curr, curr_cbf_luma,
721 left, left_cbf_luma, left_refPicList, 0);
722 if (s->sh.disable_deblocking_filter_flag == 1)
725 s->vertical_bs[((x0 + i) >> 3) + ((y0 + j) >> 2) * s->bs_width] = bs;
734 void ff_hevc_hls_filter(HEVCContext *s, int x, int y)
736 deblocking_filter_CTB(s, x, y);
737 if (s->sps->sao_enabled)
738 sao_filter_CTB(s, x, y);
741 void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size)
744 ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size);
745 if (y_ctb && x_ctb >= s->sps->width - ctb_size) {
746 ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size);
747 if (s->threads_type == FF_THREAD_FRAME )
748 ff_thread_report_progress(&s->ref->tf, y_ctb - ctb_size, 0);
750 if (x_ctb && y_ctb >= s->sps->height - ctb_size)
751 ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb);