h264: er: Copy from the previous reference only if compatible
[libav.git] / libavcodec / h264_slice.c
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"
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
47 static 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
53 static 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
59 static 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
66 static 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
85 static 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)]
105 static 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
124 static 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
133 static 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
137 static 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
146
147 static 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++) {
153 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
154 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
155 ff_h264_unref_picture(h, &h->DPB[i]);
156 }
157 }
158 }
159
160 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
161 {
162 const H264Context *h = sl->h264;
163 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
164
165 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
166 // edge emu needs blocksize + filter length - 1
167 // (= 21x21 for h264)
168 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
169
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]) {
177 av_freep(&sl->bipred_scratchpad);
178 av_freep(&sl->edge_emu_buffer);
179 av_freep(&sl->top_borders[0]);
180 av_freep(&sl->top_borders[1]);
181
182 sl->bipred_scratchpad_allocated = 0;
183 sl->edge_emu_buffer_allocated = 0;
184 sl->top_borders_allocated[0] = 0;
185 sl->top_borders_allocated[1] = 0;
186 return AVERROR(ENOMEM);
187 }
188
189 return 0;
190 }
191
192 static 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);
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);
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
219 static int alloc_picture(H264Context *h, H264Picture *pic)
220 {
221 int i, ret = 0;
222
223 av_assert0(!pic->f->data[0]);
224
225 pic->tf.f = pic->f;
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
231 if (h->avctx->hwaccel) {
232 const AVHWAccel *hwaccel = h->avctx->hwaccel;
233 av_assert0(!pic->hwaccel_picture_private);
234 if (hwaccel->frame_priv_data_size) {
235 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
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;
267 fail:
268 ff_h264_unref_picture(h, pic);
269 return (ret < 0) ? ret : AVERROR(ENOMEM);
270 }
271
272 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
273 {
274 if (!pic->f->buf[0])
275 return 1;
276 return 0;
277 }
278
279 static 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
290 return i;
291 }
292
293
294 static 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
321 static 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
347 void 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
364 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
365
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
371 static 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
386 static 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
410 static int h264_slice_header_init(H264Context *h);
411
412 int 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;
417 int need_reinit = 0;
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)) {
431 need_reinit = 1;
432 }
433
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;
458 }
459
460 /* copy block_offset since frame_start may not be called */
461 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
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;
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]);
476 if (h1->DPB[i].f->buf[0] &&
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);
483 if (h1->cur_pic.f->buf[0]) {
484 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
485 if (ret < 0)
486 return ret;
487 }
488
489 h->enable_er = h1->enable_er;
490 h->workaround_bugs = h1->workaround_bugs;
491 h->low_delay = h1->low_delay;
492 h->droppable = h1->droppable;
493
494 // extradata/NAL handling
495 h->is_avc = h1->is_avc;
496 h->nal_length_size = h1->nal_length_size;
497
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
513 copy_fields(h, h1, poc_lsb, default_ref_list);
514
515 // reference lists
516 copy_fields(h, h1, short_ref, current_slice);
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
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;
535
536 h->recovery_frame = h1->recovery_frame;
537 h->frame_recovered = h1->frame_recovered;
538
539 return err;
540 }
541
542 static 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;
559 pic->f->coded_picture_number = h->coded_picture_number++;
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 */
566 pic->f->key_frame = 0;
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
578 if (CONFIG_ERROR_RESILIENCE && h->enable_er)
579 ff_er_frame_start(&h->slice_ctx[0].er);
580
581 for (i = 0; i < 16; i++) {
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);
584 }
585 for (i = 0; i < 16; i++) {
586 h->block_offset[16 + i] =
587 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
588 h->block_offset[48 + 16 + i] =
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);
590 }
591
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
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
599 * in the next ff_mpv_frame_start().
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
612 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
613 uint8_t *src_y,
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)) {
629 if (sl->mb_y & 1) {
630 if (!MB_MBAFF(sl)) {
631 top_border = sl->top_borders[0][sl->mb_x];
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 }
665 } else if (MB_MBAFF(sl)) {
666 top_idx = 0;
667 } else
668 return;
669 }
670
671 top_border = sl->top_borders[top_idx][sl->mb_x];
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 */
714 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
715 {
716 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
717
718 for (i = 0; i < 2; i++) {
719 sl->luma_weight_flag[i] = 0;
720 sl->chroma_weight_flag[i] = 0;
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 }
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) {
731 sl->use_weight = 0;
732 sl->use_weight_chroma = 0;
733 return;
734 }
735 ref_start = 0;
736 ref_count0 = sl->ref_count[0];
737 ref_count1 = sl->ref_count[1];
738 } else {
739 cur_poc = h->cur_pic_ptr->field_poc[field];
740 ref_start = 16;
741 ref_count0 = 16 + 2 * sl->ref_count[0];
742 ref_count1 = 16 + 2 * sl->ref_count[1];
743 }
744
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;
749
750 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
751 int poc0 = sl->ref_list[0][ref0].poc;
752 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
753 int w = 32;
754 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
755 int poc1 = sl->ref_list[1][ref1].poc;
756 int td = av_clip_int8(poc1 - poc0);
757 if (td) {
758 int tb = av_clip_int8(cur_poc - poc0);
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) {
766 sl->implicit_weight[ref0][ref1][0] =
767 sl->implicit_weight[ref0][ref1][1] = w;
768 } else {
769 sl->implicit_weight[ref0][ref1][field] = w;
770 }
771 }
772 }
773 }
774
775 /**
776 * initialize scan tables
777 */
778 static 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
812 static enum AVPixelFormat get_pixel_format(H264Context *h)
813 {
814 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
815 CONFIG_H264_D3D11VA_HWACCEL + \
816 CONFIG_H264_VAAPI_HWACCEL + \
817 (CONFIG_H264_VDA_HWACCEL * 2) + \
818 CONFIG_H264_VDPAU_HWACCEL)
819 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
820 const enum AVPixelFormat *choices = pix_fmts;
821
822 switch (h->sps.bit_depth_luma) {
823 case 9:
824 if (CHROMA444(h)) {
825 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
826 *fmt++ = AV_PIX_FMT_GBRP9;
827 } else
828 *fmt++ = AV_PIX_FMT_YUV444P9;
829 } else if (CHROMA422(h))
830 *fmt++ = AV_PIX_FMT_YUV422P9;
831 else
832 *fmt++ = AV_PIX_FMT_YUV420P9;
833 break;
834 case 10:
835 if (CHROMA444(h)) {
836 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
837 *fmt++ = AV_PIX_FMT_GBRP10;
838 } else
839 *fmt++ = AV_PIX_FMT_YUV444P10;
840 } else if (CHROMA422(h))
841 *fmt++ = AV_PIX_FMT_YUV422P10;
842 else
843 *fmt++ = AV_PIX_FMT_YUV420P10;
844 break;
845 case 8:
846 #if CONFIG_H264_VDPAU_HWACCEL
847 *fmt++ = AV_PIX_FMT_VDPAU;
848 #endif
849 if (CHROMA444(h)) {
850 if (h->avctx->colorspace == AVCOL_SPC_RGB)
851 *fmt++ = AV_PIX_FMT_GBRP;
852 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
853 *fmt++ = AV_PIX_FMT_YUVJ444P;
854 else
855 *fmt++ = AV_PIX_FMT_YUV444P;
856 } else if (CHROMA422(h)) {
857 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
858 *fmt++ = AV_PIX_FMT_YUVJ422P;
859 else
860 *fmt++ = AV_PIX_FMT_YUV422P;
861 } else {
862 #if CONFIG_H264_DXVA2_HWACCEL
863 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
864 #endif
865 #if CONFIG_H264_D3D11VA_HWACCEL
866 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
867 #endif
868 #if CONFIG_H264_VAAPI_HWACCEL
869 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
870 #endif
871 #if CONFIG_H264_VDA_HWACCEL
872 *fmt++ = AV_PIX_FMT_VDA_VLD;
873 *fmt++ = AV_PIX_FMT_VDA;
874 #endif
875 if (h->avctx->codec->pix_fmts)
876 choices = h->avctx->codec->pix_fmts;
877 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
878 *fmt++ = AV_PIX_FMT_YUVJ420P;
879 else
880 *fmt++ = AV_PIX_FMT_YUV420P;
881 }
882 break;
883 default:
884 av_log(h->avctx, AV_LOG_ERROR,
885 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
886 return AVERROR_INVALIDDATA;
887 }
888
889 *fmt = AV_PIX_FMT_NONE;
890
891 return ff_get_format(h->avctx, choices);
892 }
893
894 /* export coded and cropped frame dimensions to AVCodecContext */
895 static int init_dimensions(H264Context *h)
896 {
897 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
898 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
899 int crop_present = h->sps.crop_left || h->sps.crop_top ||
900 h->sps.crop_right || h->sps.crop_bottom;
901
902 /* handle container cropping */
903 if (!crop_present &&
904 FFALIGN(h->avctx->width, 16) == h->width &&
905 FFALIGN(h->avctx->height, 16) == h->height) {
906 width = h->avctx->width;
907 height = h->avctx->height;
908 }
909
910 if (width <= 0 || height <= 0) {
911 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
912 width, height);
913 if (h->avctx->err_recognition & AV_EF_EXPLODE)
914 return AVERROR_INVALIDDATA;
915
916 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
917 h->sps.crop_bottom =
918 h->sps.crop_top =
919 h->sps.crop_right =
920 h->sps.crop_left =
921 h->sps.crop = 0;
922
923 width = h->width;
924 height = h->height;
925 }
926
927 h->avctx->coded_width = h->width;
928 h->avctx->coded_height = h->height;
929 h->avctx->width = width;
930 h->avctx->height = height;
931
932 return 0;
933 }
934
935 static int h264_slice_header_init(H264Context *h)
936 {
937 int nb_slices = (HAVE_THREADS &&
938 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
939 h->avctx->thread_count : 1;
940 int i, ret;
941
942 ff_set_sar(h->avctx, h->sps.sar);
943 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
944 &h->chroma_x_shift, &h->chroma_y_shift);
945
946 if (h->sps.timing_info_present_flag) {
947 int64_t den = h->sps.time_scale;
948 if (h->x264_build < 44U)
949 den *= 2;
950 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
951 h->sps.num_units_in_tick, den, 1 << 30);
952 }
953
954 ff_h264_free_tables(h);
955
956 h->first_field = 0;
957 h->prev_interlaced_frame = 1;
958
959 init_scan_tables(h);
960 ret = ff_h264_alloc_tables(h);
961 if (ret < 0) {
962 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
963 return ret;
964 }
965
966 if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 10) {
967 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
968 h->sps.bit_depth_luma);
969 return AVERROR_INVALIDDATA;
970 }
971
972 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
973 h->pixel_shift = h->sps.bit_depth_luma > 8;
974 h->chroma_format_idc = h->sps.chroma_format_idc;
975 h->bit_depth_luma = h->sps.bit_depth_luma;
976
977 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
978 h->sps.chroma_format_idc);
979 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
980 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
981 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
982 h->sps.chroma_format_idc);
983 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
984
985 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
986 int max_slices;
987 if (h->mb_height)
988 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
989 else
990 max_slices = H264_MAX_THREADS;
991 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
992 " reducing to %d\n", nb_slices, max_slices);
993 nb_slices = max_slices;
994 }
995 h->slice_context_count = nb_slices;
996
997 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
998 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
999 if (ret < 0) {
1000 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1001 return ret;
1002 }
1003 } else {
1004 for (i = 0; i < h->slice_context_count; i++) {
1005 H264SliceContext *sl = &h->slice_ctx[i];
1006
1007 sl->h264 = h;
1008 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1009 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1010 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1011
1012 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1013 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1014 return ret;
1015 }
1016 }
1017 }
1018
1019 h->context_initialized = 1;
1020
1021 return 0;
1022 }
1023
1024 /**
1025 * Decode a slice header.
1026 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1027 *
1028 * @param h h264context
1029 *
1030 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1031 */
1032 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1033 {
1034 unsigned int first_mb_in_slice;
1035 unsigned int pps_id;
1036 int ret;
1037 unsigned int slice_type, tmp, i, j;
1038 int default_ref_list_done = 0;
1039 int last_pic_structure, last_pic_droppable;
1040 int needs_reinit = 0;
1041 int field_pic_flag, bottom_field_flag;
1042
1043 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1044 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1045
1046 first_mb_in_slice = get_ue_golomb(&sl->gb);
1047
1048 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1049 if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1050 ff_h264_field_end(h, sl, 1);
1051 }
1052
1053 h->current_slice = 0;
1054 if (!h->first_field) {
1055 if (h->cur_pic_ptr && !h->droppable) {
1056 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1057 h->picture_structure == PICT_BOTTOM_FIELD);
1058 }
1059 h->cur_pic_ptr = NULL;
1060 }
1061 }
1062
1063 slice_type = get_ue_golomb_31(&sl->gb);
1064 if (slice_type > 9) {
1065 av_log(h->avctx, AV_LOG_ERROR,
1066 "slice type %d too large at %d\n",
1067 slice_type, first_mb_in_slice);
1068 return AVERROR_INVALIDDATA;
1069 }
1070 if (slice_type > 4) {
1071 slice_type -= 5;
1072 sl->slice_type_fixed = 1;
1073 } else
1074 sl->slice_type_fixed = 0;
1075
1076 slice_type = golomb_to_pict_type[slice_type];
1077 if (slice_type == AV_PICTURE_TYPE_I ||
1078 (h->current_slice != 0 && slice_type == h->last_slice_type)) {
1079 default_ref_list_done = 1;
1080 }
1081 sl->slice_type = slice_type;
1082 sl->slice_type_nos = slice_type & 3;
1083
1084 if (h->nal_unit_type == NAL_IDR_SLICE &&
1085 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1086 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1087 return AVERROR_INVALIDDATA;
1088 }
1089
1090 // to make a few old functions happy, it's wrong though
1091 h->pict_type = sl->slice_type;
1092
1093 pps_id = get_ue_golomb(&sl->gb);
1094 if (pps_id >= MAX_PPS_COUNT) {
1095 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1096 return AVERROR_INVALIDDATA;
1097 }
1098 if (!h->pps_buffers[pps_id]) {
1099 av_log(h->avctx, AV_LOG_ERROR,
1100 "non-existing PPS %u referenced\n",
1101 pps_id);
1102 return AVERROR_INVALIDDATA;
1103 }
1104 h->pps = *h->pps_buffers[pps_id];
1105
1106 if (!h->sps_buffers[h->pps.sps_id]) {
1107 av_log(h->avctx, AV_LOG_ERROR,
1108 "non-existing SPS %u referenced\n",
1109 h->pps.sps_id);
1110 return AVERROR_INVALIDDATA;
1111 }
1112
1113 if (h->pps.sps_id != h->sps.sps_id ||
1114 h->sps_buffers[h->pps.sps_id]->new) {
1115 h->sps_buffers[h->pps.sps_id]->new = 0;
1116
1117 h->sps = *h->sps_buffers[h->pps.sps_id];
1118
1119 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1120 h->chroma_format_idc != h->sps.chroma_format_idc)
1121 needs_reinit = 1;
1122
1123 if (h->flags & CODEC_FLAG_LOW_DELAY ||
1124 (h->sps.bitstream_restriction_flag &&
1125 !h->sps.num_reorder_frames)) {
1126 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1127 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1128 "Reenabling low delay requires a codec flush.\n");
1129 else
1130 h->low_delay = 1;
1131 }
1132
1133 if (h->avctx->has_b_frames < 2)
1134 h->avctx->has_b_frames = !h->low_delay;
1135
1136 }
1137
1138 h->avctx->profile = ff_h264_get_profile(&h->sps);
1139 h->avctx->level = h->sps.level_idc;
1140 h->avctx->refs = h->sps.ref_frame_count;
1141
1142 if (h->mb_width != h->sps.mb_width ||
1143 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1144 needs_reinit = 1;
1145
1146 h->mb_width = h->sps.mb_width;
1147 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1148 h->mb_num = h->mb_width * h->mb_height;
1149 h->mb_stride = h->mb_width + 1;
1150
1151 h->b_stride = h->mb_width * 4;
1152
1153 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1154
1155 h->width = 16 * h->mb_width;
1156 h->height = 16 * h->mb_height;
1157
1158 ret = init_dimensions(h);
1159 if (ret < 0)
1160 return ret;
1161
1162 if (h->sps.video_signal_type_present_flag) {
1163 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1164 : AVCOL_RANGE_MPEG;
1165 if (h->sps.colour_description_present_flag) {
1166 if (h->avctx->colorspace != h->sps.colorspace)
1167 needs_reinit = 1;
1168 h->avctx->color_primaries = h->sps.color_primaries;
1169 h->avctx->color_trc = h->sps.color_trc;
1170 h->avctx->colorspace = h->sps.colorspace;
1171 }
1172 }
1173
1174 if (h->context_initialized && needs_reinit) {
1175 h->context_initialized = 0;
1176 if (sl != h->slice_ctx) {
1177 av_log(h->avctx, AV_LOG_ERROR,
1178 "changing width %d -> %d / height %d -> %d on "
1179 "slice %d\n",
1180 h->width, h->avctx->coded_width,
1181 h->height, h->avctx->coded_height,
1182 h->current_slice + 1);
1183 return AVERROR_INVALIDDATA;
1184 }
1185
1186 ff_h264_flush_change(h);
1187
1188 if ((ret = get_pixel_format(h)) < 0)
1189 return ret;
1190 h->avctx->pix_fmt = ret;
1191
1192 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1193 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1194
1195 if ((ret = h264_slice_header_init(h)) < 0) {
1196 av_log(h->avctx, AV_LOG_ERROR,
1197 "h264_slice_header_init() failed\n");
1198 return ret;
1199 }
1200 }
1201 if (!h->context_initialized) {
1202 if (sl != h->slice_ctx) {
1203 av_log(h->avctx, AV_LOG_ERROR,
1204 "Cannot (re-)initialize context during parallel decoding.\n");
1205 return AVERROR_PATCHWELCOME;
1206 }
1207
1208 if ((ret = get_pixel_format(h)) < 0)
1209 return ret;
1210 h->avctx->pix_fmt = ret;
1211
1212 if ((ret = h264_slice_header_init(h)) < 0) {
1213 av_log(h->avctx, AV_LOG_ERROR,
1214 "h264_slice_header_init() failed\n");
1215 return ret;
1216 }
1217 }
1218
1219 if (sl == h->slice_ctx && h->dequant_coeff_pps != pps_id) {
1220 h->dequant_coeff_pps = pps_id;
1221 h264_init_dequant_tables(h);
1222 }
1223
1224 h->frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1225
1226 sl->mb_mbaff = 0;
1227 h->mb_aff_frame = 0;
1228 last_pic_structure = h->picture_structure;
1229 last_pic_droppable = h->droppable;
1230 h->droppable = h->nal_ref_idc == 0;
1231 if (h->sps.frame_mbs_only_flag) {
1232 h->picture_structure = PICT_FRAME;
1233 } else {
1234 field_pic_flag = get_bits1(&sl->gb);
1235 if (field_pic_flag) {
1236 bottom_field_flag = get_bits1(&sl->gb);
1237 h->picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1238 } else {
1239 h->picture_structure = PICT_FRAME;
1240 h->mb_aff_frame = h->sps.mb_aff;
1241 }
1242 }
1243 sl->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1244
1245 if (h->current_slice != 0) {
1246 if (last_pic_structure != h->picture_structure ||
1247 last_pic_droppable != h->droppable) {
1248 av_log(h->avctx, AV_LOG_ERROR,
1249 "Changing field mode (%d -> %d) between slices is not allowed\n",
1250 last_pic_structure, h->picture_structure);
1251 h->picture_structure = last_pic_structure;
1252 h->droppable = last_pic_droppable;
1253 return AVERROR_INVALIDDATA;
1254 } else if (!h->cur_pic_ptr) {
1255 av_log(h->avctx, AV_LOG_ERROR,
1256 "unset cur_pic_ptr on slice %d\n",
1257 h->current_slice + 1);
1258 return AVERROR_INVALIDDATA;
1259 }
1260 } else {
1261 /* Shorten frame num gaps so we don't have to allocate reference
1262 * frames just to throw them away */
1263 if (h->frame_num != h->prev_frame_num) {
1264 int unwrap_prev_frame_num = h->prev_frame_num;
1265 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1266
1267 if (unwrap_prev_frame_num > h->frame_num)
1268 unwrap_prev_frame_num -= max_frame_num;
1269
1270 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1271 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1272 if (unwrap_prev_frame_num < 0)
1273 unwrap_prev_frame_num += max_frame_num;
1274
1275 h->prev_frame_num = unwrap_prev_frame_num;
1276 }
1277 }
1278
1279 /* See if we have a decoded first field looking for a pair...
1280 * Here, we're using that to see if we should mark previously
1281 * decode frames as "finished".
1282 * We have to do that before the "dummy" in-between frame allocation,
1283 * since that can modify s->current_picture_ptr. */
1284 if (h->first_field) {
1285 assert(h->cur_pic_ptr);
1286 assert(h->cur_pic_ptr->f->buf[0]);
1287 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1288
1289 /* figure out if we have a complementary field pair */
1290 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1291 /* Previous field is unmatched. Don't display it, but let it
1292 * remain for reference if marked as such. */
1293 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1294 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1295 last_pic_structure == PICT_TOP_FIELD);
1296 }
1297 } else {
1298 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1299 /* This and previous field were reference, but had
1300 * different frame_nums. Consider this field first in
1301 * pair. Throw away previous field except for reference
1302 * purposes. */
1303 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1304 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1305 last_pic_structure == PICT_TOP_FIELD);
1306 }
1307 } else {
1308 /* Second field in complementary pair */
1309 if (!((last_pic_structure == PICT_TOP_FIELD &&
1310 h->picture_structure == PICT_BOTTOM_FIELD) ||
1311 (last_pic_structure == PICT_BOTTOM_FIELD &&
1312 h->picture_structure == PICT_TOP_FIELD))) {
1313 av_log(h->avctx, AV_LOG_ERROR,
1314 "Invalid field mode combination %d/%d\n",
1315 last_pic_structure, h->picture_structure);
1316 h->picture_structure = last_pic_structure;
1317 h->droppable = last_pic_droppable;
1318 return AVERROR_INVALIDDATA;
1319 } else if (last_pic_droppable != h->droppable) {
1320 avpriv_request_sample(h->avctx,
1321 "Found reference and non-reference fields in the same frame, which");
1322 h->picture_structure = last_pic_structure;
1323 h->droppable = last_pic_droppable;
1324 return AVERROR_PATCHWELCOME;
1325 }
1326 }
1327 }
1328 }
1329
1330 while (h->frame_num != h->prev_frame_num &&
1331 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1332 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1333 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1334 h->frame_num, h->prev_frame_num);
1335 ret = h264_frame_start(h);
1336 if (ret < 0) {
1337 h->first_field = 0;
1338 return ret;
1339 }
1340
1341 h->prev_frame_num++;
1342 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1343 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1344 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1345 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1346 ret = ff_generate_sliding_window_mmcos(h, 1);
1347 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1348 return ret;
1349 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1350 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1351 return ret;
1352 /* Error concealment: If a ref is missing, copy the previous ref
1353 * in its place.
1354 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1355 * many assumptions about there being no actual duplicates.
1356 * FIXME: This does not copy padding for out-of-frame motion
1357 * vectors. Given we are concealing a lost frame, this probably
1358 * is not noticeable by comparison, but it should be fixed. */
1359 if (h->short_ref_count) {
1360 if (prev &&
1361 h->short_ref[0]->f->width == prev->f->width &&
1362 h->short_ref[0]->f->height == prev->f->height &&
1363 h->short_ref[0]->f->format == prev->f->format) {
1364 av_image_copy(h->short_ref[0]->f->data,
1365 h->short_ref[0]->f->linesize,
1366 (const uint8_t **)prev->f->data,
1367 prev->f->linesize,
1368 prev->f->format,
1369 h->mb_width * 16,
1370 h->mb_height * 16);
1371 h->short_ref[0]->poc = prev->poc + 2;
1372 }
1373 h->short_ref[0]->frame_num = h->prev_frame_num;
1374 }
1375 }
1376
1377 /* See if we have a decoded first field looking for a pair...
1378 * We're using that to see whether to continue decoding in that
1379 * frame, or to allocate a new one. */
1380 if (h->first_field) {
1381 assert(h->cur_pic_ptr);
1382 assert(h->cur_pic_ptr->f->buf[0]);
1383 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1384
1385 /* figure out if we have a complementary field pair */
1386 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1387 /* Previous field is unmatched. Don't display it, but let it
1388 * remain for reference if marked as such. */
1389 h->cur_pic_ptr = NULL;
1390 h->first_field = FIELD_PICTURE(h);
1391 } else {
1392 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1393 /* This and the previous field had different frame_nums.
1394 * Consider this field first in pair. Throw away previous
1395 * one except for reference purposes. */
1396 h->first_field = 1;
1397 h->cur_pic_ptr = NULL;
1398 } else {
1399 /* Second field in complementary pair */
1400 h->first_field = 0;
1401 }
1402 }
1403 } else {
1404 /* Frame or first field in a potentially complementary pair */
1405 h->first_field = FIELD_PICTURE(h);
1406 }
1407
1408 if (!FIELD_PICTURE(h) || h->first_field) {
1409 if (h264_frame_start(h) < 0) {
1410 h->first_field = 0;
1411 return AVERROR_INVALIDDATA;
1412 }
1413 } else {
1414 release_unused_pictures(h, 0);
1415 }
1416 }
1417
1418 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1419
1420 assert(h->mb_num == h->mb_width * h->mb_height);
1421 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1422 first_mb_in_slice >= h->mb_num) {
1423 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1424 return AVERROR_INVALIDDATA;
1425 }
1426 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1427 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1428 FIELD_OR_MBAFF_PICTURE(h);
1429 if (h->picture_structure == PICT_BOTTOM_FIELD)
1430 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1431 assert(sl->mb_y < h->mb_height);
1432
1433 if (h->picture_structure == PICT_FRAME) {
1434 h->curr_pic_num = h->frame_num;
1435 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1436 } else {
1437 h->curr_pic_num = 2 * h->frame_num + 1;
1438 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1439 }
1440
1441 if (h->nal_unit_type == NAL_IDR_SLICE)
1442 get_ue_golomb(&sl->gb); /* idr_pic_id */
1443
1444 if (h->sps.poc_type == 0) {
1445 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1446
1447 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1448 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1449 }
1450
1451 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1452 h->delta_poc[0] = get_se_golomb(&sl->gb);
1453
1454 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1455 h->delta_poc[1] = get_se_golomb(&sl->gb);
1456 }
1457
1458 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1459
1460 if (h->pps.redundant_pic_cnt_present)
1461 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1462
1463 ret = ff_set_ref_count(h, sl);
1464 if (ret < 0)
1465 return ret;
1466 else if (ret == 1)
1467 default_ref_list_done = 0;
1468
1469 if (!default_ref_list_done)
1470 ff_h264_fill_default_ref_list(h, sl);
1471
1472 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1473 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1474 if (ret < 0) {
1475 sl->ref_count[1] = sl->ref_count[0] = 0;
1476 return ret;
1477 }
1478 }
1479
1480 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1481 (h->pps.weighted_bipred_idc == 1 &&
1482 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1483 ff_pred_weight_table(h, sl);
1484 else if (h->pps.weighted_bipred_idc == 2 &&
1485 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1486 implicit_weight_table(h, sl, -1);
1487 } else {
1488 sl->use_weight = 0;
1489 for (i = 0; i < 2; i++) {
1490 sl->luma_weight_flag[i] = 0;
1491 sl->chroma_weight_flag[i] = 0;
1492 }
1493 }
1494
1495 // If frame-mt is enabled, only update mmco tables for the first slice
1496 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1497 // or h->mmco, which will cause ref list mix-ups and decoding errors
1498 // further down the line. This may break decoding if the first slice is
1499 // corrupt, thus we only do this if frame-mt is enabled.
1500 if (h->nal_ref_idc) {
1501 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1502 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1503 h->current_slice == 0);
1504 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1505 return AVERROR_INVALIDDATA;
1506 }
1507
1508 if (FRAME_MBAFF(h)) {
1509 ff_h264_fill_mbaff_ref_list(h, sl);
1510
1511 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1512 implicit_weight_table(h, sl, 0);
1513 implicit_weight_table(h, sl, 1);
1514 }
1515 }
1516
1517 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1518 ff_h264_direct_dist_scale_factor(h, sl);
1519 ff_h264_direct_ref_list_init(h, sl);
1520
1521 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1522 tmp = get_ue_golomb_31(&sl->gb);
1523 if (tmp > 2) {
1524 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1525 return AVERROR_INVALIDDATA;
1526 }
1527 sl->cabac_init_idc = tmp;
1528 }
1529
1530 sl->last_qscale_diff = 0;
1531 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1532 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1533 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1534 return AVERROR_INVALIDDATA;
1535 }
1536 sl->qscale = tmp;
1537 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1538 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1539 // FIXME qscale / qp ... stuff
1540 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1541 get_bits1(&sl->gb); /* sp_for_switch_flag */
1542 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1543 sl->slice_type == AV_PICTURE_TYPE_SI)
1544 get_se_golomb(&sl->gb); /* slice_qs_delta */
1545
1546 sl->deblocking_filter = 1;
1547 sl->slice_alpha_c0_offset = 0;
1548 sl->slice_beta_offset = 0;
1549 if (h->pps.deblocking_filter_parameters_present) {
1550 tmp = get_ue_golomb_31(&sl->gb);
1551 if (tmp > 2) {
1552 av_log(h->avctx, AV_LOG_ERROR,
1553 "deblocking_filter_idc %u out of range\n", tmp);
1554 return AVERROR_INVALIDDATA;
1555 }
1556 sl->deblocking_filter = tmp;
1557 if (sl->deblocking_filter < 2)
1558 sl->deblocking_filter ^= 1; // 1<->0
1559
1560 if (sl->deblocking_filter) {
1561 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1562 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1563 if (sl->slice_alpha_c0_offset > 12 ||
1564 sl->slice_alpha_c0_offset < -12 ||
1565 sl->slice_beta_offset > 12 ||
1566 sl->slice_beta_offset < -12) {
1567 av_log(h->avctx, AV_LOG_ERROR,
1568 "deblocking filter parameters %d %d out of range\n",
1569 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1570 return AVERROR_INVALIDDATA;
1571 }
1572 }
1573 }
1574
1575 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1576 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1577 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1578 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1579 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1580 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1581 h->nal_ref_idc == 0))
1582 sl->deblocking_filter = 0;
1583
1584 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1585 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1586 /* Cheat slightly for speed:
1587 * Do not bother to deblock across slices. */
1588 sl->deblocking_filter = 2;
1589 } else {
1590 h->max_contexts = 1;
1591 if (!h->single_decode_warning) {
1592 av_log(h->avctx, AV_LOG_INFO,
1593 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1594 h->single_decode_warning = 1;
1595 }
1596 if (sl != h->slice_ctx) {
1597 av_log(h->avctx, AV_LOG_ERROR,
1598 "Deblocking switched inside frame.\n");
1599 return 1;
1600 }
1601 }
1602 }
1603 sl->qp_thresh = 15 -
1604 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1605 FFMAX3(0,
1606 h->pps.chroma_qp_index_offset[0],
1607 h->pps.chroma_qp_index_offset[1]) +
1608 6 * (h->sps.bit_depth_luma - 8);
1609
1610 h->last_slice_type = slice_type;
1611 sl->slice_num = ++h->current_slice;
1612 if (sl->slice_num >= MAX_SLICES) {
1613 av_log(h->avctx, AV_LOG_ERROR,
1614 "Too many slices, increase MAX_SLICES and recompile\n");
1615 }
1616
1617 for (j = 0; j < 2; j++) {
1618 int id_list[16];
1619 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1620 for (i = 0; i < 16; i++) {
1621 id_list[i] = 60;
1622 if (j < sl->list_count && i < sl->ref_count[j] &&
1623 sl->ref_list[j][i].parent->f->buf[0]) {
1624 int k;
1625 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1626 for (k = 0; k < h->short_ref_count; k++)
1627 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1628 id_list[i] = k;
1629 break;
1630 }
1631 for (k = 0; k < h->long_ref_count; k++)
1632 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1633 id_list[i] = h->short_ref_count + k;
1634 break;
1635 }
1636 }
1637 }
1638
1639 ref2frm[0] =
1640 ref2frm[1] = -1;
1641 for (i = 0; i < 16; i++)
1642 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1643 ref2frm[18 + 0] =
1644 ref2frm[18 + 1] = -1;
1645 for (i = 16; i < 48; i++)
1646 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1647 (sl->ref_list[j][i].reference & 3);
1648 }
1649
1650 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1651 av_log(h->avctx, AV_LOG_DEBUG,
1652 "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",
1653 sl->slice_num,
1654 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1655 first_mb_in_slice,
1656 av_get_picture_type_char(sl->slice_type),
1657 sl->slice_type_fixed ? " fix" : "",
1658 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1659 pps_id, h->frame_num,
1660 h->cur_pic_ptr->field_poc[0],
1661 h->cur_pic_ptr->field_poc[1],
1662 sl->ref_count[0], sl->ref_count[1],
1663 sl->qscale,
1664 sl->deblocking_filter,
1665 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1666 sl->use_weight,
1667 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
1668 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1669 }
1670
1671 return 0;
1672 }
1673
1674 int ff_h264_get_slice_type(const H264SliceContext *sl)
1675 {
1676 switch (sl->slice_type) {
1677 case AV_PICTURE_TYPE_P:
1678 return 0;
1679 case AV_PICTURE_TYPE_B:
1680 return 1;
1681 case AV_PICTURE_TYPE_I:
1682 return 2;
1683 case AV_PICTURE_TYPE_SP:
1684 return 3;
1685 case AV_PICTURE_TYPE_SI:
1686 return 4;
1687 default:
1688 return AVERROR_INVALIDDATA;
1689 }
1690 }
1691
1692 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1693 H264SliceContext *sl,
1694 int mb_type, int top_xy,
1695 int left_xy[LEFT_MBS],
1696 int top_type,
1697 int left_type[LEFT_MBS],
1698 int mb_xy, int list)
1699 {
1700 int b_stride = h->b_stride;
1701 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1702 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1703 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1704 if (USES_LIST(top_type, list)) {
1705 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1706 const int b8_xy = 4 * top_xy + 2;
1707 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1708 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1709 ref_cache[0 - 1 * 8] =
1710 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1711 ref_cache[2 - 1 * 8] =
1712 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1713 } else {
1714 AV_ZERO128(mv_dst - 1 * 8);
1715 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1716 }
1717
1718 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1719 if (USES_LIST(left_type[LTOP], list)) {
1720 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1721 const int b8_xy = 4 * left_xy[LTOP] + 1;
1722 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1723 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1724 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1725 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1726 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1727 ref_cache[-1 + 0] =
1728 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1729 ref_cache[-1 + 16] =
1730 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1731 } else {
1732 AV_ZERO32(mv_dst - 1 + 0);
1733 AV_ZERO32(mv_dst - 1 + 8);
1734 AV_ZERO32(mv_dst - 1 + 16);
1735 AV_ZERO32(mv_dst - 1 + 24);
1736 ref_cache[-1 + 0] =
1737 ref_cache[-1 + 8] =
1738 ref_cache[-1 + 16] =
1739 ref_cache[-1 + 24] = LIST_NOT_USED;
1740 }
1741 }
1742 }
1743
1744 if (!USES_LIST(mb_type, list)) {
1745 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1746 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1747 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1748 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1749 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1750 return;
1751 }
1752
1753 {
1754 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1755 int (*ref2frm)[64] = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1756 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1757 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1758 AV_WN32A(&ref_cache[0 * 8], ref01);
1759 AV_WN32A(&ref_cache[1 * 8], ref01);
1760 AV_WN32A(&ref_cache[2 * 8], ref23);
1761 AV_WN32A(&ref_cache[3 * 8], ref23);
1762 }
1763
1764 {
1765 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1766 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1767 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1768 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1769 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1770 }
1771 }
1772
1773 /**
1774 *
1775 * @return non zero if the loop filter can be skipped
1776 */
1777 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
1778 {
1779 const int mb_xy = sl->mb_xy;
1780 int top_xy, left_xy[LEFT_MBS];
1781 int top_type, left_type[LEFT_MBS];
1782 uint8_t *nnz;
1783 uint8_t *nnz_cache;
1784
1785 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1786
1787 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1788 * stuff, I can't imagine that these complex rules are worth it. */
1789
1790 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1791 if (FRAME_MBAFF(h)) {
1792 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1793 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1794 if (sl->mb_y & 1) {
1795 if (left_mb_field_flag != curr_mb_field_flag)
1796 left_xy[LTOP] -= h->mb_stride;
1797 } else {
1798 if (curr_mb_field_flag)
1799 top_xy += h->mb_stride &
1800 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1801 if (left_mb_field_flag != curr_mb_field_flag)
1802 left_xy[LBOT] += h->mb_stride;
1803 }
1804 }
1805
1806 sl->top_mb_xy = top_xy;
1807 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1808 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1809 {
1810 /* For sufficiently low qp, filtering wouldn't do anything.
1811 * This is a conservative estimate: could also check beta_offset
1812 * and more accurate chroma_qp. */
1813 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1814 int qp = h->cur_pic.qscale_table[mb_xy];
1815 if (qp <= qp_thresh &&
1816 (left_xy[LTOP] < 0 ||
1817 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1818 (top_xy < 0 ||
1819 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1820 if (!FRAME_MBAFF(h))
1821 return 1;
1822 if ((left_xy[LTOP] < 0 ||
1823 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1824 (top_xy < h->mb_stride ||
1825 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1826 return 1;
1827 }
1828 }
1829
1830 top_type = h->cur_pic.mb_type[top_xy];
1831 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1832 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1833 if (sl->deblocking_filter == 2) {
1834 if (h->slice_table[top_xy] != sl->slice_num)
1835 top_type = 0;
1836 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
1837 left_type[LTOP] = left_type[LBOT] = 0;
1838 } else {
1839 if (h->slice_table[top_xy] == 0xFFFF)
1840 top_type = 0;
1841 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1842 left_type[LTOP] = left_type[LBOT] = 0;
1843 }
1844 sl->top_type = top_type;
1845 sl->left_type[LTOP] = left_type[LTOP];
1846 sl->left_type[LBOT] = left_type[LBOT];
1847
1848 if (IS_INTRA(mb_type))
1849 return 0;
1850
1851 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1852 top_type, left_type, mb_xy, 0);
1853 if (sl->list_count == 2)
1854 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1855 top_type, left_type, mb_xy, 1);
1856
1857 nnz = h->non_zero_count[mb_xy];
1858 nnz_cache = sl->non_zero_count_cache;
1859 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1860 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1861 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1862 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1863 sl->cbp = h->cbp_table[mb_xy];
1864
1865 if (top_type) {
1866 nnz = h->non_zero_count[top_xy];
1867 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1868 }
1869
1870 if (left_type[LTOP]) {
1871 nnz = h->non_zero_count[left_xy[LTOP]];
1872 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
1873 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
1874 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
1875 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
1876 }
1877
1878 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
1879 * from what the loop filter needs */
1880 if (!CABAC(h) && h->pps.transform_8x8_mode) {
1881 if (IS_8x8DCT(top_type)) {
1882 nnz_cache[4 + 8 * 0] =
1883 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
1884 nnz_cache[6 + 8 * 0] =
1885 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
1886 }
1887 if (IS_8x8DCT(left_type[LTOP])) {
1888 nnz_cache[3 + 8 * 1] =
1889 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
1890 }
1891 if (IS_8x8DCT(left_type[LBOT])) {
1892 nnz_cache[3 + 8 * 3] =
1893 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
1894 }
1895
1896 if (IS_8x8DCT(mb_type)) {
1897 nnz_cache[scan8[0]] =
1898 nnz_cache[scan8[1]] =
1899 nnz_cache[scan8[2]] =
1900 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
1901
1902 nnz_cache[scan8[0 + 4]] =
1903 nnz_cache[scan8[1 + 4]] =
1904 nnz_cache[scan8[2 + 4]] =
1905 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
1906
1907 nnz_cache[scan8[0 + 8]] =
1908 nnz_cache[scan8[1 + 8]] =
1909 nnz_cache[scan8[2 + 8]] =
1910 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
1911
1912 nnz_cache[scan8[0 + 12]] =
1913 nnz_cache[scan8[1 + 12]] =
1914 nnz_cache[scan8[2 + 12]] =
1915 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
1916 }
1917 }
1918
1919 return 0;
1920 }
1921
1922 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
1923 {
1924 uint8_t *dest_y, *dest_cb, *dest_cr;
1925 int linesize, uvlinesize, mb_x, mb_y;
1926 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
1927 const int old_slice_type = sl->slice_type;
1928 const int pixel_shift = h->pixel_shift;
1929 const int block_h = 16 >> h->chroma_y_shift;
1930
1931 if (sl->deblocking_filter) {
1932 for (mb_x = start_x; mb_x < end_x; mb_x++)
1933 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
1934 int mb_xy, mb_type;
1935 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
1936 sl->slice_num = h->slice_table[mb_xy];
1937 mb_type = h->cur_pic.mb_type[mb_xy];
1938 sl->list_count = h->list_counts[mb_xy];
1939
1940 if (FRAME_MBAFF(h))
1941 sl->mb_mbaff =
1942 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
1943
1944 sl->mb_x = mb_x;
1945 sl->mb_y = mb_y;
1946 dest_y = h->cur_pic.f->data[0] +
1947 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
1948 dest_cb = h->cur_pic.f->data[1] +
1949 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1950 mb_y * sl->uvlinesize * block_h;
1951 dest_cr = h->cur_pic.f->data[2] +
1952 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1953 mb_y * sl->uvlinesize * block_h;
1954 // FIXME simplify above
1955
1956 if (MB_FIELD(sl)) {
1957 linesize = sl->mb_linesize = sl->linesize * 2;
1958 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
1959 if (mb_y & 1) { // FIXME move out of this function?
1960 dest_y -= sl->linesize * 15;
1961 dest_cb -= sl->uvlinesize * (block_h - 1);
1962 dest_cr -= sl->uvlinesize * (block_h - 1);
1963 }
1964 } else {
1965 linesize = sl->mb_linesize = sl->linesize;
1966 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
1967 }
1968 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
1969 uvlinesize, 0);
1970 if (fill_filter_caches(h, sl, mb_type))
1971 continue;
1972 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
1973 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
1974
1975 if (FRAME_MBAFF(h)) {
1976 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
1977 linesize, uvlinesize);
1978 } else {
1979 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
1980 dest_cr, linesize, uvlinesize);
1981 }
1982 }
1983 }
1984 sl->slice_type = old_slice_type;
1985 sl->mb_x = end_x;
1986 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
1987 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1988 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1989 }
1990
1991 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
1992 {
1993 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
1994 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
1995 h->cur_pic.mb_type[mb_xy - 1] :
1996 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
1997 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
1998 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1999 }
2000
2001 /**
2002 * Draw edges and report progress for the last MB row.
2003 */
2004 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2005 {
2006 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2007 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2008 int height = 16 << FRAME_MBAFF(h);
2009 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2010
2011 if (sl->deblocking_filter) {
2012 if ((top + height) >= pic_height)
2013 height += deblock_border;
2014 top -= deblock_border;
2015 }
2016
2017 if (top >= pic_height || (top + height) < 0)
2018 return;
2019
2020 height = FFMIN(height, pic_height - top);
2021 if (top < 0) {
2022 height = top + height;
2023 top = 0;
2024 }
2025
2026 ff_h264_draw_horiz_band(h, sl, top, height);
2027
2028 if (h->droppable)
2029 return;
2030
2031 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2032 h->picture_structure == PICT_BOTTOM_FIELD);
2033 }
2034
2035 static void er_add_slice(H264SliceContext *sl,
2036 int startx, int starty,
2037 int endx, int endy, int status)
2038 {
2039 #if CONFIG_ERROR_RESILIENCE
2040 ERContext *er = &sl->er;
2041
2042 if (!sl->h264->enable_er)
2043 return;
2044
2045 er->ref_count = sl->ref_count[0];
2046 ff_er_add_slice(er, startx, starty, endx, endy, status);
2047 #endif
2048 }
2049
2050 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2051 {
2052 H264SliceContext *sl = arg;
2053 const H264Context *h = sl->h264;
2054 int lf_x_start = sl->mb_x;
2055 int ret;
2056
2057 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2058 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2059
2060 ret = alloc_scratch_buffers(sl, sl->linesize);
2061 if (ret < 0)
2062 return ret;
2063
2064 sl->mb_skip_run = -1;
2065
2066 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2067 avctx->codec_id != AV_CODEC_ID_H264 ||
2068 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2069
2070 if (h->pps.cabac) {
2071 /* realign */
2072 align_get_bits(&sl->gb);
2073
2074 /* init cabac */
2075 ff_init_cabac_decoder(&sl->cabac,
2076 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2077 (get_bits_left(&sl->gb) + 7) / 8);
2078
2079 ff_h264_init_cabac_states(h, sl);
2080
2081 for (;;) {
2082 // START_TIMER
2083 int ret = ff_h264_decode_mb_cabac(h, sl);
2084 int eos;
2085 // STOP_TIMER("decode_mb_cabac")
2086
2087 if (ret >= 0)
2088 ff_h264_hl_decode_mb(h, sl);
2089
2090 // FIXME optimal? or let mb_decode decode 16x32 ?
2091 if (ret >= 0 && FRAME_MBAFF(h)) {
2092 sl->mb_y++;
2093
2094 ret = ff_h264_decode_mb_cabac(h, sl);
2095
2096 if (ret >= 0)
2097 ff_h264_hl_decode_mb(h, sl);
2098 sl->mb_y--;
2099 }
2100 eos = get_cabac_terminate(&sl->cabac);
2101
2102 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2103 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2104 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2105 sl->mb_y, ER_MB_END);
2106 if (sl->mb_x >= lf_x_start)
2107 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2108 return 0;
2109 }
2110 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2111 av_log(h->avctx, AV_LOG_ERROR,
2112 "error while decoding MB %d %d, bytestream %td\n",
2113 sl->mb_x, sl->mb_y,
2114 sl->cabac.bytestream_end - sl->cabac.bytestream);
2115 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2116 sl->mb_y, ER_MB_ERROR);
2117 return AVERROR_INVALIDDATA;
2118 }
2119
2120 if (++sl->mb_x >= h->mb_width) {
2121 loop_filter(h, sl, lf_x_start, sl->mb_x);
2122 sl->mb_x = lf_x_start = 0;
2123 decode_finish_row(h, sl);
2124 ++sl->mb_y;
2125 if (FIELD_OR_MBAFF_PICTURE(h)) {
2126 ++sl->mb_y;
2127 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2128 predict_field_decoding_flag(h, sl);
2129 }
2130 }
2131
2132 if (eos || sl->mb_y >= h->mb_height) {
2133 ff_tlog(h->avctx, "slice end %d %d\n",
2134 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2135 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2136 sl->mb_y, ER_MB_END);
2137 if (sl->mb_x > lf_x_start)
2138 loop_filter(h, sl, lf_x_start, sl->mb_x);
2139 return 0;
2140 }
2141 }
2142 } else {
2143 for (;;) {
2144 int ret = ff_h264_decode_mb_cavlc(h, sl);
2145
2146 if (ret >= 0)
2147 ff_h264_hl_decode_mb(h, sl);
2148
2149 // FIXME optimal? or let mb_decode decode 16x32 ?
2150 if (ret >= 0 && FRAME_MBAFF(h)) {
2151 sl->mb_y++;
2152 ret = ff_h264_decode_mb_cavlc(h, sl);
2153
2154 if (ret >= 0)
2155 ff_h264_hl_decode_mb(h, sl);
2156 sl->mb_y--;
2157 }
2158
2159 if (ret < 0) {
2160 av_log(h->avctx, AV_LOG_ERROR,
2161 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2162 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2163 sl->mb_y, ER_MB_ERROR);
2164 return ret;
2165 }
2166
2167 if (++sl->mb_x >= h->mb_width) {
2168 loop_filter(h, sl, lf_x_start, sl->mb_x);
2169 sl->mb_x = lf_x_start = 0;
2170 decode_finish_row(h, sl);
2171 ++sl->mb_y;
2172 if (FIELD_OR_MBAFF_PICTURE(h)) {
2173 ++sl->mb_y;
2174 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2175 predict_field_decoding_flag(h, sl);
2176 }
2177 if (sl->mb_y >= h->mb_height) {
2178 ff_tlog(h->avctx, "slice end %d %d\n",
2179 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2180
2181 if (get_bits_left(&sl->gb) == 0) {
2182 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2183 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2184
2185 return 0;
2186 } else {
2187 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2188 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2189
2190 return AVERROR_INVALIDDATA;
2191 }
2192 }
2193 }
2194
2195 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2196 ff_tlog(h->avctx, "slice end %d %d\n",
2197 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2198
2199 if (get_bits_left(&sl->gb) == 0) {
2200 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2201 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2202 if (sl->mb_x > lf_x_start)
2203 loop_filter(h, sl, lf_x_start, sl->mb_x);
2204
2205 return 0;
2206 } else {
2207 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2208 sl->mb_y, ER_MB_ERROR);
2209
2210 return AVERROR_INVALIDDATA;
2211 }
2212 }
2213 }
2214 }
2215 }
2216
2217 /**
2218 * Call decode_slice() for each context.
2219 *
2220 * @param h h264 master context
2221 * @param context_count number of contexts to execute
2222 */
2223 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2224 {
2225 AVCodecContext *const avctx = h->avctx;
2226 H264SliceContext *sl;
2227 int i;
2228
2229 if (h->avctx->hwaccel)
2230 return 0;
2231 if (context_count == 1) {
2232 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2233 h->mb_y = h->slice_ctx[0].mb_y;
2234 return ret;
2235 } else {
2236 for (i = 1; i < context_count; i++) {
2237 sl = &h->slice_ctx[i];
2238 sl->er.error_count = 0;
2239 }
2240
2241 avctx->execute(avctx, decode_slice, h->slice_ctx,
2242 NULL, context_count, sizeof(h->slice_ctx[0]));
2243
2244 /* pull back stuff from slices to master context */
2245 sl = &h->slice_ctx[context_count - 1];
2246 h->mb_y = sl->mb_y;
2247 for (i = 1; i < context_count; i++)
2248 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
2249 }
2250
2251 return 0;
2252 }