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 Libav.
10 * Libav 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 * Libav 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 Libav; 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"
36 static const uint8_t tctable[54] = {
37 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, // QP 0...18
38 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, // QP 19...37
39 5, 5, 6, 6, 7, 8, 9, 10, 11, 13, 14, 16, 18, 20, 22, 24 // QP 38...53
42 static const uint8_t betatable[52] = {
43 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6, 7, 8, // QP 0...18
44 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 20, 22, 24, 26, 28, 30, 32, 34, 36, // QP 19...37
45 38, 40, 42, 44, 46, 48, 50, 52, 54, 56, 58, 60, 62, 64 // QP 38...51
48 static int chroma_tc(HEVCContext *s, int qp_y, int c_idx, int tc_offset)
50 static const int qp_c[] = {
51 29, 30, 31, 32, 33, 33, 34, 34, 35, 35, 36, 36, 37, 37
53 int qp, qp_i, offset, idxt;
55 // slice qp offset is not used for deblocking
57 offset = s->ps.pps->cb_qp_offset;
59 offset = s->ps.pps->cr_qp_offset;
61 qp_i = av_clip(qp_y + offset, 0, 57);
69 idxt = av_clip(qp + DEFAULT_INTRA_TC_OFFSET + tc_offset, 0, 53);
73 static int get_qPy_pred(HEVCContext *s, int xC, int yC,
74 int xBase, int yBase, int log2_cb_size)
76 HEVCLocalContext *lc = &s->HEVClc;
77 int ctb_size_mask = (1 << s->ps.sps->log2_ctb_size) - 1;
78 int MinCuQpDeltaSizeMask = (1 << (s->ps.sps->log2_ctb_size -
79 s->ps.pps->diff_cu_qp_delta_depth)) - 1;
80 int xQgBase = xBase - (xBase & MinCuQpDeltaSizeMask);
81 int yQgBase = yBase - (yBase & MinCuQpDeltaSizeMask);
82 int min_cb_width = s->ps.sps->min_cb_width;
83 int min_cb_height = s->ps.sps->min_cb_height;
84 int x_cb = xQgBase >> s->ps.sps->log2_min_cb_size;
85 int y_cb = yQgBase >> s->ps.sps->log2_min_cb_size;
86 int availableA = (xBase & ctb_size_mask) &&
87 (xQgBase & ctb_size_mask);
88 int availableB = (yBase & ctb_size_mask) &&
89 (yQgBase & ctb_size_mask);
90 int qPy_pred, qPy_a, qPy_b;
93 if (lc->first_qp_group || (!xQgBase && !yQgBase)) {
94 lc->first_qp_group = !lc->tu.is_cu_qp_delta_coded;
95 qPy_pred = s->sh.slice_qp;
98 if (log2_cb_size < s->ps.sps->log2_ctb_size -
99 s->ps.pps->diff_cu_qp_delta_depth) {
100 static const int offsetX[8][8] = {
101 { -1, 1, 3, 1, 7, 1, 3, 1 },
102 { 0, 0, 0, 0, 0, 0, 0, 0 },
103 { 1, 3, 1, 3, 1, 3, 1, 3 },
104 { 2, 2, 2, 2, 2, 2, 2, 2 },
105 { 3, 5, 7, 5, 3, 5, 7, 5 },
106 { 4, 4, 4, 4, 4, 4, 4, 4 },
107 { 5, 7, 5, 7, 5, 7, 5, 7 },
108 { 6, 6, 6, 6, 6, 6, 6, 6 }
110 static const int offsetY[8][8] = {
111 { 7, 0, 1, 2, 3, 4, 5, 6 },
112 { 0, 1, 2, 3, 4, 5, 6, 7 },
113 { 1, 0, 3, 2, 5, 4, 7, 6 },
114 { 0, 1, 2, 3, 4, 5, 6, 7 },
115 { 3, 0, 1, 2, 7, 4, 5, 6 },
116 { 0, 1, 2, 3, 4, 5, 6, 7 },
117 { 1, 0, 3, 2, 5, 4, 7, 6 },
118 { 0, 1, 2, 3, 4, 5, 6, 7 }
120 int xC0b = (xC - (xC & ctb_size_mask)) >> s->ps.sps->log2_min_cb_size;
121 int yC0b = (yC - (yC & ctb_size_mask)) >> s->ps.sps->log2_min_cb_size;
122 int idxX = (xQgBase & ctb_size_mask) >> s->ps.sps->log2_min_cb_size;
123 int idxY = (yQgBase & ctb_size_mask) >> s->ps.sps->log2_min_cb_size;
124 int idx_mask = ctb_size_mask >> s->ps.sps->log2_min_cb_size;
127 x = FFMIN(xC0b + offsetX[idxX][idxY], min_cb_width - 1);
128 y = FFMIN(yC0b + (offsetY[idxX][idxY] & idx_mask), min_cb_height - 1);
130 if (xC0b == (lc->start_of_tiles_x >> s->ps.sps->log2_min_cb_size) &&
131 offsetX[idxX][idxY] == -1) {
132 x = (lc->end_of_tiles_x >> s->ps.sps->log2_min_cb_size) - 1;
135 qPy_pred = s->qp_y_tab[y * min_cb_width + x];
143 qPy_a = s->qp_y_tab[(x_cb - 1) + y_cb * min_cb_width];
149 qPy_b = s->qp_y_tab[x_cb + (y_cb - 1) * min_cb_width];
151 return (qPy_a + qPy_b + 1) >> 1;
154 void ff_hevc_set_qPy(HEVCContext *s, int xC, int yC,
155 int xBase, int yBase, int log2_cb_size)
157 int qp_y = get_qPy_pred(s, xC, yC, xBase, yBase, log2_cb_size);
159 if (s->HEVClc.tu.cu_qp_delta != 0) {
160 int off = s->ps.sps->qp_bd_offset;
161 s->HEVClc.qp_y = FFUMOD(qp_y + s->HEVClc.tu.cu_qp_delta + 52 + 2 * off,
164 s->HEVClc.qp_y = qp_y;
167 static int get_qPy(HEVCContext *s, int xC, int yC)
169 int log2_min_cb_size = s->ps.sps->log2_min_cb_size;
170 int x = xC >> log2_min_cb_size;
171 int y = yC >> log2_min_cb_size;
172 return s->qp_y_tab[x + y * s->ps.sps->min_cb_width];
175 static void copy_CTB(uint8_t *dst, uint8_t *src,
176 int width, int height, int stride)
180 for (i = 0; i < height; i++) {
181 memcpy(dst, src, width);
187 #define CTB(tab, x, y) ((tab)[(y) * s->ps.sps->ctb_width + (x)])
189 static void sao_filter_CTB(HEVCContext *s, int x, int y)
191 // TODO: This should be easily parallelizable
192 // TODO: skip CBs when (cu_transquant_bypass_flag || (pcm_loop_filter_disable_flag && pcm_flag))
194 int class = 1, class_index;
195 int edges[4]; // 0 left 1 top 2 right 3 bottom
198 int x_shift = 0, y_shift = 0;
199 int x_ctb = x >> s->ps.sps->log2_ctb_size;
200 int y_ctb = y >> s->ps.sps->log2_ctb_size;
201 int ctb_addr_rs = y_ctb * s->ps.sps->ctb_width + x_ctb;
202 int ctb_addr_ts = s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs];
204 // flags indicating unfilterable edges
205 uint8_t vert_edge[] = { 0, 0, 0, 0 };
206 uint8_t horiz_edge[] = { 0, 0, 0, 0 };
207 uint8_t diag_edge[] = { 0, 0, 0, 0 };
208 uint8_t lfase[3]; // current, above, left
209 uint8_t no_tile_filter = s->ps.pps->tiles_enabled_flag &&
210 !s->ps.pps->loop_filter_across_tiles_enabled_flag;
211 uint8_t left_tile_edge = 0, up_tile_edge = 0;
213 sao[0] = &CTB(s->sao, x_ctb, y_ctb);
214 edges[0] = x_ctb == 0;
215 edges[1] = y_ctb == 0;
216 edges[2] = x_ctb == s->ps.sps->ctb_width - 1;
217 edges[3] = y_ctb == s->ps.sps->ctb_height - 1;
218 lfase[0] = CTB(s->filter_slice_edges, x_ctb, y_ctb);
222 left_tile_edge = no_tile_filter && s->ps.pps->tile_id[ctb_addr_ts] != s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs-1]];
223 sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb);
224 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;
225 vert_edge[2] = vert_edge[0];
226 lfase[2] = CTB(s->filter_slice_edges, x_ctb - 1, y_ctb);
233 up_tile_edge = no_tile_filter && s->ps.pps->tile_id[ctb_addr_ts] != s->ps.pps->tile_id[s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs - s->ps.sps->ctb_width]];
234 sao[class] = &CTB(s->sao, x_ctb, y_ctb - 1);
235 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;
236 horiz_edge[1] = horiz_edge[0];
237 lfase[1] = CTB(s->filter_slice_edges, x_ctb, y_ctb - 1);
244 sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb - 1);
247 // Tile check here is done current CTB row/col, not above/left like you'd expect,
248 //but that is because the tile boundary always extends through the whole pic
249 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;
250 vert_edge[3] = vert_edge[1];
251 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;
252 horiz_edge[3] = horiz_edge[2];
253 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;
254 diag_edge[3] = diag_edge[0];
256 // Does left CTB comes after above CTB?
257 if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) >
258 CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
259 diag_edge[2] = !lfase[2] || left_tile_edge || up_tile_edge;
260 diag_edge[1] = diag_edge[2];
261 } else if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) <
262 CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
263 diag_edge[1] = !lfase[1] || left_tile_edge || up_tile_edge;
264 diag_edge[2] = diag_edge[1];
266 // Same slice, only consider tiles
267 diag_edge[2] = left_tile_edge || up_tile_edge;
268 diag_edge[1] = diag_edge[2];
273 for (c_idx = 0; c_idx < 3; c_idx++) {
274 int chroma = c_idx ? 1 : 0;
275 int x0 = x >> chroma;
276 int y0 = y >> chroma;
277 int stride = s->frame->linesize[c_idx];
278 int ctb_size = (1 << (s->ps.sps->log2_ctb_size)) >> s->ps.sps->hshift[c_idx];
279 int width = FFMIN(ctb_size,
280 (s->ps.sps->width >> s->ps.sps->hshift[c_idx]) - x0);
281 int height = FFMIN(ctb_size,
282 (s->ps.sps->height >> s->ps.sps->vshift[c_idx]) - y0);
284 uint8_t *src = &s->frame->data[c_idx][y0 * stride + (x0 << s->ps.sps->pixel_shift)];
285 uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride + (x0 << s->ps.sps->pixel_shift)];
286 int offset = (y_shift >> chroma) * stride + ((x_shift >> chroma) << s->ps.sps->pixel_shift);
288 copy_CTB(dst - offset, src - offset,
289 (edges[2] ? width + (x_shift >> chroma) : width) << s->ps.sps->pixel_shift,
290 (edges[3] ? height + (y_shift >> chroma) : height), stride);
292 for (class_index = 0; class_index < class; class_index++) {
294 switch (sao[class_index]->type_idx[c_idx]) {
296 s->hevcdsp.sao_band_filter[classes[class_index]](dst, src,
303 s->hevcdsp.sao_edge_filter[classes[class_index]](dst, src,
308 vert_edge[classes[class_index]],
309 horiz_edge[classes[class_index]],
310 diag_edge[classes[class_index]]);
317 static int get_pcm(HEVCContext *s, int x, int y)
319 int log2_min_pu_size = s->ps.sps->log2_min_pu_size;
325 x_pu = x >> log2_min_pu_size;
326 y_pu = y >> log2_min_pu_size;
328 if (x_pu >= s->ps.sps->min_pu_width || y_pu >= s->ps.sps->min_pu_height)
330 return s->is_pcm[y_pu * s->ps.sps->min_pu_width + x_pu];
333 #define TC_CALC(qp, bs) \
334 tctable[av_clip((qp) + DEFAULT_INTRA_TC_OFFSET * ((bs) - 1) + \
335 (tc_offset >> 1 << 1), \
336 0, MAX_QP + DEFAULT_INTRA_TC_OFFSET)]
338 static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
341 int x, y, x_end, y_end, chroma;
342 int c_tc[2], tc[2], beta;
343 uint8_t no_p[2] = { 0 };
344 uint8_t no_q[2] = { 0 };
346 int log2_ctb_size = s->ps.sps->log2_ctb_size;
347 int ctb_size = 1 << log2_ctb_size;
348 int ctb = (x0 >> log2_ctb_size) +
349 (y0 >> log2_ctb_size) * s->ps.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 tc_offset, left_tc_offset, beta_offset, left_beta_offset;
353 int pcmf = (s->ps.sps->pcm_enabled_flag &&
354 s->ps.sps->pcm.loop_filter_disable_flag) ||
355 s->ps.pps->transquant_bypass_enable_flag;
358 left_tc_offset = s->deblock[ctb - 1].tc_offset;
359 left_beta_offset = s->deblock[ctb - 1].beta_offset;
362 x_end = x0 + ctb_size;
363 if (x_end > s->ps.sps->width)
364 x_end = s->ps.sps->width;
365 y_end = y0 + ctb_size;
366 if (y_end > s->ps.sps->height)
367 y_end = s->ps.sps->height;
369 tc_offset = cur_tc_offset;
370 beta_offset = cur_beta_offset;
372 // vertical filtering luma
373 for (y = y0; y < y_end; y += 8) {
374 for (x = x0 ? x0 : 8; x < x_end; x += 8) {
375 const int bs0 = s->vertical_bs[(x >> 3) + (y >> 2) * s->bs_width];
376 const int bs1 = s->vertical_bs[(x >> 3) + ((y + 4) >> 2) * s->bs_width];
378 const int qp = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
380 beta = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
382 tc[0] = bs0 ? TC_CALC(qp, bs0) : 0;
383 tc[1] = bs1 ? TC_CALC(qp, bs1) : 0;
384 src = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->ps.sps->pixel_shift)];
386 no_p[0] = get_pcm(s, x - 1, y);
387 no_p[1] = get_pcm(s, x - 1, y + 4);
388 no_q[0] = get_pcm(s, x, y);
389 no_q[1] = get_pcm(s, x, y + 4);
390 s->hevcdsp.hevc_v_loop_filter_luma_c(src,
391 s->frame->linesize[LUMA],
392 beta, tc, no_p, no_q);
394 s->hevcdsp.hevc_v_loop_filter_luma(src,
395 s->frame->linesize[LUMA],
396 beta, tc, no_p, no_q);
401 // vertical filtering chroma
402 for (chroma = 1; chroma <= 2; chroma++) {
403 for (y = y0; y < y_end; y += 16) {
404 for (x = x0 ? x0 : 16; x < x_end; x += 16) {
405 const int bs0 = s->vertical_bs[(x >> 3) + (y >> 2) * s->bs_width];
406 const int bs1 = s->vertical_bs[(x >> 3) + ((y + 8) >> 2) * s->bs_width];
407 if ((bs0 == 2) || (bs1 == 2)) {
408 const int qp0 = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
409 const int qp1 = (get_qPy(s, x - 1, y + 8) + get_qPy(s, x, y + 8) + 1) >> 1;
411 c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
412 c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
413 src = &s->frame->data[chroma][y / 2 * s->frame->linesize[chroma] + ((x / 2) << s->ps.sps->pixel_shift)];
415 no_p[0] = get_pcm(s, x - 1, y);
416 no_p[1] = get_pcm(s, x - 1, y + 8);
417 no_q[0] = get_pcm(s, x, y);
418 no_q[1] = get_pcm(s, x, y + 8);
419 s->hevcdsp.hevc_v_loop_filter_chroma_c(src,
420 s->frame->linesize[chroma],
423 s->hevcdsp.hevc_v_loop_filter_chroma(src,
424 s->frame->linesize[chroma],
431 // horizontal filtering luma
432 if (x_end != s->ps.sps->width)
434 for (y = y0 ? y0 : 8; y < y_end; y += 8) {
435 for (x = x0 ? x0 - 8 : 0; x < x_end; x += 8) {
436 const int bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
437 const int bs1 = s->horizontal_bs[(x + 4 + y * s->bs_width) >> 2];
439 const int qp = (get_qPy(s, x, y - 1) + get_qPy(s, x, y) + 1) >> 1;
441 tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
442 beta_offset = x >= x0 ? cur_beta_offset : left_beta_offset;
444 beta = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
445 tc[0] = bs0 ? TC_CALC(qp, bs0) : 0;
446 tc[1] = bs1 ? TC_CALC(qp, bs1) : 0;
447 src = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->ps.sps->pixel_shift)];
449 no_p[0] = get_pcm(s, x, y - 1);
450 no_p[1] = get_pcm(s, x + 4, y - 1);
451 no_q[0] = get_pcm(s, x, y);
452 no_q[1] = get_pcm(s, x + 4, y);
453 s->hevcdsp.hevc_h_loop_filter_luma_c(src,
454 s->frame->linesize[LUMA],
455 beta, tc, no_p, no_q);
457 s->hevcdsp.hevc_h_loop_filter_luma(src,
458 s->frame->linesize[LUMA],
459 beta, tc, no_p, no_q);
464 // horizontal filtering chroma
465 for (chroma = 1; chroma <= 2; chroma++) {
466 for (y = y0 ? y0 : 16; y < y_end; y += 16) {
467 for (x = x0 - 8; x < x_end; x += 16) {
469 // to make sure no memory access over boundary when x = -8
470 // TODO: simplify with row based deblocking
473 bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
474 } else if (x >= x_end - 8) {
475 bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
478 bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
479 bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
482 if ((bs0 == 2) || (bs1 == 2)) {
483 const int qp0 = bs0 == 2 ? (get_qPy(s, x, y - 1) + get_qPy(s, x, y) + 1) >> 1 : 0;
484 const int qp1 = bs1 == 2 ? (get_qPy(s, x + 8, y - 1) + get_qPy(s, x + 8, y) + 1) >> 1 : 0;
486 tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
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 / 2 * s->frame->linesize[chroma] + ((x / 2) << s->ps.sps->pixel_shift)];
491 no_p[0] = get_pcm(s, x, y - 1);
492 no_p[1] = get_pcm(s, x + 8, y - 1);
493 no_q[0] = get_pcm(s, x, y);
494 no_q[1] = get_pcm(s, x + 8, y);
495 s->hevcdsp.hevc_h_loop_filter_chroma_c(src,
496 s->frame->linesize[chroma],
499 s->hevcdsp.hevc_h_loop_filter_chroma(src,
500 s->frame->linesize[chroma],
508 static int boundary_strength(HEVCContext *s, MvField *curr,
509 uint8_t curr_cbf_luma, MvField *neigh,
510 uint8_t neigh_cbf_luma,
511 RefPicList *neigh_refPicList,
514 int mvs = curr->pred_flag[0] + curr->pred_flag[1];
517 if (curr->is_intra || neigh->is_intra)
519 if (curr_cbf_luma || neigh_cbf_luma)
523 if (mvs == neigh->pred_flag[0] + neigh->pred_flag[1]) {
526 if (s->ref->refPicList[0].list[curr->ref_idx[0]] == neigh_refPicList[0].list[neigh->ref_idx[0]] &&
527 s->ref->refPicList[0].list[curr->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]] &&
528 neigh_refPicList[0].list[neigh->ref_idx[0]] == neigh_refPicList[1].list[neigh->ref_idx[1]]) {
529 if ((abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
530 abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4) &&
531 (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
532 abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4))
536 } else if (neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
537 neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
538 if (abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
539 abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4)
543 } else if (neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
544 neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
545 if (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
546 abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4)
557 if (curr->pred_flag[0]) {
559 ref_A = s->ref->refPicList[0].list[curr->ref_idx[0]];
562 ref_A = s->ref->refPicList[1].list[curr->ref_idx[1]];
565 if (neigh->pred_flag[0]) {
567 ref_B = neigh_refPicList[0].list[neigh->ref_idx[0]];
570 ref_B = neigh_refPicList[1].list[neigh->ref_idx[1]];
573 if (ref_A == ref_B) {
574 if (abs(A.x - B.x) >= 4 || abs(A.y - B.y) >= 4)
586 void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
589 HEVCLocalContext *lc = &s->HEVClc;
590 MvField *tab_mvf = s->ref->tab_mvf;
591 int log2_min_pu_size = s->ps.sps->log2_min_pu_size;
592 int log2_min_tu_size = s->ps.sps->log2_min_tb_size;
593 int min_pu_width = s->ps.sps->min_pu_width;
594 int min_tu_width = s->ps.sps->min_tb_width;
595 int is_intra = tab_mvf[(y0 >> log2_min_pu_size) * min_pu_width +
596 (x0 >> log2_min_pu_size)].is_intra;
597 int boundary_upper, boundary_left;
600 boundary_upper = y0 > 0 && !(y0 & 7);
601 if (boundary_upper &&
602 ((!s->sh.slice_loop_filter_across_slices_enabled_flag &&
603 lc->boundary_flags & BOUNDARY_UPPER_SLICE &&
604 (y0 % (1 << s->ps.sps->log2_ctb_size)) == 0) ||
605 (!s->ps.pps->loop_filter_across_tiles_enabled_flag &&
606 lc->boundary_flags & BOUNDARY_UPPER_TILE &&
607 (y0 % (1 << s->ps.sps->log2_ctb_size)) == 0)))
610 if (boundary_upper) {
611 RefPicList *rpl_top = (lc->boundary_flags & BOUNDARY_UPPER_SLICE) ?
612 ff_hevc_get_ref_list(s, s->ref, x0, y0 - 1) :
615 int yp_pu = (y0 - 1) >> log2_min_pu_size;
616 int yq_pu = y0 >> log2_min_pu_size;
617 int yp_tu = (y0 - 1) >> log2_min_tu_size;
618 int yq_tu = y0 >> log2_min_tu_size;
620 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
621 int x_pu = (x0 + i) >> log2_min_pu_size;
622 int x_tu = (x0 + i) >> log2_min_tu_size;
623 MvField *top = &tab_mvf[yp_pu * min_pu_width + x_pu];
624 MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
625 uint8_t top_cbf_luma = s->cbf_luma[yp_tu * min_tu_width + x_tu];
626 uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
628 bs = boundary_strength(s, curr, curr_cbf_luma,
629 top, top_cbf_luma, rpl_top, 1);
631 s->horizontal_bs[((x0 + i) + y0 * s->bs_width) >> 2] = bs;
635 // bs for TU internal horizontal PU boundaries
636 if (log2_trafo_size > s->ps.sps->log2_min_pu_size && !is_intra) {
637 RefPicList *rpl = s->ref->refPicList;
639 for (j = 8; j < (1 << log2_trafo_size); j += 8) {
640 int yp_pu = (y0 + j - 1) >> log2_min_pu_size;
641 int yq_pu = (y0 + j) >> log2_min_pu_size;
642 int yp_tu = (y0 + j - 1) >> log2_min_tu_size;
643 int yq_tu = (y0 + j) >> log2_min_tu_size;
645 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
646 int x_pu = (x0 + i) >> log2_min_pu_size;
647 int x_tu = (x0 + i) >> log2_min_tu_size;
648 MvField *top = &tab_mvf[yp_pu * min_pu_width + x_pu];
649 MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
650 uint8_t top_cbf_luma = s->cbf_luma[yp_tu * min_tu_width + x_tu];
651 uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
653 bs = boundary_strength(s, curr, curr_cbf_luma,
654 top, top_cbf_luma, rpl, 0);
656 s->horizontal_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
661 // bs for vertical TU boundaries
662 boundary_left = x0 > 0 && !(x0 & 7);
664 ((!s->sh.slice_loop_filter_across_slices_enabled_flag &&
665 lc->boundary_flags & BOUNDARY_LEFT_SLICE &&
666 (x0 % (1 << s->ps.sps->log2_ctb_size)) == 0) ||
667 (!s->ps.pps->loop_filter_across_tiles_enabled_flag &&
668 lc->boundary_flags & BOUNDARY_LEFT_TILE &&
669 (x0 % (1 << s->ps.sps->log2_ctb_size)) == 0)))
673 RefPicList *rpl_left = (lc->boundary_flags & BOUNDARY_LEFT_SLICE) ?
674 ff_hevc_get_ref_list(s, s->ref, x0 - 1, y0) :
677 int xp_pu = (x0 - 1) >> log2_min_pu_size;
678 int xq_pu = x0 >> log2_min_pu_size;
679 int xp_tu = (x0 - 1) >> log2_min_tu_size;
680 int xq_tu = x0 >> log2_min_tu_size;
682 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
683 int y_pu = (y0 + i) >> log2_min_pu_size;
684 int y_tu = (y0 + i) >> log2_min_tu_size;
685 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
686 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
688 uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
689 uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
691 bs = boundary_strength(s, curr, curr_cbf_luma,
692 left, left_cbf_luma, rpl_left, 1);
694 s->vertical_bs[(x0 >> 3) + ((y0 + i) >> 2) * s->bs_width] = bs;
698 // bs for TU internal vertical PU boundaries
699 if (log2_trafo_size > log2_min_pu_size && !is_intra) {
700 RefPicList *rpl = s->ref->refPicList;
702 for (j = 0; j < (1 << log2_trafo_size); j += 4) {
703 int y_pu = (y0 + j) >> log2_min_pu_size;
704 int y_tu = (y0 + j) >> log2_min_tu_size;
706 for (i = 8; i < (1 << log2_trafo_size); i += 8) {
707 int xp_pu = (x0 + i - 1) >> log2_min_pu_size;
708 int xq_pu = (x0 + i) >> log2_min_pu_size;
709 int xp_tu = (x0 + i - 1) >> log2_min_tu_size;
710 int xq_tu = (x0 + i) >> log2_min_tu_size;
711 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
712 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
713 uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
714 uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
716 bs = boundary_strength(s, curr, curr_cbf_luma,
717 left, left_cbf_luma, rpl, 0);
719 s->vertical_bs[((x0 + i) >> 3) + ((y0 + j) >> 2) * s->bs_width] = bs;
729 void ff_hevc_hls_filter(HEVCContext *s, int x, int y)
731 deblocking_filter_CTB(s, x, y);
732 if (s->ps.sps->sao_enabled)
733 sao_filter_CTB(s, x, y);
736 void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size)
739 ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size);
740 if (y_ctb && x_ctb >= s->ps.sps->width - ctb_size) {
741 ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size);
742 ff_thread_report_progress(&s->ref->tf, y_ctb - ctb_size, 0);
744 if (x_ctb && y_ctb >= s->ps.sps->height - ctb_size)
745 ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb);