2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg 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.
12 * FFmpeg 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.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/avassert.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/timer.h"
33 #include "cabac_functions.h"
34 #include "error_resilience.h"
38 #include "h264chroma.h"
39 #include "h264_mvpred.h"
42 #include "mpegutils.h"
43 #include "rectangle.h"
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,
51 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
55 static const uint8_t div6[QP_MAX_NUM + 1] = {
56 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
57 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
58 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
59 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
63 static const uint8_t field_scan[16+1] = {
64 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
65 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
66 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
67 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
70 static const uint8_t field_scan8x8[64+1] = {
71 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
72 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
73 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
74 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
75 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
76 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
77 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
78 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
79 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
80 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
81 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
82 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
83 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
84 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
85 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
86 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
89 static const uint8_t field_scan8x8_cavlc[64+1] = {
90 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
91 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
92 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
93 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
94 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
95 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
96 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
97 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
98 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
99 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
100 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
101 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
102 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
103 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
104 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
105 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
108 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
109 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
110 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
111 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
112 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
113 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
114 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
115 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
116 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
117 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
118 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
119 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
120 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
121 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
122 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
123 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
124 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
125 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
128 static const uint8_t dequant4_coeff_init[6][3] = {
137 static const uint8_t dequant8_coeff_init_scan[16] = {
138 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
141 static const uint8_t dequant8_coeff_init[6][6] = {
142 { 20, 18, 32, 19, 25, 24 },
143 { 22, 19, 35, 21, 28, 26 },
144 { 26, 23, 42, 24, 33, 31 },
145 { 28, 25, 45, 26, 35, 33 },
146 { 32, 28, 51, 30, 40, 38 },
147 { 36, 32, 58, 34, 46, 43 },
151 static void release_unused_pictures(H264Context *h, int remove_current)
155 /* release non reference frames */
156 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
157 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
158 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
159 ff_h264_unref_picture(h, &h->DPB[i]);
164 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
166 const H264Context *h = sl->h264;
167 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
169 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
170 // edge emu needs blocksize + filter length - 1
171 // (= 21x21 for h264)
172 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
174 av_fast_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
175 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
176 av_fast_malloc(&sl->top_borders[1], &sl->top_borders_allocated[1],
177 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
179 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
180 !sl->top_borders[0] || !sl->top_borders[1]) {
181 av_freep(&sl->bipred_scratchpad);
182 av_freep(&sl->edge_emu_buffer);
183 av_freep(&sl->top_borders[0]);
184 av_freep(&sl->top_borders[1]);
186 sl->bipred_scratchpad_allocated = 0;
187 sl->edge_emu_buffer_allocated = 0;
188 sl->top_borders_allocated[0] = 0;
189 sl->top_borders_allocated[1] = 0;
190 return AVERROR(ENOMEM);
196 static int init_table_pools(H264Context *h)
198 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
199 const int mb_array_size = h->mb_stride * h->mb_height;
200 const int b4_stride = h->mb_width * 4 + 1;
201 const int b4_array_size = b4_stride * h->mb_height * 4;
203 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
205 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
206 sizeof(uint32_t), av_buffer_allocz);
207 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
208 sizeof(int16_t), av_buffer_allocz);
209 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
211 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
212 !h->ref_index_pool) {
213 av_buffer_pool_uninit(&h->qscale_table_pool);
214 av_buffer_pool_uninit(&h->mb_type_pool);
215 av_buffer_pool_uninit(&h->motion_val_pool);
216 av_buffer_pool_uninit(&h->ref_index_pool);
217 return AVERROR(ENOMEM);
223 static int alloc_picture(H264Context *h, H264Picture *pic)
227 av_assert0(!pic->f->data[0]);
230 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
231 AV_GET_BUFFER_FLAG_REF : 0);
235 pic->crop = h->sps.crop;
236 pic->crop_top = h->sps.crop_top;
237 pic->crop_left= h->sps.crop_left;
239 if (h->avctx->hwaccel) {
240 const AVHWAccel *hwaccel = h->avctx->hwaccel;
241 av_assert0(!pic->hwaccel_picture_private);
242 if (hwaccel->frame_priv_data_size) {
243 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
244 if (!pic->hwaccel_priv_buf)
245 return AVERROR(ENOMEM);
246 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
249 if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & CODEC_FLAG_GRAY && pic->f->data[2]) {
250 int h_chroma_shift, v_chroma_shift;
251 av_pix_fmt_get_chroma_sub_sample(pic->f->format,
252 &h_chroma_shift, &v_chroma_shift);
254 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
255 memset(pic->f->data[1] + pic->f->linesize[1]*i,
256 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
257 memset(pic->f->data[2] + pic->f->linesize[2]*i,
258 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
262 if (!h->qscale_table_pool) {
263 ret = init_table_pools(h);
268 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
269 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
270 if (!pic->qscale_table_buf || !pic->mb_type_buf)
273 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
274 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
276 for (i = 0; i < 2; i++) {
277 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
278 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
279 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
282 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
283 pic->ref_index[i] = pic->ref_index_buf[i]->data;
288 ff_h264_unref_picture(h, pic);
289 return (ret < 0) ? ret : AVERROR(ENOMEM);
292 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
299 static int find_unused_picture(H264Context *h)
303 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
304 if (pic_is_unused(h, &h->DPB[i]))
307 if (i == H264_MAX_PICTURE_COUNT)
308 return AVERROR_INVALIDDATA;
314 static void init_dequant8_coeff_table(H264Context *h)
317 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
319 for (i = 0; i < 6; i++) {
320 h->dequant8_coeff[i] = h->dequant8_buffer[i];
321 for (j = 0; j < i; j++)
322 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
323 64 * sizeof(uint8_t))) {
324 h->dequant8_coeff[i] = h->dequant8_buffer[j];
330 for (q = 0; q < max_qp + 1; q++) {
333 for (x = 0; x < 64; x++)
334 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
335 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
336 h->pps.scaling_matrix8[i][x]) << shift;
341 static void init_dequant4_coeff_table(H264Context *h)
344 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
345 for (i = 0; i < 6; i++) {
346 h->dequant4_coeff[i] = h->dequant4_buffer[i];
347 for (j = 0; j < i; j++)
348 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
349 16 * sizeof(uint8_t))) {
350 h->dequant4_coeff[i] = h->dequant4_buffer[j];
356 for (q = 0; q < max_qp + 1; q++) {
357 int shift = div6[q] + 2;
359 for (x = 0; x < 16; x++)
360 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
361 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
362 h->pps.scaling_matrix4[i][x]) << shift;
367 void ff_h264_init_dequant_tables(H264Context *h)
370 init_dequant4_coeff_table(h);
371 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
373 if (h->pps.transform_8x8_mode)
374 init_dequant8_coeff_table(h);
375 if (h->sps.transform_bypass) {
376 for (i = 0; i < 6; i++)
377 for (x = 0; x < 16; x++)
378 h->dequant4_coeff[i][0][x] = 1 << 6;
379 if (h->pps.transform_8x8_mode)
380 for (i = 0; i < 6; i++)
381 for (x = 0; x < 64; x++)
382 h->dequant8_coeff[i][0][x] = 1 << 6;
386 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
388 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
389 (((pic) && (pic) >= (old_ctx)->DPB && \
390 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
391 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
393 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
394 H264Context *new_base,
395 H264Context *old_base)
399 for (i = 0; i < count; i++) {
400 av_assert1((IN_RANGE(from[i], old_base, 1) ||
401 IN_RANGE(from[i], old_base->DPB,
402 H264_MAX_PICTURE_COUNT) ||
404 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
408 static int copy_parameter_set(void **to, void **from, int count, int size)
412 for (i = 0; i < count; i++) {
413 if (to[i] && !from[i]) {
415 } else if (from[i] && !to[i]) {
416 to[i] = av_malloc(size);
418 return AVERROR(ENOMEM);
422 memcpy(to[i], from[i], size);
428 #define copy_fields(to, from, start_field, end_field) \
429 memcpy(&(to)->start_field, &(from)->start_field, \
430 (char *)&(to)->end_field - (char *)&(to)->start_field)
432 static int h264_slice_header_init(H264Context *h);
434 int ff_h264_update_thread_context(AVCodecContext *dst,
435 const AVCodecContext *src)
437 H264Context *h = dst->priv_data, *h1 = src->priv_data;
438 int inited = h->context_initialized, err = 0;
446 (h->width != h1->width ||
447 h->height != h1->height ||
448 h->mb_width != h1->mb_width ||
449 h->mb_height != h1->mb_height ||
450 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
451 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
452 h->sps.colorspace != h1->sps.colorspace)) {
457 /* copy block_offset since frame_start may not be called */
458 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
461 if ((ret = copy_parameter_set((void **)h->sps_buffers,
462 (void **)h1->sps_buffers,
463 MAX_SPS_COUNT, sizeof(SPS))) < 0)
466 if ((ret = copy_parameter_set((void **)h->pps_buffers,
467 (void **)h1->pps_buffers,
468 MAX_PPS_COUNT, sizeof(PPS))) < 0)
472 if (need_reinit || !inited) {
473 h->width = h1->width;
474 h->height = h1->height;
475 h->mb_height = h1->mb_height;
476 h->mb_width = h1->mb_width;
477 h->mb_num = h1->mb_num;
478 h->mb_stride = h1->mb_stride;
479 h->b_stride = h1->b_stride;
481 if (h->context_initialized || h1->context_initialized) {
482 if ((err = h264_slice_header_init(h)) < 0) {
483 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
487 /* copy block_offset since frame_start may not be called */
488 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
491 h->avctx->coded_height = h1->avctx->coded_height;
492 h->avctx->coded_width = h1->avctx->coded_width;
493 h->avctx->width = h1->avctx->width;
494 h->avctx->height = h1->avctx->height;
495 h->coded_picture_number = h1->coded_picture_number;
496 h->first_field = h1->first_field;
497 h->picture_structure = h1->picture_structure;
498 h->droppable = h1->droppable;
499 h->low_delay = h1->low_delay;
501 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
502 ff_h264_unref_picture(h, &h->DPB[i]);
503 if (h1->DPB[i].f->buf[0] &&
504 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
508 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
509 ff_h264_unref_picture(h, &h->cur_pic);
510 if (h1->cur_pic.f->buf[0]) {
511 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
516 h->enable_er = h1->enable_er;
517 h->workaround_bugs = h1->workaround_bugs;
518 h->low_delay = h1->low_delay;
519 h->droppable = h1->droppable;
521 // extradata/NAL handling
522 h->is_avc = h1->is_avc;
523 h->nal_length_size = h1->nal_length_size;
524 h->x264_build = h1->x264_build;
526 // Dequantization matrices
527 // FIXME these are big - can they be only copied when PPS changes?
528 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
530 for (i = 0; i < 6; i++)
531 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
532 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
534 for (i = 0; i < 6; i++)
535 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
536 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
538 h->dequant_coeff_pps = h1->dequant_coeff_pps;
541 copy_fields(h, h1, poc_lsb, default_ref_list);
544 copy_fields(h, h1, short_ref, current_slice);
546 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
547 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
548 copy_picture_range(h->delayed_pic, h1->delayed_pic,
549 MAX_DELAYED_PIC_COUNT + 2, h, h1);
551 h->frame_recovered = h1->frame_recovered;
557 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
558 h->prev_poc_msb = h->poc_msb;
559 h->prev_poc_lsb = h->poc_lsb;
561 h->prev_frame_num_offset = h->frame_num_offset;
562 h->prev_frame_num = h->frame_num;
564 h->recovery_frame = h1->recovery_frame;
569 static int h264_frame_start(H264Context *h)
573 const int pixel_shift = h->pixel_shift;
575 1<<(h->sps.bit_depth_luma-1),
576 1<<(h->sps.bit_depth_chroma-1),
577 1<<(h->sps.bit_depth_chroma-1),
581 if (!ff_thread_can_start_frame(h->avctx)) {
582 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
586 release_unused_pictures(h, 1);
587 h->cur_pic_ptr = NULL;
589 i = find_unused_picture(h);
591 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
596 pic->reference = h->droppable ? 0 : h->picture_structure;
597 pic->f->coded_picture_number = h->coded_picture_number++;
598 pic->field_picture = h->picture_structure != PICT_FRAME;
601 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
603 * See decode_nal_units().
605 pic->f->key_frame = 0;
608 pic->invalid_gap = 0;
609 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
611 if ((ret = alloc_picture(h, pic)) < 0)
613 if(!h->frame_recovered && !h->avctx->hwaccel &&
614 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
615 avpriv_color_frame(pic->f, c);
617 h->cur_pic_ptr = pic;
618 ff_h264_unref_picture(h, &h->cur_pic);
619 if (CONFIG_ERROR_RESILIENCE) {
620 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
623 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
626 for (i = 0; i < h->nb_slice_ctx; i++) {
627 h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
628 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
631 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
632 ff_er_frame_start(&h->slice_ctx[0].er);
633 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
634 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
637 for (i = 0; i < 16; i++) {
638 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
639 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
641 for (i = 0; i < 16; i++) {
642 h->block_offset[16 + i] =
643 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
644 h->block_offset[48 + 16 + i] =
645 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
648 /* We mark the current picture as non-reference after allocating it, so
649 * that if we break out due to an error it can be released automatically
650 * in the next ff_mpv_frame_start().
652 h->cur_pic_ptr->reference = 0;
654 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
656 h->next_output_pic = NULL;
658 assert(h->cur_pic_ptr->long_ref == 0);
663 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
665 uint8_t *src_cb, uint8_t *src_cr,
666 int linesize, int uvlinesize,
671 const int pixel_shift = h->pixel_shift;
672 int chroma444 = CHROMA444(h);
673 int chroma422 = CHROMA422(h);
676 src_cb -= uvlinesize;
677 src_cr -= uvlinesize;
679 if (!simple && FRAME_MBAFF(h)) {
682 top_border = sl->top_borders[0][sl->mb_x];
683 AV_COPY128(top_border, src_y + 15 * linesize);
685 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
686 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
689 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
690 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
691 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
692 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
694 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
695 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
697 } else if (chroma422) {
699 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
700 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
702 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
703 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
707 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
708 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
710 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
711 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
716 } else if (MB_MBAFF(sl)) {
722 top_border = sl->top_borders[top_idx][sl->mb_x];
723 /* There are two lines saved, the line above the top macroblock
724 * of a pair, and the line above the bottom macroblock. */
725 AV_COPY128(top_border, src_y + 16 * linesize);
727 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
729 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
732 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
733 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
734 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
735 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
737 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
738 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
740 } else if (chroma422) {
742 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
743 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
745 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
746 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
750 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
751 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
753 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
754 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
761 * Initialize implicit_weight table.
762 * @param field 0/1 initialize the weight for interlaced MBAFF
763 * -1 initializes the rest
765 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
767 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
769 for (i = 0; i < 2; i++) {
770 sl->luma_weight_flag[i] = 0;
771 sl->chroma_weight_flag[i] = 0;
775 if (h->picture_structure == PICT_FRAME) {
776 cur_poc = h->cur_pic_ptr->poc;
778 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
780 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
781 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
783 sl->use_weight_chroma = 0;
787 ref_count0 = sl->ref_count[0];
788 ref_count1 = sl->ref_count[1];
790 cur_poc = h->cur_pic_ptr->field_poc[field];
792 ref_count0 = 16 + 2 * sl->ref_count[0];
793 ref_count1 = 16 + 2 * sl->ref_count[1];
797 sl->use_weight_chroma = 2;
798 sl->luma_log2_weight_denom = 5;
799 sl->chroma_log2_weight_denom = 5;
801 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
802 int poc0 = sl->ref_list[0][ref0].poc;
803 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
805 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
806 int poc1 = sl->ref_list[1][ref1].poc;
807 int td = av_clip_int8(poc1 - poc0);
809 int tb = av_clip_int8(cur_poc - poc0);
810 int tx = (16384 + (FFABS(td) >> 1)) / td;
811 int dist_scale_factor = (tb * tx + 32) >> 8;
812 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
813 w = 64 - dist_scale_factor;
817 sl->implicit_weight[ref0][ref1][0] =
818 sl->implicit_weight[ref0][ref1][1] = w;
820 sl->implicit_weight[ref0][ref1][field] = w;
827 * initialize scan tables
829 static void init_scan_tables(H264Context *h)
832 for (i = 0; i < 16; i++) {
833 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
834 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
835 h->field_scan[i] = TRANSPOSE(field_scan[i]);
838 for (i = 0; i < 64; i++) {
839 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
840 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
841 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
842 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
843 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
846 if (h->sps.transform_bypass) { // FIXME same ugly
847 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
848 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
849 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
850 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
851 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
852 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
854 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
855 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
856 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
857 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
858 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
859 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
863 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
865 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
866 CONFIG_H264_VAAPI_HWACCEL + \
867 (CONFIG_H264_VDA_HWACCEL * 2) + \
868 CONFIG_H264_VDPAU_HWACCEL)
869 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
870 const enum AVPixelFormat *choices = pix_fmts;
873 switch (h->sps.bit_depth_luma) {
876 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
877 *fmt++ = AV_PIX_FMT_GBRP9;
879 *fmt++ = AV_PIX_FMT_YUV444P9;
880 } else if (CHROMA422(h))
881 *fmt++ = AV_PIX_FMT_YUV422P9;
883 *fmt++ = AV_PIX_FMT_YUV420P9;
887 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
888 *fmt++ = AV_PIX_FMT_GBRP10;
890 *fmt++ = AV_PIX_FMT_YUV444P10;
891 } else if (CHROMA422(h))
892 *fmt++ = AV_PIX_FMT_YUV422P10;
894 *fmt++ = AV_PIX_FMT_YUV420P10;
898 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
899 *fmt++ = AV_PIX_FMT_GBRP12;
901 *fmt++ = AV_PIX_FMT_YUV444P12;
902 } else if (CHROMA422(h))
903 *fmt++ = AV_PIX_FMT_YUV422P12;
905 *fmt++ = AV_PIX_FMT_YUV420P12;
909 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
910 *fmt++ = AV_PIX_FMT_GBRP14;
912 *fmt++ = AV_PIX_FMT_YUV444P14;
913 } else if (CHROMA422(h))
914 *fmt++ = AV_PIX_FMT_YUV422P14;
916 *fmt++ = AV_PIX_FMT_YUV420P14;
919 #if CONFIG_H264_VDPAU_HWACCEL
920 *fmt++ = AV_PIX_FMT_VDPAU;
923 if (h->avctx->colorspace == AVCOL_SPC_RGB)
924 *fmt++ = AV_PIX_FMT_GBRP;
925 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
926 *fmt++ = AV_PIX_FMT_YUVJ444P;
928 *fmt++ = AV_PIX_FMT_YUV444P;
929 } else if (CHROMA422(h)) {
930 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
931 *fmt++ = AV_PIX_FMT_YUVJ422P;
933 *fmt++ = AV_PIX_FMT_YUV422P;
935 #if CONFIG_H264_DXVA2_HWACCEL
936 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
938 #if CONFIG_H264_D3D11VA_HWACCEL
939 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
941 #if CONFIG_H264_VAAPI_HWACCEL
942 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
944 #if CONFIG_H264_VDA_HWACCEL
945 *fmt++ = AV_PIX_FMT_VDA_VLD;
946 *fmt++ = AV_PIX_FMT_VDA;
948 if (h->avctx->codec->pix_fmts)
949 choices = h->avctx->codec->pix_fmts;
950 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
951 *fmt++ = AV_PIX_FMT_YUVJ420P;
953 *fmt++ = AV_PIX_FMT_YUV420P;
957 av_log(h->avctx, AV_LOG_ERROR,
958 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
959 return AVERROR_INVALIDDATA;
962 *fmt = AV_PIX_FMT_NONE;
964 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
965 if (choices[i] == h->avctx->pix_fmt && !force_callback)
967 return ff_thread_get_format(h->avctx, choices);
970 /* export coded and cropped frame dimensions to AVCodecContext */
971 static int init_dimensions(H264Context *h)
973 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
974 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
975 int crop_present = h->sps.crop_left || h->sps.crop_top ||
976 h->sps.crop_right || h->sps.crop_bottom;
977 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
978 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
980 /* handle container cropping */
982 FFALIGN(h->avctx->width, 16) == h->width &&
983 FFALIGN(h->avctx->height, 16) == h->height) {
984 width = h->avctx->width;
985 height = h->avctx->height;
988 if (width <= 0 || height <= 0) {
989 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
991 if (h->avctx->err_recognition & AV_EF_EXPLODE)
992 return AVERROR_INVALIDDATA;
994 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1005 h->avctx->coded_width = h->width;
1006 h->avctx->coded_height = h->height;
1007 h->avctx->width = width;
1008 h->avctx->height = height;
1013 static int h264_slice_header_init(H264Context *h)
1015 int nb_slices = (HAVE_THREADS &&
1016 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1017 h->avctx->thread_count : 1;
1020 ff_set_sar(h->avctx, h->sps.sar);
1021 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1022 &h->chroma_x_shift, &h->chroma_y_shift);
1024 if (h->sps.timing_info_present_flag) {
1025 int64_t den = h->sps.time_scale;
1026 if (h->x264_build < 44U)
1028 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1029 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1032 ff_h264_free_tables(h);
1035 h->prev_interlaced_frame = 1;
1037 init_scan_tables(h);
1038 ret = ff_h264_alloc_tables(h);
1040 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1044 if (h->avctx->codec &&
1045 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU &&
1046 (h->sps.bit_depth_luma != 8 || h->sps.chroma_format_idc > 1)) {
1047 av_log(h->avctx, AV_LOG_ERROR,
1048 "VDPAU decoding does not support video colorspace.\n");
1049 ret = AVERROR_INVALIDDATA;
1053 if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 14 ||
1054 h->sps.bit_depth_luma == 11 || h->sps.bit_depth_luma == 13
1056 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
1057 h->sps.bit_depth_luma);
1058 ret = AVERROR_INVALIDDATA;
1062 h->cur_bit_depth_luma =
1063 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
1064 h->cur_chroma_format_idc = h->sps.chroma_format_idc;
1065 h->pixel_shift = h->sps.bit_depth_luma > 8;
1066 h->chroma_format_idc = h->sps.chroma_format_idc;
1067 h->bit_depth_luma = h->sps.bit_depth_luma;
1069 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
1070 h->sps.chroma_format_idc);
1071 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
1072 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
1073 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
1074 h->sps.chroma_format_idc);
1075 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
1077 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1080 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1082 max_slices = H264_MAX_THREADS;
1083 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1084 " reducing to %d\n", nb_slices, max_slices);
1085 nb_slices = max_slices;
1087 h->slice_context_count = nb_slices;
1089 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1090 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
1092 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1096 for (i = 0; i < h->slice_context_count; i++) {
1097 H264SliceContext *sl = &h->slice_ctx[i];
1100 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1101 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1102 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1104 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1105 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1111 h->context_initialized = 1;
1115 ff_h264_free_tables(h);
1116 h->context_initialized = 0;
1120 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1123 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1124 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1125 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1132 * Decode a slice header.
1133 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1135 * @param h h264context
1137 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1139 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1141 unsigned int first_mb_in_slice;
1142 unsigned int pps_id;
1144 unsigned int slice_type, tmp, i, j;
1145 int last_pic_structure, last_pic_droppable;
1147 int needs_reinit = 0;
1148 int field_pic_flag, bottom_field_flag;
1149 int first_slice = sl == h->slice_ctx && !h->current_slice;
1150 int frame_num, picture_structure, droppable;
1151 int mb_aff_frame, last_mb_aff_frame;
1154 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1155 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1157 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1159 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1160 if (h->current_slice) {
1161 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
1162 ff_h264_field_end(h, h->slice_ctx, 1);
1163 h->current_slice = 0;
1164 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == NAL_IDR_SLICE) {
1165 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
1166 ff_h264_field_end(h, h->slice_ctx, 1);
1167 h->current_slice = 0;
1168 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1169 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1170 h->cur_pic_ptr = NULL;
1172 return AVERROR_INVALIDDATA;
1175 if (!h->first_field) {
1176 if (h->cur_pic_ptr && !h->droppable) {
1177 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1178 h->picture_structure == PICT_BOTTOM_FIELD);
1180 h->cur_pic_ptr = NULL;
1184 slice_type = get_ue_golomb_31(&sl->gb);
1185 if (slice_type > 9) {
1186 av_log(h->avctx, AV_LOG_ERROR,
1187 "slice type %d too large at %d\n",
1188 slice_type, first_mb_in_slice);
1189 return AVERROR_INVALIDDATA;
1191 if (slice_type > 4) {
1193 sl->slice_type_fixed = 1;
1195 sl->slice_type_fixed = 0;
1197 slice_type = golomb_to_pict_type[slice_type];
1199 sl->slice_type = slice_type;
1200 sl->slice_type_nos = slice_type & 3;
1202 if (h->nal_unit_type == NAL_IDR_SLICE &&
1203 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1204 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1205 return AVERROR_INVALIDDATA;
1209 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1210 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1211 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1212 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1213 h->avctx->skip_frame >= AVDISCARD_ALL) {
1214 return SLICE_SKIPED;
1217 // to make a few old functions happy, it's wrong though
1218 h->pict_type = sl->slice_type;
1220 pps_id = get_ue_golomb(&sl->gb);
1221 if (pps_id >= MAX_PPS_COUNT) {
1222 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1223 return AVERROR_INVALIDDATA;
1225 if (!h->pps_buffers[pps_id]) {
1226 av_log(h->avctx, AV_LOG_ERROR,
1227 "non-existing PPS %u referenced\n",
1229 return AVERROR_INVALIDDATA;
1231 if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
1232 av_log(h->avctx, AV_LOG_ERROR,
1233 "PPS change from %d to %d forbidden\n",
1234 h->au_pps_id, pps_id);
1235 return AVERROR_INVALIDDATA;
1238 pps = h->pps_buffers[pps_id];
1240 if (!h->sps_buffers[pps->sps_id]) {
1241 av_log(h->avctx, AV_LOG_ERROR,
1242 "non-existing SPS %u referenced\n",
1244 return AVERROR_INVALIDDATA;
1247 h->pps = *h->pps_buffers[pps_id];
1249 if (pps->sps_id != h->sps.sps_id ||
1250 pps->sps_id != h->current_sps_id ||
1251 h->sps_buffers[pps->sps_id]->new) {
1254 av_log(h->avctx, AV_LOG_ERROR,
1255 "SPS changed in the middle of the frame\n");
1256 return AVERROR_INVALIDDATA;
1259 h->sps = *h->sps_buffers[h->pps.sps_id];
1261 if (h->mb_width != h->sps.mb_width ||
1262 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1263 h->cur_bit_depth_luma != h->sps.bit_depth_luma ||
1264 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1268 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1269 h->chroma_format_idc != h->sps.chroma_format_idc)
1272 if (h->flags & CODEC_FLAG_LOW_DELAY ||
1273 (h->sps.bitstream_restriction_flag &&
1274 !h->sps.num_reorder_frames)) {
1275 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1276 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1277 "Reenabling low delay requires a codec flush.\n");
1282 if (h->avctx->has_b_frames < 2)
1283 h->avctx->has_b_frames = !h->low_delay;
1287 h->avctx->profile = ff_h264_get_profile(&h->sps);
1288 h->avctx->level = h->sps.level_idc;
1289 h->avctx->refs = h->sps.ref_frame_count;
1291 must_reinit = (h->context_initialized &&
1292 ( 16*h->sps.mb_width != h->avctx->coded_width
1293 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1294 || h->cur_bit_depth_luma != h->sps.bit_depth_luma
1295 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1296 || h->mb_width != h->sps.mb_width
1297 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1299 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1300 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1303 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1306 h->mb_width = h->sps.mb_width;
1307 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1308 h->mb_num = h->mb_width * h->mb_height;
1309 h->mb_stride = h->mb_width + 1;
1311 h->b_stride = h->mb_width * 4;
1313 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1315 h->width = 16 * h->mb_width;
1316 h->height = 16 * h->mb_height;
1318 ret = init_dimensions(h);
1322 if (h->sps.video_signal_type_present_flag) {
1323 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1325 if (h->sps.colour_description_present_flag) {
1326 if (h->avctx->colorspace != h->sps.colorspace)
1328 h->avctx->color_primaries = h->sps.color_primaries;
1329 h->avctx->color_trc = h->sps.color_trc;
1330 h->avctx->colorspace = h->sps.colorspace;
1334 if (h->context_initialized &&
1335 (must_reinit || needs_reinit)) {
1336 if (sl != h->slice_ctx) {
1337 av_log(h->avctx, AV_LOG_ERROR,
1338 "changing width %d -> %d / height %d -> %d on "
1340 h->width, h->avctx->coded_width,
1341 h->height, h->avctx->coded_height,
1342 h->current_slice + 1);
1343 return AVERROR_INVALIDDATA;
1346 av_assert1(first_slice);
1348 ff_h264_flush_change(h);
1350 if ((ret = get_pixel_format(h, 1)) < 0)
1352 h->avctx->pix_fmt = ret;
1354 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1355 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1357 if ((ret = h264_slice_header_init(h)) < 0) {
1358 av_log(h->avctx, AV_LOG_ERROR,
1359 "h264_slice_header_init() failed\n");
1363 if (!h->context_initialized) {
1364 if (sl != h->slice_ctx) {
1365 av_log(h->avctx, AV_LOG_ERROR,
1366 "Cannot (re-)initialize context during parallel decoding.\n");
1367 return AVERROR_PATCHWELCOME;
1370 if ((ret = get_pixel_format(h, 1)) < 0)
1372 h->avctx->pix_fmt = ret;
1374 if ((ret = h264_slice_header_init(h)) < 0) {
1375 av_log(h->avctx, AV_LOG_ERROR,
1376 "h264_slice_header_init() failed\n");
1381 if (first_slice && h->dequant_coeff_pps != pps_id) {
1382 h->dequant_coeff_pps = pps_id;
1383 ff_h264_init_dequant_tables(h);
1386 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1388 if (h->frame_num != frame_num) {
1389 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1390 h->frame_num, frame_num);
1391 return AVERROR_INVALIDDATA;
1397 last_mb_aff_frame = h->mb_aff_frame;
1398 last_pic_structure = h->picture_structure;
1399 last_pic_droppable = h->droppable;
1400 droppable = h->nal_ref_idc == 0;
1401 if (h->sps.frame_mbs_only_flag) {
1402 picture_structure = PICT_FRAME;
1404 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1405 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1408 field_pic_flag = get_bits1(&sl->gb);
1410 if (field_pic_flag) {
1411 bottom_field_flag = get_bits1(&sl->gb);
1412 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1414 picture_structure = PICT_FRAME;
1415 mb_aff_frame = h->sps.mb_aff;
1418 if (h->current_slice) {
1419 if (last_pic_structure != picture_structure ||
1420 last_pic_droppable != droppable ||
1421 last_mb_aff_frame != mb_aff_frame) {
1422 av_log(h->avctx, AV_LOG_ERROR,
1423 "Changing field mode (%d -> %d) between slices is not allowed\n",
1424 last_pic_structure, h->picture_structure);
1425 return AVERROR_INVALIDDATA;
1426 } else if (!h->cur_pic_ptr) {
1427 av_log(h->avctx, AV_LOG_ERROR,
1428 "unset cur_pic_ptr on slice %d\n",
1429 h->current_slice + 1);
1430 return AVERROR_INVALIDDATA;
1434 h->picture_structure = picture_structure;
1435 h->droppable = droppable;
1436 h->frame_num = frame_num;
1437 h->mb_aff_frame = mb_aff_frame;
1438 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1440 if (h->current_slice == 0) {
1441 /* Shorten frame num gaps so we don't have to allocate reference
1442 * frames just to throw them away */
1443 if (h->frame_num != h->prev_frame_num) {
1444 int unwrap_prev_frame_num = h->prev_frame_num;
1445 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1447 if (unwrap_prev_frame_num > h->frame_num)
1448 unwrap_prev_frame_num -= max_frame_num;
1450 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1451 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1452 if (unwrap_prev_frame_num < 0)
1453 unwrap_prev_frame_num += max_frame_num;
1455 h->prev_frame_num = unwrap_prev_frame_num;
1459 /* See if we have a decoded first field looking for a pair...
1460 * Here, we're using that to see if we should mark previously
1461 * decode frames as "finished".
1462 * We have to do that before the "dummy" in-between frame allocation,
1463 * since that can modify h->cur_pic_ptr. */
1464 if (h->first_field) {
1465 av_assert0(h->cur_pic_ptr);
1466 av_assert0(h->cur_pic_ptr->f->buf[0]);
1467 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1469 /* Mark old field/frame as completed */
1470 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1471 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1472 last_pic_structure == PICT_BOTTOM_FIELD);
1475 /* figure out if we have a complementary field pair */
1476 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1477 /* Previous field is unmatched. Don't display it, but let it
1478 * remain for reference if marked as such. */
1479 if (last_pic_structure != PICT_FRAME) {
1480 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1481 last_pic_structure == PICT_TOP_FIELD);
1484 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1485 /* This and previous field were reference, but had
1486 * different frame_nums. Consider this field first in
1487 * pair. Throw away previous field except for reference
1489 if (last_pic_structure != PICT_FRAME) {
1490 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1491 last_pic_structure == PICT_TOP_FIELD);
1494 /* Second field in complementary pair */
1495 if (!((last_pic_structure == PICT_TOP_FIELD &&
1496 h->picture_structure == PICT_BOTTOM_FIELD) ||
1497 (last_pic_structure == PICT_BOTTOM_FIELD &&
1498 h->picture_structure == PICT_TOP_FIELD))) {
1499 av_log(h->avctx, AV_LOG_ERROR,
1500 "Invalid field mode combination %d/%d\n",
1501 last_pic_structure, h->picture_structure);
1502 h->picture_structure = last_pic_structure;
1503 h->droppable = last_pic_droppable;
1504 return AVERROR_INVALIDDATA;
1505 } else if (last_pic_droppable != h->droppable) {
1506 avpriv_request_sample(h->avctx,
1507 "Found reference and non-reference fields in the same frame, which");
1508 h->picture_structure = last_pic_structure;
1509 h->droppable = last_pic_droppable;
1510 return AVERROR_PATCHWELCOME;
1516 while (h->frame_num != h->prev_frame_num && !h->first_field &&
1517 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1518 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1519 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1520 h->frame_num, h->prev_frame_num);
1521 if (!h->sps.gaps_in_frame_num_allowed_flag)
1522 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1523 h->last_pocs[i] = INT_MIN;
1524 ret = h264_frame_start(h);
1530 h->prev_frame_num++;
1531 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1532 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1533 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1534 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1535 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1536 ret = ff_generate_sliding_window_mmcos(h, 1);
1537 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1539 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1540 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1542 /* Error concealment: If a ref is missing, copy the previous ref
1544 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1545 * many assumptions about there being no actual duplicates.
1546 * FIXME: This does not copy padding for out-of-frame motion
1547 * vectors. Given we are concealing a lost frame, this probably
1548 * is not noticeable by comparison, but it should be fixed. */
1549 if (h->short_ref_count) {
1551 av_image_copy(h->short_ref[0]->f->data,
1552 h->short_ref[0]->f->linesize,
1553 (const uint8_t **)prev->f->data,
1558 h->short_ref[0]->poc = prev->poc + 2;
1560 h->short_ref[0]->frame_num = h->prev_frame_num;
1564 /* See if we have a decoded first field looking for a pair...
1565 * We're using that to see whether to continue decoding in that
1566 * frame, or to allocate a new one. */
1567 if (h->first_field) {
1568 av_assert0(h->cur_pic_ptr);
1569 av_assert0(h->cur_pic_ptr->f->buf[0]);
1570 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1572 /* figure out if we have a complementary field pair */
1573 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1574 /* Previous field is unmatched. Don't display it, but let it
1575 * remain for reference if marked as such. */
1576 h->missing_fields ++;
1577 h->cur_pic_ptr = NULL;
1578 h->first_field = FIELD_PICTURE(h);
1580 h->missing_fields = 0;
1581 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1582 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1583 h->picture_structure==PICT_BOTTOM_FIELD);
1584 /* This and the previous field had different frame_nums.
1585 * Consider this field first in pair. Throw away previous
1586 * one except for reference purposes. */
1588 h->cur_pic_ptr = NULL;
1590 /* Second field in complementary pair */
1595 /* Frame or first field in a potentially complementary pair */
1596 h->first_field = FIELD_PICTURE(h);
1599 if (!FIELD_PICTURE(h) || h->first_field) {
1600 if (h264_frame_start(h) < 0) {
1602 return AVERROR_INVALIDDATA;
1605 release_unused_pictures(h, 0);
1607 /* Some macroblocks can be accessed before they're available in case
1608 * of lost slices, MBAFF or threading. */
1609 if (FIELD_PICTURE(h)) {
1610 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1611 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1613 memset(h->slice_table, -1,
1614 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1616 h->last_slice_type = -1;
1620 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1622 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1623 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1624 first_mb_in_slice >= h->mb_num) {
1625 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1626 return AVERROR_INVALIDDATA;
1628 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1629 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1630 FIELD_OR_MBAFF_PICTURE(h);
1631 if (h->picture_structure == PICT_BOTTOM_FIELD)
1632 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1633 av_assert1(sl->mb_y < h->mb_height);
1635 if (h->picture_structure == PICT_FRAME) {
1636 h->curr_pic_num = h->frame_num;
1637 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1639 h->curr_pic_num = 2 * h->frame_num + 1;
1640 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1643 if (h->nal_unit_type == NAL_IDR_SLICE)
1644 get_ue_golomb(&sl->gb); /* idr_pic_id */
1646 if (h->sps.poc_type == 0) {
1647 h->poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1649 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1650 h->delta_poc_bottom = get_se_golomb(&sl->gb);
1653 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1654 h->delta_poc[0] = get_se_golomb(&sl->gb);
1656 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1657 h->delta_poc[1] = get_se_golomb(&sl->gb);
1660 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1662 if (h->pps.redundant_pic_cnt_present)
1663 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1665 ret = ff_set_ref_count(h, sl);
1669 if (slice_type != AV_PICTURE_TYPE_I &&
1670 (h->current_slice == 0 ||
1671 slice_type != h->last_slice_type ||
1672 memcmp(h->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
1674 ff_h264_fill_default_ref_list(h, sl);
1677 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1678 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1680 sl->ref_count[1] = sl->ref_count[0] = 0;
1685 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1686 (h->pps.weighted_bipred_idc == 1 &&
1687 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1688 ff_pred_weight_table(h, sl);
1689 else if (h->pps.weighted_bipred_idc == 2 &&
1690 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1691 implicit_weight_table(h, sl, -1);
1694 for (i = 0; i < 2; i++) {
1695 sl->luma_weight_flag[i] = 0;
1696 sl->chroma_weight_flag[i] = 0;
1700 // If frame-mt is enabled, only update mmco tables for the first slice
1701 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1702 // or h->mmco, which will cause ref list mix-ups and decoding errors
1703 // further down the line. This may break decoding if the first slice is
1704 // corrupt, thus we only do this if frame-mt is enabled.
1705 if (h->nal_ref_idc) {
1706 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1707 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1708 h->current_slice == 0);
1709 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1710 return AVERROR_INVALIDDATA;
1713 if (FRAME_MBAFF(h)) {
1714 ff_h264_fill_mbaff_ref_list(h, sl);
1716 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1717 implicit_weight_table(h, sl, 0);
1718 implicit_weight_table(h, sl, 1);
1722 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1723 ff_h264_direct_dist_scale_factor(h, sl);
1724 ff_h264_direct_ref_list_init(h, sl);
1726 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1727 tmp = get_ue_golomb_31(&sl->gb);
1729 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1730 return AVERROR_INVALIDDATA;
1732 sl->cabac_init_idc = tmp;
1735 sl->last_qscale_diff = 0;
1736 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1737 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1738 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1739 return AVERROR_INVALIDDATA;
1742 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1743 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1744 // FIXME qscale / qp ... stuff
1745 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1746 get_bits1(&sl->gb); /* sp_for_switch_flag */
1747 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1748 sl->slice_type == AV_PICTURE_TYPE_SI)
1749 get_se_golomb(&sl->gb); /* slice_qs_delta */
1751 sl->deblocking_filter = 1;
1752 sl->slice_alpha_c0_offset = 0;
1753 sl->slice_beta_offset = 0;
1754 if (h->pps.deblocking_filter_parameters_present) {
1755 tmp = get_ue_golomb_31(&sl->gb);
1757 av_log(h->avctx, AV_LOG_ERROR,
1758 "deblocking_filter_idc %u out of range\n", tmp);
1759 return AVERROR_INVALIDDATA;
1761 sl->deblocking_filter = tmp;
1762 if (sl->deblocking_filter < 2)
1763 sl->deblocking_filter ^= 1; // 1<->0
1765 if (sl->deblocking_filter) {
1766 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1767 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1768 if (sl->slice_alpha_c0_offset > 12 ||
1769 sl->slice_alpha_c0_offset < -12 ||
1770 sl->slice_beta_offset > 12 ||
1771 sl->slice_beta_offset < -12) {
1772 av_log(h->avctx, AV_LOG_ERROR,
1773 "deblocking filter parameters %d %d out of range\n",
1774 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1775 return AVERROR_INVALIDDATA;
1780 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1781 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1782 h->nal_unit_type != NAL_IDR_SLICE) ||
1783 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1784 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1785 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1786 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1787 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1788 h->nal_ref_idc == 0))
1789 sl->deblocking_filter = 0;
1791 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1792 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1793 /* Cheat slightly for speed:
1794 * Do not bother to deblock across slices. */
1795 sl->deblocking_filter = 2;
1797 h->max_contexts = 1;
1798 if (!h->single_decode_warning) {
1799 av_log(h->avctx, AV_LOG_INFO,
1800 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1801 "To parallelize slice decoding you need video encoded with disable_deblocking_filter_idc set to 2 (deblock only edges that do not cross slices).\n"
1802 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1803 "but will generate non-standard-compliant output.\n");
1804 h->single_decode_warning = 1;
1806 if (sl != h->slice_ctx) {
1807 av_log(h->avctx, AV_LOG_ERROR,
1808 "Deblocking switched inside frame.\n");
1809 return SLICE_SINGLETHREAD;
1813 sl->qp_thresh = 15 -
1814 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1816 h->pps.chroma_qp_index_offset[0],
1817 h->pps.chroma_qp_index_offset[1]) +
1818 6 * (h->sps.bit_depth_luma - 8);
1820 h->last_slice_type = slice_type;
1821 memcpy(h->last_ref_count, sl->ref_count, sizeof(h->last_ref_count));
1822 sl->slice_num = ++h->current_slice;
1825 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1826 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1827 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1828 && sl->slice_num >= MAX_SLICES) {
1829 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1830 av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", sl->slice_num, MAX_SLICES);
1833 for (j = 0; j < 2; j++) {
1835 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1836 for (i = 0; i < 16; i++) {
1838 if (j < sl->list_count && i < sl->ref_count[j] &&
1839 sl->ref_list[j][i].parent->f->buf[0]) {
1841 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1842 for (k = 0; k < h->short_ref_count; k++)
1843 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1847 for (k = 0; k < h->long_ref_count; k++)
1848 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1849 id_list[i] = h->short_ref_count + k;
1857 for (i = 0; i < 16; i++)
1858 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1860 ref2frm[18 + 1] = -1;
1861 for (i = 16; i < 48; i++)
1862 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1863 (sl->ref_list[j][i].reference & 3);
1866 h->au_pps_id = pps_id;
1868 h->sps_buffers[h->pps.sps_id]->new = 0;
1869 h->current_sps_id = h->pps.sps_id;
1871 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1872 av_log(h->avctx, AV_LOG_DEBUG,
1873 "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",
1875 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1877 av_get_picture_type_char(sl->slice_type),
1878 sl->slice_type_fixed ? " fix" : "",
1879 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1880 pps_id, h->frame_num,
1881 h->cur_pic_ptr->field_poc[0],
1882 h->cur_pic_ptr->field_poc[1],
1883 sl->ref_count[0], sl->ref_count[1],
1885 sl->deblocking_filter,
1886 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1888 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
1889 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1895 int ff_h264_get_slice_type(const H264SliceContext *sl)
1897 switch (sl->slice_type) {
1898 case AV_PICTURE_TYPE_P:
1900 case AV_PICTURE_TYPE_B:
1902 case AV_PICTURE_TYPE_I:
1904 case AV_PICTURE_TYPE_SP:
1906 case AV_PICTURE_TYPE_SI:
1909 return AVERROR_INVALIDDATA;
1913 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1914 H264SliceContext *sl,
1915 int mb_type, int top_xy,
1916 int left_xy[LEFT_MBS],
1918 int left_type[LEFT_MBS],
1919 int mb_xy, int list)
1921 int b_stride = h->b_stride;
1922 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1923 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1924 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1925 if (USES_LIST(top_type, list)) {
1926 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1927 const int b8_xy = 4 * top_xy + 2;
1928 int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1929 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1930 ref_cache[0 - 1 * 8] =
1931 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1932 ref_cache[2 - 1 * 8] =
1933 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1935 AV_ZERO128(mv_dst - 1 * 8);
1936 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1939 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1940 if (USES_LIST(left_type[LTOP], list)) {
1941 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1942 const int b8_xy = 4 * left_xy[LTOP] + 1;
1943 int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1944 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1945 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1946 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1947 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1949 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1950 ref_cache[-1 + 16] =
1951 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1953 AV_ZERO32(mv_dst - 1 + 0);
1954 AV_ZERO32(mv_dst - 1 + 8);
1955 AV_ZERO32(mv_dst - 1 + 16);
1956 AV_ZERO32(mv_dst - 1 + 24);
1959 ref_cache[-1 + 16] =
1960 ref_cache[-1 + 24] = LIST_NOT_USED;
1965 if (!USES_LIST(mb_type, list)) {
1966 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1967 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1968 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1969 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1970 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1975 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1976 int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2));
1977 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1978 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1979 AV_WN32A(&ref_cache[0 * 8], ref01);
1980 AV_WN32A(&ref_cache[1 * 8], ref01);
1981 AV_WN32A(&ref_cache[2 * 8], ref23);
1982 AV_WN32A(&ref_cache[3 * 8], ref23);
1986 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1987 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1988 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1989 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1990 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1996 * @return non zero if the loop filter can be skipped
1998 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2000 const int mb_xy = sl->mb_xy;
2001 int top_xy, left_xy[LEFT_MBS];
2002 int top_type, left_type[LEFT_MBS];
2006 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2008 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2009 * stuff, I can't imagine that these complex rules are worth it. */
2011 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2012 if (FRAME_MBAFF(h)) {
2013 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2014 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2016 if (left_mb_field_flag != curr_mb_field_flag)
2017 left_xy[LTOP] -= h->mb_stride;
2019 if (curr_mb_field_flag)
2020 top_xy += h->mb_stride &
2021 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2022 if (left_mb_field_flag != curr_mb_field_flag)
2023 left_xy[LBOT] += h->mb_stride;
2027 sl->top_mb_xy = top_xy;
2028 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2029 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2031 /* For sufficiently low qp, filtering wouldn't do anything.
2032 * This is a conservative estimate: could also check beta_offset
2033 * and more accurate chroma_qp. */
2034 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2035 int qp = h->cur_pic.qscale_table[mb_xy];
2036 if (qp <= qp_thresh &&
2037 (left_xy[LTOP] < 0 ||
2038 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2040 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2041 if (!FRAME_MBAFF(h))
2043 if ((left_xy[LTOP] < 0 ||
2044 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2045 (top_xy < h->mb_stride ||
2046 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2051 top_type = h->cur_pic.mb_type[top_xy];
2052 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2053 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2054 if (sl->deblocking_filter == 2) {
2055 if (h->slice_table[top_xy] != sl->slice_num)
2057 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2058 left_type[LTOP] = left_type[LBOT] = 0;
2060 if (h->slice_table[top_xy] == 0xFFFF)
2062 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2063 left_type[LTOP] = left_type[LBOT] = 0;
2065 sl->top_type = top_type;
2066 sl->left_type[LTOP] = left_type[LTOP];
2067 sl->left_type[LBOT] = left_type[LBOT];
2069 if (IS_INTRA(mb_type))
2072 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2073 top_type, left_type, mb_xy, 0);
2074 if (sl->list_count == 2)
2075 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2076 top_type, left_type, mb_xy, 1);
2078 nnz = h->non_zero_count[mb_xy];
2079 nnz_cache = sl->non_zero_count_cache;
2080 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2081 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2082 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2083 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2084 sl->cbp = h->cbp_table[mb_xy];
2087 nnz = h->non_zero_count[top_xy];
2088 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2091 if (left_type[LTOP]) {
2092 nnz = h->non_zero_count[left_xy[LTOP]];
2093 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2094 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2095 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2096 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2099 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2100 * from what the loop filter needs */
2101 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2102 if (IS_8x8DCT(top_type)) {
2103 nnz_cache[4 + 8 * 0] =
2104 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2105 nnz_cache[6 + 8 * 0] =
2106 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2108 if (IS_8x8DCT(left_type[LTOP])) {
2109 nnz_cache[3 + 8 * 1] =
2110 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2112 if (IS_8x8DCT(left_type[LBOT])) {
2113 nnz_cache[3 + 8 * 3] =
2114 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2117 if (IS_8x8DCT(mb_type)) {
2118 nnz_cache[scan8[0]] =
2119 nnz_cache[scan8[1]] =
2120 nnz_cache[scan8[2]] =
2121 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2123 nnz_cache[scan8[0 + 4]] =
2124 nnz_cache[scan8[1 + 4]] =
2125 nnz_cache[scan8[2 + 4]] =
2126 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2128 nnz_cache[scan8[0 + 8]] =
2129 nnz_cache[scan8[1 + 8]] =
2130 nnz_cache[scan8[2 + 8]] =
2131 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2133 nnz_cache[scan8[0 + 12]] =
2134 nnz_cache[scan8[1 + 12]] =
2135 nnz_cache[scan8[2 + 12]] =
2136 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2143 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2145 uint8_t *dest_y, *dest_cb, *dest_cr;
2146 int linesize, uvlinesize, mb_x, mb_y;
2147 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2148 const int old_slice_type = sl->slice_type;
2149 const int pixel_shift = h->pixel_shift;
2150 const int block_h = 16 >> h->chroma_y_shift;
2152 if (sl->deblocking_filter) {
2153 for (mb_x = start_x; mb_x < end_x; mb_x++)
2154 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2156 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2157 sl->slice_num = h->slice_table[mb_xy];
2158 mb_type = h->cur_pic.mb_type[mb_xy];
2159 sl->list_count = h->list_counts[mb_xy];
2163 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2167 dest_y = h->cur_pic.f->data[0] +
2168 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2169 dest_cb = h->cur_pic.f->data[1] +
2170 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2171 mb_y * sl->uvlinesize * block_h;
2172 dest_cr = h->cur_pic.f->data[2] +
2173 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2174 mb_y * sl->uvlinesize * block_h;
2175 // FIXME simplify above
2178 linesize = sl->mb_linesize = sl->linesize * 2;
2179 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2180 if (mb_y & 1) { // FIXME move out of this function?
2181 dest_y -= sl->linesize * 15;
2182 dest_cb -= sl->uvlinesize * (block_h - 1);
2183 dest_cr -= sl->uvlinesize * (block_h - 1);
2186 linesize = sl->mb_linesize = sl->linesize;
2187 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2189 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2191 if (fill_filter_caches(h, sl, mb_type))
2193 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2194 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2196 if (FRAME_MBAFF(h)) {
2197 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2198 linesize, uvlinesize);
2200 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2201 dest_cr, linesize, uvlinesize);
2205 sl->slice_type = old_slice_type;
2207 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2208 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2209 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2212 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2214 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2215 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2216 h->cur_pic.mb_type[mb_xy - 1] :
2217 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2218 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2219 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2223 * Draw edges and report progress for the last MB row.
2225 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2227 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2228 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2229 int height = 16 << FRAME_MBAFF(h);
2230 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2232 if (sl->deblocking_filter) {
2233 if ((top + height) >= pic_height)
2234 height += deblock_border;
2235 top -= deblock_border;
2238 if (top >= pic_height || (top + height) < 0)
2241 height = FFMIN(height, pic_height - top);
2243 height = top + height;
2247 ff_h264_draw_horiz_band(h, sl, top, height);
2249 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2252 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2253 h->picture_structure == PICT_BOTTOM_FIELD);
2256 static void er_add_slice(H264SliceContext *sl,
2257 int startx, int starty,
2258 int endx, int endy, int status)
2260 if (!sl->h264->enable_er)
2263 if (CONFIG_ERROR_RESILIENCE) {
2264 ERContext *er = &sl->h264->slice_ctx[0].er;
2266 ff_er_add_slice(er, startx, starty, endx, endy, status);
2270 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2272 H264SliceContext *sl = arg;
2273 const H264Context *h = sl->h264;
2274 int lf_x_start = sl->mb_x;
2277 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2278 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2280 ret = alloc_scratch_buffers(sl, sl->linesize);
2284 sl->mb_skip_run = -1;
2286 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2288 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2289 avctx->codec_id != AV_CODEC_ID_H264 ||
2290 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2292 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2293 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2295 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2296 prev_status &= ~ VP_START;
2297 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2298 h->slice_ctx[0].er.error_occurred = 1;
2304 align_get_bits(&sl->gb);
2307 ff_init_cabac_decoder(&sl->cabac,
2308 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2309 (get_bits_left(&sl->gb) + 7) / 8);
2311 ff_h264_init_cabac_states(h, sl);
2317 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->mb_index_end) {
2318 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps next at %d\n",
2320 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2321 sl->mb_y, ER_MB_ERROR);
2322 return AVERROR_INVALIDDATA;
2325 ret = ff_h264_decode_mb_cabac(h, sl);
2326 // STOP_TIMER("decode_mb_cabac")
2329 ff_h264_hl_decode_mb(h, sl);
2331 // FIXME optimal? or let mb_decode decode 16x32 ?
2332 if (ret >= 0 && FRAME_MBAFF(h)) {
2335 ret = ff_h264_decode_mb_cabac(h, sl);
2338 ff_h264_hl_decode_mb(h, sl);
2341 eos = get_cabac_terminate(&sl->cabac);
2343 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2344 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2345 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2346 sl->mb_y, ER_MB_END);
2347 if (sl->mb_x >= lf_x_start)
2348 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2351 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2352 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2353 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2354 av_log(h->avctx, AV_LOG_ERROR,
2355 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2357 sl->cabac.bytestream_end - sl->cabac.bytestream);
2358 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2359 sl->mb_y, ER_MB_ERROR);
2360 return AVERROR_INVALIDDATA;
2363 if (++sl->mb_x >= h->mb_width) {
2364 loop_filter(h, sl, lf_x_start, sl->mb_x);
2365 sl->mb_x = lf_x_start = 0;
2366 decode_finish_row(h, sl);
2368 if (FIELD_OR_MBAFF_PICTURE(h)) {
2370 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2371 predict_field_decoding_flag(h, sl);
2375 if (eos || sl->mb_y >= h->mb_height) {
2376 ff_tlog(h->avctx, "slice end %d %d\n",
2377 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2378 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2379 sl->mb_y, ER_MB_END);
2380 if (sl->mb_x > lf_x_start)
2381 loop_filter(h, sl, lf_x_start, sl->mb_x);
2389 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->mb_index_end) {
2390 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps next at %d\n",
2392 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2393 sl->mb_y, ER_MB_ERROR);
2394 return AVERROR_INVALIDDATA;
2397 ret = ff_h264_decode_mb_cavlc(h, sl);
2400 ff_h264_hl_decode_mb(h, sl);
2402 // FIXME optimal? or let mb_decode decode 16x32 ?
2403 if (ret >= 0 && FRAME_MBAFF(h)) {
2405 ret = ff_h264_decode_mb_cavlc(h, sl);
2408 ff_h264_hl_decode_mb(h, sl);
2413 av_log(h->avctx, AV_LOG_ERROR,
2414 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2415 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2416 sl->mb_y, ER_MB_ERROR);
2420 if (++sl->mb_x >= h->mb_width) {
2421 loop_filter(h, sl, lf_x_start, sl->mb_x);
2422 sl->mb_x = lf_x_start = 0;
2423 decode_finish_row(h, sl);
2425 if (FIELD_OR_MBAFF_PICTURE(h)) {
2427 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2428 predict_field_decoding_flag(h, sl);
2430 if (sl->mb_y >= h->mb_height) {
2431 ff_tlog(h->avctx, "slice end %d %d\n",
2432 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2434 if ( get_bits_left(&sl->gb) == 0
2435 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2436 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2437 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2441 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2442 sl->mb_x, sl->mb_y, ER_MB_END);
2444 return AVERROR_INVALIDDATA;
2449 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2450 ff_tlog(h->avctx, "slice end %d %d\n",
2451 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2453 if (get_bits_left(&sl->gb) == 0) {
2454 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2455 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2456 if (sl->mb_x > lf_x_start)
2457 loop_filter(h, sl, lf_x_start, sl->mb_x);
2461 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2462 sl->mb_y, ER_MB_ERROR);
2464 return AVERROR_INVALIDDATA;
2472 * Call decode_slice() for each context.
2474 * @param h h264 master context
2475 * @param context_count number of contexts to execute
2477 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2479 AVCodecContext *const avctx = h->avctx;
2480 H264SliceContext *sl;
2483 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2485 h->slice_ctx[0].mb_index_end = INT_MAX;
2487 if (h->avctx->hwaccel ||
2488 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2490 if (context_count == 1) {
2491 int ret = decode_slice(avctx, &h->slice_ctx[0]);
2492 h->mb_y = h->slice_ctx[0].mb_y;
2496 av_assert0(context_count > 0);
2497 for (i = 0; i < context_count; i++) {
2498 int mb_index_end = h->mb_width * h->mb_height;
2499 sl = &h->slice_ctx[i];
2500 mb_index = sl->resync_mb_x + sl->resync_mb_y * h->mb_width;
2501 if (CONFIG_ERROR_RESILIENCE) {
2502 sl->er.error_count = 0;
2504 for (j = 0; j < context_count; j++) {
2505 H264SliceContext *sl2 = &h->slice_ctx[j];
2506 int mb_index2 = sl2->resync_mb_x + sl2->resync_mb_y * h->mb_width;
2508 if (i==j || mb_index > mb_index2)
2510 mb_index_end = FFMIN(mb_index_end, mb_index2);
2512 sl->mb_index_end = mb_index_end;
2515 avctx->execute(avctx, decode_slice, h->slice_ctx,
2516 NULL, context_count, sizeof(h->slice_ctx[0]));
2518 /* pull back stuff from slices to master context */
2519 sl = &h->slice_ctx[context_count - 1];
2521 if (CONFIG_ERROR_RESILIENCE) {
2522 for (i = 1; i < context_count; i++)
2523 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;