hevc: Change type of array stride parameters to ptrdiff_t
[libav.git] / 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 Libav.
9 *
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.
14 *
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.
19 *
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
23 */
24
25 #include "libavutil/common.h"
26 #include "libavutil/internal.h"
27
28 #include "cabac_functions.h"
29 #include "hevc.h"
30
31 #define LUMA 0
32 #define CB 1
33 #define CR 2
34
35 static const uint8_t tctable[54] = {
36 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, // QP 0...18
37 1, 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 3, 3, 3, 3, 4, 4, 4, // QP 19...37
38 5, 5, 6, 6, 7, 8, 9, 10, 11, 13, 14, 16, 18, 20, 22, 24 // QP 38...53
39 };
40
41 static const uint8_t betatable[52] = {
42 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 6, 7, 8, // QP 0...18
43 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 20, 22, 24, 26, 28, 30, 32, 34, 36, // QP 19...37
44 38, 40, 42, 44, 46, 48, 50, 52, 54, 56, 58, 60, 62, 64 // QP 38...51
45 };
46
47 static int chroma_tc(HEVCContext *s, int qp_y, int c_idx, int tc_offset)
48 {
49 static const int qp_c[] = {
50 29, 30, 31, 32, 33, 33, 34, 34, 35, 35, 36, 36, 37, 37
51 };
52 int qp, qp_i, offset, idxt;
53
54 // slice qp offset is not used for deblocking
55 if (c_idx == 1)
56 offset = s->ps.pps->cb_qp_offset;
57 else
58 offset = s->ps.pps->cr_qp_offset;
59
60 qp_i = av_clip(qp_y + offset, 0, 57);
61 if (qp_i < 30)
62 qp = qp_i;
63 else if (qp_i > 43)
64 qp = qp_i - 6;
65 else
66 qp = qp_c[qp_i - 30];
67
68 idxt = av_clip(qp + DEFAULT_INTRA_TC_OFFSET + tc_offset, 0, 53);
69 return tctable[idxt];
70 }
71
72 static int get_qPy_pred(HEVCContext *s, int xC, int yC,
73 int xBase, int yBase, int log2_cb_size)
74 {
75 HEVCLocalContext *lc = &s->HEVClc;
76 int ctb_size_mask = (1 << s->ps.sps->log2_ctb_size) - 1;
77 int MinCuQpDeltaSizeMask = (1 << (s->ps.sps->log2_ctb_size -
78 s->ps.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->ps.sps->min_cb_width;
82 int min_cb_height = s->ps.sps->min_cb_height;
83 int x_cb = xQgBase >> s->ps.sps->log2_min_cb_size;
84 int y_cb = yQgBase >> s->ps.sps->log2_min_cb_size;
85 int availableA = (xBase & ctb_size_mask) &&
86 (xQgBase & ctb_size_mask);
87 int availableB = (yBase & ctb_size_mask) &&
88 (yQgBase & ctb_size_mask);
89 int qPy_pred, qPy_a, qPy_b;
90
91 // qPy_pred
92 if (lc->first_qp_group || (!xQgBase && !yQgBase)) {
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->ps.sps->log2_ctb_size -
98 s->ps.pps->diff_cu_qp_delta_depth) {
99 static const int offsetX[8][8] = {
100 { -1, 1, 3, 1, 7, 1, 3, 1 },
101 { 0, 0, 0, 0, 0, 0, 0, 0 },
102 { 1, 3, 1, 3, 1, 3, 1, 3 },
103 { 2, 2, 2, 2, 2, 2, 2, 2 },
104 { 3, 5, 7, 5, 3, 5, 7, 5 },
105 { 4, 4, 4, 4, 4, 4, 4, 4 },
106 { 5, 7, 5, 7, 5, 7, 5, 7 },
107 { 6, 6, 6, 6, 6, 6, 6, 6 }
108 };
109 static const int offsetY[8][8] = {
110 { 7, 0, 1, 2, 3, 4, 5, 6 },
111 { 0, 1, 2, 3, 4, 5, 6, 7 },
112 { 1, 0, 3, 2, 5, 4, 7, 6 },
113 { 0, 1, 2, 3, 4, 5, 6, 7 },
114 { 3, 0, 1, 2, 7, 4, 5, 6 },
115 { 0, 1, 2, 3, 4, 5, 6, 7 },
116 { 1, 0, 3, 2, 5, 4, 7, 6 },
117 { 0, 1, 2, 3, 4, 5, 6, 7 }
118 };
119 int xC0b = (xC - (xC & ctb_size_mask)) >> s->ps.sps->log2_min_cb_size;
120 int yC0b = (yC - (yC & ctb_size_mask)) >> s->ps.sps->log2_min_cb_size;
121 int idxX = (xQgBase & ctb_size_mask) >> s->ps.sps->log2_min_cb_size;
122 int idxY = (yQgBase & ctb_size_mask) >> s->ps.sps->log2_min_cb_size;
123 int idx_mask = ctb_size_mask >> s->ps.sps->log2_min_cb_size;
124 int x, y;
125
126 x = FFMIN(xC0b + offsetX[idxX][idxY], min_cb_width - 1);
127 y = FFMIN(yC0b + (offsetY[idxX][idxY] & idx_mask), min_cb_height - 1);
128
129 if (xC0b == (lc->start_of_tiles_x >> s->ps.sps->log2_min_cb_size) &&
130 offsetX[idxX][idxY] == -1) {
131 x = (lc->end_of_tiles_x >> s->ps.sps->log2_min_cb_size) - 1;
132 y = yC0b - 1;
133 }
134 qPy_pred = s->qp_y_tab[y * min_cb_width + x];
135 }
136 }
137
138 // qPy_a
139 if (availableA == 0)
140 qPy_a = qPy_pred;
141 else
142 qPy_a = s->qp_y_tab[(x_cb - 1) + y_cb * min_cb_width];
143
144 // qPy_b
145 if (availableB == 0)
146 qPy_b = qPy_pred;
147 else
148 qPy_b = s->qp_y_tab[x_cb + (y_cb - 1) * min_cb_width];
149
150 return (qPy_a + qPy_b + 1) >> 1;
151 }
152
153 void ff_hevc_set_qPy(HEVCContext *s, int xC, int yC,
154 int xBase, int yBase, int log2_cb_size)
155 {
156 int qp_y = get_qPy_pred(s, xC, yC, xBase, yBase, log2_cb_size);
157
158 if (s->HEVClc.tu.cu_qp_delta != 0) {
159 int off = s->ps.sps->qp_bd_offset;
160 s->HEVClc.qp_y = FFUMOD(qp_y + s->HEVClc.tu.cu_qp_delta + 52 + 2 * off,
161 52 + off) - off;
162 } else
163 s->HEVClc.qp_y = qp_y;
164 }
165
166 static int get_qPy(HEVCContext *s, int xC, int yC)
167 {
168 int log2_min_cb_size = s->ps.sps->log2_min_cb_size;
169 int x = xC >> log2_min_cb_size;
170 int y = yC >> log2_min_cb_size;
171 return s->qp_y_tab[x + y * s->ps.sps->min_cb_width];
172 }
173
174 static void copy_CTB(uint8_t *dst, uint8_t *src,
175 int width, int height, ptrdiff_t stride)
176 {
177 int i;
178
179 for (i = 0; i < height; i++) {
180 memcpy(dst, src, width);
181 dst += stride;
182 src += stride;
183 }
184 }
185
186 #define CTB(tab, x, y) ((tab)[(y) * s->ps.sps->ctb_width + (x)])
187
188 static void sao_filter_CTB(HEVCContext *s, int x, int y)
189 {
190 // TODO: This should be easily parallelizable
191 // TODO: skip CBs when (cu_transquant_bypass_flag || (pcm_loop_filter_disable_flag && pcm_flag))
192 int c_idx = 0;
193 int class = 1, class_index;
194 int edges[4]; // 0 left 1 top 2 right 3 bottom
195 SAOParams *sao[4];
196 int classes[4];
197 int x_shift = 0, y_shift = 0;
198 int x_ctb = x >> s->ps.sps->log2_ctb_size;
199 int y_ctb = y >> s->ps.sps->log2_ctb_size;
200 int ctb_addr_rs = y_ctb * s->ps.sps->ctb_width + x_ctb;
201 int ctb_addr_ts = s->ps.pps->ctb_addr_rs_to_ts[ctb_addr_rs];
202
203 // flags indicating unfilterable edges
204 uint8_t vert_edge[] = { 0, 0, 0, 0 };
205 uint8_t horiz_edge[] = { 0, 0, 0, 0 };
206 uint8_t diag_edge[] = { 0, 0, 0, 0 };
207 uint8_t lfase[3]; // current, above, left
208 uint8_t no_tile_filter = s->ps.pps->tiles_enabled_flag &&
209 !s->ps.pps->loop_filter_across_tiles_enabled_flag;
210 uint8_t left_tile_edge = 0, up_tile_edge = 0;
211
212 sao[0] = &CTB(s->sao, x_ctb, y_ctb);
213 edges[0] = x_ctb == 0;
214 edges[1] = y_ctb == 0;
215 edges[2] = x_ctb == s->ps.sps->ctb_width - 1;
216 edges[3] = y_ctb == s->ps.sps->ctb_height - 1;
217 lfase[0] = CTB(s->filter_slice_edges, x_ctb, y_ctb);
218 classes[0] = 0;
219
220 if (!edges[0]) {
221 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]];
222 sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb);
223 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;
224 vert_edge[2] = vert_edge[0];
225 lfase[2] = CTB(s->filter_slice_edges, x_ctb - 1, y_ctb);
226 classes[class] = 2;
227 class++;
228 x_shift = 8;
229 }
230
231 if (!edges[1]) {
232 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]];
233 sao[class] = &CTB(s->sao, x_ctb, y_ctb - 1);
234 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;
235 horiz_edge[1] = horiz_edge[0];
236 lfase[1] = CTB(s->filter_slice_edges, x_ctb, y_ctb - 1);
237 classes[class] = 1;
238 class++;
239 y_shift = 4;
240
241 if (!edges[0]) {
242 classes[class] = 3;
243 sao[class] = &CTB(s->sao, x_ctb - 1, y_ctb - 1);
244 class++;
245
246 // Tile check here is done current CTB row/col, not above/left like you'd expect,
247 //but that is because the tile boundary always extends through the whole pic
248 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;
249 vert_edge[3] = vert_edge[1];
250 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;
251 horiz_edge[3] = horiz_edge[2];
252 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;
253 diag_edge[3] = diag_edge[0];
254
255 // Does left CTB comes after above CTB?
256 if (CTB(s->tab_slice_address, x_ctb - 1, y_ctb) >
257 CTB(s->tab_slice_address, x_ctb, y_ctb - 1)) {
258 diag_edge[2] = !lfase[2] || left_tile_edge || up_tile_edge;
259 diag_edge[1] = diag_edge[2];
260 } else 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[1] = !lfase[1] || left_tile_edge || up_tile_edge;
263 diag_edge[2] = diag_edge[1];
264 } else {
265 // Same slice, only consider tiles
266 diag_edge[2] = left_tile_edge || up_tile_edge;
267 diag_edge[1] = diag_edge[2];
268 }
269 }
270 }
271
272 for (c_idx = 0; c_idx < 3; c_idx++) {
273 int chroma = c_idx ? 1 : 0;
274 int x0 = x >> chroma;
275 int y0 = y >> chroma;
276 ptrdiff_t stride = s->frame->linesize[c_idx];
277 int ctb_size = (1 << (s->ps.sps->log2_ctb_size)) >> s->ps.sps->hshift[c_idx];
278 int width = FFMIN(ctb_size,
279 (s->ps.sps->width >> s->ps.sps->hshift[c_idx]) - x0);
280 int height = FFMIN(ctb_size,
281 (s->ps.sps->height >> s->ps.sps->vshift[c_idx]) - y0);
282
283 uint8_t *src = &s->frame->data[c_idx][y0 * stride + (x0 << s->ps.sps->pixel_shift)];
284 uint8_t *dst = &s->sao_frame->data[c_idx][y0 * stride + (x0 << s->ps.sps->pixel_shift)];
285 int offset = (y_shift >> chroma) * stride + ((x_shift >> chroma) << s->ps.sps->pixel_shift);
286
287 copy_CTB(dst - offset, src - offset,
288 (edges[2] ? width + (x_shift >> chroma) : width) << s->ps.sps->pixel_shift,
289 (edges[3] ? height + (y_shift >> chroma) : height), stride);
290
291 for (class_index = 0; class_index < class; class_index++) {
292
293 switch (sao[class_index]->type_idx[c_idx]) {
294 case SAO_BAND:
295 s->hevcdsp.sao_band_filter[classes[class_index]](dst, src,
296 stride,
297 sao[class_index],
298 edges, width,
299 height, c_idx);
300 break;
301 case SAO_EDGE:
302 s->hevcdsp.sao_edge_filter[classes[class_index]](dst, src,
303 stride,
304 sao[class_index],
305 edges, width,
306 height, c_idx,
307 vert_edge[classes[class_index]],
308 horiz_edge[classes[class_index]],
309 diag_edge[classes[class_index]]);
310 break;
311 }
312 }
313 }
314 }
315
316 static int get_pcm(HEVCContext *s, int x, int y)
317 {
318 int log2_min_pu_size = s->ps.sps->log2_min_pu_size;
319 int x_pu, y_pu;
320
321 if (x < 0 || y < 0)
322 return 2;
323
324 x_pu = x >> log2_min_pu_size;
325 y_pu = y >> log2_min_pu_size;
326
327 if (x_pu >= s->ps.sps->min_pu_width || y_pu >= s->ps.sps->min_pu_height)
328 return 2;
329 return s->is_pcm[y_pu * s->ps.sps->min_pu_width + x_pu];
330 }
331
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)]
336
337 static void deblocking_filter_CTB(HEVCContext *s, int x0, int y0)
338 {
339 uint8_t *src;
340 int x, y, x_end, y_end, chroma;
341 int c_tc[2], tc[2], beta;
342 uint8_t no_p[2] = { 0 };
343 uint8_t no_q[2] = { 0 };
344
345 int log2_ctb_size = s->ps.sps->log2_ctb_size;
346 int ctb_size = 1 << log2_ctb_size;
347 int ctb = (x0 >> log2_ctb_size) +
348 (y0 >> log2_ctb_size) * s->ps.sps->ctb_width;
349 int cur_tc_offset = s->deblock[ctb].tc_offset;
350 int cur_beta_offset = s->deblock[ctb].beta_offset;
351 int tc_offset, left_tc_offset, beta_offset, left_beta_offset;
352 int pcmf = (s->ps.sps->pcm_enabled_flag &&
353 s->ps.sps->pcm.loop_filter_disable_flag) ||
354 s->ps.pps->transquant_bypass_enable_flag;
355
356 if (x0) {
357 left_tc_offset = s->deblock[ctb - 1].tc_offset;
358 left_beta_offset = s->deblock[ctb - 1].beta_offset;
359 }
360
361 x_end = x0 + ctb_size;
362 if (x_end > s->ps.sps->width)
363 x_end = s->ps.sps->width;
364 y_end = y0 + ctb_size;
365 if (y_end > s->ps.sps->height)
366 y_end = s->ps.sps->height;
367
368 tc_offset = cur_tc_offset;
369 beta_offset = cur_beta_offset;
370
371 // vertical filtering luma
372 for (y = y0; y < y_end; y += 8) {
373 for (x = x0 ? x0 : 8; x < x_end; x += 8) {
374 const int bs0 = s->vertical_bs[(x >> 3) + (y >> 2) * s->bs_width];
375 const int bs1 = s->vertical_bs[(x >> 3) + ((y + 4) >> 2) * s->bs_width];
376 if (bs0 || bs1) {
377 const int qp = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
378
379 beta = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
380
381 tc[0] = bs0 ? TC_CALC(qp, bs0) : 0;
382 tc[1] = bs1 ? TC_CALC(qp, bs1) : 0;
383 src = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->ps.sps->pixel_shift)];
384 if (pcmf) {
385 no_p[0] = get_pcm(s, x - 1, y);
386 no_p[1] = get_pcm(s, x - 1, y + 4);
387 no_q[0] = get_pcm(s, x, y);
388 no_q[1] = get_pcm(s, x, y + 4);
389 s->hevcdsp.hevc_v_loop_filter_luma_c(src,
390 s->frame->linesize[LUMA],
391 beta, tc, no_p, no_q);
392 } else
393 s->hevcdsp.hevc_v_loop_filter_luma(src,
394 s->frame->linesize[LUMA],
395 beta, tc, no_p, no_q);
396 }
397 }
398 }
399
400 // vertical filtering chroma
401 for (chroma = 1; chroma <= 2; chroma++) {
402 for (y = y0; y < y_end; y += 16) {
403 for (x = x0 ? x0 : 16; x < x_end; x += 16) {
404 const int bs0 = s->vertical_bs[(x >> 3) + (y >> 2) * s->bs_width];
405 const int bs1 = s->vertical_bs[(x >> 3) + ((y + 8) >> 2) * s->bs_width];
406 if ((bs0 == 2) || (bs1 == 2)) {
407 const int qp0 = (get_qPy(s, x - 1, y) + get_qPy(s, x, y) + 1) >> 1;
408 const int qp1 = (get_qPy(s, x - 1, y + 8) + get_qPy(s, x, y + 8) + 1) >> 1;
409
410 c_tc[0] = (bs0 == 2) ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
411 c_tc[1] = (bs1 == 2) ? chroma_tc(s, qp1, chroma, tc_offset) : 0;
412 src = &s->frame->data[chroma][y / 2 * s->frame->linesize[chroma] + ((x / 2) << s->ps.sps->pixel_shift)];
413 if (pcmf) {
414 no_p[0] = get_pcm(s, x - 1, y);
415 no_p[1] = get_pcm(s, x - 1, y + 8);
416 no_q[0] = get_pcm(s, x, y);
417 no_q[1] = get_pcm(s, x, y + 8);
418 s->hevcdsp.hevc_v_loop_filter_chroma_c(src,
419 s->frame->linesize[chroma],
420 c_tc, no_p, no_q);
421 } else
422 s->hevcdsp.hevc_v_loop_filter_chroma(src,
423 s->frame->linesize[chroma],
424 c_tc, no_p, no_q);
425 }
426 }
427 }
428 }
429
430 // horizontal filtering luma
431 if (x_end != s->ps.sps->width)
432 x_end -= 8;
433 for (y = y0 ? y0 : 8; y < y_end; y += 8) {
434 for (x = x0 ? x0 - 8 : 0; x < x_end; x += 8) {
435 const int bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
436 const int bs1 = s->horizontal_bs[(x + 4 + y * s->bs_width) >> 2];
437 if (bs0 || bs1) {
438 const int qp = (get_qPy(s, x, y - 1) + get_qPy(s, x, y) + 1) >> 1;
439
440 tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
441 beta_offset = x >= x0 ? cur_beta_offset : left_beta_offset;
442
443 beta = betatable[av_clip(qp + beta_offset, 0, MAX_QP)];
444 tc[0] = bs0 ? TC_CALC(qp, bs0) : 0;
445 tc[1] = bs1 ? TC_CALC(qp, bs1) : 0;
446 src = &s->frame->data[LUMA][y * s->frame->linesize[LUMA] + (x << s->ps.sps->pixel_shift)];
447 if (pcmf) {
448 no_p[0] = get_pcm(s, x, y - 1);
449 no_p[1] = get_pcm(s, x + 4, y - 1);
450 no_q[0] = get_pcm(s, x, y);
451 no_q[1] = get_pcm(s, x + 4, y);
452 s->hevcdsp.hevc_h_loop_filter_luma_c(src,
453 s->frame->linesize[LUMA],
454 beta, tc, no_p, no_q);
455 } else
456 s->hevcdsp.hevc_h_loop_filter_luma(src,
457 s->frame->linesize[LUMA],
458 beta, tc, no_p, no_q);
459 }
460 }
461 }
462
463 // horizontal filtering chroma
464 for (chroma = 1; chroma <= 2; chroma++) {
465 for (y = y0 ? y0 : 16; y < y_end; y += 16) {
466 for (x = x0 - 8; x < x_end; x += 16) {
467 int bs0, bs1;
468 // to make sure no memory access over boundary when x = -8
469 // TODO: simplify with row based deblocking
470 if (x < 0) {
471 bs0 = 0;
472 bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
473 } else if (x >= x_end - 8) {
474 bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
475 bs1 = 0;
476 } else {
477 bs0 = s->horizontal_bs[(x + y * s->bs_width) >> 2];
478 bs1 = s->horizontal_bs[(x + 8 + y * s->bs_width) >> 2];
479 }
480
481 if ((bs0 == 2) || (bs1 == 2)) {
482 const int qp0 = bs0 == 2 ? (get_qPy(s, x, y - 1) + get_qPy(s, x, y) + 1) >> 1 : 0;
483 const int qp1 = bs1 == 2 ? (get_qPy(s, x + 8, y - 1) + get_qPy(s, x + 8, y) + 1) >> 1 : 0;
484
485 tc_offset = x >= x0 ? cur_tc_offset : left_tc_offset;
486 c_tc[0] = bs0 == 2 ? chroma_tc(s, qp0, chroma, tc_offset) : 0;
487 c_tc[1] = bs1 == 2 ? chroma_tc(s, qp1, chroma, cur_tc_offset) : 0;
488 src = &s->frame->data[chroma][y / 2 * s->frame->linesize[chroma] + ((x / 2) << s->ps.sps->pixel_shift)];
489 if (pcmf) {
490 no_p[0] = get_pcm(s, x, y - 1);
491 no_p[1] = get_pcm(s, x + 8, y - 1);
492 no_q[0] = get_pcm(s, x, y);
493 no_q[1] = get_pcm(s, x + 8, y);
494 s->hevcdsp.hevc_h_loop_filter_chroma_c(src,
495 s->frame->linesize[chroma],
496 c_tc, no_p, no_q);
497 } else
498 s->hevcdsp.hevc_h_loop_filter_chroma(src,
499 s->frame->linesize[chroma],
500 c_tc, no_p, no_q);
501 }
502 }
503 }
504 }
505 }
506
507 static int boundary_strength(HEVCContext *s, MvField *curr,
508 uint8_t curr_cbf_luma, MvField *neigh,
509 uint8_t neigh_cbf_luma,
510 RefPicList *neigh_refPicList,
511 int tu_border)
512 {
513 int mvs = curr->pred_flag[0] + curr->pred_flag[1];
514
515 if (tu_border) {
516 if (curr->is_intra || neigh->is_intra)
517 return 2;
518 if (curr_cbf_luma || neigh_cbf_luma)
519 return 1;
520 }
521
522 if (mvs == neigh->pred_flag[0] + neigh->pred_flag[1]) {
523 if (mvs == 2) {
524 // same L0 and L1
525 if (s->ref->refPicList[0].list[curr->ref_idx[0]] == neigh_refPicList[0].list[neigh->ref_idx[0]] &&
526 s->ref->refPicList[0].list[curr->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]] &&
527 neigh_refPicList[0].list[neigh->ref_idx[0]] == neigh_refPicList[1].list[neigh->ref_idx[1]]) {
528 if ((abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
529 abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4) &&
530 (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
531 abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4))
532 return 1;
533 else
534 return 0;
535 } else if (neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
536 neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
537 if (abs(neigh->mv[0].x - curr->mv[0].x) >= 4 || abs(neigh->mv[0].y - curr->mv[0].y) >= 4 ||
538 abs(neigh->mv[1].x - curr->mv[1].x) >= 4 || abs(neigh->mv[1].y - curr->mv[1].y) >= 4)
539 return 1;
540 else
541 return 0;
542 } else if (neigh_refPicList[1].list[neigh->ref_idx[1]] == s->ref->refPicList[0].list[curr->ref_idx[0]] &&
543 neigh_refPicList[0].list[neigh->ref_idx[0]] == s->ref->refPicList[1].list[curr->ref_idx[1]]) {
544 if (abs(neigh->mv[1].x - curr->mv[0].x) >= 4 || abs(neigh->mv[1].y - curr->mv[0].y) >= 4 ||
545 abs(neigh->mv[0].x - curr->mv[1].x) >= 4 || abs(neigh->mv[0].y - curr->mv[1].y) >= 4)
546 return 1;
547 else
548 return 0;
549 } else {
550 return 1;
551 }
552 } else { // 1 MV
553 Mv A, B;
554 int ref_A, ref_B;
555
556 if (curr->pred_flag[0]) {
557 A = curr->mv[0];
558 ref_A = s->ref->refPicList[0].list[curr->ref_idx[0]];
559 } else {
560 A = curr->mv[1];
561 ref_A = s->ref->refPicList[1].list[curr->ref_idx[1]];
562 }
563
564 if (neigh->pred_flag[0]) {
565 B = neigh->mv[0];
566 ref_B = neigh_refPicList[0].list[neigh->ref_idx[0]];
567 } else {
568 B = neigh->mv[1];
569 ref_B = neigh_refPicList[1].list[neigh->ref_idx[1]];
570 }
571
572 if (ref_A == ref_B) {
573 if (abs(A.x - B.x) >= 4 || abs(A.y - B.y) >= 4)
574 return 1;
575 else
576 return 0;
577 } else
578 return 1;
579 }
580 }
581
582 return 1;
583 }
584
585 void ff_hevc_deblocking_boundary_strengths(HEVCContext *s, int x0, int y0,
586 int log2_trafo_size)
587 {
588 HEVCLocalContext *lc = &s->HEVClc;
589 MvField *tab_mvf = s->ref->tab_mvf;
590 int log2_min_pu_size = s->ps.sps->log2_min_pu_size;
591 int log2_min_tu_size = s->ps.sps->log2_min_tb_size;
592 int min_pu_width = s->ps.sps->min_pu_width;
593 int min_tu_width = s->ps.sps->min_tb_width;
594 int is_intra = tab_mvf[(y0 >> log2_min_pu_size) * min_pu_width +
595 (x0 >> log2_min_pu_size)].is_intra;
596 int boundary_upper, boundary_left;
597 int i, j, bs;
598
599 boundary_upper = y0 > 0 && !(y0 & 7);
600 if (boundary_upper &&
601 ((!s->sh.slice_loop_filter_across_slices_enabled_flag &&
602 lc->boundary_flags & BOUNDARY_UPPER_SLICE &&
603 (y0 % (1 << s->ps.sps->log2_ctb_size)) == 0) ||
604 (!s->ps.pps->loop_filter_across_tiles_enabled_flag &&
605 lc->boundary_flags & BOUNDARY_UPPER_TILE &&
606 (y0 % (1 << s->ps.sps->log2_ctb_size)) == 0)))
607 boundary_upper = 0;
608
609 if (boundary_upper) {
610 RefPicList *rpl_top = (lc->boundary_flags & BOUNDARY_UPPER_SLICE) ?
611 ff_hevc_get_ref_list(s, s->ref, x0, y0 - 1) :
612 s->ref->refPicList;
613
614 int yp_pu = (y0 - 1) >> log2_min_pu_size;
615 int yq_pu = y0 >> log2_min_pu_size;
616 int yp_tu = (y0 - 1) >> log2_min_tu_size;
617 int yq_tu = y0 >> log2_min_tu_size;
618
619 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
620 int x_pu = (x0 + i) >> log2_min_pu_size;
621 int x_tu = (x0 + i) >> log2_min_tu_size;
622 MvField *top = &tab_mvf[yp_pu * min_pu_width + x_pu];
623 MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
624 uint8_t top_cbf_luma = s->cbf_luma[yp_tu * min_tu_width + x_tu];
625 uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
626
627 bs = boundary_strength(s, curr, curr_cbf_luma,
628 top, top_cbf_luma, rpl_top, 1);
629 if (bs)
630 s->horizontal_bs[((x0 + i) + y0 * s->bs_width) >> 2] = bs;
631 }
632 }
633
634 // bs for TU internal horizontal PU boundaries
635 if (log2_trafo_size > s->ps.sps->log2_min_pu_size && !is_intra) {
636 RefPicList *rpl = s->ref->refPicList;
637
638 for (j = 8; j < (1 << log2_trafo_size); j += 8) {
639 int yp_pu = (y0 + j - 1) >> log2_min_pu_size;
640 int yq_pu = (y0 + j) >> log2_min_pu_size;
641 int yp_tu = (y0 + j - 1) >> log2_min_tu_size;
642 int yq_tu = (y0 + j) >> log2_min_tu_size;
643
644 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
645 int x_pu = (x0 + i) >> log2_min_pu_size;
646 int x_tu = (x0 + i) >> log2_min_tu_size;
647 MvField *top = &tab_mvf[yp_pu * min_pu_width + x_pu];
648 MvField *curr = &tab_mvf[yq_pu * min_pu_width + x_pu];
649 uint8_t top_cbf_luma = s->cbf_luma[yp_tu * min_tu_width + x_tu];
650 uint8_t curr_cbf_luma = s->cbf_luma[yq_tu * min_tu_width + x_tu];
651
652 bs = boundary_strength(s, curr, curr_cbf_luma,
653 top, top_cbf_luma, rpl, 0);
654 if (bs)
655 s->horizontal_bs[((x0 + i) + (y0 + j) * s->bs_width) >> 2] = bs;
656 }
657 }
658 }
659
660 // bs for vertical TU boundaries
661 boundary_left = x0 > 0 && !(x0 & 7);
662 if (boundary_left &&
663 ((!s->sh.slice_loop_filter_across_slices_enabled_flag &&
664 lc->boundary_flags & BOUNDARY_LEFT_SLICE &&
665 (x0 % (1 << s->ps.sps->log2_ctb_size)) == 0) ||
666 (!s->ps.pps->loop_filter_across_tiles_enabled_flag &&
667 lc->boundary_flags & BOUNDARY_LEFT_TILE &&
668 (x0 % (1 << s->ps.sps->log2_ctb_size)) == 0)))
669 boundary_left = 0;
670
671 if (boundary_left) {
672 RefPicList *rpl_left = (lc->boundary_flags & BOUNDARY_LEFT_SLICE) ?
673 ff_hevc_get_ref_list(s, s->ref, x0 - 1, y0) :
674 s->ref->refPicList;
675
676 int xp_pu = (x0 - 1) >> log2_min_pu_size;
677 int xq_pu = x0 >> log2_min_pu_size;
678 int xp_tu = (x0 - 1) >> log2_min_tu_size;
679 int xq_tu = x0 >> log2_min_tu_size;
680
681 for (i = 0; i < (1 << log2_trafo_size); i += 4) {
682 int y_pu = (y0 + i) >> log2_min_pu_size;
683 int y_tu = (y0 + i) >> log2_min_tu_size;
684 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
685 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
686
687 uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
688 uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
689
690 bs = boundary_strength(s, curr, curr_cbf_luma,
691 left, left_cbf_luma, rpl_left, 1);
692 if (bs)
693 s->vertical_bs[(x0 >> 3) + ((y0 + i) >> 2) * s->bs_width] = bs;
694 }
695 }
696
697 // bs for TU internal vertical PU boundaries
698 if (log2_trafo_size > log2_min_pu_size && !is_intra) {
699 RefPicList *rpl = s->ref->refPicList;
700
701 for (j = 0; j < (1 << log2_trafo_size); j += 4) {
702 int y_pu = (y0 + j) >> log2_min_pu_size;
703 int y_tu = (y0 + j) >> log2_min_tu_size;
704
705 for (i = 8; i < (1 << log2_trafo_size); i += 8) {
706 int xp_pu = (x0 + i - 1) >> log2_min_pu_size;
707 int xq_pu = (x0 + i) >> log2_min_pu_size;
708 int xp_tu = (x0 + i - 1) >> log2_min_tu_size;
709 int xq_tu = (x0 + i) >> log2_min_tu_size;
710 MvField *left = &tab_mvf[y_pu * min_pu_width + xp_pu];
711 MvField *curr = &tab_mvf[y_pu * min_pu_width + xq_pu];
712 uint8_t left_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xp_tu];
713 uint8_t curr_cbf_luma = s->cbf_luma[y_tu * min_tu_width + xq_tu];
714
715 bs = boundary_strength(s, curr, curr_cbf_luma,
716 left, left_cbf_luma, rpl, 0);
717 if (bs)
718 s->vertical_bs[((x0 + i) >> 3) + ((y0 + j) >> 2) * s->bs_width] = bs;
719 }
720 }
721 }
722 }
723
724 #undef LUMA
725 #undef CB
726 #undef CR
727
728 void ff_hevc_hls_filter(HEVCContext *s, int x, int y)
729 {
730 deblocking_filter_CTB(s, x, y);
731 if (s->ps.sps->sao_enabled)
732 sao_filter_CTB(s, x, y);
733 }
734
735 void ff_hevc_hls_filters(HEVCContext *s, int x_ctb, int y_ctb, int ctb_size)
736 {
737 if (y_ctb && x_ctb)
738 ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb - ctb_size);
739 if (y_ctb && x_ctb >= s->ps.sps->width - ctb_size) {
740 ff_hevc_hls_filter(s, x_ctb, y_ctb - ctb_size);
741 ff_thread_report_progress(&s->ref->tf, y_ctb - ctb_size, 0);
742 }
743 if (x_ctb && y_ctb >= s->ps.sps->height - ctb_size)
744 ff_hevc_hls_filter(s, x_ctb - ctb_size, y_ctb);
745 }