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 / MPEG-4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/avassert.h"
29 #include "libavutil/display.h"
30 #include "libavutil/imgutils.h"
31 #include "libavutil/stereo3d.h"
32 #include "libavutil/timer.h"
35 #include "cabac_functions.h"
36 #include "error_resilience.h"
41 #include "h264chroma.h"
42 #include "h264_mvpred.h"
46 #include "mpegutils.h"
47 #include "rectangle.h"
50 static const uint8_t field_scan[16+1] = {
51 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
52 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
53 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
54 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
57 static const uint8_t field_scan8x8[64+1] = {
58 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
59 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
60 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
61 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
62 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
63 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
64 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
65 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
66 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
67 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
68 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
69 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
70 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
71 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
72 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
73 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
76 static const uint8_t field_scan8x8_cavlc[64+1] = {
77 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
78 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
79 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
80 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
81 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
82 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
83 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
84 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
85 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
86 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
87 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
88 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
89 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
90 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
91 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
92 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
95 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
96 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
97 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
98 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
99 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
100 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
101 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
102 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
103 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
104 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
105 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
106 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
107 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
108 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
109 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
110 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
111 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
112 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
115 static void release_unused_pictures(H264Context *h, int remove_current)
119 /* release non reference frames */
120 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
121 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
122 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
123 ff_h264_unref_picture(h, &h->DPB[i]);
128 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
130 const H264Context *h = sl->h264;
131 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
133 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
134 // edge emu needs blocksize + filter length - 1
135 // (= 21x21 for H.264)
136 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
138 av_fast_mallocz(&sl->top_borders[0], &sl->top_borders_allocated[0],
139 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
140 av_fast_mallocz(&sl->top_borders[1], &sl->top_borders_allocated[1],
141 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
143 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
144 !sl->top_borders[0] || !sl->top_borders[1]) {
145 av_freep(&sl->bipred_scratchpad);
146 av_freep(&sl->edge_emu_buffer);
147 av_freep(&sl->top_borders[0]);
148 av_freep(&sl->top_borders[1]);
150 sl->bipred_scratchpad_allocated = 0;
151 sl->edge_emu_buffer_allocated = 0;
152 sl->top_borders_allocated[0] = 0;
153 sl->top_borders_allocated[1] = 0;
154 return AVERROR(ENOMEM);
160 static int init_table_pools(H264Context *h)
162 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
163 const int mb_array_size = h->mb_stride * h->mb_height;
164 const int b4_stride = h->mb_width * 4 + 1;
165 const int b4_array_size = b4_stride * h->mb_height * 4;
167 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
169 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
170 sizeof(uint32_t), av_buffer_allocz);
171 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
172 sizeof(int16_t), av_buffer_allocz);
173 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
175 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
176 !h->ref_index_pool) {
177 av_buffer_pool_uninit(&h->qscale_table_pool);
178 av_buffer_pool_uninit(&h->mb_type_pool);
179 av_buffer_pool_uninit(&h->motion_val_pool);
180 av_buffer_pool_uninit(&h->ref_index_pool);
181 return AVERROR(ENOMEM);
187 static int alloc_picture(H264Context *h, H264Picture *pic)
191 av_assert0(!pic->f->data[0]);
194 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
195 AV_GET_BUFFER_FLAG_REF : 0);
199 pic->crop = h->ps.sps->crop;
200 pic->crop_top = h->ps.sps->crop_top;
201 pic->crop_left= h->ps.sps->crop_left;
203 if (h->avctx->hwaccel) {
204 const AVHWAccel *hwaccel = h->avctx->hwaccel;
205 av_assert0(!pic->hwaccel_picture_private);
206 if (hwaccel->frame_priv_data_size) {
207 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
208 if (!pic->hwaccel_priv_buf)
209 return AVERROR(ENOMEM);
210 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
213 if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
214 int h_chroma_shift, v_chroma_shift;
215 av_pix_fmt_get_chroma_sub_sample(pic->f->format,
216 &h_chroma_shift, &v_chroma_shift);
218 for(i=0; i<AV_CEIL_RSHIFT(pic->f->height, v_chroma_shift); i++) {
219 memset(pic->f->data[1] + pic->f->linesize[1]*i,
220 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
221 memset(pic->f->data[2] + pic->f->linesize[2]*i,
222 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
226 if (!h->qscale_table_pool) {
227 ret = init_table_pools(h);
232 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
233 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
234 if (!pic->qscale_table_buf || !pic->mb_type_buf)
237 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
238 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
240 for (i = 0; i < 2; i++) {
241 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
242 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
243 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
246 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
247 pic->ref_index[i] = pic->ref_index_buf[i]->data;
252 ff_h264_unref_picture(h, pic);
253 return (ret < 0) ? ret : AVERROR(ENOMEM);
256 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
263 static int find_unused_picture(H264Context *h)
267 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
268 if (pic_is_unused(h, &h->DPB[i]))
271 if (i == H264_MAX_PICTURE_COUNT)
272 return AVERROR_INVALIDDATA;
278 #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
280 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
281 (((pic) && (pic) >= (old_ctx)->DPB && \
282 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
283 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
285 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
286 H264Context *new_base,
287 H264Context *old_base)
291 for (i = 0; i < count; i++) {
292 av_assert1(!from[i] ||
293 IN_RANGE(from[i], old_base, 1) ||
294 IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
295 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
299 static int h264_slice_header_init(H264Context *h);
301 int ff_h264_update_thread_context(AVCodecContext *dst,
302 const AVCodecContext *src)
304 H264Context *h = dst->priv_data, *h1 = src->priv_data;
305 int inited = h->context_initialized, err = 0;
312 // We can't fail if SPS isn't set at it breaks current skip_frame code
314 // return AVERROR_INVALIDDATA;
317 (h->width != h1->width ||
318 h->height != h1->height ||
319 h->mb_width != h1->mb_width ||
320 h->mb_height != h1->mb_height ||
322 h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
323 h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
324 h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
328 /* copy block_offset since frame_start may not be called */
329 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
332 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
333 av_buffer_unref(&h->ps.sps_list[i]);
334 if (h1->ps.sps_list[i]) {
335 h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
336 if (!h->ps.sps_list[i])
337 return AVERROR(ENOMEM);
340 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
341 av_buffer_unref(&h->ps.pps_list[i]);
342 if (h1->ps.pps_list[i]) {
343 h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
344 if (!h->ps.pps_list[i])
345 return AVERROR(ENOMEM);
349 av_buffer_unref(&h->ps.pps_ref);
350 av_buffer_unref(&h->ps.sps_ref);
353 if (h1->ps.pps_ref) {
354 h->ps.pps_ref = av_buffer_ref(h1->ps.pps_ref);
356 return AVERROR(ENOMEM);
357 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
359 if (h1->ps.sps_ref) {
360 h->ps.sps_ref = av_buffer_ref(h1->ps.sps_ref);
362 return AVERROR(ENOMEM);
363 h->ps.sps = (const SPS*)h->ps.sps_ref->data;
366 if (need_reinit || !inited) {
367 h->width = h1->width;
368 h->height = h1->height;
369 h->mb_height = h1->mb_height;
370 h->mb_width = h1->mb_width;
371 h->mb_num = h1->mb_num;
372 h->mb_stride = h1->mb_stride;
373 h->b_stride = h1->b_stride;
375 if (h->context_initialized || h1->context_initialized) {
376 if ((err = h264_slice_header_init(h)) < 0) {
377 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
382 /* copy block_offset since frame_start may not be called */
383 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
386 h->avctx->coded_height = h1->avctx->coded_height;
387 h->avctx->coded_width = h1->avctx->coded_width;
388 h->avctx->width = h1->avctx->width;
389 h->avctx->height = h1->avctx->height;
390 h->coded_picture_number = h1->coded_picture_number;
391 h->first_field = h1->first_field;
392 h->picture_structure = h1->picture_structure;
393 h->mb_aff_frame = h1->mb_aff_frame;
394 h->droppable = h1->droppable;
395 h->backup_width = h1->backup_width;
396 h->backup_height = h1->backup_height;
397 h->backup_pix_fmt = h1->backup_pix_fmt;
399 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
400 ff_h264_unref_picture(h, &h->DPB[i]);
401 if (h1->DPB[i].f->buf[0] &&
402 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
406 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
407 ff_h264_unref_picture(h, &h->cur_pic);
408 if (h1->cur_pic.f->buf[0]) {
409 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
414 h->enable_er = h1->enable_er;
415 h->workaround_bugs = h1->workaround_bugs;
416 h->droppable = h1->droppable;
418 // extradata/NAL handling
419 h->is_avc = h1->is_avc;
420 h->nal_length_size = h1->nal_length_size;
421 h->sei.unregistered.x264_build = h1->sei.unregistered.x264_build;
423 memcpy(&h->poc, &h1->poc, sizeof(h->poc));
425 memcpy(h->default_ref, h1->default_ref, sizeof(h->default_ref));
426 memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
427 memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
428 memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
429 memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
431 h->next_output_pic = h1->next_output_pic;
432 h->next_outputed_poc = h1->next_outputed_poc;
434 memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
435 h->nb_mmco = h1->nb_mmco;
436 h->mmco_reset = h1->mmco_reset;
437 h->explicit_ref_marking = h1->explicit_ref_marking;
438 h->long_ref_count = h1->long_ref_count;
439 h->short_ref_count = h1->short_ref_count;
441 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
442 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
443 copy_picture_range(h->delayed_pic, h1->delayed_pic,
444 MAX_DELAYED_PIC_COUNT + 2, h, h1);
446 h->frame_recovered = h1->frame_recovered;
452 err = ff_h264_execute_ref_pic_marking(h);
453 h->poc.prev_poc_msb = h->poc.poc_msb;
454 h->poc.prev_poc_lsb = h->poc.poc_lsb;
456 h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
457 h->poc.prev_frame_num = h->poc.frame_num;
459 h->recovery_frame = h1->recovery_frame;
464 static int h264_frame_start(H264Context *h)
468 const int pixel_shift = h->pixel_shift;
470 1<<(h->ps.sps->bit_depth_luma-1),
471 1<<(h->ps.sps->bit_depth_chroma-1),
472 1<<(h->ps.sps->bit_depth_chroma-1),
476 if (!ff_thread_can_start_frame(h->avctx)) {
477 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
481 release_unused_pictures(h, 1);
482 h->cur_pic_ptr = NULL;
484 i = find_unused_picture(h);
486 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
491 pic->reference = h->droppable ? 0 : h->picture_structure;
492 pic->f->coded_picture_number = h->coded_picture_number++;
493 pic->field_picture = h->picture_structure != PICT_FRAME;
494 pic->frame_num = h->poc.frame_num;
496 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
498 * See decode_nal_units().
500 pic->f->key_frame = 0;
503 pic->invalid_gap = 0;
504 pic->sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
506 pic->f->pict_type = h->slice_ctx[0].slice_type;
508 if ((ret = alloc_picture(h, pic)) < 0)
510 if(!h->frame_recovered && !h->avctx->hwaccel
512 && !(h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
515 ff_color_frame(pic->f, c);
517 h->cur_pic_ptr = pic;
518 ff_h264_unref_picture(h, &h->cur_pic);
519 if (CONFIG_ERROR_RESILIENCE) {
520 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
523 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
526 for (i = 0; i < h->nb_slice_ctx; i++) {
527 h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
528 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
531 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
532 ff_er_frame_start(&h->slice_ctx[0].er);
533 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
534 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
537 for (i = 0; i < 16; i++) {
538 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
539 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
541 for (i = 0; i < 16; i++) {
542 h->block_offset[16 + i] =
543 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
544 h->block_offset[48 + 16 + i] =
545 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
548 /* We mark the current picture as non-reference after allocating it, so
549 * that if we break out due to an error it can be released automatically
550 * in the next ff_mpv_frame_start().
552 h->cur_pic_ptr->reference = 0;
554 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
556 h->next_output_pic = NULL;
558 h->postpone_filter = 0;
560 h->mb_aff_frame = h->ps.sps->mb_aff && (h->picture_structure == PICT_FRAME);
562 assert(h->cur_pic_ptr->long_ref == 0);
567 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
569 uint8_t *src_cb, uint8_t *src_cr,
570 int linesize, int uvlinesize,
575 const int pixel_shift = h->pixel_shift;
576 int chroma444 = CHROMA444(h);
577 int chroma422 = CHROMA422(h);
580 src_cb -= uvlinesize;
581 src_cr -= uvlinesize;
583 if (!simple && FRAME_MBAFF(h)) {
586 top_border = sl->top_borders[0][sl->mb_x];
587 AV_COPY128(top_border, src_y + 15 * linesize);
589 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
590 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
593 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
594 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
595 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
596 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
598 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
599 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
601 } else if (chroma422) {
603 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
604 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
606 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
607 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
611 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
612 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
614 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
615 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
620 } else if (MB_MBAFF(sl)) {
626 top_border = sl->top_borders[top_idx][sl->mb_x];
627 /* There are two lines saved, the line above the top macroblock
628 * of a pair, and the line above the bottom macroblock. */
629 AV_COPY128(top_border, src_y + 16 * linesize);
631 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
633 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
636 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
637 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
638 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
639 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
641 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
642 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
644 } else if (chroma422) {
646 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
647 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
649 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
650 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
654 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
655 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
657 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
658 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
665 * Initialize implicit_weight table.
666 * @param field 0/1 initialize the weight for interlaced MBAFF
667 * -1 initializes the rest
669 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
671 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
673 for (i = 0; i < 2; i++) {
674 sl->pwt.luma_weight_flag[i] = 0;
675 sl->pwt.chroma_weight_flag[i] = 0;
679 if (h->picture_structure == PICT_FRAME) {
680 cur_poc = h->cur_pic_ptr->poc;
682 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
684 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
685 sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2 * cur_poc) {
686 sl->pwt.use_weight = 0;
687 sl->pwt.use_weight_chroma = 0;
691 ref_count0 = sl->ref_count[0];
692 ref_count1 = sl->ref_count[1];
694 cur_poc = h->cur_pic_ptr->field_poc[field];
696 ref_count0 = 16 + 2 * sl->ref_count[0];
697 ref_count1 = 16 + 2 * sl->ref_count[1];
700 sl->pwt.use_weight = 2;
701 sl->pwt.use_weight_chroma = 2;
702 sl->pwt.luma_log2_weight_denom = 5;
703 sl->pwt.chroma_log2_weight_denom = 5;
705 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
706 int64_t poc0 = sl->ref_list[0][ref0].poc;
707 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
709 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
710 int poc1 = sl->ref_list[1][ref1].poc;
711 int td = av_clip_int8(poc1 - poc0);
713 int tb = av_clip_int8(cur_poc - poc0);
714 int tx = (16384 + (FFABS(td) >> 1)) / td;
715 int dist_scale_factor = (tb * tx + 32) >> 8;
716 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
717 w = 64 - dist_scale_factor;
721 sl->pwt.implicit_weight[ref0][ref1][0] =
722 sl->pwt.implicit_weight[ref0][ref1][1] = w;
724 sl->pwt.implicit_weight[ref0][ref1][field] = w;
731 * initialize scan tables
733 static void init_scan_tables(H264Context *h)
736 for (i = 0; i < 16; i++) {
737 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
738 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
739 h->field_scan[i] = TRANSPOSE(field_scan[i]);
742 for (i = 0; i < 64; i++) {
743 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
744 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
745 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
746 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
747 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
750 if (h->ps.sps->transform_bypass) { // FIXME same ugly
751 memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
752 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
753 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
754 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
755 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
756 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
758 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
759 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
760 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
761 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
762 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
763 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
767 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
769 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
770 CONFIG_H264_D3D11VA_HWACCEL + \
771 CONFIG_H264_VAAPI_HWACCEL + \
772 (CONFIG_H264_VDA_HWACCEL * 2) + \
773 CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
774 CONFIG_H264_VDPAU_HWACCEL)
775 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
776 const enum AVPixelFormat *choices = pix_fmts;
779 switch (h->ps.sps->bit_depth_luma) {
782 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
783 *fmt++ = AV_PIX_FMT_GBRP9;
785 *fmt++ = AV_PIX_FMT_YUV444P9;
786 } else if (CHROMA422(h))
787 *fmt++ = AV_PIX_FMT_YUV422P9;
789 *fmt++ = AV_PIX_FMT_YUV420P9;
793 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
794 *fmt++ = AV_PIX_FMT_GBRP10;
796 *fmt++ = AV_PIX_FMT_YUV444P10;
797 } else if (CHROMA422(h))
798 *fmt++ = AV_PIX_FMT_YUV422P10;
800 *fmt++ = AV_PIX_FMT_YUV420P10;
804 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
805 *fmt++ = AV_PIX_FMT_GBRP12;
807 *fmt++ = AV_PIX_FMT_YUV444P12;
808 } else if (CHROMA422(h))
809 *fmt++ = AV_PIX_FMT_YUV422P12;
811 *fmt++ = AV_PIX_FMT_YUV420P12;
815 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
816 *fmt++ = AV_PIX_FMT_GBRP14;
818 *fmt++ = AV_PIX_FMT_YUV444P14;
819 } else if (CHROMA422(h))
820 *fmt++ = AV_PIX_FMT_YUV422P14;
822 *fmt++ = AV_PIX_FMT_YUV420P14;
825 #if CONFIG_H264_VDPAU_HWACCEL
826 *fmt++ = AV_PIX_FMT_VDPAU;
829 if (h->avctx->colorspace == AVCOL_SPC_RGB)
830 *fmt++ = AV_PIX_FMT_GBRP;
831 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
832 *fmt++ = AV_PIX_FMT_YUVJ444P;
834 *fmt++ = AV_PIX_FMT_YUV444P;
835 } else if (CHROMA422(h)) {
836 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
837 *fmt++ = AV_PIX_FMT_YUVJ422P;
839 *fmt++ = AV_PIX_FMT_YUV422P;
841 #if CONFIG_H264_DXVA2_HWACCEL
842 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
844 #if CONFIG_H264_D3D11VA_HWACCEL
845 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
847 #if CONFIG_H264_VAAPI_HWACCEL
848 *fmt++ = AV_PIX_FMT_VAAPI;
850 #if CONFIG_H264_VDA_HWACCEL
851 *fmt++ = AV_PIX_FMT_VDA_VLD;
852 *fmt++ = AV_PIX_FMT_VDA;
854 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
855 *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
857 if (h->avctx->codec->pix_fmts)
858 choices = h->avctx->codec->pix_fmts;
859 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
860 *fmt++ = AV_PIX_FMT_YUVJ420P;
862 *fmt++ = AV_PIX_FMT_YUV420P;
866 av_log(h->avctx, AV_LOG_ERROR,
867 "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
868 return AVERROR_INVALIDDATA;
871 *fmt = AV_PIX_FMT_NONE;
873 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
874 if (choices[i] == h->avctx->pix_fmt && !force_callback)
876 return ff_thread_get_format(h->avctx, choices);
879 /* export coded and cropped frame dimensions to AVCodecContext */
880 static int init_dimensions(H264Context *h)
882 const SPS *sps = (const SPS*)h->ps.sps;
883 int width = h->width - (sps->crop_right + sps->crop_left);
884 int height = h->height - (sps->crop_top + sps->crop_bottom);
885 av_assert0(sps->crop_right + sps->crop_left < (unsigned)h->width);
886 av_assert0(sps->crop_top + sps->crop_bottom < (unsigned)h->height);
888 /* handle container cropping */
889 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
890 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16) &&
891 h->avctx->width <= width &&
892 h->avctx->height <= height
894 width = h->avctx->width;
895 height = h->avctx->height;
898 h->avctx->coded_width = h->width;
899 h->avctx->coded_height = h->height;
900 h->avctx->width = width;
901 h->avctx->height = height;
906 static int h264_slice_header_init(H264Context *h)
908 const SPS *sps = h->ps.sps;
911 ff_set_sar(h->avctx, sps->sar);
912 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
913 &h->chroma_x_shift, &h->chroma_y_shift);
915 if (sps->timing_info_present_flag) {
916 int64_t den = sps->time_scale;
917 if (h->sei.unregistered.x264_build < 44U)
919 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
920 sps->num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
923 ff_h264_free_tables(h);
926 h->prev_interlaced_frame = 1;
929 ret = ff_h264_alloc_tables(h);
931 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
936 if (h->avctx->codec &&
937 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU &&
938 (sps->bit_depth_luma != 8 || sps->chroma_format_idc > 1)) {
939 av_log(h->avctx, AV_LOG_ERROR,
940 "VDPAU decoding does not support video colorspace.\n");
941 ret = AVERROR_INVALIDDATA;
946 if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 14 ||
947 sps->bit_depth_luma == 11 || sps->bit_depth_luma == 13
949 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
950 sps->bit_depth_luma);
951 ret = AVERROR_INVALIDDATA;
955 h->cur_bit_depth_luma =
956 h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
957 h->cur_chroma_format_idc = sps->chroma_format_idc;
958 h->pixel_shift = sps->bit_depth_luma > 8;
959 h->chroma_format_idc = sps->chroma_format_idc;
960 h->bit_depth_luma = sps->bit_depth_luma;
962 ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
963 sps->chroma_format_idc);
964 ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
965 ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
966 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
967 sps->chroma_format_idc);
968 ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
970 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
971 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
973 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
977 for (i = 0; i < h->nb_slice_ctx; i++) {
978 H264SliceContext *sl = &h->slice_ctx[i];
981 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
982 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
983 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
985 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
986 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
992 h->context_initialized = 1;
996 ff_h264_free_tables(h);
997 h->context_initialized = 0;
1001 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1004 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1005 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1006 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1012 static int h264_init_ps(H264Context *h, const H264SliceContext *sl, int first_slice)
1015 int needs_reinit = 0, must_reinit, ret;
1018 av_buffer_unref(&h->ps.pps_ref);
1020 h->ps.pps_ref = av_buffer_ref(h->ps.pps_list[sl->pps_id]);
1022 return AVERROR(ENOMEM);
1023 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
1026 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
1027 av_buffer_unref(&h->ps.sps_ref);
1029 h->ps.sps_ref = av_buffer_ref(h->ps.sps_list[h->ps.pps->sps_id]);
1031 return AVERROR(ENOMEM);
1032 h->ps.sps = (const SPS*)h->ps.sps_ref->data;
1034 if (h->mb_width != h->ps.sps->mb_width ||
1035 h->mb_height != h->ps.sps->mb_height ||
1036 h->cur_bit_depth_luma != h->ps.sps->bit_depth_luma ||
1037 h->cur_chroma_format_idc != h->ps.sps->chroma_format_idc
1041 if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
1042 h->chroma_format_idc != h->ps.sps->chroma_format_idc)
1047 must_reinit = (h->context_initialized &&
1048 ( 16*sps->mb_width != h->avctx->coded_width
1049 || 16*sps->mb_height != h->avctx->coded_height
1050 || h->cur_bit_depth_luma != sps->bit_depth_luma
1051 || h->cur_chroma_format_idc != sps->chroma_format_idc
1052 || h->mb_width != sps->mb_width
1053 || h->mb_height != sps->mb_height
1055 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1056 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1059 if (first_slice && av_cmp_q(sps->sar, h->avctx->sample_aspect_ratio))
1062 if (!h->setup_finished) {
1063 h->avctx->profile = ff_h264_get_profile(sps);
1064 h->avctx->level = sps->level_idc;
1065 h->avctx->refs = sps->ref_frame_count;
1067 h->mb_width = sps->mb_width;
1068 h->mb_height = sps->mb_height;
1069 h->mb_num = h->mb_width * h->mb_height;
1070 h->mb_stride = h->mb_width + 1;
1072 h->b_stride = h->mb_width * 4;
1074 h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
1076 h->width = 16 * h->mb_width;
1077 h->height = 16 * h->mb_height;
1079 ret = init_dimensions(h);
1083 if (sps->video_signal_type_present_flag) {
1084 h->avctx->color_range = sps->full_range > 0 ? AVCOL_RANGE_JPEG
1086 if (sps->colour_description_present_flag) {
1087 if (h->avctx->colorspace != sps->colorspace)
1089 h->avctx->color_primaries = sps->color_primaries;
1090 h->avctx->color_trc = sps->color_trc;
1091 h->avctx->colorspace = sps->colorspace;
1096 if (!h->context_initialized || must_reinit || needs_reinit) {
1097 int flush_changes = h->context_initialized;
1098 h->context_initialized = 0;
1099 if (sl != h->slice_ctx) {
1100 av_log(h->avctx, AV_LOG_ERROR,
1101 "changing width %d -> %d / height %d -> %d on "
1103 h->width, h->avctx->coded_width,
1104 h->height, h->avctx->coded_height,
1105 h->current_slice + 1);
1106 return AVERROR_INVALIDDATA;
1109 av_assert1(first_slice);
1112 ff_h264_flush_change(h);
1114 if ((ret = get_pixel_format(h, 1)) < 0)
1116 h->avctx->pix_fmt = ret;
1118 av_log(h->avctx, AV_LOG_VERBOSE, "Reinit context to %dx%d, "
1119 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1121 if ((ret = h264_slice_header_init(h)) < 0) {
1122 av_log(h->avctx, AV_LOG_ERROR,
1123 "h264_slice_header_init() failed\n");
1131 static int h264_export_frame_props(H264Context *h)
1133 const SPS *sps = h->ps.sps;
1134 H264Picture *cur = h->cur_pic_ptr;
1136 cur->f->interlaced_frame = 0;
1137 cur->f->repeat_pict = 0;
1139 /* Signal interlacing information externally. */
1140 /* Prioritize picture timing SEI information over used
1141 * decoding process if it exists. */
1143 if (sps->pic_struct_present_flag) {
1144 H264SEIPictureTiming *pt = &h->sei.picture_timing;
1145 switch (pt->pic_struct) {
1146 case SEI_PIC_STRUCT_FRAME:
1148 case SEI_PIC_STRUCT_TOP_FIELD:
1149 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1150 cur->f->interlaced_frame = 1;
1152 case SEI_PIC_STRUCT_TOP_BOTTOM:
1153 case SEI_PIC_STRUCT_BOTTOM_TOP:
1154 if (FIELD_OR_MBAFF_PICTURE(h))
1155 cur->f->interlaced_frame = 1;
1157 // try to flag soft telecine progressive
1158 cur->f->interlaced_frame = h->prev_interlaced_frame;
1160 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1161 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1162 /* Signal the possibility of telecined film externally
1163 * (pic_struct 5,6). From these hints, let the applications
1164 * decide if they apply deinterlacing. */
1165 cur->f->repeat_pict = 1;
1167 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1168 cur->f->repeat_pict = 2;
1170 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1171 cur->f->repeat_pict = 4;
1175 if ((pt->ct_type & 3) &&
1176 pt->pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1177 cur->f->interlaced_frame = (pt->ct_type & (1 << 1)) != 0;
1179 /* Derive interlacing flag from used decoding process. */
1180 cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
1182 h->prev_interlaced_frame = cur->f->interlaced_frame;
1184 if (cur->field_poc[0] != cur->field_poc[1]) {
1185 /* Derive top_field_first from field pocs. */
1186 cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
1188 if (sps->pic_struct_present_flag) {
1189 /* Use picture timing SEI information. Even if it is a
1190 * information of a past frame, better than nothing. */
1191 if (h->sei.picture_timing.pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1192 h->sei.picture_timing.pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1193 cur->f->top_field_first = 1;
1195 cur->f->top_field_first = 0;
1196 } else if (cur->f->interlaced_frame) {
1197 /* Default to top field first when pic_struct_present_flag
1198 * is not set but interlaced frame detected */
1199 cur->f->top_field_first = 1;
1201 /* Most likely progressive */
1202 cur->f->top_field_first = 0;
1206 if (h->sei.frame_packing.present &&
1207 h->sei.frame_packing.frame_packing_arrangement_type <= 6 &&
1208 h->sei.frame_packing.content_interpretation_type > 0 &&
1209 h->sei.frame_packing.content_interpretation_type < 3) {
1210 H264SEIFramePacking *fp = &h->sei.frame_packing;
1211 AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
1213 switch (fp->frame_packing_arrangement_type) {
1215 stereo->type = AV_STEREO3D_CHECKERBOARD;
1218 stereo->type = AV_STEREO3D_COLUMNS;
1221 stereo->type = AV_STEREO3D_LINES;
1224 if (fp->quincunx_sampling_flag)
1225 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
1227 stereo->type = AV_STEREO3D_SIDEBYSIDE;
1230 stereo->type = AV_STEREO3D_TOPBOTTOM;
1233 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
1236 stereo->type = AV_STEREO3D_2D;
1240 if (fp->content_interpretation_type == 2)
1241 stereo->flags = AV_STEREO3D_FLAG_INVERT;
1245 if (h->sei.display_orientation.present &&
1246 (h->sei.display_orientation.anticlockwise_rotation ||
1247 h->sei.display_orientation.hflip ||
1248 h->sei.display_orientation.vflip)) {
1249 H264SEIDisplayOrientation *o = &h->sei.display_orientation;
1250 double angle = o->anticlockwise_rotation * 360 / (double) (1 << 16);
1251 AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
1252 AV_FRAME_DATA_DISPLAYMATRIX,
1253 sizeof(int32_t) * 9);
1255 av_display_rotation_set((int32_t *)rotation->data, angle);
1256 av_display_matrix_flip((int32_t *)rotation->data,
1257 o->hflip, o->vflip);
1261 if (h->sei.afd.present) {
1262 AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
1266 *sd->data = h->sei.afd.active_format_description;
1267 h->sei.afd.present = 0;
1271 if (h->sei.a53_caption.a53_caption) {
1272 H264SEIA53Caption *a53 = &h->sei.a53_caption;
1273 AVFrameSideData *sd = av_frame_new_side_data(cur->f,
1274 AV_FRAME_DATA_A53_CC,
1275 a53->a53_caption_size);
1277 memcpy(sd->data, a53->a53_caption, a53->a53_caption_size);
1278 av_freep(&a53->a53_caption);
1279 a53->a53_caption_size = 0;
1280 h->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
1286 /* This function is called right after decoding the slice header for a first
1287 * slice in a field (or a frame). It decides whether we are decoding a new frame
1288 * or a second field in a pair and does the necessary setup.
1290 static int h264_field_start(H264Context *h, const H264SliceContext *sl,
1291 const H2645NAL *nal, int first_slice)
1296 int last_pic_structure, last_pic_droppable, ret;
1298 ret = h264_init_ps(h, sl, first_slice);
1304 last_pic_droppable = h->droppable;
1305 last_pic_structure = h->picture_structure;
1306 h->droppable = (nal->ref_idc == 0);
1307 h->picture_structure = sl->picture_structure;
1309 h->poc.frame_num = sl->frame_num;
1310 h->poc.poc_lsb = sl->poc_lsb;
1311 h->poc.delta_poc_bottom = sl->delta_poc_bottom;
1312 h->poc.delta_poc[0] = sl->delta_poc[0];
1313 h->poc.delta_poc[1] = sl->delta_poc[1];
1315 /* Shorten frame num gaps so we don't have to allocate reference
1316 * frames just to throw them away */
1317 if (h->poc.frame_num != h->poc.prev_frame_num) {
1318 int unwrap_prev_frame_num = h->poc.prev_frame_num;
1319 int max_frame_num = 1 << sps->log2_max_frame_num;
1321 if (unwrap_prev_frame_num > h->poc.frame_num)
1322 unwrap_prev_frame_num -= max_frame_num;
1324 if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
1325 unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
1326 if (unwrap_prev_frame_num < 0)
1327 unwrap_prev_frame_num += max_frame_num;
1329 h->poc.prev_frame_num = unwrap_prev_frame_num;
1333 /* See if we have a decoded first field looking for a pair...
1334 * Here, we're using that to see if we should mark previously
1335 * decode frames as "finished".
1336 * We have to do that before the "dummy" in-between frame allocation,
1337 * since that can modify h->cur_pic_ptr. */
1338 if (h->first_field) {
1339 av_assert0(h->cur_pic_ptr);
1340 av_assert0(h->cur_pic_ptr->f->buf[0]);
1341 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1343 /* Mark old field/frame as completed */
1344 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1345 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1346 last_pic_structure == PICT_BOTTOM_FIELD);
1349 /* figure out if we have a complementary field pair */
1350 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1351 /* Previous field is unmatched. Don't display it, but let it
1352 * remain for reference if marked as such. */
1353 if (last_pic_structure != PICT_FRAME) {
1354 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1355 last_pic_structure == PICT_TOP_FIELD);
1358 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1359 /* This and previous field were reference, but had
1360 * different frame_nums. Consider this field first in
1361 * pair. Throw away previous field except for reference
1363 if (last_pic_structure != PICT_FRAME) {
1364 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1365 last_pic_structure == PICT_TOP_FIELD);
1368 /* Second field in complementary pair */
1369 if (!((last_pic_structure == PICT_TOP_FIELD &&
1370 h->picture_structure == PICT_BOTTOM_FIELD) ||
1371 (last_pic_structure == PICT_BOTTOM_FIELD &&
1372 h->picture_structure == PICT_TOP_FIELD))) {
1373 av_log(h->avctx, AV_LOG_ERROR,
1374 "Invalid field mode combination %d/%d\n",
1375 last_pic_structure, h->picture_structure);
1376 h->picture_structure = last_pic_structure;
1377 h->droppable = last_pic_droppable;
1378 return AVERROR_INVALIDDATA;
1379 } else if (last_pic_droppable != h->droppable) {
1380 avpriv_request_sample(h->avctx,
1381 "Found reference and non-reference fields in the same frame, which");
1382 h->picture_structure = last_pic_structure;
1383 h->droppable = last_pic_droppable;
1384 return AVERROR_PATCHWELCOME;
1390 while (h->poc.frame_num != h->poc.prev_frame_num && !h->first_field &&
1391 h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
1392 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1393 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1394 h->poc.frame_num, h->poc.prev_frame_num);
1395 if (!sps->gaps_in_frame_num_allowed_flag)
1396 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1397 h->last_pocs[i] = INT_MIN;
1398 ret = h264_frame_start(h);
1404 h->poc.prev_frame_num++;
1405 h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
1406 h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
1407 h->cur_pic_ptr->invalid_gap = !sps->gaps_in_frame_num_allowed_flag;
1408 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1409 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1411 h->explicit_ref_marking = 0;
1412 ret = ff_h264_execute_ref_pic_marking(h);
1413 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1415 /* Error concealment: If a ref is missing, copy the previous ref
1417 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1418 * many assumptions about there being no actual duplicates.
1419 * FIXME: This does not copy padding for out-of-frame motion
1420 * vectors. Given we are concealing a lost frame, this probably
1421 * is not noticeable by comparison, but it should be fixed. */
1422 if (h->short_ref_count) {
1424 h->short_ref[0]->f->width == prev->f->width &&
1425 h->short_ref[0]->f->height == prev->f->height &&
1426 h->short_ref[0]->f->format == prev->f->format) {
1427 av_image_copy(h->short_ref[0]->f->data,
1428 h->short_ref[0]->f->linesize,
1429 (const uint8_t **)prev->f->data,
1434 h->short_ref[0]->poc = prev->poc + 2;
1436 h->short_ref[0]->frame_num = h->poc.prev_frame_num;
1440 /* See if we have a decoded first field looking for a pair...
1441 * We're using that to see whether to continue decoding in that
1442 * frame, or to allocate a new one. */
1443 if (h->first_field) {
1444 av_assert0(h->cur_pic_ptr);
1445 av_assert0(h->cur_pic_ptr->f->buf[0]);
1446 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1448 /* figure out if we have a complementary field pair */
1449 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1450 /* Previous field is unmatched. Don't display it, but let it
1451 * remain for reference if marked as such. */
1452 h->missing_fields ++;
1453 h->cur_pic_ptr = NULL;
1454 h->first_field = FIELD_PICTURE(h);
1456 h->missing_fields = 0;
1457 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1458 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1459 h->picture_structure==PICT_BOTTOM_FIELD);
1460 /* This and the previous field had different frame_nums.
1461 * Consider this field first in pair. Throw away previous
1462 * one except for reference purposes. */
1464 h->cur_pic_ptr = NULL;
1466 /* Second field in complementary pair */
1471 /* Frame or first field in a potentially complementary pair */
1472 h->first_field = FIELD_PICTURE(h);
1475 if (!FIELD_PICTURE(h) || h->first_field) {
1476 if (h264_frame_start(h) < 0) {
1478 return AVERROR_INVALIDDATA;
1481 release_unused_pictures(h, 0);
1483 /* Some macroblocks can be accessed before they're available in case
1484 * of lost slices, MBAFF or threading. */
1485 if (FIELD_PICTURE(h)) {
1486 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1487 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1489 memset(h->slice_table, -1,
1490 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1493 ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
1494 h->ps.sps, &h->poc, h->picture_structure, nal->ref_idc);
1496 memcpy(h->mmco, sl->mmco, sl->nb_mmco * sizeof(*h->mmco));
1497 h->nb_mmco = sl->nb_mmco;
1498 h->explicit_ref_marking = sl->explicit_ref_marking;
1500 h->picture_idr = nal->type == H264_NAL_IDR_SLICE;
1502 if (h->sei.recovery_point.recovery_frame_cnt >= 0) {
1503 const int sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
1505 if (h->poc.frame_num != sei_recovery_frame_cnt || sl->slice_type_nos != AV_PICTURE_TYPE_I)
1506 h->valid_recovery_point = 1;
1508 if ( h->recovery_frame < 0
1509 || av_mod_uintp2(h->recovery_frame - h->poc.frame_num, h->ps.sps->log2_max_frame_num) > sei_recovery_frame_cnt) {
1510 h->recovery_frame = av_mod_uintp2(h->poc.frame_num + sei_recovery_frame_cnt, h->ps.sps->log2_max_frame_num);
1512 if (!h->valid_recovery_point)
1513 h->recovery_frame = h->poc.frame_num;
1517 h->cur_pic_ptr->f->key_frame |= (nal->type == H264_NAL_IDR_SLICE);
1519 if (nal->type == H264_NAL_IDR_SLICE ||
1520 (h->recovery_frame == h->poc.frame_num && nal->ref_idc)) {
1521 h->recovery_frame = -1;
1522 h->cur_pic_ptr->recovered = 1;
1524 // If we have an IDR, all frames after it in decoded order are
1526 if (nal->type == H264_NAL_IDR_SLICE)
1527 h->frame_recovered |= FRAME_RECOVERED_IDR;
1529 h->cur_pic_ptr->recovered |= h->frame_recovered;
1531 h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
1534 /* Set the frame properties/side data. Only done for the second field in
1535 * field coded frames, since some SEI information is present for each field
1536 * and is merged by the SEI parsing code. */
1537 if (!FIELD_PICTURE(h) || !h->first_field) {
1538 ret = h264_export_frame_props(h);
1546 static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl,
1547 const H2645NAL *nal)
1552 unsigned int slice_type, tmp, i;
1553 int field_pic_flag, bottom_field_flag;
1554 int first_slice = sl == h->slice_ctx && !h->current_slice;
1555 int picture_structure;
1558 av_assert0(!h->setup_finished);
1560 sl->first_mb_addr = get_ue_golomb_long(&sl->gb);
1562 slice_type = get_ue_golomb_31(&sl->gb);
1563 if (slice_type > 9) {
1564 av_log(h->avctx, AV_LOG_ERROR,
1565 "slice type %d too large at %d\n",
1566 slice_type, sl->first_mb_addr);
1567 return AVERROR_INVALIDDATA;
1569 if (slice_type > 4) {
1571 sl->slice_type_fixed = 1;
1573 sl->slice_type_fixed = 0;
1575 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1576 sl->slice_type = slice_type;
1577 sl->slice_type_nos = slice_type & 3;
1579 if (nal->type == H264_NAL_IDR_SLICE &&
1580 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1581 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1582 return AVERROR_INVALIDDATA;
1585 sl->pps_id = get_ue_golomb(&sl->gb);
1586 if (sl->pps_id >= MAX_PPS_COUNT) {
1587 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", sl->pps_id);
1588 return AVERROR_INVALIDDATA;
1590 if (!h->ps.pps_list[sl->pps_id]) {
1591 av_log(h->avctx, AV_LOG_ERROR,
1592 "non-existing PPS %u referenced\n",
1594 return AVERROR_INVALIDDATA;
1596 pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
1598 if (!h->ps.sps_list[pps->sps_id]) {
1599 av_log(h->avctx, AV_LOG_ERROR,
1600 "non-existing SPS %u referenced\n", pps->sps_id);
1601 return AVERROR_INVALIDDATA;
1603 sps = (const SPS*)h->ps.sps_list[pps->sps_id]->data;
1605 sl->frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
1607 if (h->poc.frame_num != sl->frame_num) {
1608 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1609 h->poc.frame_num, sl->frame_num);
1610 return AVERROR_INVALIDDATA;
1616 if (sps->frame_mbs_only_flag) {
1617 picture_structure = PICT_FRAME;
1619 if (!sps->direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1620 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1623 field_pic_flag = get_bits1(&sl->gb);
1624 if (field_pic_flag) {
1625 bottom_field_flag = get_bits1(&sl->gb);
1626 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1628 picture_structure = PICT_FRAME;
1631 sl->picture_structure = picture_structure;
1632 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1634 if (picture_structure == PICT_FRAME) {
1635 sl->curr_pic_num = sl->frame_num;
1636 sl->max_pic_num = 1 << sps->log2_max_frame_num;
1638 sl->curr_pic_num = 2 * sl->frame_num + 1;
1639 sl->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
1642 if (nal->type == H264_NAL_IDR_SLICE)
1643 get_ue_golomb_long(&sl->gb); /* idr_pic_id */
1645 if (sps->poc_type == 0) {
1646 sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
1648 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1649 sl->delta_poc_bottom = get_se_golomb(&sl->gb);
1652 if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
1653 sl->delta_poc[0] = get_se_golomb(&sl->gb);
1655 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1656 sl->delta_poc[1] = get_se_golomb(&sl->gb);
1659 sl->redundant_pic_count = 0;
1660 if (pps->redundant_pic_cnt_present)
1661 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1663 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1664 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1666 ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
1667 &sl->gb, pps, sl->slice_type_nos,
1668 picture_structure, h->avctx);
1672 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1673 ret = ff_h264_decode_ref_pic_list_reordering(sl, h->avctx);
1675 sl->ref_count[1] = sl->ref_count[0] = 0;
1680 sl->pwt.use_weight = 0;
1681 for (i = 0; i < 2; i++) {
1682 sl->pwt.luma_weight_flag[i] = 0;
1683 sl->pwt.chroma_weight_flag[i] = 0;
1685 if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1686 (pps->weighted_bipred_idc == 1 &&
1687 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1688 ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
1689 sl->slice_type_nos, &sl->pwt, h->avctx);
1691 sl->explicit_ref_marking = 0;
1693 ret = ff_h264_decode_ref_pic_marking(sl, &sl->gb, nal, h->avctx);
1694 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1695 return AVERROR_INVALIDDATA;
1698 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
1699 tmp = get_ue_golomb_31(&sl->gb);
1701 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1702 return AVERROR_INVALIDDATA;
1704 sl->cabac_init_idc = tmp;
1707 sl->last_qscale_diff = 0;
1708 tmp = pps->init_qp + get_se_golomb(&sl->gb);
1709 if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
1710 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1711 return AVERROR_INVALIDDATA;
1714 sl->chroma_qp[0] = get_chroma_qp(pps, 0, sl->qscale);
1715 sl->chroma_qp[1] = get_chroma_qp(pps, 1, sl->qscale);
1716 // FIXME qscale / qp ... stuff
1717 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1718 get_bits1(&sl->gb); /* sp_for_switch_flag */
1719 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1720 sl->slice_type == AV_PICTURE_TYPE_SI)
1721 get_se_golomb(&sl->gb); /* slice_qs_delta */
1723 sl->deblocking_filter = 1;
1724 sl->slice_alpha_c0_offset = 0;
1725 sl->slice_beta_offset = 0;
1726 if (pps->deblocking_filter_parameters_present) {
1727 tmp = get_ue_golomb_31(&sl->gb);
1729 av_log(h->avctx, AV_LOG_ERROR,
1730 "deblocking_filter_idc %u out of range\n", tmp);
1731 return AVERROR_INVALIDDATA;
1733 sl->deblocking_filter = tmp;
1734 if (sl->deblocking_filter < 2)
1735 sl->deblocking_filter ^= 1; // 1<->0
1737 if (sl->deblocking_filter) {
1738 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1739 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1740 if (sl->slice_alpha_c0_offset > 12 ||
1741 sl->slice_alpha_c0_offset < -12 ||
1742 sl->slice_beta_offset > 12 ||
1743 sl->slice_beta_offset < -12) {
1744 av_log(h->avctx, AV_LOG_ERROR,
1745 "deblocking filter parameters %d %d out of range\n",
1746 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1747 return AVERROR_INVALIDDATA;
1756 * Decode a slice header.
1757 * This will (re)initialize the decoder and call h264_frame_start() as needed.
1759 * @param h h264context
1761 * @return 0 if okay, <0 if an error occurred
1763 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl,
1764 const H2645NAL *nal)
1767 int first_slice = sl == h->slice_ctx && !h->current_slice;
1769 ret = h264_slice_header_parse(h, sl, nal);
1773 // discard redundant pictures
1774 if (sl->redundant_pic_count > 0)
1777 if (sl->first_mb_addr == 0 || !h->current_slice) {
1778 if (h->setup_finished) {
1779 av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
1780 return AVERROR_INVALIDDATA;
1784 if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
1785 if (h->current_slice) {
1786 if (h->max_contexts > 1) {
1787 if (!h->single_decode_warning) {
1788 av_log(h->avctx, AV_LOG_WARNING, "Cannot decode multiple access units as slice threads\n");
1789 h->single_decode_warning = 1;
1791 h->max_contexts = 1;
1792 return SLICE_SINGLETHREAD;
1795 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
1796 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1797 h->current_slice = 0;
1800 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == H264_NAL_IDR_SLICE) {
1801 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
1802 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1803 h->current_slice = 0;
1804 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1805 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1806 h->cur_pic_ptr = NULL;
1810 return AVERROR_INVALIDDATA;
1813 if (!h->first_field) {
1814 if (h->cur_pic_ptr && !h->droppable) {
1815 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1816 h->picture_structure == PICT_BOTTOM_FIELD);
1818 h->cur_pic_ptr = NULL;
1822 if (!h->current_slice)
1823 av_assert0(sl == h->slice_ctx);
1825 if (h->current_slice == 0 && !h->first_field) {
1827 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1828 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1829 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1830 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != H264_NAL_IDR_SLICE && h->sei.recovery_point.recovery_frame_cnt < 0) ||
1831 h->avctx->skip_frame >= AVDISCARD_ALL) {
1832 return SLICE_SKIPED;
1837 const PPS *pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
1839 if (h->ps.pps->sps_id != pps->sps_id ||
1840 h->ps.pps->transform_8x8_mode != pps->transform_8x8_mode /*||
1841 (h->setup_finished && h->ps.pps != pps)*/) {
1842 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1843 return AVERROR_INVALIDDATA;
1845 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
1846 av_log(h->avctx, AV_LOG_ERROR,
1847 "SPS changed in the middle of the frame\n");
1848 return AVERROR_INVALIDDATA;
1852 if (h->current_slice == 0) {
1853 ret = h264_field_start(h, sl, nal, first_slice);
1857 if (h->picture_structure != sl->picture_structure ||
1858 h->droppable != (nal->ref_idc == 0)) {
1859 av_log(h->avctx, AV_LOG_ERROR,
1860 "Changing field mode (%d -> %d) between slices is not allowed\n",
1861 h->picture_structure, sl->picture_structure);
1862 return AVERROR_INVALIDDATA;
1863 } else if (!h->cur_pic_ptr) {
1864 av_log(h->avctx, AV_LOG_ERROR,
1865 "unset cur_pic_ptr on slice %d\n",
1866 h->current_slice + 1);
1867 return AVERROR_INVALIDDATA;
1871 if (h->picture_idr && nal->type != H264_NAL_IDR_SLICE) {
1872 av_log(h->avctx, AV_LOG_ERROR, "Invalid mix of IDR and non-IDR slices\n");
1873 return AVERROR_INVALIDDATA;
1876 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1877 if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1878 sl->first_mb_addr >= h->mb_num) {
1879 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1880 return AVERROR_INVALIDDATA;
1882 sl->resync_mb_x = sl->mb_x = sl->first_mb_addr % h->mb_width;
1883 sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
1884 FIELD_OR_MBAFF_PICTURE(h);
1885 if (h->picture_structure == PICT_BOTTOM_FIELD)
1886 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1887 av_assert1(sl->mb_y < h->mb_height);
1889 ret = ff_h264_build_ref_list(h, sl);
1893 if (h->ps.pps->weighted_bipred_idc == 2 &&
1894 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1895 implicit_weight_table(h, sl, -1);
1896 if (FRAME_MBAFF(h)) {
1897 implicit_weight_table(h, sl, 0);
1898 implicit_weight_table(h, sl, 1);
1902 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1903 ff_h264_direct_dist_scale_factor(h, sl);
1904 ff_h264_direct_ref_list_init(h, sl);
1906 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1907 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1908 h->nal_unit_type != H264_NAL_IDR_SLICE) ||
1909 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1910 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1911 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1912 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1913 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1915 sl->deblocking_filter = 0;
1917 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1918 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1919 /* Cheat slightly for speed:
1920 * Do not bother to deblock across slices. */
1921 sl->deblocking_filter = 2;
1923 h->postpone_filter = 1;
1926 sl->qp_thresh = 15 -
1927 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1929 h->ps.pps->chroma_qp_index_offset[0],
1930 h->ps.pps->chroma_qp_index_offset[1]) +
1931 6 * (h->ps.sps->bit_depth_luma - 8);
1933 sl->slice_num = ++h->current_slice;
1936 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1937 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1938 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1939 && sl->slice_num >= MAX_SLICES) {
1940 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1941 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);
1944 for (j = 0; j < 2; j++) {
1946 int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1947 for (i = 0; i < 16; i++) {
1949 if (j < sl->list_count && i < sl->ref_count[j] &&
1950 sl->ref_list[j][i].parent->f->buf[0]) {
1952 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1953 for (k = 0; k < h->short_ref_count; k++)
1954 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1958 for (k = 0; k < h->long_ref_count; k++)
1959 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1960 id_list[i] = h->short_ref_count + k;
1968 for (i = 0; i < 16; i++)
1969 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1971 ref2frm[18 + 1] = -1;
1972 for (i = 16; i < 48; i++)
1973 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1974 (sl->ref_list[j][i].reference & 3);
1977 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1978 av_log(h->avctx, AV_LOG_DEBUG,
1979 "slice:%d %s mb:%d %c%s%s frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
1981 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1982 sl->mb_y * h->mb_width + sl->mb_x,
1983 av_get_picture_type_char(sl->slice_type),
1984 sl->slice_type_fixed ? " fix" : "",
1985 nal->type == H264_NAL_IDR_SLICE ? " IDR" : "",
1987 h->cur_pic_ptr->field_poc[0],
1988 h->cur_pic_ptr->field_poc[1],
1989 sl->ref_count[0], sl->ref_count[1],
1991 sl->deblocking_filter,
1992 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1994 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1995 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2001 int ff_h264_get_slice_type(const H264SliceContext *sl)
2003 switch (sl->slice_type) {
2004 case AV_PICTURE_TYPE_P:
2006 case AV_PICTURE_TYPE_B:
2008 case AV_PICTURE_TYPE_I:
2010 case AV_PICTURE_TYPE_SP:
2012 case AV_PICTURE_TYPE_SI:
2015 return AVERROR_INVALIDDATA;
2019 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
2020 H264SliceContext *sl,
2021 int mb_type, int top_xy,
2022 int left_xy[LEFT_MBS],
2024 int left_type[LEFT_MBS],
2025 int mb_xy, int list)
2027 int b_stride = h->b_stride;
2028 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
2029 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
2030 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2031 if (USES_LIST(top_type, list)) {
2032 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2033 const int b8_xy = 4 * top_xy + 2;
2034 const int *ref2frm = &h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2035 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2036 ref_cache[0 - 1 * 8] =
2037 ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
2038 ref_cache[2 - 1 * 8] =
2039 ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
2041 AV_ZERO128(mv_dst - 1 * 8);
2042 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2045 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2046 if (USES_LIST(left_type[LTOP], list)) {
2047 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2048 const int b8_xy = 4 * left_xy[LTOP] + 1;
2049 const int *ref2frm = &h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2050 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2051 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2052 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2053 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2055 ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2056 ref_cache[-1 + 16] =
2057 ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2059 AV_ZERO32(mv_dst - 1 + 0);
2060 AV_ZERO32(mv_dst - 1 + 8);
2061 AV_ZERO32(mv_dst - 1 + 16);
2062 AV_ZERO32(mv_dst - 1 + 24);
2065 ref_cache[-1 + 16] =
2066 ref_cache[-1 + 24] = LIST_NOT_USED;
2071 if (!USES_LIST(mb_type, list)) {
2072 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2073 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2074 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2075 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2076 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2081 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2082 const int *ref2frm = &h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2083 uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
2084 uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
2085 AV_WN32A(&ref_cache[0 * 8], ref01);
2086 AV_WN32A(&ref_cache[1 * 8], ref01);
2087 AV_WN32A(&ref_cache[2 * 8], ref23);
2088 AV_WN32A(&ref_cache[3 * 8], ref23);
2092 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2093 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2094 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2095 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2096 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2101 * @return non zero if the loop filter can be skipped
2103 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2105 const int mb_xy = sl->mb_xy;
2106 int top_xy, left_xy[LEFT_MBS];
2107 int top_type, left_type[LEFT_MBS];
2111 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2113 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2114 if (FRAME_MBAFF(h)) {
2115 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2116 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2118 if (left_mb_field_flag != curr_mb_field_flag)
2119 left_xy[LTOP] -= h->mb_stride;
2121 if (curr_mb_field_flag)
2122 top_xy += h->mb_stride &
2123 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2124 if (left_mb_field_flag != curr_mb_field_flag)
2125 left_xy[LBOT] += h->mb_stride;
2129 sl->top_mb_xy = top_xy;
2130 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2131 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2133 /* For sufficiently low qp, filtering wouldn't do anything.
2134 * This is a conservative estimate: could also check beta_offset
2135 * and more accurate chroma_qp. */
2136 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2137 int qp = h->cur_pic.qscale_table[mb_xy];
2138 if (qp <= qp_thresh &&
2139 (left_xy[LTOP] < 0 ||
2140 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2142 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2143 if (!FRAME_MBAFF(h))
2145 if ((left_xy[LTOP] < 0 ||
2146 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2147 (top_xy < h->mb_stride ||
2148 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2153 top_type = h->cur_pic.mb_type[top_xy];
2154 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2155 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2156 if (sl->deblocking_filter == 2) {
2157 if (h->slice_table[top_xy] != sl->slice_num)
2159 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2160 left_type[LTOP] = left_type[LBOT] = 0;
2162 if (h->slice_table[top_xy] == 0xFFFF)
2164 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2165 left_type[LTOP] = left_type[LBOT] = 0;
2167 sl->top_type = top_type;
2168 sl->left_type[LTOP] = left_type[LTOP];
2169 sl->left_type[LBOT] = left_type[LBOT];
2171 if (IS_INTRA(mb_type))
2174 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2175 top_type, left_type, mb_xy, 0);
2176 if (sl->list_count == 2)
2177 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2178 top_type, left_type, mb_xy, 1);
2180 nnz = h->non_zero_count[mb_xy];
2181 nnz_cache = sl->non_zero_count_cache;
2182 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2183 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2184 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2185 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2186 sl->cbp = h->cbp_table[mb_xy];
2189 nnz = h->non_zero_count[top_xy];
2190 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2193 if (left_type[LTOP]) {
2194 nnz = h->non_zero_count[left_xy[LTOP]];
2195 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2196 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2197 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2198 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2201 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2202 * from what the loop filter needs */
2203 if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
2204 if (IS_8x8DCT(top_type)) {
2205 nnz_cache[4 + 8 * 0] =
2206 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2207 nnz_cache[6 + 8 * 0] =
2208 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2210 if (IS_8x8DCT(left_type[LTOP])) {
2211 nnz_cache[3 + 8 * 1] =
2212 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2214 if (IS_8x8DCT(left_type[LBOT])) {
2215 nnz_cache[3 + 8 * 3] =
2216 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2219 if (IS_8x8DCT(mb_type)) {
2220 nnz_cache[scan8[0]] =
2221 nnz_cache[scan8[1]] =
2222 nnz_cache[scan8[2]] =
2223 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2225 nnz_cache[scan8[0 + 4]] =
2226 nnz_cache[scan8[1 + 4]] =
2227 nnz_cache[scan8[2 + 4]] =
2228 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2230 nnz_cache[scan8[0 + 8]] =
2231 nnz_cache[scan8[1 + 8]] =
2232 nnz_cache[scan8[2 + 8]] =
2233 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2235 nnz_cache[scan8[0 + 12]] =
2236 nnz_cache[scan8[1 + 12]] =
2237 nnz_cache[scan8[2 + 12]] =
2238 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2245 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2247 uint8_t *dest_y, *dest_cb, *dest_cr;
2248 int linesize, uvlinesize, mb_x, mb_y;
2249 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2250 const int old_slice_type = sl->slice_type;
2251 const int pixel_shift = h->pixel_shift;
2252 const int block_h = 16 >> h->chroma_y_shift;
2254 if (h->postpone_filter)
2257 if (sl->deblocking_filter) {
2258 for (mb_x = start_x; mb_x < end_x; mb_x++)
2259 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2261 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2262 mb_type = h->cur_pic.mb_type[mb_xy];
2266 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2270 dest_y = h->cur_pic.f->data[0] +
2271 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2272 dest_cb = h->cur_pic.f->data[1] +
2273 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2274 mb_y * sl->uvlinesize * block_h;
2275 dest_cr = h->cur_pic.f->data[2] +
2276 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2277 mb_y * sl->uvlinesize * block_h;
2278 // FIXME simplify above
2281 linesize = sl->mb_linesize = sl->linesize * 2;
2282 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2283 if (mb_y & 1) { // FIXME move out of this function?
2284 dest_y -= sl->linesize * 15;
2285 dest_cb -= sl->uvlinesize * (block_h - 1);
2286 dest_cr -= sl->uvlinesize * (block_h - 1);
2289 linesize = sl->mb_linesize = sl->linesize;
2290 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2292 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2294 if (fill_filter_caches(h, sl, mb_type))
2296 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mb_xy]);
2297 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mb_xy]);
2299 if (FRAME_MBAFF(h)) {
2300 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2301 linesize, uvlinesize);
2303 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2304 dest_cr, linesize, uvlinesize);
2308 sl->slice_type = old_slice_type;
2310 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2311 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
2312 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
2315 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2317 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2318 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2319 h->cur_pic.mb_type[mb_xy - 1] :
2320 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2321 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2322 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2326 * Draw edges and report progress for the last MB row.
2328 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2330 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2331 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2332 int height = 16 << FRAME_MBAFF(h);
2333 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2335 if (sl->deblocking_filter) {
2336 if ((top + height) >= pic_height)
2337 height += deblock_border;
2338 top -= deblock_border;
2341 if (top >= pic_height || (top + height) < 0)
2344 height = FFMIN(height, pic_height - top);
2346 height = top + height;
2350 ff_h264_draw_horiz_band(h, sl, top, height);
2352 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2355 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2356 h->picture_structure == PICT_BOTTOM_FIELD);
2359 static void er_add_slice(H264SliceContext *sl,
2360 int startx, int starty,
2361 int endx, int endy, int status)
2363 if (!sl->h264->enable_er)
2366 if (CONFIG_ERROR_RESILIENCE) {
2367 ERContext *er = &sl->h264->slice_ctx[0].er;
2369 ff_er_add_slice(er, startx, starty, endx, endy, status);
2373 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2375 H264SliceContext *sl = arg;
2376 const H264Context *h = sl->h264;
2377 int lf_x_start = sl->mb_x;
2378 int orig_deblock = sl->deblocking_filter;
2381 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2382 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2384 ret = alloc_scratch_buffers(sl, sl->linesize);
2388 sl->mb_skip_run = -1;
2390 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2392 if (h->postpone_filter)
2393 sl->deblocking_filter = 0;
2395 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2396 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2398 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2399 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2401 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2402 prev_status &= ~ VP_START;
2403 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2404 h->slice_ctx[0].er.error_occurred = 1;
2408 if (h->ps.pps->cabac) {
2410 align_get_bits(&sl->gb);
2413 ret = ff_init_cabac_decoder(&sl->cabac,
2414 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2415 (get_bits_left(&sl->gb) + 7) / 8);
2419 ff_h264_init_cabac_states(h, sl);
2424 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2425 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2426 sl->next_slice_idx);
2427 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2428 sl->mb_y, ER_MB_ERROR);
2429 return AVERROR_INVALIDDATA;
2432 ret = ff_h264_decode_mb_cabac(h, sl);
2433 // STOP_TIMER("decode_mb_cabac")
2436 ff_h264_hl_decode_mb(h, sl);
2438 // FIXME optimal? or let mb_decode decode 16x32 ?
2439 if (ret >= 0 && FRAME_MBAFF(h)) {
2442 ret = ff_h264_decode_mb_cabac(h, sl);
2445 ff_h264_hl_decode_mb(h, sl);
2448 eos = get_cabac_terminate(&sl->cabac);
2450 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2451 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2452 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2453 sl->mb_y, ER_MB_END);
2454 if (sl->mb_x >= lf_x_start)
2455 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2458 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2459 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2460 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2461 av_log(h->avctx, AV_LOG_ERROR,
2462 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2464 sl->cabac.bytestream_end - sl->cabac.bytestream);
2465 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2466 sl->mb_y, ER_MB_ERROR);
2467 return AVERROR_INVALIDDATA;
2470 if (++sl->mb_x >= h->mb_width) {
2471 loop_filter(h, sl, lf_x_start, sl->mb_x);
2472 sl->mb_x = lf_x_start = 0;
2473 decode_finish_row(h, sl);
2475 if (FIELD_OR_MBAFF_PICTURE(h)) {
2477 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2478 predict_field_decoding_flag(h, sl);
2482 if (eos || sl->mb_y >= h->mb_height) {
2483 ff_tlog(h->avctx, "slice end %d %d\n",
2484 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2485 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2486 sl->mb_y, ER_MB_END);
2487 if (sl->mb_x > lf_x_start)
2488 loop_filter(h, sl, lf_x_start, sl->mb_x);
2496 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2497 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2498 sl->next_slice_idx);
2499 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2500 sl->mb_y, ER_MB_ERROR);
2501 return AVERROR_INVALIDDATA;
2504 ret = ff_h264_decode_mb_cavlc(h, sl);
2507 ff_h264_hl_decode_mb(h, sl);
2509 // FIXME optimal? or let mb_decode decode 16x32 ?
2510 if (ret >= 0 && FRAME_MBAFF(h)) {
2512 ret = ff_h264_decode_mb_cavlc(h, sl);
2515 ff_h264_hl_decode_mb(h, sl);
2520 av_log(h->avctx, AV_LOG_ERROR,
2521 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2522 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2523 sl->mb_y, ER_MB_ERROR);
2527 if (++sl->mb_x >= h->mb_width) {
2528 loop_filter(h, sl, lf_x_start, sl->mb_x);
2529 sl->mb_x = lf_x_start = 0;
2530 decode_finish_row(h, sl);
2532 if (FIELD_OR_MBAFF_PICTURE(h)) {
2534 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2535 predict_field_decoding_flag(h, sl);
2537 if (sl->mb_y >= h->mb_height) {
2538 ff_tlog(h->avctx, "slice end %d %d\n",
2539 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2541 if ( get_bits_left(&sl->gb) == 0
2542 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2543 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2544 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2548 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2549 sl->mb_x, sl->mb_y, ER_MB_END);
2551 return AVERROR_INVALIDDATA;
2556 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2557 ff_tlog(h->avctx, "slice end %d %d\n",
2558 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2560 if (get_bits_left(&sl->gb) == 0) {
2561 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2562 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2563 if (sl->mb_x > lf_x_start)
2564 loop_filter(h, sl, lf_x_start, sl->mb_x);
2568 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2569 sl->mb_y, ER_MB_ERROR);
2571 return AVERROR_INVALIDDATA;
2578 sl->deblocking_filter = orig_deblock;
2583 * Call decode_slice() for each context.
2585 * @param h h264 master context
2586 * @param context_count number of contexts to execute
2588 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2590 AVCodecContext *const avctx = h->avctx;
2591 H264SliceContext *sl;
2594 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2596 h->slice_ctx[0].next_slice_idx = INT_MAX;
2598 if (h->avctx->hwaccel
2599 #if FF_API_CAP_VDPAU
2600 || h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU
2604 if (context_count == 1) {
2607 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2608 h->postpone_filter = 0;
2610 ret = decode_slice(avctx, &h->slice_ctx[0]);
2611 h->mb_y = h->slice_ctx[0].mb_y;
2614 av_assert0(context_count > 0);
2615 for (i = 0; i < context_count; i++) {
2616 int next_slice_idx = h->mb_width * h->mb_height;
2619 sl = &h->slice_ctx[i];
2620 if (CONFIG_ERROR_RESILIENCE) {
2621 sl->er.error_count = 0;
2624 /* make sure none of those slices overlap */
2625 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2626 for (j = 0; j < context_count; j++) {
2627 H264SliceContext *sl2 = &h->slice_ctx[j];
2628 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2630 if (i == j || slice_idx2 < slice_idx)
2632 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2634 sl->next_slice_idx = next_slice_idx;
2637 avctx->execute(avctx, decode_slice, h->slice_ctx,
2638 NULL, context_count, sizeof(h->slice_ctx[0]));
2640 /* pull back stuff from slices to master context */
2641 sl = &h->slice_ctx[context_count - 1];
2643 if (CONFIG_ERROR_RESILIENCE) {
2644 for (i = 1; i < context_count; i++)
2645 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
2648 if (h->postpone_filter) {
2649 h->postpone_filter = 0;
2651 for (i = 0; i < context_count; i++) {
2654 sl = &h->slice_ctx[i];
2655 y_end = FFMIN(sl->mb_y + 1, h->mb_height);
2656 x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
2658 for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
2660 loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
2661 j == y_end - 1 ? x_end : h->mb_width);