D3D11va: add a Direct3D11 video decoder similar to DXVA2
[libav.git] / libavcodec / h264_slice.c
CommitLineData
f51d0f39
LB
1/*
2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
4 *
5 * This file is part of Libav.
6 *
7 * Libav is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
11 *
12 * Libav is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
22/**
23 * @file
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
26 */
27
28#include "libavutil/avassert.h"
29#include "libavutil/imgutils.h"
30#include "libavutil/timer.h"
31#include "internal.h"
32#include "cabac.h"
33#include "cabac_functions.h"
f51d0f39
LB
34#include "error_resilience.h"
35#include "avcodec.h"
36#include "h264.h"
37#include "h264data.h"
38#include "h264chroma.h"
39#include "h264_mvpred.h"
40#include "golomb.h"
41#include "mathops.h"
42#include "mpegutils.h"
43#include "rectangle.h"
44#include "thread.h"
45
46
47static const uint8_t rem6[QP_MAX_NUM + 1] = {
48 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
49 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
50 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3,
51};
52
53static const uint8_t div6[QP_MAX_NUM + 1] = {
54 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
55 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
56 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10, 10,
57};
58
59static const uint8_t field_scan[16] = {
60 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
61 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
62 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
63 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
64};
65
66static const uint8_t field_scan8x8[64] = {
67 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
68 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
69 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
70 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
71 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
72 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
73 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
74 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
75 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
76 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
77 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
78 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
79 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
80 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
81 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
82 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
83};
84
85static const uint8_t field_scan8x8_cavlc[64] = {
86 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
87 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
88 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
89 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
90 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
91 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
92 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
93 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
94 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
95 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
96 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
97 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
98 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
99 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
100 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
101 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
102};
103
104// zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
105static const uint8_t zigzag_scan8x8_cavlc[64] = {
106 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
107 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
108 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
109 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
110 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
111 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
112 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
113 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
114 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
115 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
116 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
117 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
118 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
119 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
120 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
121 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
122};
123
124static const uint8_t dequant4_coeff_init[6][3] = {
125 { 10, 13, 16 },
126 { 11, 14, 18 },
127 { 13, 16, 20 },
128 { 14, 18, 23 },
129 { 16, 20, 25 },
130 { 18, 23, 29 },
131};
132
133static const uint8_t dequant8_coeff_init_scan[16] = {
134 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
135};
136
137static const uint8_t dequant8_coeff_init[6][6] = {
138 { 20, 18, 32, 19, 25, 24 },
139 { 22, 19, 35, 21, 28, 26 },
140 { 26, 23, 42, 24, 33, 31 },
141 { 28, 25, 45, 26, 35, 33 },
142 { 32, 28, 51, 30, 40, 38 },
143 { 36, 32, 58, 34, 46, 43 },
144};
145
f51d0f39
LB
146
147static void release_unused_pictures(H264Context *h, int remove_current)
148{
149 int i;
150
151 /* release non reference frames */
152 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
a0f29460 153 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
f51d0f39
LB
154 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
155 ff_h264_unref_picture(h, &h->DPB[i]);
156 }
157 }
158}
159
36d04801 160static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
f51d0f39 161{
c377e04d 162 const H264Context *h = sl->h264;
f51d0f39
LB
163 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
164
36d04801 165 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
f51d0f39
LB
166 // edge emu needs blocksize + filter length - 1
167 // (= 21x21 for h264)
36d04801 168 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
f51d0f39 169
c377e04d
AK
170 av_fast_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
171 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
172 av_fast_malloc(&sl->top_borders[1], &sl->top_borders_allocated[1],
173 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
174
175 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
176 !sl->top_borders[0] || !sl->top_borders[1]) {
36d04801
AK
177 av_freep(&sl->bipred_scratchpad);
178 av_freep(&sl->edge_emu_buffer);
c377e04d
AK
179 av_freep(&sl->top_borders[0]);
180 av_freep(&sl->top_borders[1]);
181
36d04801
AK
182 sl->bipred_scratchpad_allocated = 0;
183 sl->edge_emu_buffer_allocated = 0;
c377e04d
AK
184 sl->top_borders_allocated[0] = 0;
185 sl->top_borders_allocated[1] = 0;
f51d0f39
LB
186 return AVERROR(ENOMEM);
187 }
188
189 return 0;
190}
191
192static int init_table_pools(H264Context *h)
193{
194 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
195 const int mb_array_size = h->mb_stride * h->mb_height;
196 const int b4_stride = h->mb_width * 4 + 1;
197 const int b4_array_size = b4_stride * h->mb_height * 4;
198
199 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
200 av_buffer_allocz);
201 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
202 sizeof(uint32_t), av_buffer_allocz);
2db953f8
LB
203 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
204 sizeof(int16_t), av_buffer_allocz);
205 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
f51d0f39
LB
206
207 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
208 !h->ref_index_pool) {
209 av_buffer_pool_uninit(&h->qscale_table_pool);
210 av_buffer_pool_uninit(&h->mb_type_pool);
211 av_buffer_pool_uninit(&h->motion_val_pool);
212 av_buffer_pool_uninit(&h->ref_index_pool);
213 return AVERROR(ENOMEM);
214 }
215
216 return 0;
217}
218
219static int alloc_picture(H264Context *h, H264Picture *pic)
220{
221 int i, ret = 0;
222
a0f29460 223 av_assert0(!pic->f->data[0]);
f51d0f39 224
a0f29460 225 pic->tf.f = pic->f;
f51d0f39
LB
226 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
227 AV_GET_BUFFER_FLAG_REF : 0);
228 if (ret < 0)
229 goto fail;
230
f51d0f39
LB
231 if (h->avctx->hwaccel) {
232 const AVHWAccel *hwaccel = h->avctx->hwaccel;
233 av_assert0(!pic->hwaccel_picture_private);
a871ef0c
AK
234 if (hwaccel->frame_priv_data_size) {
235 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
f51d0f39
LB
236 if (!pic->hwaccel_priv_buf)
237 return AVERROR(ENOMEM);
238 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
239 }
240 }
241
242 if (!h->qscale_table_pool) {
243 ret = init_table_pools(h);
244 if (ret < 0)
245 goto fail;
246 }
247
248 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
249 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
250 if (!pic->qscale_table_buf || !pic->mb_type_buf)
251 goto fail;
252
253 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
254 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
255
256 for (i = 0; i < 2; i++) {
257 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
258 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
259 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
260 goto fail;
261
262 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
263 pic->ref_index[i] = pic->ref_index_buf[i]->data;
264 }
265
266 return 0;
267fail:
268 ff_h264_unref_picture(h, pic);
269 return (ret < 0) ? ret : AVERROR(ENOMEM);
270}
271
272static inline int pic_is_unused(H264Context *h, H264Picture *pic)
273{
a0f29460 274 if (!pic->f->buf[0])
f51d0f39 275 return 1;
f51d0f39
LB
276 return 0;
277}
278
279static int find_unused_picture(H264Context *h)
280{
281 int i;
282
283 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
284 if (pic_is_unused(h, &h->DPB[i]))
285 break;
286 }
287 if (i == H264_MAX_PICTURE_COUNT)
288 return AVERROR_INVALIDDATA;
289
f51d0f39
LB
290 return i;
291}
292
293
294static void init_dequant8_coeff_table(H264Context *h)
295{
296 int i, j, q, x;
297 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
298
299 for (i = 0; i < 6; i++) {
300 h->dequant8_coeff[i] = h->dequant8_buffer[i];
301 for (j = 0; j < i; j++)
302 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
303 64 * sizeof(uint8_t))) {
304 h->dequant8_coeff[i] = h->dequant8_buffer[j];
305 break;
306 }
307 if (j < i)
308 continue;
309
310 for (q = 0; q < max_qp + 1; q++) {
311 int shift = div6[q];
312 int idx = rem6[q];
313 for (x = 0; x < 64; x++)
314 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
315 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
316 h->pps.scaling_matrix8[i][x]) << shift;
317 }
318 }
319}
320
321static void init_dequant4_coeff_table(H264Context *h)
322{
323 int i, j, q, x;
324 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
325 for (i = 0; i < 6; i++) {
326 h->dequant4_coeff[i] = h->dequant4_buffer[i];
327 for (j = 0; j < i; j++)
328 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
329 16 * sizeof(uint8_t))) {
330 h->dequant4_coeff[i] = h->dequant4_buffer[j];
331 break;
332 }
333 if (j < i)
334 continue;
335
336 for (q = 0; q < max_qp + 1; q++) {
337 int shift = div6[q] + 2;
338 int idx = rem6[q];
339 for (x = 0; x < 16; x++)
340 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
341 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
342 h->pps.scaling_matrix4[i][x]) << shift;
343 }
344 }
345}
346
347void h264_init_dequant_tables(H264Context *h)
348{
349 int i, x;
350 init_dequant4_coeff_table(h);
351 if (h->pps.transform_8x8_mode)
352 init_dequant8_coeff_table(h);
353 if (h->sps.transform_bypass) {
354 for (i = 0; i < 6; i++)
355 for (x = 0; x < 16; x++)
356 h->dequant4_coeff[i][0][x] = 1 << 6;
357 if (h->pps.transform_8x8_mode)
358 for (i = 0; i < 6; i++)
359 for (x = 0; x < 64; x++)
360 h->dequant8_coeff[i][0][x] = 1 << 6;
361 }
362}
363
f51d0f39 364#define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
d75190aa 365
f51d0f39
LB
366#define REBASE_PICTURE(pic, new_ctx, old_ctx) \
367 ((pic && pic >= old_ctx->DPB && \
368 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
369 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
370
371static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
372 H264Context *new_base,
373 H264Context *old_base)
374{
375 int i;
376
377 for (i = 0; i < count; i++) {
378 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
379 IN_RANGE(from[i], old_base->DPB,
380 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
381 !from[i]));
382 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
383 }
384}
385
386static int copy_parameter_set(void **to, void **from, int count, int size)
387{
388 int i;
389
390 for (i = 0; i < count; i++) {
391 if (to[i] && !from[i]) {
392 av_freep(&to[i]);
393 } else if (from[i] && !to[i]) {
394 to[i] = av_malloc(size);
395 if (!to[i])
396 return AVERROR(ENOMEM);
397 }
398
399 if (from[i])
400 memcpy(to[i], from[i], size);
401 }
402
403 return 0;
404}
405
406#define copy_fields(to, from, start_field, end_field) \
407 memcpy(&to->start_field, &from->start_field, \
408 (char *)&to->end_field - (char *)&to->start_field)
409
8a66fd40 410static int h264_slice_header_init(H264Context *h);
f51d0f39
LB
411
412int ff_h264_update_thread_context(AVCodecContext *dst,
413 const AVCodecContext *src)
414{
415 H264Context *h = dst->priv_data, *h1 = src->priv_data;
416 int inited = h->context_initialized, err = 0;
54986d6d 417 int need_reinit = 0;
f51d0f39
LB
418 int i, ret;
419
420 if (dst == src || !h1->context_initialized)
421 return 0;
422
423 if (inited &&
424 (h->width != h1->width ||
425 h->height != h1->height ||
426 h->mb_width != h1->mb_width ||
427 h->mb_height != h1->mb_height ||
428 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
429 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
430 h->sps.colorspace != h1->sps.colorspace)) {
54986d6d 431 need_reinit = 1;
f51d0f39
LB
432 }
433
88c612e3
AK
434 // SPS/PPS
435 if ((ret = copy_parameter_set((void **)h->sps_buffers,
436 (void **)h1->sps_buffers,
437 MAX_SPS_COUNT, sizeof(SPS))) < 0)
438 return ret;
439 h->sps = h1->sps;
440 if ((ret = copy_parameter_set((void **)h->pps_buffers,
441 (void **)h1->pps_buffers,
442 MAX_PPS_COUNT, sizeof(PPS))) < 0)
443 return ret;
444 h->pps = h1->pps;
445
446 if (need_reinit || !inited) {
447 h->width = h1->width;
448 h->height = h1->height;
449 h->mb_height = h1->mb_height;
450 h->mb_width = h1->mb_width;
451 h->mb_num = h1->mb_num;
452 h->mb_stride = h1->mb_stride;
453 h->b_stride = h1->b_stride;
454
455 if ((err = h264_slice_header_init(h)) < 0) {
456 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
457 return err;
f51d0f39
LB
458 }
459
88c612e3
AK
460 /* copy block_offset since frame_start may not be called */
461 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
f51d0f39
LB
462 }
463
464 h->avctx->coded_height = h1->avctx->coded_height;
465 h->avctx->coded_width = h1->avctx->coded_width;
466 h->avctx->width = h1->avctx->width;
467 h->avctx->height = h1->avctx->height;
468 h->coded_picture_number = h1->coded_picture_number;
469 h->first_field = h1->first_field;
470 h->picture_structure = h1->picture_structure;
f51d0f39
LB
471 h->droppable = h1->droppable;
472 h->low_delay = h1->low_delay;
473
474 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
475 ff_h264_unref_picture(h, &h->DPB[i]);
a0f29460 476 if (h1->DPB[i].f->buf[0] &&
f51d0f39
LB
477 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
478 return ret;
479 }
480
481 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
482 ff_h264_unref_picture(h, &h->cur_pic);
a0f29460 483 if (h1->cur_pic.f->buf[0]) {
0dea4c77
AK
484 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
485 if (ret < 0)
486 return ret;
487 }
f51d0f39 488
a4d34e21 489 h->enable_er = h1->enable_er;
f51d0f39
LB
490 h->workaround_bugs = h1->workaround_bugs;
491 h->low_delay = h1->low_delay;
492 h->droppable = h1->droppable;
493
f51d0f39
LB
494 // extradata/NAL handling
495 h->is_avc = h1->is_avc;
29dcde2b 496 h->nal_length_size = h1->nal_length_size;
f51d0f39 497
f51d0f39
LB
498 // Dequantization matrices
499 // FIXME these are big - can they be only copied when PPS changes?
500 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
501
502 for (i = 0; i < 6; i++)
503 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
504 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
505
506 for (i = 0; i < 6; i++)
507 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
508 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
509
510 h->dequant_coeff_pps = h1->dequant_coeff_pps;
511
512 // POC timing
9951907f 513 copy_fields(h, h1, poc_lsb, default_ref_list);
f51d0f39
LB
514
515 // reference lists
5bf3c0fa 516 copy_fields(h, h1, short_ref, current_slice);
f51d0f39
LB
517
518 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
519 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
520 copy_picture_range(h->delayed_pic, h1->delayed_pic,
521 MAX_DELAYED_PIC_COUNT + 2, h, h1);
522
523 h->last_slice_type = h1->last_slice_type;
524
f51d0f39
LB
525 if (!h->cur_pic_ptr)
526 return 0;
527
528 if (!h->droppable) {
529 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
530 h->prev_poc_msb = h->poc_msb;
531 h->prev_poc_lsb = h->poc_lsb;
532 }
533 h->prev_frame_num_offset = h->frame_num_offset;
534 h->prev_frame_num = h->frame_num;
f51d0f39
LB
535
536 h->recovery_frame = h1->recovery_frame;
537 h->frame_recovered = h1->frame_recovered;
538
539 return err;
540}
541
542static int h264_frame_start(H264Context *h)
543{
544 H264Picture *pic;
545 int i, ret;
546 const int pixel_shift = h->pixel_shift;
547
548 release_unused_pictures(h, 1);
549 h->cur_pic_ptr = NULL;
550
551 i = find_unused_picture(h);
552 if (i < 0) {
553 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
554 return i;
555 }
556 pic = &h->DPB[i];
557
558 pic->reference = h->droppable ? 0 : h->picture_structure;
a0f29460 559 pic->f->coded_picture_number = h->coded_picture_number++;
f51d0f39
LB
560 pic->field_picture = h->picture_structure != PICT_FRAME;
561 /*
562 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
563 * in later.
564 * See decode_nal_units().
565 */
a0f29460 566 pic->f->key_frame = 0;
f51d0f39
LB
567 pic->mmco_reset = 0;
568 pic->recovered = 0;
569
570 if ((ret = alloc_picture(h, pic)) < 0)
571 return ret;
572
573 h->cur_pic_ptr = pic;
574 ff_h264_unref_picture(h, &h->cur_pic);
575 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
576 return ret;
577
a4d34e21 578 if (CONFIG_ERROR_RESILIENCE && h->enable_er)
582683b6 579 ff_er_frame_start(&h->slice_ctx[0].er);
f51d0f39 580
f51d0f39 581 for (i = 0; i < 16; i++) {
a0f29460
AK
582 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
583 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
f51d0f39
LB
584 }
585 for (i = 0; i < 16; i++) {
586 h->block_offset[16 + i] =
a0f29460 587 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
f51d0f39 588 h->block_offset[48 + 16 + i] =
a0f29460 589 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
f51d0f39
LB
590 }
591
f51d0f39
LB
592 /* Some macroblocks can be accessed before they're available in case
593 * of lost slices, MBAFF or threading. */
594 memset(h->slice_table, -1,
595 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
596
f51d0f39
LB
597 /* We mark the current picture as non-reference after allocating it, so
598 * that if we break out due to an error it can be released automatically
835f798c 599 * in the next ff_mpv_frame_start().
f51d0f39
LB
600 */
601 h->cur_pic_ptr->reference = 0;
602
603 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
604
605 h->next_output_pic = NULL;
606
607 assert(h->cur_pic_ptr->long_ref == 0);
608
609 return 0;
610}
611
92e0b734 612static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
d4d9068c 613 uint8_t *src_y,
f51d0f39
LB
614 uint8_t *src_cb, uint8_t *src_cr,
615 int linesize, int uvlinesize,
616 int simple)
617{
618 uint8_t *top_border;
619 int top_idx = 1;
620 const int pixel_shift = h->pixel_shift;
621 int chroma444 = CHROMA444(h);
622 int chroma422 = CHROMA422(h);
623
624 src_y -= linesize;
625 src_cb -= uvlinesize;
626 src_cr -= uvlinesize;
627
628 if (!simple && FRAME_MBAFF(h)) {
d4d9068c 629 if (sl->mb_y & 1) {
e9b2383b 630 if (!MB_MBAFF(sl)) {
c377e04d 631 top_border = sl->top_borders[0][sl->mb_x];
f51d0f39
LB
632 AV_COPY128(top_border, src_y + 15 * linesize);
633 if (pixel_shift)
634 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
635 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
636 if (chroma444) {
637 if (pixel_shift) {
638 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
639 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
640 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
641 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
642 } else {
643 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
644 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
645 }
646 } else if (chroma422) {
647 if (pixel_shift) {
648 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
649 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
650 } else {
651 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
652 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
653 }
654 } else {
655 if (pixel_shift) {
656 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
657 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
658 } else {
659 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
660 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
661 }
662 }
663 }
664 }
e9b2383b 665 } else if (MB_MBAFF(sl)) {
f51d0f39
LB
666 top_idx = 0;
667 } else
668 return;
669 }
670
c377e04d 671 top_border = sl->top_borders[top_idx][sl->mb_x];
f51d0f39
LB
672 /* There are two lines saved, the line above the top macroblock
673 * of a pair, and the line above the bottom macroblock. */
674 AV_COPY128(top_border, src_y + 16 * linesize);
675 if (pixel_shift)
676 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
677
678 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
679 if (chroma444) {
680 if (pixel_shift) {
681 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
682 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
683 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
684 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
685 } else {
686 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
687 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
688 }
689 } else if (chroma422) {
690 if (pixel_shift) {
691 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
692 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
693 } else {
694 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
695 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
696 }
697 } else {
698 if (pixel_shift) {
699 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
700 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
701 } else {
702 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
703 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
704 }
705 }
706 }
707}
708
709/**
710 * Initialize implicit_weight table.
711 * @param field 0/1 initialize the weight for interlaced MBAFF
712 * -1 initializes the rest
713 */
92e0b734 714static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
f51d0f39
LB
715{
716 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
717
718 for (i = 0; i < 2; i++) {
92c6c2a6
AK
719 sl->luma_weight_flag[i] = 0;
720 sl->chroma_weight_flag[i] = 0;
f51d0f39
LB
721 }
722
723 if (field < 0) {
724 if (h->picture_structure == PICT_FRAME) {
725 cur_poc = h->cur_pic_ptr->poc;
726 } else {
727 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
728 }
95eb35f3
AK
729 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
730 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
92c6c2a6
AK
731 sl->use_weight = 0;
732 sl->use_weight_chroma = 0;
f51d0f39
LB
733 return;
734 }
735 ref_start = 0;
95eb35f3
AK
736 ref_count0 = sl->ref_count[0];
737 ref_count1 = sl->ref_count[1];
f51d0f39
LB
738 } else {
739 cur_poc = h->cur_pic_ptr->field_poc[field];
740 ref_start = 16;
95eb35f3
AK
741 ref_count0 = 16 + 2 * sl->ref_count[0];
742 ref_count1 = 16 + 2 * sl->ref_count[1];
f51d0f39
LB
743 }
744
92c6c2a6
AK
745 sl->use_weight = 2;
746 sl->use_weight_chroma = 2;
747 sl->luma_log2_weight_denom = 5;
748 sl->chroma_log2_weight_denom = 5;
f51d0f39
LB
749
750 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
95eb35f3 751 int poc0 = sl->ref_list[0][ref0].poc;
f51d0f39
LB
752 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
753 int w = 32;
a12d3188 754 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
95eb35f3 755 int poc1 = sl->ref_list[1][ref1].poc;
9abc80f1 756 int td = av_clip_int8(poc1 - poc0);
f51d0f39 757 if (td) {
9abc80f1 758 int tb = av_clip_int8(cur_poc - poc0);
f51d0f39
LB
759 int tx = (16384 + (FFABS(td) >> 1)) / td;
760 int dist_scale_factor = (tb * tx + 32) >> 8;
761 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
762 w = 64 - dist_scale_factor;
763 }
764 }
765 if (field < 0) {
92c6c2a6
AK
766 sl->implicit_weight[ref0][ref1][0] =
767 sl->implicit_weight[ref0][ref1][1] = w;
f51d0f39 768 } else {
92c6c2a6 769 sl->implicit_weight[ref0][ref1][field] = w;
f51d0f39
LB
770 }
771 }
772 }
773}
774
775/**
776 * initialize scan tables
777 */
778static void init_scan_tables(H264Context *h)
779{
780 int i;
781 for (i = 0; i < 16; i++) {
782#define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
783 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
784 h->field_scan[i] = TRANSPOSE(field_scan[i]);
785#undef TRANSPOSE
786 }
787 for (i = 0; i < 64; i++) {
788#define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
789 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
790 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
791 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
792 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
793#undef TRANSPOSE
794 }
795 if (h->sps.transform_bypass) { // FIXME same ugly
796 h->zigzag_scan_q0 = zigzag_scan;
797 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
798 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
799 h->field_scan_q0 = field_scan;
800 h->field_scan8x8_q0 = field_scan8x8;
801 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
802 } else {
803 h->zigzag_scan_q0 = h->zigzag_scan;
804 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
805 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
806 h->field_scan_q0 = h->field_scan;
807 h->field_scan8x8_q0 = h->field_scan8x8;
808 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
809 }
810}
811
f51d0f39
LB
812static enum AVPixelFormat get_pixel_format(H264Context *h)
813{
4cfbeef3
RDC
814#define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
815 CONFIG_H264_VAAPI_HWACCEL + \
816 (CONFIG_H264_VDA_HWACCEL * 2) + \
817 CONFIG_H264_VDPAU_HWACCEL)
818 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
ce423163
RDC
819 const enum AVPixelFormat *choices = pix_fmts;
820
f51d0f39
LB
821 switch (h->sps.bit_depth_luma) {
822 case 9:
823 if (CHROMA444(h)) {
824 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
4cfbeef3 825 *fmt++ = AV_PIX_FMT_GBRP9;
f51d0f39 826 } else
4cfbeef3 827 *fmt++ = AV_PIX_FMT_YUV444P9;
f51d0f39 828 } else if (CHROMA422(h))
4cfbeef3 829 *fmt++ = AV_PIX_FMT_YUV422P9;
f51d0f39 830 else
4cfbeef3 831 *fmt++ = AV_PIX_FMT_YUV420P9;
f51d0f39
LB
832 break;
833 case 10:
834 if (CHROMA444(h)) {
835 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
4cfbeef3 836 *fmt++ = AV_PIX_FMT_GBRP10;
f51d0f39 837 } else
4cfbeef3 838 *fmt++ = AV_PIX_FMT_YUV444P10;
f51d0f39 839 } else if (CHROMA422(h))
4cfbeef3 840 *fmt++ = AV_PIX_FMT_YUV422P10;
f51d0f39 841 else
4cfbeef3 842 *fmt++ = AV_PIX_FMT_YUV420P10;
f51d0f39
LB
843 break;
844 case 8:
ebd5320a
RDC
845#if CONFIG_H264_VDPAU_HWACCEL
846 *fmt++ = AV_PIX_FMT_VDPAU;
847#endif
f51d0f39 848 if (CHROMA444(h)) {
ce423163 849 if (h->avctx->colorspace == AVCOL_SPC_RGB)
4cfbeef3 850 *fmt++ = AV_PIX_FMT_GBRP;
ce423163 851 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
4cfbeef3 852 *fmt++ = AV_PIX_FMT_YUVJ444P;
ce423163 853 else
4cfbeef3 854 *fmt++ = AV_PIX_FMT_YUV444P;
f51d0f39 855 } else if (CHROMA422(h)) {
ce423163 856 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
4cfbeef3 857 *fmt++ = AV_PIX_FMT_YUVJ422P;
ce423163 858 else
4cfbeef3 859 *fmt++ = AV_PIX_FMT_YUV422P;
f51d0f39 860 } else {
4cfbeef3
RDC
861#if CONFIG_H264_DXVA2_HWACCEL
862 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
863#endif
d8039ef8
SL
864#if CONFIG_H264_D3D11VA_HWACCEL
865 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
866#endif
4cfbeef3
RDC
867#if CONFIG_H264_VAAPI_HWACCEL
868 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
869#endif
870#if CONFIG_H264_VDA_HWACCEL
871 *fmt++ = AV_PIX_FMT_VDA_VLD;
872 *fmt++ = AV_PIX_FMT_VDA;
873#endif
ce423163
RDC
874 if (h->avctx->codec->pix_fmts)
875 choices = h->avctx->codec->pix_fmts;
876 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
4cfbeef3 877 *fmt++ = AV_PIX_FMT_YUVJ420P;
ce423163 878 else
4cfbeef3 879 *fmt++ = AV_PIX_FMT_YUV420P;
f51d0f39
LB
880 }
881 break;
882 default:
883 av_log(h->avctx, AV_LOG_ERROR,
884 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
885 return AVERROR_INVALIDDATA;
886 }
ce423163 887
4cfbeef3
RDC
888 *fmt = AV_PIX_FMT_NONE;
889
ce423163 890 return ff_get_format(h->avctx, choices);
f51d0f39
LB
891}
892
893/* export coded and cropped frame dimensions to AVCodecContext */
894static int init_dimensions(H264Context *h)
895{
896 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
897 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
e87f5e4e
VG
898 int crop_present = h->sps.crop_left || h->sps.crop_top ||
899 h->sps.crop_right || h->sps.crop_bottom;
f51d0f39
LB
900
901 /* handle container cropping */
e87f5e4e 902 if (!crop_present &&
f51d0f39
LB
903 FFALIGN(h->avctx->width, 16) == h->width &&
904 FFALIGN(h->avctx->height, 16) == h->height) {
905 width = h->avctx->width;
906 height = h->avctx->height;
907 }
908
909 if (width <= 0 || height <= 0) {
910 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
911 width, height);
912 if (h->avctx->err_recognition & AV_EF_EXPLODE)
913 return AVERROR_INVALIDDATA;
914
915 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
2db953f8
LB
916 h->sps.crop_bottom =
917 h->sps.crop_top =
918 h->sps.crop_right =
919 h->sps.crop_left =
f51d0f39
LB
920 h->sps.crop = 0;
921
922 width = h->width;
923 height = h->height;
924 }
925
926 h->avctx->coded_width = h->width;
927 h->avctx->coded_height = h->height;
928 h->avctx->width = width;
929 h->avctx->height = height;
930
931 return 0;
932}
933
8a66fd40 934static int h264_slice_header_init(H264Context *h)
f51d0f39
LB
935{
936 int nb_slices = (HAVE_THREADS &&
937 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
938 h->avctx->thread_count : 1;
939 int i, ret;
940
9e500efd 941 ff_set_sar(h->avctx, h->sps.sar);
f51d0f39
LB
942 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
943 &h->chroma_x_shift, &h->chroma_y_shift);
944
945 if (h->sps.timing_info_present_flag) {
946 int64_t den = h->sps.time_scale;
947 if (h->x264_build < 44U)
948 den *= 2;
7ea1b347 949 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
f51d0f39
LB
950 h->sps.num_units_in_tick, den, 1 << 30);
951 }
952
8a66fd40
AK
953 ff_h264_free_tables(h);
954
f51d0f39
LB
955 h->first_field = 0;
956 h->prev_interlaced_frame = 1;
957
958 init_scan_tables(h);
959 ret = ff_h264_alloc_tables(h);
960 if (ret < 0) {
961 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
962 return ret;
963 }
964
ee62b364
AK
965 if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 10) {
966 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
967 h->sps.bit_depth_luma);
968 return AVERROR_INVALIDDATA;
969 }
970
971 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
972 h->pixel_shift = h->sps.bit_depth_luma > 8;
c9ccbc73
AK
973 h->chroma_format_idc = h->sps.chroma_format_idc;
974 h->bit_depth_luma = h->sps.bit_depth_luma;
ee62b364
AK
975
976 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
977 h->sps.chroma_format_idc);
978 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
979 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
980 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
981 h->sps.chroma_format_idc);
982 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
983
f51d0f39
LB
984 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
985 int max_slices;
986 if (h->mb_height)
987 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
988 else
989 max_slices = H264_MAX_THREADS;
990 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
991 " reducing to %d\n", nb_slices, max_slices);
992 nb_slices = max_slices;
993 }
994 h->slice_context_count = nb_slices;
995
996 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
582683b6 997 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
f51d0f39
LB
998 if (ret < 0) {
999 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1000 return ret;
1001 }
1002 } else {
5bf3c0fa
AK
1003 for (i = 0; i < h->slice_context_count; i++) {
1004 H264SliceContext *sl = &h->slice_ctx[i];
f51d0f39 1005
5bf3c0fa
AK
1006 sl->h264 = h;
1007 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1008 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1009 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1010
1011 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
f51d0f39
LB
1012 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1013 return ret;
1014 }
5bf3c0fa 1015 }
f51d0f39
LB
1016 }
1017
1018 h->context_initialized = 1;
1019
1020 return 0;
1021}
1022
1023/**
1024 * Decode a slice header.
1025 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1026 *
1027 * @param h h264context
f51d0f39
LB
1028 *
1029 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1030 */
51d8725a 1031int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
f51d0f39
LB
1032{
1033 unsigned int first_mb_in_slice;
1034 unsigned int pps_id;
1035 int ret;
1036 unsigned int slice_type, tmp, i, j;
1037 int default_ref_list_done = 0;
1038 int last_pic_structure, last_pic_droppable;
1039 int needs_reinit = 0;
1040 int field_pic_flag, bottom_field_flag;
1041
1042 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1043 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1044
f42485db 1045 first_mb_in_slice = get_ue_golomb(&sl->gb);
f51d0f39
LB
1046
1047 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
51d8725a 1048 if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
95eb35f3 1049 ff_h264_field_end(h, sl, 1);
f51d0f39
LB
1050 }
1051
51d8725a
AK
1052 h->current_slice = 0;
1053 if (!h->first_field) {
f51d0f39
LB
1054 if (h->cur_pic_ptr && !h->droppable) {
1055 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1056 h->picture_structure == PICT_BOTTOM_FIELD);
1057 }
1058 h->cur_pic_ptr = NULL;
1059 }
1060 }
1061
f42485db 1062 slice_type = get_ue_golomb_31(&sl->gb);
f51d0f39
LB
1063 if (slice_type > 9) {
1064 av_log(h->avctx, AV_LOG_ERROR,
d4d9068c
AK
1065 "slice type %d too large at %d\n",
1066 slice_type, first_mb_in_slice);
f51d0f39
LB
1067 return AVERROR_INVALIDDATA;
1068 }
1069 if (slice_type > 4) {
1070 slice_type -= 5;
56febc99 1071 sl->slice_type_fixed = 1;
f51d0f39 1072 } else
56febc99 1073 sl->slice_type_fixed = 0;
f51d0f39
LB
1074
1075 slice_type = golomb_to_pict_type[slice_type];
1076 if (slice_type == AV_PICTURE_TYPE_I ||
51d8725a 1077 (h->current_slice != 0 && slice_type == h->last_slice_type)) {
f51d0f39
LB
1078 default_ref_list_done = 1;
1079 }
56febc99
AK
1080 sl->slice_type = slice_type;
1081 sl->slice_type_nos = slice_type & 3;
f51d0f39
LB
1082
1083 if (h->nal_unit_type == NAL_IDR_SLICE &&
56febc99 1084 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
f51d0f39
LB
1085 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1086 return AVERROR_INVALIDDATA;
1087 }
1088
1089 // to make a few old functions happy, it's wrong though
56febc99 1090 h->pict_type = sl->slice_type;
f51d0f39 1091
f42485db 1092 pps_id = get_ue_golomb(&sl->gb);
f51d0f39
LB
1093 if (pps_id >= MAX_PPS_COUNT) {
1094 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1095 return AVERROR_INVALIDDATA;
1096 }
51d8725a 1097 if (!h->pps_buffers[pps_id]) {
f51d0f39
LB
1098 av_log(h->avctx, AV_LOG_ERROR,
1099 "non-existing PPS %u referenced\n",
1100 pps_id);
1101 return AVERROR_INVALIDDATA;
1102 }
51d8725a 1103 h->pps = *h->pps_buffers[pps_id];
f51d0f39 1104
51d8725a 1105 if (!h->sps_buffers[h->pps.sps_id]) {
f51d0f39
LB
1106 av_log(h->avctx, AV_LOG_ERROR,
1107 "non-existing SPS %u referenced\n",
1108 h->pps.sps_id);
1109 return AVERROR_INVALIDDATA;
1110 }
1111
1112 if (h->pps.sps_id != h->sps.sps_id ||
51d8725a
AK
1113 h->sps_buffers[h->pps.sps_id]->new) {
1114 h->sps_buffers[h->pps.sps_id]->new = 0;
f51d0f39 1115
51d8725a 1116 h->sps = *h->sps_buffers[h->pps.sps_id];
f51d0f39
LB
1117
1118 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
c9ccbc73 1119 h->chroma_format_idc != h->sps.chroma_format_idc)
f51d0f39 1120 needs_reinit = 1;
ee62b364
AK
1121
1122 if (h->flags & CODEC_FLAG_LOW_DELAY ||
1123 (h->sps.bitstream_restriction_flag &&
1124 !h->sps.num_reorder_frames)) {
1125 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1126 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1127 "Reenabling low delay requires a codec flush.\n");
1128 else
1129 h->low_delay = 1;
1130 }
1131
1132 if (h->avctx->has_b_frames < 2)
1133 h->avctx->has_b_frames = !h->low_delay;
1134
f51d0f39
LB
1135 }
1136
1137 h->avctx->profile = ff_h264_get_profile(&h->sps);
1138 h->avctx->level = h->sps.level_idc;
1139 h->avctx->refs = h->sps.ref_frame_count;
1140
1141 if (h->mb_width != h->sps.mb_width ||
1142 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1143 needs_reinit = 1;
1144
1145 h->mb_width = h->sps.mb_width;
1146 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1147 h->mb_num = h->mb_width * h->mb_height;
1148 h->mb_stride = h->mb_width + 1;
1149
1150 h->b_stride = h->mb_width * 4;
1151
1152 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1153
1154 h->width = 16 * h->mb_width;
1155 h->height = 16 * h->mb_height;
1156
1157 ret = init_dimensions(h);
1158 if (ret < 0)
1159 return ret;
1160
1161 if (h->sps.video_signal_type_present_flag) {
1162 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1163 : AVCOL_RANGE_MPEG;
1164 if (h->sps.colour_description_present_flag) {
1165 if (h->avctx->colorspace != h->sps.colorspace)
1166 needs_reinit = 1;
1167 h->avctx->color_primaries = h->sps.color_primaries;
1168 h->avctx->color_trc = h->sps.color_trc;
1169 h->avctx->colorspace = h->sps.colorspace;
1170 }
1171 }
1172
34c5a666 1173 if (h->context_initialized && needs_reinit) {
51d8725a 1174 if (sl != h->slice_ctx) {
f51d0f39
LB
1175 av_log(h->avctx, AV_LOG_ERROR,
1176 "changing width %d -> %d / height %d -> %d on "
1177 "slice %d\n",
1178 h->width, h->avctx->coded_width,
1179 h->height, h->avctx->coded_height,
51d8725a 1180 h->current_slice + 1);
f51d0f39
LB
1181 return AVERROR_INVALIDDATA;
1182 }
1183
1184 ff_h264_flush_change(h);
1185
1186 if ((ret = get_pixel_format(h)) < 0)
1187 return ret;
1188 h->avctx->pix_fmt = ret;
1189
1190 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1191 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1192
8a66fd40 1193 if ((ret = h264_slice_header_init(h)) < 0) {
f51d0f39
LB
1194 av_log(h->avctx, AV_LOG_ERROR,
1195 "h264_slice_header_init() failed\n");
1196 return ret;
1197 }
1198 }
1199 if (!h->context_initialized) {
51d8725a 1200 if (sl != h->slice_ctx) {
f51d0f39
LB
1201 av_log(h->avctx, AV_LOG_ERROR,
1202 "Cannot (re-)initialize context during parallel decoding.\n");
1203 return AVERROR_PATCHWELCOME;
1204 }
1205
1206 if ((ret = get_pixel_format(h)) < 0)
1207 return ret;
1208 h->avctx->pix_fmt = ret;
1209
8a66fd40 1210 if ((ret = h264_slice_header_init(h)) < 0) {
f51d0f39
LB
1211 av_log(h->avctx, AV_LOG_ERROR,
1212 "h264_slice_header_init() failed\n");
1213 return ret;
1214 }
1215 }
1216
51d8725a 1217 if (sl == h->slice_ctx && h->dequant_coeff_pps != pps_id) {
f51d0f39
LB
1218 h->dequant_coeff_pps = pps_id;
1219 h264_init_dequant_tables(h);
1220 }
1221
f42485db 1222 h->frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
f51d0f39 1223
e9b2383b 1224 sl->mb_mbaff = 0;
f51d0f39 1225 h->mb_aff_frame = 0;
51d8725a
AK
1226 last_pic_structure = h->picture_structure;
1227 last_pic_droppable = h->droppable;
f51d0f39
LB
1228 h->droppable = h->nal_ref_idc == 0;
1229 if (h->sps.frame_mbs_only_flag) {
1230 h->picture_structure = PICT_FRAME;
1231 } else {
f42485db 1232 field_pic_flag = get_bits1(&sl->gb);
f51d0f39 1233 if (field_pic_flag) {
f42485db 1234 bottom_field_flag = get_bits1(&sl->gb);
f51d0f39
LB
1235 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1236 } else {
1237 h->picture_structure = PICT_FRAME;
1238 h->mb_aff_frame = h->sps.mb_aff;
1239 }
1240 }
bc98e8c0 1241 sl->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
f51d0f39 1242
51d8725a 1243 if (h->current_slice != 0) {
f51d0f39
LB
1244 if (last_pic_structure != h->picture_structure ||
1245 last_pic_droppable != h->droppable) {
1246 av_log(h->avctx, AV_LOG_ERROR,
1247 "Changing field mode (%d -> %d) between slices is not allowed\n",
1248 last_pic_structure, h->picture_structure);
1249 h->picture_structure = last_pic_structure;
1250 h->droppable = last_pic_droppable;
1251 return AVERROR_INVALIDDATA;
51d8725a 1252 } else if (!h->cur_pic_ptr) {
f51d0f39
LB
1253 av_log(h->avctx, AV_LOG_ERROR,
1254 "unset cur_pic_ptr on slice %d\n",
51d8725a 1255 h->current_slice + 1);
f51d0f39
LB
1256 return AVERROR_INVALIDDATA;
1257 }
1258 } else {
1259 /* Shorten frame num gaps so we don't have to allocate reference
1260 * frames just to throw them away */
1261 if (h->frame_num != h->prev_frame_num) {
1262 int unwrap_prev_frame_num = h->prev_frame_num;
1263 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1264
1265 if (unwrap_prev_frame_num > h->frame_num)
1266 unwrap_prev_frame_num -= max_frame_num;
1267
1268 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1269 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1270 if (unwrap_prev_frame_num < 0)
1271 unwrap_prev_frame_num += max_frame_num;
1272
1273 h->prev_frame_num = unwrap_prev_frame_num;
1274 }
1275 }
1276
1277 /* See if we have a decoded first field looking for a pair...
1278 * Here, we're using that to see if we should mark previously
1279 * decode frames as "finished".
1280 * We have to do that before the "dummy" in-between frame allocation,
1281 * since that can modify s->current_picture_ptr. */
51d8725a
AK
1282 if (h->first_field) {
1283 assert(h->cur_pic_ptr);
a0f29460 1284 assert(h->cur_pic_ptr->f->buf[0]);
51d8725a 1285 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
f51d0f39
LB
1286
1287 /* figure out if we have a complementary field pair */
1288 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1289 /* Previous field is unmatched. Don't display it, but let it
1290 * remain for reference if marked as such. */
1291 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
51d8725a 1292 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
f51d0f39
LB
1293 last_pic_structure == PICT_TOP_FIELD);
1294 }
1295 } else {
51d8725a 1296 if (h->cur_pic_ptr->frame_num != h->frame_num) {
f51d0f39
LB
1297 /* This and previous field were reference, but had
1298 * different frame_nums. Consider this field first in
1299 * pair. Throw away previous field except for reference
1300 * purposes. */
1301 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
51d8725a 1302 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
f51d0f39
LB
1303 last_pic_structure == PICT_TOP_FIELD);
1304 }
1305 } else {
1306 /* Second field in complementary pair */
1307 if (!((last_pic_structure == PICT_TOP_FIELD &&
1308 h->picture_structure == PICT_BOTTOM_FIELD) ||
1309 (last_pic_structure == PICT_BOTTOM_FIELD &&
1310 h->picture_structure == PICT_TOP_FIELD))) {
1311 av_log(h->avctx, AV_LOG_ERROR,
1312 "Invalid field mode combination %d/%d\n",
1313 last_pic_structure, h->picture_structure);
1314 h->picture_structure = last_pic_structure;
1315 h->droppable = last_pic_droppable;
1316 return AVERROR_INVALIDDATA;
1317 } else if (last_pic_droppable != h->droppable) {
1318 avpriv_request_sample(h->avctx,
1319 "Found reference and non-reference fields in the same frame, which");
1320 h->picture_structure = last_pic_structure;
1321 h->droppable = last_pic_droppable;
1322 return AVERROR_PATCHWELCOME;
1323 }
1324 }
1325 }
1326 }
1327
1328 while (h->frame_num != h->prev_frame_num &&
1329 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1330 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1331 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1332 h->frame_num, h->prev_frame_num);
1333 ret = h264_frame_start(h);
1334 if (ret < 0) {
51d8725a 1335 h->first_field = 0;
f51d0f39
LB
1336 return ret;
1337 }
1338
1339 h->prev_frame_num++;
1340 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1341 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1342 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1343 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1344 ret = ff_generate_sliding_window_mmcos(h, 1);
1345 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1346 return ret;
1347 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1348 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1349 return ret;
1350 /* Error concealment: If a ref is missing, copy the previous ref
1351 * in its place.
1352 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1353 * many assumptions about there being no actual duplicates.
1354 * FIXME: This does not copy padding for out-of-frame motion
1355 * vectors. Given we are concealing a lost frame, this probably
1356 * is not noticeable by comparison, but it should be fixed. */
1357 if (h->short_ref_count) {
1358 if (prev) {
a0f29460
AK
1359 av_image_copy(h->short_ref[0]->f->data,
1360 h->short_ref[0]->f->linesize,
1361 (const uint8_t **)prev->f->data,
1362 prev->f->linesize,
f51d0f39
LB
1363 h->avctx->pix_fmt,
1364 h->mb_width * 16,
1365 h->mb_height * 16);
1366 h->short_ref[0]->poc = prev->poc + 2;
1367 }
1368 h->short_ref[0]->frame_num = h->prev_frame_num;
1369 }
1370 }
1371
1372 /* See if we have a decoded first field looking for a pair...
1373 * We're using that to see whether to continue decoding in that
1374 * frame, or to allocate a new one. */
51d8725a
AK
1375 if (h->first_field) {
1376 assert(h->cur_pic_ptr);
a0f29460 1377 assert(h->cur_pic_ptr->f->buf[0]);
51d8725a 1378 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
f51d0f39
LB
1379
1380 /* figure out if we have a complementary field pair */
1381 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1382 /* Previous field is unmatched. Don't display it, but let it
1383 * remain for reference if marked as such. */
51d8725a
AK
1384 h->cur_pic_ptr = NULL;
1385 h->first_field = FIELD_PICTURE(h);
f51d0f39 1386 } else {
51d8725a 1387 if (h->cur_pic_ptr->frame_num != h->frame_num) {
f51d0f39
LB
1388 /* This and the previous field had different frame_nums.
1389 * Consider this field first in pair. Throw away previous
1390 * one except for reference purposes. */
51d8725a
AK
1391 h->first_field = 1;
1392 h->cur_pic_ptr = NULL;
f51d0f39
LB
1393 } else {
1394 /* Second field in complementary pair */
51d8725a 1395 h->first_field = 0;
f51d0f39
LB
1396 }
1397 }
1398 } else {
1399 /* Frame or first field in a potentially complementary pair */
51d8725a 1400 h->first_field = FIELD_PICTURE(h);
f51d0f39
LB
1401 }
1402
51d8725a 1403 if (!FIELD_PICTURE(h) || h->first_field) {
f51d0f39 1404 if (h264_frame_start(h) < 0) {
51d8725a 1405 h->first_field = 0;
f51d0f39
LB
1406 return AVERROR_INVALIDDATA;
1407 }
1408 } else {
1409 release_unused_pictures(h, 0);
1410 }
1411 }
f51d0f39
LB
1412
1413 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1414
1415 assert(h->mb_num == h->mb_width * h->mb_height);
1416 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1417 first_mb_in_slice >= h->mb_num) {
1418 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1419 return AVERROR_INVALIDDATA;
1420 }
7bbc254d
AK
1421 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1422 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1423 FIELD_OR_MBAFF_PICTURE(h);
f51d0f39 1424 if (h->picture_structure == PICT_BOTTOM_FIELD)
7bbc254d 1425 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
d4d9068c 1426 assert(sl->mb_y < h->mb_height);
f51d0f39
LB
1427
1428 if (h->picture_structure == PICT_FRAME) {
1429 h->curr_pic_num = h->frame_num;
1430 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1431 } else {
1432 h->curr_pic_num = 2 * h->frame_num + 1;
1433 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1434 }
1435
1436 if (h->nal_unit_type == NAL_IDR_SLICE)
f42485db 1437 get_ue_golomb(&sl->gb); /* idr_pic_id */
f51d0f39
LB
1438
1439 if (h->sps.poc_type == 0) {
f42485db 1440 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
f51d0f39
LB
1441
1442 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
f42485db 1443 h->delta_poc_bottom = get_se_golomb(&sl->gb);
f51d0f39
LB
1444 }
1445
1446 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
f42485db 1447 h->delta_poc[0] = get_se_golomb(&sl->gb);
f51d0f39
LB
1448
1449 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
f42485db 1450 h->delta_poc[1] = get_se_golomb(&sl->gb);
f51d0f39
LB
1451 }
1452
1453 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1454
1455 if (h->pps.redundant_pic_cnt_present)
f42485db 1456 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
f51d0f39 1457
56febc99 1458 ret = ff_set_ref_count(h, sl);
f51d0f39
LB
1459 if (ret < 0)
1460 return ret;
1461 else if (ret == 1)
1462 default_ref_list_done = 0;
1463
1464 if (!default_ref_list_done)
56febc99 1465 ff_h264_fill_default_ref_list(h, sl);
f51d0f39 1466
56febc99 1467 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
95eb35f3 1468 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
f51d0f39 1469 if (ret < 0) {
95eb35f3 1470 sl->ref_count[1] = sl->ref_count[0] = 0;
f51d0f39
LB
1471 return ret;
1472 }
1473 }
1474
56febc99 1475 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
f51d0f39 1476 (h->pps.weighted_bipred_idc == 1 &&
56febc99 1477 sl->slice_type_nos == AV_PICTURE_TYPE_B))
92c6c2a6 1478 ff_pred_weight_table(h, sl);
f51d0f39 1479 else if (h->pps.weighted_bipred_idc == 2 &&
56febc99 1480 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
92c6c2a6 1481 implicit_weight_table(h, sl, -1);
f51d0f39 1482 } else {
92c6c2a6 1483 sl->use_weight = 0;
f51d0f39 1484 for (i = 0; i < 2; i++) {
92c6c2a6
AK
1485 sl->luma_weight_flag[i] = 0;
1486 sl->chroma_weight_flag[i] = 0;
f51d0f39
LB
1487 }
1488 }
1489
1490 // If frame-mt is enabled, only update mmco tables for the first slice
1491 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1492 // or h->mmco, which will cause ref list mix-ups and decoding errors
1493 // further down the line. This may break decoding if the first slice is
1494 // corrupt, thus we only do this if frame-mt is enabled.
1495 if (h->nal_ref_idc) {
51d8725a 1496 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
f51d0f39 1497 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
51d8725a 1498 h->current_slice == 0);
f51d0f39
LB
1499 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1500 return AVERROR_INVALIDDATA;
1501 }
1502
1503 if (FRAME_MBAFF(h)) {
92c6c2a6 1504 ff_h264_fill_mbaff_ref_list(h, sl);
f51d0f39 1505
56febc99 1506 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
92c6c2a6
AK
1507 implicit_weight_table(h, sl, 0);
1508 implicit_weight_table(h, sl, 1);
f51d0f39
LB
1509 }
1510 }
1511
066aafce 1512 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
ee0d774d 1513 ff_h264_direct_dist_scale_factor(h, sl);
56febc99 1514 ff_h264_direct_ref_list_init(h, sl);
f51d0f39 1515
56febc99 1516 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
f42485db 1517 tmp = get_ue_golomb_31(&sl->gb);
f51d0f39
LB
1518 if (tmp > 2) {
1519 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1520 return AVERROR_INVALIDDATA;
1521 }
a9b201ca 1522 sl->cabac_init_idc = tmp;
f51d0f39
LB
1523 }
1524
5c8280c3 1525 sl->last_qscale_diff = 0;
f42485db 1526 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
f51d0f39
LB
1527 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1528 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1529 return AVERROR_INVALIDDATA;
1530 }
d231e84b
AK
1531 sl->qscale = tmp;
1532 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1533 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
f51d0f39 1534 // FIXME qscale / qp ... stuff
56febc99 1535 if (sl->slice_type == AV_PICTURE_TYPE_SP)
f42485db 1536 get_bits1(&sl->gb); /* sp_for_switch_flag */
56febc99
AK
1537 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1538 sl->slice_type == AV_PICTURE_TYPE_SI)
f42485db 1539 get_se_golomb(&sl->gb); /* slice_qs_delta */
f51d0f39 1540
e6c90ce9
AK
1541 sl->deblocking_filter = 1;
1542 sl->slice_alpha_c0_offset = 0;
1543 sl->slice_beta_offset = 0;
f51d0f39 1544 if (h->pps.deblocking_filter_parameters_present) {
f42485db 1545 tmp = get_ue_golomb_31(&sl->gb);
f51d0f39
LB
1546 if (tmp > 2) {
1547 av_log(h->avctx, AV_LOG_ERROR,
1548 "deblocking_filter_idc %u out of range\n", tmp);
1549 return AVERROR_INVALIDDATA;
1550 }
e6c90ce9
AK
1551 sl->deblocking_filter = tmp;
1552 if (sl->deblocking_filter < 2)
1553 sl->deblocking_filter ^= 1; // 1<->0
1554
1555 if (sl->deblocking_filter) {
f42485db
AK
1556 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1557 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
e6c90ce9
AK
1558 if (sl->slice_alpha_c0_offset > 12 ||
1559 sl->slice_alpha_c0_offset < -12 ||
1560 sl->slice_beta_offset > 12 ||
1561 sl->slice_beta_offset < -12) {
f51d0f39
LB
1562 av_log(h->avctx, AV_LOG_ERROR,
1563 "deblocking filter parameters %d %d out of range\n",
e6c90ce9 1564 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
f51d0f39
LB
1565 return AVERROR_INVALIDDATA;
1566 }
1567 }
1568 }
1569
1570 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1571 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
56febc99 1572 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
f51d0f39 1573 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
56febc99 1574 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
f51d0f39
LB
1575 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1576 h->nal_ref_idc == 0))
e6c90ce9 1577 sl->deblocking_filter = 0;
f51d0f39 1578
51d8725a 1579 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
f51d0f39
LB
1580 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1581 /* Cheat slightly for speed:
1582 * Do not bother to deblock across slices. */
e6c90ce9 1583 sl->deblocking_filter = 2;
f51d0f39 1584 } else {
51d8725a
AK
1585 h->max_contexts = 1;
1586 if (!h->single_decode_warning) {
f51d0f39
LB
1587 av_log(h->avctx, AV_LOG_INFO,
1588 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
51d8725a 1589 h->single_decode_warning = 1;
f51d0f39 1590 }
51d8725a 1591 if (sl != h->slice_ctx) {
f51d0f39
LB
1592 av_log(h->avctx, AV_LOG_ERROR,
1593 "Deblocking switched inside frame.\n");
1594 return 1;
1595 }
1596 }
1597 }
06789ad3 1598 sl->qp_thresh = 15 -
e6c90ce9 1599 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
f51d0f39
LB
1600 FFMAX3(0,
1601 h->pps.chroma_qp_index_offset[0],
1602 h->pps.chroma_qp_index_offset[1]) +
1603 6 * (h->sps.bit_depth_luma - 8);
1604
51d8725a
AK
1605 h->last_slice_type = slice_type;
1606 sl->slice_num = ++h->current_slice;
56febc99 1607 if (sl->slice_num >= MAX_SLICES) {
f51d0f39
LB
1608 av_log(h->avctx, AV_LOG_ERROR,
1609 "Too many slices, increase MAX_SLICES and recompile\n");
1610 }
1611
1612 for (j = 0; j < 2; j++) {
1613 int id_list[16];
95eb35f3 1614 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
f51d0f39
LB
1615 for (i = 0; i < 16; i++) {
1616 id_list[i] = 60;
95eb35f3 1617 if (j < sl->list_count && i < sl->ref_count[j] &&
a0f29460 1618 sl->ref_list[j][i].parent->f->buf[0]) {
f51d0f39 1619 int k;
a0f29460 1620 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
f51d0f39 1621 for (k = 0; k < h->short_ref_count; k++)
a0f29460 1622 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
f51d0f39
LB
1623 id_list[i] = k;
1624 break;
1625 }
1626 for (k = 0; k < h->long_ref_count; k++)
a0f29460 1627 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
f51d0f39
LB
1628 id_list[i] = h->short_ref_count + k;
1629 break;
1630 }
1631 }
1632 }
1633
1634 ref2frm[0] =
1635 ref2frm[1] = -1;
1636 for (i = 0; i < 16; i++)
95eb35f3 1637 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
f51d0f39
LB
1638 ref2frm[18 + 0] =
1639 ref2frm[18 + 1] = -1;
1640 for (i = 16; i < 48; i++)
1641 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
95eb35f3 1642 (sl->ref_list[j][i].reference & 3);
f51d0f39
LB
1643 }
1644
1645 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1646 av_log(h->avctx, AV_LOG_DEBUG,
1647 "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
56febc99 1648 sl->slice_num,
f51d0f39
LB
1649 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1650 first_mb_in_slice,
56febc99
AK
1651 av_get_picture_type_char(sl->slice_type),
1652 sl->slice_type_fixed ? " fix" : "",
f51d0f39
LB
1653 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1654 pps_id, h->frame_num,
1655 h->cur_pic_ptr->field_poc[0],
1656 h->cur_pic_ptr->field_poc[1],
95eb35f3 1657 sl->ref_count[0], sl->ref_count[1],
d231e84b 1658 sl->qscale,
e6c90ce9
AK
1659 sl->deblocking_filter,
1660 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
92c6c2a6
AK
1661 sl->use_weight,
1662 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
066aafce 1663 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
f51d0f39
LB
1664 }
1665
1666 return 0;
1667}
1668
56febc99 1669int ff_h264_get_slice_type(const H264SliceContext *sl)
f51d0f39 1670{
56febc99 1671 switch (sl->slice_type) {
f51d0f39
LB
1672 case AV_PICTURE_TYPE_P:
1673 return 0;
1674 case AV_PICTURE_TYPE_B:
1675 return 1;
1676 case AV_PICTURE_TYPE_I:
1677 return 2;
1678 case AV_PICTURE_TYPE_SP:
1679 return 3;
1680 case AV_PICTURE_TYPE_SI:
1681 return 4;
1682 default:
1683 return AVERROR_INVALIDDATA;
1684 }
1685}
1686
92e0b734 1687static av_always_inline void fill_filter_caches_inter(const H264Context *h,
e6287f07 1688 H264SliceContext *sl,
f51d0f39
LB
1689 int mb_type, int top_xy,
1690 int left_xy[LEFT_MBS],
1691 int top_type,
1692 int left_type[LEFT_MBS],
1693 int mb_xy, int list)
1694{
1695 int b_stride = h->b_stride;
e6287f07
AK
1696 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1697 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
f51d0f39
LB
1698 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1699 if (USES_LIST(top_type, list)) {
1700 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1701 const int b8_xy = 4 * top_xy + 2;
e9b2383b 1702 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
f51d0f39
LB
1703 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1704 ref_cache[0 - 1 * 8] =
1705 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1706 ref_cache[2 - 1 * 8] =
1707 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1708 } else {
1709 AV_ZERO128(mv_dst - 1 * 8);
1710 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1711 }
1712
1713 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1714 if (USES_LIST(left_type[LTOP], list)) {
1715 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1716 const int b8_xy = 4 * left_xy[LTOP] + 1;
e9b2383b 1717 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
f51d0f39
LB
1718 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1719 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1720 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1721 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1722 ref_cache[-1 + 0] =
1723 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1724 ref_cache[-1 + 16] =
1725 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1726 } else {
1727 AV_ZERO32(mv_dst - 1 + 0);
1728 AV_ZERO32(mv_dst - 1 + 8);
1729 AV_ZERO32(mv_dst - 1 + 16);
1730 AV_ZERO32(mv_dst - 1 + 24);
1731 ref_cache[-1 + 0] =
1732 ref_cache[-1 + 8] =
1733 ref_cache[-1 + 16] =
1734 ref_cache[-1 + 24] = LIST_NOT_USED;
1735 }
1736 }
1737 }
1738
1739 if (!USES_LIST(mb_type, list)) {
1740 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1741 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1742 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1743 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1744 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1745 return;
1746 }
1747
1748 {
1749 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
e9b2383b 1750 int (*ref2frm)[64] = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
f51d0f39
LB
1751 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1752 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1753 AV_WN32A(&ref_cache[0 * 8], ref01);
1754 AV_WN32A(&ref_cache[1 * 8], ref01);
1755 AV_WN32A(&ref_cache[2 * 8], ref23);
1756 AV_WN32A(&ref_cache[3 * 8], ref23);
1757 }
1758
1759 {
d4d9068c 1760 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
f51d0f39
LB
1761 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1762 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1763 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1764 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1765 }
1766}
1767
1768/**
1769 *
1770 * @return non zero if the loop filter can be skipped
1771 */
92e0b734 1772static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
f51d0f39 1773{
0edbe6fa 1774 const int mb_xy = sl->mb_xy;
f51d0f39
LB
1775 int top_xy, left_xy[LEFT_MBS];
1776 int top_type, left_type[LEFT_MBS];
1777 uint8_t *nnz;
1778 uint8_t *nnz_cache;
1779
bc98e8c0 1780 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
f51d0f39
LB
1781
1782 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1783 * stuff, I can't imagine that these complex rules are worth it. */
1784
1785 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1786 if (FRAME_MBAFF(h)) {
1787 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1788 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
d4d9068c 1789 if (sl->mb_y & 1) {
f51d0f39
LB
1790 if (left_mb_field_flag != curr_mb_field_flag)
1791 left_xy[LTOP] -= h->mb_stride;
1792 } else {
1793 if (curr_mb_field_flag)
1794 top_xy += h->mb_stride &
1795 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1796 if (left_mb_field_flag != curr_mb_field_flag)
1797 left_xy[LBOT] += h->mb_stride;
1798 }
1799 }
1800
8b00f4df
AK
1801 sl->top_mb_xy = top_xy;
1802 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1803 sl->left_mb_xy[LBOT] = left_xy[LBOT];
f51d0f39
LB
1804 {
1805 /* For sufficiently low qp, filtering wouldn't do anything.
1806 * This is a conservative estimate: could also check beta_offset
1807 * and more accurate chroma_qp. */
06789ad3 1808 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
f51d0f39
LB
1809 int qp = h->cur_pic.qscale_table[mb_xy];
1810 if (qp <= qp_thresh &&
1811 (left_xy[LTOP] < 0 ||
1812 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1813 (top_xy < 0 ||
1814 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1815 if (!FRAME_MBAFF(h))
1816 return 1;
1817 if ((left_xy[LTOP] < 0 ||
1818 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1819 (top_xy < h->mb_stride ||
1820 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1821 return 1;
1822 }
1823 }
1824
1825 top_type = h->cur_pic.mb_type[top_xy];
1826 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1827 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
e6c90ce9 1828 if (sl->deblocking_filter == 2) {
56febc99 1829 if (h->slice_table[top_xy] != sl->slice_num)
f51d0f39 1830 top_type = 0;
56febc99 1831 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
f51d0f39
LB
1832 left_type[LTOP] = left_type[LBOT] = 0;
1833 } else {
1834 if (h->slice_table[top_xy] == 0xFFFF)
1835 top_type = 0;
1836 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1837 left_type[LTOP] = left_type[LBOT] = 0;
1838 }
8b00f4df
AK
1839 sl->top_type = top_type;
1840 sl->left_type[LTOP] = left_type[LTOP];
1841 sl->left_type[LBOT] = left_type[LBOT];
f51d0f39
LB
1842
1843 if (IS_INTRA(mb_type))
1844 return 0;
1845
e6287f07 1846 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
f51d0f39 1847 top_type, left_type, mb_xy, 0);
95eb35f3 1848 if (sl->list_count == 2)
e6287f07 1849 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
f51d0f39
LB
1850 top_type, left_type, mb_xy, 1);
1851
1852 nnz = h->non_zero_count[mb_xy];
f69574cf 1853 nnz_cache = sl->non_zero_count_cache;
f51d0f39
LB
1854 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1855 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1856 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1857 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
e7226984 1858 sl->cbp = h->cbp_table[mb_xy];
f51d0f39
LB
1859
1860 if (top_type) {
1861 nnz = h->non_zero_count[top_xy];
1862 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1863 }
1864
1865 if (left_type[LTOP]) {
1866 nnz = h->non_zero_count[left_xy[LTOP]];
1867 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
1868 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
1869 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
1870 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
1871 }
1872
1873 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
1874 * from what the loop filter needs */
1875 if (!CABAC(h) && h->pps.transform_8x8_mode) {
1876 if (IS_8x8DCT(top_type)) {
1877 nnz_cache[4 + 8 * 0] =
1878 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
1879 nnz_cache[6 + 8 * 0] =
1880 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
1881 }
1882 if (IS_8x8DCT(left_type[LTOP])) {
1883 nnz_cache[3 + 8 * 1] =
1884 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
1885 }
1886 if (IS_8x8DCT(left_type[LBOT])) {
1887 nnz_cache[3 + 8 * 3] =
1888 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
1889 }
1890
1891 if (IS_8x8DCT(mb_type)) {
1892 nnz_cache[scan8[0]] =
1893 nnz_cache[scan8[1]] =
1894 nnz_cache[scan8[2]] =
e7226984 1895 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
f51d0f39
LB
1896
1897 nnz_cache[scan8[0 + 4]] =
1898 nnz_cache[scan8[1 + 4]] =
1899 nnz_cache[scan8[2 + 4]] =
e7226984 1900 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
f51d0f39
LB
1901
1902 nnz_cache[scan8[0 + 8]] =
1903 nnz_cache[scan8[1 + 8]] =
1904 nnz_cache[scan8[2 + 8]] =
e7226984 1905 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
f51d0f39
LB
1906
1907 nnz_cache[scan8[0 + 12]] =
1908 nnz_cache[scan8[1 + 12]] =
1909 nnz_cache[scan8[2 + 12]] =
e7226984 1910 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
f51d0f39
LB
1911 }
1912 }
1913
1914 return 0;
1915}
1916
92e0b734 1917static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
f51d0f39
LB
1918{
1919 uint8_t *dest_y, *dest_cb, *dest_cr;
1920 int linesize, uvlinesize, mb_x, mb_y;
d4d9068c 1921 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
56febc99 1922 const int old_slice_type = sl->slice_type;
f51d0f39
LB
1923 const int pixel_shift = h->pixel_shift;
1924 const int block_h = 16 >> h->chroma_y_shift;
1925
e6c90ce9 1926 if (sl->deblocking_filter) {
f51d0f39
LB
1927 for (mb_x = start_x; mb_x < end_x; mb_x++)
1928 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
1929 int mb_xy, mb_type;
0edbe6fa 1930 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
56febc99 1931 sl->slice_num = h->slice_table[mb_xy];
f51d0f39 1932 mb_type = h->cur_pic.mb_type[mb_xy];
95eb35f3 1933 sl->list_count = h->list_counts[mb_xy];
f51d0f39
LB
1934
1935 if (FRAME_MBAFF(h))
e9b2383b 1936 sl->mb_mbaff =
bc98e8c0 1937 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
f51d0f39 1938
d4d9068c
AK
1939 sl->mb_x = mb_x;
1940 sl->mb_y = mb_y;
a0f29460 1941 dest_y = h->cur_pic.f->data[0] +
c28ed1d7 1942 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
a0f29460 1943 dest_cb = h->cur_pic.f->data[1] +
f51d0f39 1944 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
c28ed1d7 1945 mb_y * sl->uvlinesize * block_h;
a0f29460 1946 dest_cr = h->cur_pic.f->data[2] +
f51d0f39 1947 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
c28ed1d7 1948 mb_y * sl->uvlinesize * block_h;
f51d0f39
LB
1949 // FIXME simplify above
1950
bc98e8c0 1951 if (MB_FIELD(sl)) {
c28ed1d7
AK
1952 linesize = sl->mb_linesize = sl->linesize * 2;
1953 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
f51d0f39 1954 if (mb_y & 1) { // FIXME move out of this function?
c28ed1d7
AK
1955 dest_y -= sl->linesize * 15;
1956 dest_cb -= sl->uvlinesize * (block_h - 1);
1957 dest_cr -= sl->uvlinesize * (block_h - 1);
f51d0f39
LB
1958 }
1959 } else {
c28ed1d7
AK
1960 linesize = sl->mb_linesize = sl->linesize;
1961 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
f51d0f39 1962 }
d4d9068c 1963 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
f51d0f39 1964 uvlinesize, 0);
06789ad3 1965 if (fill_filter_caches(h, sl, mb_type))
f51d0f39 1966 continue;
d231e84b
AK
1967 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
1968 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
f51d0f39
LB
1969
1970 if (FRAME_MBAFF(h)) {
d231e84b 1971 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
f51d0f39
LB
1972 linesize, uvlinesize);
1973 } else {
d231e84b 1974 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
f51d0f39
LB
1975 dest_cr, linesize, uvlinesize);
1976 }
1977 }
1978 }
56febc99 1979 sl->slice_type = old_slice_type;
d4d9068c
AK
1980 sl->mb_x = end_x;
1981 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
d231e84b
AK
1982 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1983 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
f51d0f39
LB
1984}
1985
92e0b734 1986static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
f51d0f39 1987{
d4d9068c 1988 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
56febc99 1989 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
f51d0f39 1990 h->cur_pic.mb_type[mb_xy - 1] :
56febc99 1991 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
f51d0f39 1992 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
e9b2383b 1993 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
f51d0f39
LB
1994}
1995
1996/**
1997 * Draw edges and report progress for the last MB row.
1998 */
92e0b734 1999static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
f51d0f39 2000{
d4d9068c 2001 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
f51d0f39
LB
2002 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2003 int height = 16 << FRAME_MBAFF(h);
2004 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2005
e6c90ce9 2006 if (sl->deblocking_filter) {
f51d0f39
LB
2007 if ((top + height) >= pic_height)
2008 height += deblock_border;
2009 top -= deblock_border;
2010 }
2011
2012 if (top >= pic_height || (top + height) < 0)
2013 return;
2014
2015 height = FFMIN(height, pic_height - top);
2016 if (top < 0) {
2017 height = top + height;
2018 top = 0;
2019 }
2020
95eb35f3 2021 ff_h264_draw_horiz_band(h, sl, top, height);
f51d0f39
LB
2022
2023 if (h->droppable)
2024 return;
2025
2026 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2027 h->picture_structure == PICT_BOTTOM_FIELD);
2028}
2029
582683b6 2030static void er_add_slice(H264SliceContext *sl,
95eb35f3 2031 int startx, int starty,
f51d0f39
LB
2032 int endx, int endy, int status)
2033{
2034#if CONFIG_ERROR_RESILIENCE
582683b6 2035 ERContext *er = &sl->er;
f51d0f39 2036
a4d34e21
AK
2037 if (!sl->h264->enable_er)
2038 return;
2039
95eb35f3 2040 er->ref_count = sl->ref_count[0];
f51d0f39
LB
2041 ff_er_add_slice(er, startx, starty, endx, endy, status);
2042#endif
2043}
2044
2045static int decode_slice(struct AVCodecContext *avctx, void *arg)
2046{
92c6c2a6 2047 H264SliceContext *sl = arg;
92e0b734 2048 const H264Context *h = sl->h264;
d4d9068c 2049 int lf_x_start = sl->mb_x;
36d04801
AK
2050 int ret;
2051
a0f29460
AK
2052 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2053 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
c28ed1d7
AK
2054
2055 ret = alloc_scratch_buffers(sl, sl->linesize);
36d04801
AK
2056 if (ret < 0)
2057 return ret;
f51d0f39 2058
47a0d393 2059 sl->mb_skip_run = -1;
f51d0f39 2060
07c5ca55
AK
2061 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2062 avctx->codec_id != AV_CODEC_ID_H264 ||
2063 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
f51d0f39
LB
2064
2065 if (h->pps.cabac) {
2066 /* realign */
f42485db 2067 align_get_bits(&sl->gb);
f51d0f39
LB
2068
2069 /* init cabac */
30da98ad 2070 ff_init_cabac_decoder(&sl->cabac,
f42485db
AK
2071 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2072 (get_bits_left(&sl->gb) + 7) / 8);
f51d0f39 2073
d231e84b 2074 ff_h264_init_cabac_states(h, sl);
f51d0f39
LB
2075
2076 for (;;) {
2077 // START_TIMER
d231e84b 2078 int ret = ff_h264_decode_mb_cabac(h, sl);
f51d0f39
LB
2079 int eos;
2080 // STOP_TIMER("decode_mb_cabac")
2081
2082 if (ret >= 0)
92c6c2a6 2083 ff_h264_hl_decode_mb(h, sl);
f51d0f39
LB
2084
2085 // FIXME optimal? or let mb_decode decode 16x32 ?
2086 if (ret >= 0 && FRAME_MBAFF(h)) {
d4d9068c 2087 sl->mb_y++;
f51d0f39 2088
d231e84b 2089 ret = ff_h264_decode_mb_cabac(h, sl);
f51d0f39
LB
2090
2091 if (ret >= 0)
92c6c2a6 2092 ff_h264_hl_decode_mb(h, sl);
d4d9068c 2093 sl->mb_y--;
f51d0f39 2094 }
30da98ad 2095 eos = get_cabac_terminate(&sl->cabac);
f51d0f39
LB
2096
2097 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
30da98ad 2098 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
582683b6 2099 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
d4d9068c
AK
2100 sl->mb_y, ER_MB_END);
2101 if (sl->mb_x >= lf_x_start)
2102 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
f51d0f39
LB
2103 return 0;
2104 }
30da98ad 2105 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
f51d0f39
LB
2106 av_log(h->avctx, AV_LOG_ERROR,
2107 "error while decoding MB %d %d, bytestream %td\n",
d4d9068c 2108 sl->mb_x, sl->mb_y,
30da98ad 2109 sl->cabac.bytestream_end - sl->cabac.bytestream);
582683b6 2110 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
d4d9068c 2111 sl->mb_y, ER_MB_ERROR);
f51d0f39
LB
2112 return AVERROR_INVALIDDATA;
2113 }
2114
d4d9068c
AK
2115 if (++sl->mb_x >= h->mb_width) {
2116 loop_filter(h, sl, lf_x_start, sl->mb_x);
2117 sl->mb_x = lf_x_start = 0;
95eb35f3 2118 decode_finish_row(h, sl);
d4d9068c 2119 ++sl->mb_y;
f51d0f39 2120 if (FIELD_OR_MBAFF_PICTURE(h)) {
d4d9068c
AK
2121 ++sl->mb_y;
2122 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
56febc99 2123 predict_field_decoding_flag(h, sl);
f51d0f39
LB
2124 }
2125 }
2126
d4d9068c 2127 if (eos || sl->mb_y >= h->mb_height) {
6a85dfc8 2128 ff_tlog(h->avctx, "slice end %d %d\n",
f42485db 2129 get_bits_count(&sl->gb), sl->gb.size_in_bits);
582683b6 2130 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
d4d9068c
AK
2131 sl->mb_y, ER_MB_END);
2132 if (sl->mb_x > lf_x_start)
2133 loop_filter(h, sl, lf_x_start, sl->mb_x);
f51d0f39
LB
2134 return 0;
2135 }
2136 }
2137 } else {
2138 for (;;) {
d231e84b 2139 int ret = ff_h264_decode_mb_cavlc(h, sl);
f51d0f39
LB
2140
2141 if (ret >= 0)
92c6c2a6 2142 ff_h264_hl_decode_mb(h, sl);
f51d0f39
LB
2143
2144 // FIXME optimal? or let mb_decode decode 16x32 ?
2145 if (ret >= 0 && FRAME_MBAFF(h)) {
d4d9068c 2146 sl->mb_y++;
d231e84b 2147 ret = ff_h264_decode_mb_cavlc(h, sl);
f51d0f39
LB
2148
2149 if (ret >= 0)
92c6c2a6 2150 ff_h264_hl_decode_mb(h, sl);
d4d9068c 2151 sl->mb_y--;
f51d0f39
LB
2152 }
2153
2154 if (ret < 0) {
2155 av_log(h->avctx, AV_LOG_ERROR,
d4d9068c 2156 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
582683b6 2157 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
d4d9068c 2158 sl->mb_y, ER_MB_ERROR);
f51d0f39
LB
2159 return ret;
2160 }
2161
d4d9068c
AK
2162 if (++sl->mb_x >= h->mb_width) {
2163 loop_filter(h, sl, lf_x_start, sl->mb_x);
2164 sl->mb_x = lf_x_start = 0;
95eb35f3 2165 decode_finish_row(h, sl);
d4d9068c 2166 ++sl->mb_y;
f51d0f39 2167 if (FIELD_OR_MBAFF_PICTURE(h)) {
d4d9068c
AK
2168 ++sl->mb_y;
2169 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
56febc99 2170 predict_field_decoding_flag(h, sl);
f51d0f39 2171 }
d4d9068c 2172 if (sl->mb_y >= h->mb_height) {
6a85dfc8 2173 ff_tlog(h->avctx, "slice end %d %d\n",
f42485db 2174 get_bits_count(&sl->gb), sl->gb.size_in_bits);
f51d0f39 2175
f42485db 2176 if (get_bits_left(&sl->gb) == 0) {
582683b6 2177 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
d4d9068c 2178 sl->mb_x - 1, sl->mb_y, ER_MB_END);
f51d0f39
LB
2179
2180 return 0;
2181 } else {
582683b6 2182 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
d4d9068c 2183 sl->mb_x - 1, sl->mb_y, ER_MB_END);
f51d0f39
LB
2184
2185 return AVERROR_INVALIDDATA;
2186 }
2187 }
2188 }
2189
f42485db 2190 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
6a85dfc8 2191 ff_tlog(h->avctx, "slice end %d %d\n",
f42485db 2192 get_bits_count(&sl->gb), sl->gb.size_in_bits);
f51d0f39 2193
f42485db 2194 if (get_bits_left(&sl->gb) == 0) {
582683b6 2195 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
d4d9068c
AK
2196 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2197 if (sl->mb_x > lf_x_start)
2198 loop_filter(h, sl, lf_x_start, sl->mb_x);
f51d0f39
LB
2199
2200 return 0;
2201 } else {
582683b6 2202 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
d4d9068c 2203 sl->mb_y, ER_MB_ERROR);
f51d0f39
LB
2204
2205 return AVERROR_INVALIDDATA;
2206 }
2207 }
2208 }
2209 }
2210}
2211
2212/**
2213 * Call decode_slice() for each context.
2214 *
2215 * @param h h264 master context
2216 * @param context_count number of contexts to execute
2217 */
2218int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2219{
2220 AVCodecContext *const avctx = h->avctx;
d4d9068c 2221 H264SliceContext *sl;
f51d0f39
LB
2222 int i;
2223
f51d0f39
LB
2224 if (h->avctx->hwaccel)
2225 return 0;
2226 if (context_count == 1) {
d4d9068c
AK
2227 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2228 h->mb_y = h->slice_ctx[0].mb_y;
2229 return ret;
f51d0f39
LB
2230 } else {
2231 for (i = 1; i < context_count; i++) {
582683b6
AK
2232 sl = &h->slice_ctx[i];
2233 sl->er.error_count = 0;
f51d0f39
LB
2234 }
2235
92c6c2a6
AK
2236 avctx->execute(avctx, decode_slice, h->slice_ctx,
2237 NULL, context_count, sizeof(h->slice_ctx[0]));
f51d0f39
LB
2238
2239 /* pull back stuff from slices to master context */
d4d9068c
AK
2240 sl = &h->slice_ctx[context_count - 1];
2241 h->mb_y = sl->mb_y;
f51d0f39 2242 for (i = 1; i < context_count; i++)
582683b6 2243 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
f51d0f39
LB
2244 }
2245
2246 return 0;
2247}