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 "mpegvideo.h"
48 #include "rectangle.h"
51 static const uint8_t field_scan[16+1] = {
52 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
53 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
54 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
55 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
58 static const uint8_t field_scan8x8[64+1] = {
59 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
60 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
61 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
62 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
63 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
64 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
65 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
66 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
67 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
68 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
69 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
70 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
71 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
72 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
73 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
74 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
77 static const uint8_t field_scan8x8_cavlc[64+1] = {
78 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
79 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
80 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
81 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
82 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
83 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
84 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
85 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
86 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
87 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
88 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
89 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
90 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
91 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
92 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
93 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
96 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
97 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
98 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
99 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
100 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
101 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
102 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
103 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
104 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
105 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
106 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
107 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
108 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
109 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
110 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
111 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
112 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
113 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
116 static void release_unused_pictures(H264Context *h, int remove_current)
120 /* release non reference frames */
121 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
122 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
123 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
124 ff_h264_unref_picture(h, &h->DPB[i]);
129 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
131 const H264Context *h = sl->h264;
132 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
134 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
135 // edge emu needs blocksize + filter length - 1
136 // (= 21x21 for H.264)
137 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
139 av_fast_mallocz(&sl->top_borders[0], &sl->top_borders_allocated[0],
140 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
141 av_fast_mallocz(&sl->top_borders[1], &sl->top_borders_allocated[1],
142 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
144 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
145 !sl->top_borders[0] || !sl->top_borders[1]) {
146 av_freep(&sl->bipred_scratchpad);
147 av_freep(&sl->edge_emu_buffer);
148 av_freep(&sl->top_borders[0]);
149 av_freep(&sl->top_borders[1]);
151 sl->bipred_scratchpad_allocated = 0;
152 sl->edge_emu_buffer_allocated = 0;
153 sl->top_borders_allocated[0] = 0;
154 sl->top_borders_allocated[1] = 0;
155 return AVERROR(ENOMEM);
161 static int init_table_pools(H264Context *h)
163 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
164 const int mb_array_size = h->mb_stride * h->mb_height;
165 const int b4_stride = h->mb_width * 4 + 1;
166 const int b4_array_size = b4_stride * h->mb_height * 4;
168 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
170 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
171 sizeof(uint32_t), av_buffer_allocz);
172 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
173 sizeof(int16_t), av_buffer_allocz);
174 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
176 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
177 !h->ref_index_pool) {
178 av_buffer_pool_uninit(&h->qscale_table_pool);
179 av_buffer_pool_uninit(&h->mb_type_pool);
180 av_buffer_pool_uninit(&h->motion_val_pool);
181 av_buffer_pool_uninit(&h->ref_index_pool);
182 return AVERROR(ENOMEM);
188 static int alloc_picture(H264Context *h, H264Picture *pic)
192 av_assert0(!pic->f->data[0]);
195 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
196 AV_GET_BUFFER_FLAG_REF : 0);
200 pic->crop = h->ps.sps->crop;
201 pic->crop_top = h->ps.sps->crop_top;
202 pic->crop_left= h->ps.sps->crop_left;
204 if (h->avctx->hwaccel) {
205 const AVHWAccel *hwaccel = h->avctx->hwaccel;
206 av_assert0(!pic->hwaccel_picture_private);
207 if (hwaccel->frame_priv_data_size) {
208 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
209 if (!pic->hwaccel_priv_buf)
210 return AVERROR(ENOMEM);
211 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
214 if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
215 int h_chroma_shift, v_chroma_shift;
216 av_pix_fmt_get_chroma_sub_sample(pic->f->format,
217 &h_chroma_shift, &v_chroma_shift);
219 for(i=0; i<AV_CEIL_RSHIFT(pic->f->height, v_chroma_shift); i++) {
220 memset(pic->f->data[1] + pic->f->linesize[1]*i,
221 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
222 memset(pic->f->data[2] + pic->f->linesize[2]*i,
223 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
227 if (!h->qscale_table_pool) {
228 ret = init_table_pools(h);
233 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
234 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
235 if (!pic->qscale_table_buf || !pic->mb_type_buf)
238 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
239 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
241 for (i = 0; i < 2; i++) {
242 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
243 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
244 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
247 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
248 pic->ref_index[i] = pic->ref_index_buf[i]->data;
253 ff_h264_unref_picture(h, pic);
254 return (ret < 0) ? ret : AVERROR(ENOMEM);
257 static int find_unused_picture(H264Context *h)
261 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
262 if (!h->DPB[i].f->buf[0])
265 return AVERROR_INVALIDDATA;
269 #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
271 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
272 (((pic) && (pic) >= (old_ctx)->DPB && \
273 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
274 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
276 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
277 H264Context *new_base,
278 H264Context *old_base)
282 for (i = 0; i < count; i++) {
283 av_assert1(!from[i] ||
284 IN_RANGE(from[i], old_base, 1) ||
285 IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
286 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
290 static int h264_slice_header_init(H264Context *h);
292 int ff_h264_update_thread_context(AVCodecContext *dst,
293 const AVCodecContext *src)
295 H264Context *h = dst->priv_data, *h1 = src->priv_data;
296 int inited = h->context_initialized, err = 0;
303 // We can't fail if SPS isn't set at it breaks current skip_frame code
305 // return AVERROR_INVALIDDATA;
308 (h->width != h1->width ||
309 h->height != h1->height ||
310 h->mb_width != h1->mb_width ||
311 h->mb_height != h1->mb_height ||
313 h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
314 h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
315 h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
319 /* copy block_offset since frame_start may not be called */
320 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
323 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
324 av_buffer_unref(&h->ps.sps_list[i]);
325 if (h1->ps.sps_list[i]) {
326 h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
327 if (!h->ps.sps_list[i])
328 return AVERROR(ENOMEM);
331 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
332 av_buffer_unref(&h->ps.pps_list[i]);
333 if (h1->ps.pps_list[i]) {
334 h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
335 if (!h->ps.pps_list[i])
336 return AVERROR(ENOMEM);
340 av_buffer_unref(&h->ps.pps_ref);
341 av_buffer_unref(&h->ps.sps_ref);
344 if (h1->ps.pps_ref) {
345 h->ps.pps_ref = av_buffer_ref(h1->ps.pps_ref);
347 return AVERROR(ENOMEM);
348 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
350 if (h1->ps.sps_ref) {
351 h->ps.sps_ref = av_buffer_ref(h1->ps.sps_ref);
353 return AVERROR(ENOMEM);
354 h->ps.sps = (const SPS*)h->ps.sps_ref->data;
357 if (need_reinit || !inited) {
358 h->width = h1->width;
359 h->height = h1->height;
360 h->mb_height = h1->mb_height;
361 h->mb_width = h1->mb_width;
362 h->mb_num = h1->mb_num;
363 h->mb_stride = h1->mb_stride;
364 h->b_stride = h1->b_stride;
366 if (h->context_initialized || h1->context_initialized) {
367 if ((err = h264_slice_header_init(h)) < 0) {
368 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
373 /* copy block_offset since frame_start may not be called */
374 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
377 h->avctx->coded_height = h1->avctx->coded_height;
378 h->avctx->coded_width = h1->avctx->coded_width;
379 h->avctx->width = h1->avctx->width;
380 h->avctx->height = h1->avctx->height;
381 h->coded_picture_number = h1->coded_picture_number;
382 h->first_field = h1->first_field;
383 h->picture_structure = h1->picture_structure;
384 h->mb_aff_frame = h1->mb_aff_frame;
385 h->droppable = h1->droppable;
386 h->backup_width = h1->backup_width;
387 h->backup_height = h1->backup_height;
388 h->backup_pix_fmt = h1->backup_pix_fmt;
390 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
391 ff_h264_unref_picture(h, &h->DPB[i]);
392 if (h1->DPB[i].f->buf[0] &&
393 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
397 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
398 ff_h264_unref_picture(h, &h->cur_pic);
399 if (h1->cur_pic.f->buf[0]) {
400 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
405 h->enable_er = h1->enable_er;
406 h->workaround_bugs = h1->workaround_bugs;
407 h->droppable = h1->droppable;
409 // extradata/NAL handling
410 h->is_avc = h1->is_avc;
411 h->nal_length_size = h1->nal_length_size;
412 h->sei.unregistered.x264_build = h1->sei.unregistered.x264_build;
414 memcpy(&h->poc, &h1->poc, sizeof(h->poc));
416 memcpy(h->default_ref, h1->default_ref, sizeof(h->default_ref));
417 memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
418 memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
419 memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
420 memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
422 h->next_output_pic = h1->next_output_pic;
423 h->next_outputed_poc = h1->next_outputed_poc;
425 memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
426 h->nb_mmco = h1->nb_mmco;
427 h->mmco_reset = h1->mmco_reset;
428 h->explicit_ref_marking = h1->explicit_ref_marking;
429 h->long_ref_count = h1->long_ref_count;
430 h->short_ref_count = h1->short_ref_count;
432 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
433 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
434 copy_picture_range(h->delayed_pic, h1->delayed_pic,
435 MAX_DELAYED_PIC_COUNT + 2, h, h1);
437 h->frame_recovered = h1->frame_recovered;
443 err = ff_h264_execute_ref_pic_marking(h);
444 h->poc.prev_poc_msb = h->poc.poc_msb;
445 h->poc.prev_poc_lsb = h->poc.poc_lsb;
447 h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
448 h->poc.prev_frame_num = h->poc.frame_num;
450 h->recovery_frame = h1->recovery_frame;
455 static int h264_frame_start(H264Context *h)
459 const int pixel_shift = h->pixel_shift;
461 1<<(h->ps.sps->bit_depth_luma-1),
462 1<<(h->ps.sps->bit_depth_chroma-1),
463 1<<(h->ps.sps->bit_depth_chroma-1),
467 if (!ff_thread_can_start_frame(h->avctx)) {
468 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
472 release_unused_pictures(h, 1);
473 h->cur_pic_ptr = NULL;
475 i = find_unused_picture(h);
477 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
482 pic->reference = h->droppable ? 0 : h->picture_structure;
483 pic->f->coded_picture_number = h->coded_picture_number++;
484 pic->field_picture = h->picture_structure != PICT_FRAME;
485 pic->frame_num = h->poc.frame_num;
487 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
489 * See decode_nal_units().
491 pic->f->key_frame = 0;
494 pic->invalid_gap = 0;
495 pic->sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
497 pic->f->pict_type = h->slice_ctx[0].slice_type;
499 if ((ret = alloc_picture(h, pic)) < 0)
501 if(!h->frame_recovered && !h->avctx->hwaccel
503 && !(h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
506 ff_color_frame(pic->f, c);
508 h->cur_pic_ptr = pic;
509 ff_h264_unref_picture(h, &h->cur_pic);
510 if (CONFIG_ERROR_RESILIENCE) {
511 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
514 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
517 for (i = 0; i < h->nb_slice_ctx; i++) {
518 h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
519 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
522 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
523 ff_er_frame_start(&h->slice_ctx[0].er);
524 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
525 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
528 for (i = 0; i < 16; i++) {
529 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
530 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
532 for (i = 0; i < 16; i++) {
533 h->block_offset[16 + i] =
534 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
535 h->block_offset[48 + 16 + i] =
536 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
539 /* We mark the current picture as non-reference after allocating it, so
540 * that if we break out due to an error it can be released automatically
541 * in the next ff_mpv_frame_start().
543 h->cur_pic_ptr->reference = 0;
545 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
547 h->next_output_pic = NULL;
549 h->postpone_filter = 0;
551 h->mb_aff_frame = h->ps.sps->mb_aff && (h->picture_structure == PICT_FRAME);
553 assert(h->cur_pic_ptr->long_ref == 0);
558 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
560 uint8_t *src_cb, uint8_t *src_cr,
561 int linesize, int uvlinesize,
566 const int pixel_shift = h->pixel_shift;
567 int chroma444 = CHROMA444(h);
568 int chroma422 = CHROMA422(h);
571 src_cb -= uvlinesize;
572 src_cr -= uvlinesize;
574 if (!simple && FRAME_MBAFF(h)) {
577 top_border = sl->top_borders[0][sl->mb_x];
578 AV_COPY128(top_border, src_y + 15 * linesize);
580 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
581 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
584 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
585 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
586 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
587 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
589 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
590 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
592 } else if (chroma422) {
594 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
595 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
597 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
598 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
602 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
603 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
605 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
606 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
611 } else if (MB_MBAFF(sl)) {
617 top_border = sl->top_borders[top_idx][sl->mb_x];
618 /* There are two lines saved, the line above the top macroblock
619 * of a pair, and the line above the bottom macroblock. */
620 AV_COPY128(top_border, src_y + 16 * linesize);
622 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
624 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
627 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
628 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
629 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
630 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
632 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
633 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
635 } else if (chroma422) {
637 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
638 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
640 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
641 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
645 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
646 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
648 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
649 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
656 * Initialize implicit_weight table.
657 * @param field 0/1 initialize the weight for interlaced MBAFF
658 * -1 initializes the rest
660 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
662 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
664 for (i = 0; i < 2; i++) {
665 sl->pwt.luma_weight_flag[i] = 0;
666 sl->pwt.chroma_weight_flag[i] = 0;
670 if (h->picture_structure == PICT_FRAME) {
671 cur_poc = h->cur_pic_ptr->poc;
673 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
675 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
676 sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2 * cur_poc) {
677 sl->pwt.use_weight = 0;
678 sl->pwt.use_weight_chroma = 0;
682 ref_count0 = sl->ref_count[0];
683 ref_count1 = sl->ref_count[1];
685 cur_poc = h->cur_pic_ptr->field_poc[field];
687 ref_count0 = 16 + 2 * sl->ref_count[0];
688 ref_count1 = 16 + 2 * sl->ref_count[1];
691 sl->pwt.use_weight = 2;
692 sl->pwt.use_weight_chroma = 2;
693 sl->pwt.luma_log2_weight_denom = 5;
694 sl->pwt.chroma_log2_weight_denom = 5;
696 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
697 int64_t poc0 = sl->ref_list[0][ref0].poc;
698 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
700 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
701 int poc1 = sl->ref_list[1][ref1].poc;
702 int td = av_clip_int8(poc1 - poc0);
704 int tb = av_clip_int8(cur_poc - poc0);
705 int tx = (16384 + (FFABS(td) >> 1)) / td;
706 int dist_scale_factor = (tb * tx + 32) >> 8;
707 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
708 w = 64 - dist_scale_factor;
712 sl->pwt.implicit_weight[ref0][ref1][0] =
713 sl->pwt.implicit_weight[ref0][ref1][1] = w;
715 sl->pwt.implicit_weight[ref0][ref1][field] = w;
722 * initialize scan tables
724 static void init_scan_tables(H264Context *h)
727 for (i = 0; i < 16; i++) {
728 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
729 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
730 h->field_scan[i] = TRANSPOSE(field_scan[i]);
733 for (i = 0; i < 64; i++) {
734 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
735 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
736 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
737 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
738 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
741 if (h->ps.sps->transform_bypass) { // FIXME same ugly
742 memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
743 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
744 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
745 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
746 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
747 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
749 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
750 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
751 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
752 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
753 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
754 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
758 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
760 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
761 CONFIG_H264_D3D11VA_HWACCEL + \
762 CONFIG_H264_VAAPI_HWACCEL + \
763 (CONFIG_H264_VDA_HWACCEL * 2) + \
764 CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
765 CONFIG_H264_VDPAU_HWACCEL)
766 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
767 const enum AVPixelFormat *choices = pix_fmts;
770 switch (h->ps.sps->bit_depth_luma) {
773 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
774 *fmt++ = AV_PIX_FMT_GBRP9;
776 *fmt++ = AV_PIX_FMT_YUV444P9;
777 } else if (CHROMA422(h))
778 *fmt++ = AV_PIX_FMT_YUV422P9;
780 *fmt++ = AV_PIX_FMT_YUV420P9;
784 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
785 *fmt++ = AV_PIX_FMT_GBRP10;
787 *fmt++ = AV_PIX_FMT_YUV444P10;
788 } else if (CHROMA422(h))
789 *fmt++ = AV_PIX_FMT_YUV422P10;
791 *fmt++ = AV_PIX_FMT_YUV420P10;
795 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
796 *fmt++ = AV_PIX_FMT_GBRP12;
798 *fmt++ = AV_PIX_FMT_YUV444P12;
799 } else if (CHROMA422(h))
800 *fmt++ = AV_PIX_FMT_YUV422P12;
802 *fmt++ = AV_PIX_FMT_YUV420P12;
806 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
807 *fmt++ = AV_PIX_FMT_GBRP14;
809 *fmt++ = AV_PIX_FMT_YUV444P14;
810 } else if (CHROMA422(h))
811 *fmt++ = AV_PIX_FMT_YUV422P14;
813 *fmt++ = AV_PIX_FMT_YUV420P14;
816 #if CONFIG_H264_VDPAU_HWACCEL
817 *fmt++ = AV_PIX_FMT_VDPAU;
820 if (h->avctx->colorspace == AVCOL_SPC_RGB)
821 *fmt++ = AV_PIX_FMT_GBRP;
822 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
823 *fmt++ = AV_PIX_FMT_YUVJ444P;
825 *fmt++ = AV_PIX_FMT_YUV444P;
826 } else if (CHROMA422(h)) {
827 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
828 *fmt++ = AV_PIX_FMT_YUVJ422P;
830 *fmt++ = AV_PIX_FMT_YUV422P;
832 #if CONFIG_H264_DXVA2_HWACCEL
833 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
835 #if CONFIG_H264_D3D11VA_HWACCEL
836 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
838 #if CONFIG_H264_VAAPI_HWACCEL
839 *fmt++ = AV_PIX_FMT_VAAPI;
841 #if CONFIG_H264_VDA_HWACCEL
842 *fmt++ = AV_PIX_FMT_VDA_VLD;
843 *fmt++ = AV_PIX_FMT_VDA;
845 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
846 *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
848 if (h->avctx->codec->pix_fmts)
849 choices = h->avctx->codec->pix_fmts;
850 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
851 *fmt++ = AV_PIX_FMT_YUVJ420P;
853 *fmt++ = AV_PIX_FMT_YUV420P;
857 av_log(h->avctx, AV_LOG_ERROR,
858 "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
859 return AVERROR_INVALIDDATA;
862 *fmt = AV_PIX_FMT_NONE;
864 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
865 if (choices[i] == h->avctx->pix_fmt && !force_callback)
867 return ff_thread_get_format(h->avctx, choices);
870 /* export coded and cropped frame dimensions to AVCodecContext */
871 static int init_dimensions(H264Context *h)
873 const SPS *sps = (const SPS*)h->ps.sps;
874 int width = h->width - (sps->crop_right + sps->crop_left);
875 int height = h->height - (sps->crop_top + sps->crop_bottom);
876 av_assert0(sps->crop_right + sps->crop_left < (unsigned)h->width);
877 av_assert0(sps->crop_top + sps->crop_bottom < (unsigned)h->height);
879 /* handle container cropping */
880 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
881 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16) &&
882 h->avctx->width <= width &&
883 h->avctx->height <= height
885 width = h->avctx->width;
886 height = h->avctx->height;
889 h->avctx->coded_width = h->width;
890 h->avctx->coded_height = h->height;
891 h->avctx->width = width;
892 h->avctx->height = height;
897 static int h264_slice_header_init(H264Context *h)
899 const SPS *sps = h->ps.sps;
902 ff_set_sar(h->avctx, sps->sar);
903 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
904 &h->chroma_x_shift, &h->chroma_y_shift);
906 if (sps->timing_info_present_flag) {
907 int64_t den = sps->time_scale;
908 if (h->sei.unregistered.x264_build < 44U)
910 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
911 sps->num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
914 ff_h264_free_tables(h);
917 h->prev_interlaced_frame = 1;
920 ret = ff_h264_alloc_tables(h);
922 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
927 if (h->avctx->codec &&
928 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU &&
929 (sps->bit_depth_luma != 8 || sps->chroma_format_idc > 1)) {
930 av_log(h->avctx, AV_LOG_ERROR,
931 "VDPAU decoding does not support video colorspace.\n");
932 ret = AVERROR_INVALIDDATA;
937 if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 14 ||
938 sps->bit_depth_luma == 11 || sps->bit_depth_luma == 13
940 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
941 sps->bit_depth_luma);
942 ret = AVERROR_INVALIDDATA;
946 h->cur_bit_depth_luma =
947 h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
948 h->cur_chroma_format_idc = sps->chroma_format_idc;
949 h->pixel_shift = sps->bit_depth_luma > 8;
950 h->chroma_format_idc = sps->chroma_format_idc;
951 h->bit_depth_luma = sps->bit_depth_luma;
953 ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
954 sps->chroma_format_idc);
955 ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
956 ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
957 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
958 sps->chroma_format_idc);
959 ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
961 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
962 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
964 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
968 for (i = 0; i < h->nb_slice_ctx; i++) {
969 H264SliceContext *sl = &h->slice_ctx[i];
972 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
973 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
974 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
976 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
977 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
983 h->context_initialized = 1;
987 ff_h264_free_tables(h);
988 h->context_initialized = 0;
992 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
995 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
996 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
997 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1003 static int h264_init_ps(H264Context *h, const H264SliceContext *sl, int first_slice)
1006 int needs_reinit = 0, must_reinit, ret;
1009 av_buffer_unref(&h->ps.pps_ref);
1011 h->ps.pps_ref = av_buffer_ref(h->ps.pps_list[sl->pps_id]);
1013 return AVERROR(ENOMEM);
1014 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
1017 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
1018 av_buffer_unref(&h->ps.sps_ref);
1020 h->ps.sps_ref = av_buffer_ref(h->ps.sps_list[h->ps.pps->sps_id]);
1022 return AVERROR(ENOMEM);
1023 h->ps.sps = (const SPS*)h->ps.sps_ref->data;
1025 if (h->mb_width != h->ps.sps->mb_width ||
1026 h->mb_height != h->ps.sps->mb_height ||
1027 h->cur_bit_depth_luma != h->ps.sps->bit_depth_luma ||
1028 h->cur_chroma_format_idc != h->ps.sps->chroma_format_idc
1032 if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
1033 h->chroma_format_idc != h->ps.sps->chroma_format_idc)
1038 must_reinit = (h->context_initialized &&
1039 ( 16*sps->mb_width != h->avctx->coded_width
1040 || 16*sps->mb_height != h->avctx->coded_height
1041 || h->cur_bit_depth_luma != sps->bit_depth_luma
1042 || h->cur_chroma_format_idc != sps->chroma_format_idc
1043 || h->mb_width != sps->mb_width
1044 || h->mb_height != sps->mb_height
1046 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1047 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1050 if (first_slice && av_cmp_q(sps->sar, h->avctx->sample_aspect_ratio))
1053 if (!h->setup_finished) {
1054 h->avctx->profile = ff_h264_get_profile(sps);
1055 h->avctx->level = sps->level_idc;
1056 h->avctx->refs = sps->ref_frame_count;
1058 h->mb_width = sps->mb_width;
1059 h->mb_height = sps->mb_height;
1060 h->mb_num = h->mb_width * h->mb_height;
1061 h->mb_stride = h->mb_width + 1;
1063 h->b_stride = h->mb_width * 4;
1065 h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
1067 h->width = 16 * h->mb_width;
1068 h->height = 16 * h->mb_height;
1070 ret = init_dimensions(h);
1074 if (sps->video_signal_type_present_flag) {
1075 h->avctx->color_range = sps->full_range > 0 ? AVCOL_RANGE_JPEG
1077 if (sps->colour_description_present_flag) {
1078 if (h->avctx->colorspace != sps->colorspace)
1080 h->avctx->color_primaries = sps->color_primaries;
1081 h->avctx->color_trc = sps->color_trc;
1082 h->avctx->colorspace = sps->colorspace;
1087 if (!h->context_initialized || must_reinit || needs_reinit) {
1088 int flush_changes = h->context_initialized;
1089 h->context_initialized = 0;
1090 if (sl != h->slice_ctx) {
1091 av_log(h->avctx, AV_LOG_ERROR,
1092 "changing width %d -> %d / height %d -> %d on "
1094 h->width, h->avctx->coded_width,
1095 h->height, h->avctx->coded_height,
1096 h->current_slice + 1);
1097 return AVERROR_INVALIDDATA;
1100 av_assert1(first_slice);
1103 ff_h264_flush_change(h);
1105 if ((ret = get_pixel_format(h, 1)) < 0)
1107 h->avctx->pix_fmt = ret;
1109 av_log(h->avctx, AV_LOG_VERBOSE, "Reinit context to %dx%d, "
1110 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1112 if ((ret = h264_slice_header_init(h)) < 0) {
1113 av_log(h->avctx, AV_LOG_ERROR,
1114 "h264_slice_header_init() failed\n");
1122 static int h264_export_frame_props(H264Context *h)
1124 const SPS *sps = h->ps.sps;
1125 H264Picture *cur = h->cur_pic_ptr;
1127 cur->f->interlaced_frame = 0;
1128 cur->f->repeat_pict = 0;
1130 /* Signal interlacing information externally. */
1131 /* Prioritize picture timing SEI information over used
1132 * decoding process if it exists. */
1134 if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
1135 H264SEIPictureTiming *pt = &h->sei.picture_timing;
1136 switch (pt->pic_struct) {
1137 case SEI_PIC_STRUCT_FRAME:
1139 case SEI_PIC_STRUCT_TOP_FIELD:
1140 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1141 cur->f->interlaced_frame = 1;
1143 case SEI_PIC_STRUCT_TOP_BOTTOM:
1144 case SEI_PIC_STRUCT_BOTTOM_TOP:
1145 if (FIELD_OR_MBAFF_PICTURE(h))
1146 cur->f->interlaced_frame = 1;
1148 // try to flag soft telecine progressive
1149 cur->f->interlaced_frame = h->prev_interlaced_frame;
1151 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1152 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1153 /* Signal the possibility of telecined film externally
1154 * (pic_struct 5,6). From these hints, let the applications
1155 * decide if they apply deinterlacing. */
1156 cur->f->repeat_pict = 1;
1158 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1159 cur->f->repeat_pict = 2;
1161 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1162 cur->f->repeat_pict = 4;
1166 if ((pt->ct_type & 3) &&
1167 pt->pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1168 cur->f->interlaced_frame = (pt->ct_type & (1 << 1)) != 0;
1170 /* Derive interlacing flag from used decoding process. */
1171 cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
1173 h->prev_interlaced_frame = cur->f->interlaced_frame;
1175 if (cur->field_poc[0] != cur->field_poc[1]) {
1176 /* Derive top_field_first from field pocs. */
1177 cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
1179 if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
1180 /* Use picture timing SEI information. Even if it is a
1181 * information of a past frame, better than nothing. */
1182 if (h->sei.picture_timing.pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1183 h->sei.picture_timing.pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1184 cur->f->top_field_first = 1;
1186 cur->f->top_field_first = 0;
1187 } else if (cur->f->interlaced_frame) {
1188 /* Default to top field first when pic_struct_present_flag
1189 * is not set but interlaced frame detected */
1190 cur->f->top_field_first = 1;
1192 /* Most likely progressive */
1193 cur->f->top_field_first = 0;
1197 if (h->sei.frame_packing.present &&
1198 h->sei.frame_packing.frame_packing_arrangement_type <= 6 &&
1199 h->sei.frame_packing.content_interpretation_type > 0 &&
1200 h->sei.frame_packing.content_interpretation_type < 3) {
1201 H264SEIFramePacking *fp = &h->sei.frame_packing;
1202 AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
1204 switch (fp->frame_packing_arrangement_type) {
1206 stereo->type = AV_STEREO3D_CHECKERBOARD;
1209 stereo->type = AV_STEREO3D_COLUMNS;
1212 stereo->type = AV_STEREO3D_LINES;
1215 if (fp->quincunx_sampling_flag)
1216 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
1218 stereo->type = AV_STEREO3D_SIDEBYSIDE;
1221 stereo->type = AV_STEREO3D_TOPBOTTOM;
1224 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
1227 stereo->type = AV_STEREO3D_2D;
1231 if (fp->content_interpretation_type == 2)
1232 stereo->flags = AV_STEREO3D_FLAG_INVERT;
1236 if (h->sei.display_orientation.present &&
1237 (h->sei.display_orientation.anticlockwise_rotation ||
1238 h->sei.display_orientation.hflip ||
1239 h->sei.display_orientation.vflip)) {
1240 H264SEIDisplayOrientation *o = &h->sei.display_orientation;
1241 double angle = o->anticlockwise_rotation * 360 / (double) (1 << 16);
1242 AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
1243 AV_FRAME_DATA_DISPLAYMATRIX,
1244 sizeof(int32_t) * 9);
1246 av_display_rotation_set((int32_t *)rotation->data, angle);
1247 av_display_matrix_flip((int32_t *)rotation->data,
1248 o->hflip, o->vflip);
1252 if (h->sei.afd.present) {
1253 AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
1257 *sd->data = h->sei.afd.active_format_description;
1258 h->sei.afd.present = 0;
1262 if (h->sei.a53_caption.a53_caption) {
1263 H264SEIA53Caption *a53 = &h->sei.a53_caption;
1264 AVFrameSideData *sd = av_frame_new_side_data(cur->f,
1265 AV_FRAME_DATA_A53_CC,
1266 a53->a53_caption_size);
1268 memcpy(sd->data, a53->a53_caption, a53->a53_caption_size);
1269 av_freep(&a53->a53_caption);
1270 a53->a53_caption_size = 0;
1271 h->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
1277 static int h264_select_output_frame(H264Context *h)
1279 const SPS *sps = h->ps.sps;
1280 H264Picture *out = h->cur_pic_ptr;
1281 H264Picture *cur = h->cur_pic_ptr;
1282 int i, pics, out_of_order, out_idx;
1284 cur->mmco_reset = h->mmco_reset;
1287 if (sps->bitstream_restriction_flag ||
1288 h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT) {
1289 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, sps->num_reorder_frames);
1292 for (i = 0; 1; i++) {
1293 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1295 h->last_pocs[i-1] = cur->poc;
1298 h->last_pocs[i-1]= h->last_pocs[i];
1301 out_of_order = MAX_DELAYED_PIC_COUNT - i;
1302 if( cur->f->pict_type == AV_PICTURE_TYPE_B
1303 || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
1304 out_of_order = FFMAX(out_of_order, 1);
1305 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
1306 av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
1307 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
1308 h->last_pocs[i] = INT_MIN;
1309 h->last_pocs[0] = cur->poc;
1310 cur->mmco_reset = 1;
1311 } else if(h->avctx->has_b_frames < out_of_order && !sps->bitstream_restriction_flag){
1312 int loglevel = h->avctx->frame_number > 1 ? AV_LOG_WARNING : AV_LOG_VERBOSE;
1313 av_log(h->avctx, loglevel, "Increasing reorder buffer to %d\n", out_of_order);
1314 h->avctx->has_b_frames = out_of_order;
1318 while (h->delayed_pic[pics])
1321 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1323 h->delayed_pic[pics++] = cur;
1324 if (cur->reference == 0)
1325 cur->reference = DELAYED_PIC_REF;
1327 out = h->delayed_pic[0];
1329 for (i = 1; h->delayed_pic[i] &&
1330 !h->delayed_pic[i]->f->key_frame &&
1331 !h->delayed_pic[i]->mmco_reset;
1333 if (h->delayed_pic[i]->poc < out->poc) {
1334 out = h->delayed_pic[i];
1337 if (h->avctx->has_b_frames == 0 &&
1338 (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset))
1339 h->next_outputed_poc = INT_MIN;
1340 out_of_order = out->poc < h->next_outputed_poc;
1342 if (out_of_order || pics > h->avctx->has_b_frames) {
1343 out->reference &= ~DELAYED_PIC_REF;
1344 for (i = out_idx; h->delayed_pic[i]; i++)
1345 h->delayed_pic[i] = h->delayed_pic[i + 1];
1347 if (!out_of_order && pics > h->avctx->has_b_frames) {
1348 h->next_output_pic = out;
1349 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset)) {
1350 h->next_outputed_poc = INT_MIN;
1352 h->next_outputed_poc = out->poc;
1354 if (out->recovered) {
1355 // We have reached an recovery point and all frames after it in
1356 // display order are "recovered".
1357 h->frame_recovered |= FRAME_RECOVERED_SEI;
1359 out->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
1361 if (!out->recovered) {
1362 if (!(h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) &&
1363 !(h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL)) {
1364 h->next_output_pic = NULL;
1366 out->f->flags |= AV_FRAME_FLAG_CORRUPT;
1370 av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1376 /* This function is called right after decoding the slice header for a first
1377 * slice in a field (or a frame). It decides whether we are decoding a new frame
1378 * or a second field in a pair and does the necessary setup.
1380 static int h264_field_start(H264Context *h, const H264SliceContext *sl,
1381 const H2645NAL *nal, int first_slice)
1386 int last_pic_structure, last_pic_droppable, ret;
1388 ret = h264_init_ps(h, sl, first_slice);
1394 last_pic_droppable = h->droppable;
1395 last_pic_structure = h->picture_structure;
1396 h->droppable = (nal->ref_idc == 0);
1397 h->picture_structure = sl->picture_structure;
1399 h->poc.frame_num = sl->frame_num;
1400 h->poc.poc_lsb = sl->poc_lsb;
1401 h->poc.delta_poc_bottom = sl->delta_poc_bottom;
1402 h->poc.delta_poc[0] = sl->delta_poc[0];
1403 h->poc.delta_poc[1] = sl->delta_poc[1];
1405 /* Shorten frame num gaps so we don't have to allocate reference
1406 * frames just to throw them away */
1407 if (h->poc.frame_num != h->poc.prev_frame_num) {
1408 int unwrap_prev_frame_num = h->poc.prev_frame_num;
1409 int max_frame_num = 1 << sps->log2_max_frame_num;
1411 if (unwrap_prev_frame_num > h->poc.frame_num)
1412 unwrap_prev_frame_num -= max_frame_num;
1414 if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
1415 unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
1416 if (unwrap_prev_frame_num < 0)
1417 unwrap_prev_frame_num += max_frame_num;
1419 h->poc.prev_frame_num = unwrap_prev_frame_num;
1423 /* See if we have a decoded first field looking for a pair...
1424 * Here, we're using that to see if we should mark previously
1425 * decode frames as "finished".
1426 * We have to do that before the "dummy" in-between frame allocation,
1427 * since that can modify h->cur_pic_ptr. */
1428 if (h->first_field) {
1429 av_assert0(h->cur_pic_ptr);
1430 av_assert0(h->cur_pic_ptr->f->buf[0]);
1431 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1433 /* Mark old field/frame as completed */
1434 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1435 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1436 last_pic_structure == PICT_BOTTOM_FIELD);
1439 /* figure out if we have a complementary field pair */
1440 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1441 /* Previous field is unmatched. Don't display it, but let it
1442 * remain for reference if marked as such. */
1443 if (last_pic_structure != PICT_FRAME) {
1444 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1445 last_pic_structure == PICT_TOP_FIELD);
1448 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1449 /* This and previous field were reference, but had
1450 * different frame_nums. Consider this field first in
1451 * pair. Throw away previous field except for reference
1453 if (last_pic_structure != PICT_FRAME) {
1454 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1455 last_pic_structure == PICT_TOP_FIELD);
1458 /* Second field in complementary pair */
1459 if (!((last_pic_structure == PICT_TOP_FIELD &&
1460 h->picture_structure == PICT_BOTTOM_FIELD) ||
1461 (last_pic_structure == PICT_BOTTOM_FIELD &&
1462 h->picture_structure == PICT_TOP_FIELD))) {
1463 av_log(h->avctx, AV_LOG_ERROR,
1464 "Invalid field mode combination %d/%d\n",
1465 last_pic_structure, h->picture_structure);
1466 h->picture_structure = last_pic_structure;
1467 h->droppable = last_pic_droppable;
1468 return AVERROR_INVALIDDATA;
1469 } else if (last_pic_droppable != h->droppable) {
1470 avpriv_request_sample(h->avctx,
1471 "Found reference and non-reference fields in the same frame, which");
1472 h->picture_structure = last_pic_structure;
1473 h->droppable = last_pic_droppable;
1474 return AVERROR_PATCHWELCOME;
1480 while (h->poc.frame_num != h->poc.prev_frame_num && !h->first_field &&
1481 h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
1482 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1483 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1484 h->poc.frame_num, h->poc.prev_frame_num);
1485 if (!sps->gaps_in_frame_num_allowed_flag)
1486 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1487 h->last_pocs[i] = INT_MIN;
1488 ret = h264_frame_start(h);
1494 h->poc.prev_frame_num++;
1495 h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
1496 h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
1497 h->cur_pic_ptr->invalid_gap = !sps->gaps_in_frame_num_allowed_flag;
1498 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1499 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1501 h->explicit_ref_marking = 0;
1502 ret = ff_h264_execute_ref_pic_marking(h);
1503 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1505 /* Error concealment: If a ref is missing, copy the previous ref
1507 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1508 * many assumptions about there being no actual duplicates.
1509 * FIXME: This does not copy padding for out-of-frame motion
1510 * vectors. Given we are concealing a lost frame, this probably
1511 * is not noticeable by comparison, but it should be fixed. */
1512 if (h->short_ref_count) {
1514 h->short_ref[0]->f->width == prev->f->width &&
1515 h->short_ref[0]->f->height == prev->f->height &&
1516 h->short_ref[0]->f->format == prev->f->format) {
1517 ff_thread_await_progress(&prev->tf, INT_MAX, 0);
1518 if (prev->field_picture)
1519 ff_thread_await_progress(&prev->tf, INT_MAX, 1);
1520 av_image_copy(h->short_ref[0]->f->data,
1521 h->short_ref[0]->f->linesize,
1522 (const uint8_t **)prev->f->data,
1527 h->short_ref[0]->poc = prev->poc + 2;
1529 h->short_ref[0]->frame_num = h->poc.prev_frame_num;
1533 /* See if we have a decoded first field looking for a pair...
1534 * We're using that to see whether to continue decoding in that
1535 * frame, or to allocate a new one. */
1536 if (h->first_field) {
1537 av_assert0(h->cur_pic_ptr);
1538 av_assert0(h->cur_pic_ptr->f->buf[0]);
1539 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1541 /* figure out if we have a complementary field pair */
1542 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1543 /* Previous field is unmatched. Don't display it, but let it
1544 * remain for reference if marked as such. */
1545 h->missing_fields ++;
1546 h->cur_pic_ptr = NULL;
1547 h->first_field = FIELD_PICTURE(h);
1549 h->missing_fields = 0;
1550 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1551 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1552 h->picture_structure==PICT_BOTTOM_FIELD);
1553 /* This and the previous field had different frame_nums.
1554 * Consider this field first in pair. Throw away previous
1555 * one except for reference purposes. */
1557 h->cur_pic_ptr = NULL;
1559 /* Second field in complementary pair */
1564 /* Frame or first field in a potentially complementary pair */
1565 h->first_field = FIELD_PICTURE(h);
1568 if (!FIELD_PICTURE(h) || h->first_field) {
1569 if (h264_frame_start(h) < 0) {
1571 return AVERROR_INVALIDDATA;
1574 release_unused_pictures(h, 0);
1576 /* Some macroblocks can be accessed before they're available in case
1577 * of lost slices, MBAFF or threading. */
1578 if (FIELD_PICTURE(h)) {
1579 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1580 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1582 memset(h->slice_table, -1,
1583 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1586 ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
1587 h->ps.sps, &h->poc, h->picture_structure, nal->ref_idc);
1589 memcpy(h->mmco, sl->mmco, sl->nb_mmco * sizeof(*h->mmco));
1590 h->nb_mmco = sl->nb_mmco;
1591 h->explicit_ref_marking = sl->explicit_ref_marking;
1593 h->picture_idr = nal->type == H264_NAL_IDR_SLICE;
1595 if (h->sei.recovery_point.recovery_frame_cnt >= 0) {
1596 const int sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
1598 if (h->poc.frame_num != sei_recovery_frame_cnt || sl->slice_type_nos != AV_PICTURE_TYPE_I)
1599 h->valid_recovery_point = 1;
1601 if ( h->recovery_frame < 0
1602 || av_mod_uintp2(h->recovery_frame - h->poc.frame_num, h->ps.sps->log2_max_frame_num) > sei_recovery_frame_cnt) {
1603 h->recovery_frame = av_mod_uintp2(h->poc.frame_num + sei_recovery_frame_cnt, h->ps.sps->log2_max_frame_num);
1605 if (!h->valid_recovery_point)
1606 h->recovery_frame = h->poc.frame_num;
1610 h->cur_pic_ptr->f->key_frame |= (nal->type == H264_NAL_IDR_SLICE);
1612 if (nal->type == H264_NAL_IDR_SLICE ||
1613 (h->recovery_frame == h->poc.frame_num && nal->ref_idc)) {
1614 h->recovery_frame = -1;
1615 h->cur_pic_ptr->recovered = 1;
1617 // If we have an IDR, all frames after it in decoded order are
1619 if (nal->type == H264_NAL_IDR_SLICE)
1620 h->frame_recovered |= FRAME_RECOVERED_IDR;
1622 h->cur_pic_ptr->recovered |= h->frame_recovered;
1624 h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
1627 /* Set the frame properties/side data. Only done for the second field in
1628 * field coded frames, since some SEI information is present for each field
1629 * and is merged by the SEI parsing code. */
1630 if (!FIELD_PICTURE(h) || !h->first_field || h->missing_fields > 1) {
1631 ret = h264_export_frame_props(h);
1635 ret = h264_select_output_frame(h);
1643 static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl,
1644 const H2645NAL *nal)
1649 unsigned int slice_type, tmp, i;
1650 int field_pic_flag, bottom_field_flag;
1651 int first_slice = sl == h->slice_ctx && !h->current_slice;
1652 int picture_structure;
1655 av_assert0(!h->setup_finished);
1657 sl->first_mb_addr = get_ue_golomb_long(&sl->gb);
1659 slice_type = get_ue_golomb_31(&sl->gb);
1660 if (slice_type > 9) {
1661 av_log(h->avctx, AV_LOG_ERROR,
1662 "slice type %d too large at %d\n",
1663 slice_type, sl->first_mb_addr);
1664 return AVERROR_INVALIDDATA;
1666 if (slice_type > 4) {
1668 sl->slice_type_fixed = 1;
1670 sl->slice_type_fixed = 0;
1672 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1673 sl->slice_type = slice_type;
1674 sl->slice_type_nos = slice_type & 3;
1676 if (nal->type == H264_NAL_IDR_SLICE &&
1677 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1678 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1679 return AVERROR_INVALIDDATA;
1682 sl->pps_id = get_ue_golomb(&sl->gb);
1683 if (sl->pps_id >= MAX_PPS_COUNT) {
1684 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", sl->pps_id);
1685 return AVERROR_INVALIDDATA;
1687 if (!h->ps.pps_list[sl->pps_id]) {
1688 av_log(h->avctx, AV_LOG_ERROR,
1689 "non-existing PPS %u referenced\n",
1691 return AVERROR_INVALIDDATA;
1693 pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
1695 if (!h->ps.sps_list[pps->sps_id]) {
1696 av_log(h->avctx, AV_LOG_ERROR,
1697 "non-existing SPS %u referenced\n", pps->sps_id);
1698 return AVERROR_INVALIDDATA;
1700 sps = (const SPS*)h->ps.sps_list[pps->sps_id]->data;
1702 sl->frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
1704 if (h->poc.frame_num != sl->frame_num) {
1705 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1706 h->poc.frame_num, sl->frame_num);
1707 return AVERROR_INVALIDDATA;
1713 if (sps->frame_mbs_only_flag) {
1714 picture_structure = PICT_FRAME;
1716 if (!sps->direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1717 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1720 field_pic_flag = get_bits1(&sl->gb);
1721 if (field_pic_flag) {
1722 bottom_field_flag = get_bits1(&sl->gb);
1723 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1725 picture_structure = PICT_FRAME;
1728 sl->picture_structure = picture_structure;
1729 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1731 if (picture_structure == PICT_FRAME) {
1732 sl->curr_pic_num = sl->frame_num;
1733 sl->max_pic_num = 1 << sps->log2_max_frame_num;
1735 sl->curr_pic_num = 2 * sl->frame_num + 1;
1736 sl->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
1739 if (nal->type == H264_NAL_IDR_SLICE)
1740 get_ue_golomb_long(&sl->gb); /* idr_pic_id */
1742 if (sps->poc_type == 0) {
1743 sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
1745 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1746 sl->delta_poc_bottom = get_se_golomb(&sl->gb);
1749 if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
1750 sl->delta_poc[0] = get_se_golomb(&sl->gb);
1752 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1753 sl->delta_poc[1] = get_se_golomb(&sl->gb);
1756 sl->redundant_pic_count = 0;
1757 if (pps->redundant_pic_cnt_present)
1758 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1760 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1761 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1763 ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
1764 &sl->gb, pps, sl->slice_type_nos,
1765 picture_structure, h->avctx);
1769 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1770 ret = ff_h264_decode_ref_pic_list_reordering(sl, h->avctx);
1772 sl->ref_count[1] = sl->ref_count[0] = 0;
1777 sl->pwt.use_weight = 0;
1778 for (i = 0; i < 2; i++) {
1779 sl->pwt.luma_weight_flag[i] = 0;
1780 sl->pwt.chroma_weight_flag[i] = 0;
1782 if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1783 (pps->weighted_bipred_idc == 1 &&
1784 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1785 ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
1786 sl->slice_type_nos, &sl->pwt, h->avctx);
1788 sl->explicit_ref_marking = 0;
1790 ret = ff_h264_decode_ref_pic_marking(sl, &sl->gb, nal, h->avctx);
1791 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1792 return AVERROR_INVALIDDATA;
1795 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
1796 tmp = get_ue_golomb_31(&sl->gb);
1798 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1799 return AVERROR_INVALIDDATA;
1801 sl->cabac_init_idc = tmp;
1804 sl->last_qscale_diff = 0;
1805 tmp = pps->init_qp + get_se_golomb(&sl->gb);
1806 if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
1807 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1808 return AVERROR_INVALIDDATA;
1811 sl->chroma_qp[0] = get_chroma_qp(pps, 0, sl->qscale);
1812 sl->chroma_qp[1] = get_chroma_qp(pps, 1, sl->qscale);
1813 // FIXME qscale / qp ... stuff
1814 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1815 get_bits1(&sl->gb); /* sp_for_switch_flag */
1816 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1817 sl->slice_type == AV_PICTURE_TYPE_SI)
1818 get_se_golomb(&sl->gb); /* slice_qs_delta */
1820 sl->deblocking_filter = 1;
1821 sl->slice_alpha_c0_offset = 0;
1822 sl->slice_beta_offset = 0;
1823 if (pps->deblocking_filter_parameters_present) {
1824 tmp = get_ue_golomb_31(&sl->gb);
1826 av_log(h->avctx, AV_LOG_ERROR,
1827 "deblocking_filter_idc %u out of range\n", tmp);
1828 return AVERROR_INVALIDDATA;
1830 sl->deblocking_filter = tmp;
1831 if (sl->deblocking_filter < 2)
1832 sl->deblocking_filter ^= 1; // 1<->0
1834 if (sl->deblocking_filter) {
1835 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1836 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1837 if (sl->slice_alpha_c0_offset > 12 ||
1838 sl->slice_alpha_c0_offset < -12 ||
1839 sl->slice_beta_offset > 12 ||
1840 sl->slice_beta_offset < -12) {
1841 av_log(h->avctx, AV_LOG_ERROR,
1842 "deblocking filter parameters %d %d out of range\n",
1843 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1844 return AVERROR_INVALIDDATA;
1852 /* do all the per-slice initialization needed before we can start decoding the
1854 static int h264_slice_init(H264Context *h, H264SliceContext *sl,
1855 const H2645NAL *nal)
1859 if (h->picture_idr && nal->type != H264_NAL_IDR_SLICE) {
1860 av_log(h->avctx, AV_LOG_ERROR, "Invalid mix of IDR and non-IDR slices\n");
1861 return AVERROR_INVALIDDATA;
1864 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1865 if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1866 sl->first_mb_addr >= h->mb_num) {
1867 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1868 return AVERROR_INVALIDDATA;
1870 sl->resync_mb_x = sl->mb_x = sl->first_mb_addr % h->mb_width;
1871 sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
1872 FIELD_OR_MBAFF_PICTURE(h);
1873 if (h->picture_structure == PICT_BOTTOM_FIELD)
1874 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1875 av_assert1(sl->mb_y < h->mb_height);
1877 ret = ff_h264_build_ref_list(h, sl);
1881 if (h->ps.pps->weighted_bipred_idc == 2 &&
1882 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1883 implicit_weight_table(h, sl, -1);
1884 if (FRAME_MBAFF(h)) {
1885 implicit_weight_table(h, sl, 0);
1886 implicit_weight_table(h, sl, 1);
1890 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1891 ff_h264_direct_dist_scale_factor(h, sl);
1892 ff_h264_direct_ref_list_init(h, sl);
1894 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1895 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1896 h->nal_unit_type != H264_NAL_IDR_SLICE) ||
1897 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1898 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1899 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1900 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1901 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1903 sl->deblocking_filter = 0;
1905 if (sl->deblocking_filter == 1 && h->nb_slice_ctx > 1) {
1906 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1907 /* Cheat slightly for speed:
1908 * Do not bother to deblock across slices. */
1909 sl->deblocking_filter = 2;
1911 h->postpone_filter = 1;
1914 sl->qp_thresh = 15 -
1915 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1917 h->ps.pps->chroma_qp_index_offset[0],
1918 h->ps.pps->chroma_qp_index_offset[1]) +
1919 6 * (h->ps.sps->bit_depth_luma - 8);
1921 sl->slice_num = ++h->current_slice;
1924 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1925 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1926 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1927 && sl->slice_num >= MAX_SLICES) {
1928 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1929 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);
1932 for (j = 0; j < 2; j++) {
1934 int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1935 for (i = 0; i < 16; i++) {
1937 if (j < sl->list_count && i < sl->ref_count[j] &&
1938 sl->ref_list[j][i].parent->f->buf[0]) {
1940 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1941 for (k = 0; k < h->short_ref_count; k++)
1942 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1946 for (k = 0; k < h->long_ref_count; k++)
1947 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1948 id_list[i] = h->short_ref_count + k;
1956 for (i = 0; i < 16; i++)
1957 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1959 ref2frm[18 + 1] = -1;
1960 for (i = 16; i < 48; i++)
1961 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1962 (sl->ref_list[j][i].reference & 3);
1965 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1966 av_log(h->avctx, AV_LOG_DEBUG,
1967 "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",
1969 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1970 sl->mb_y * h->mb_width + sl->mb_x,
1971 av_get_picture_type_char(sl->slice_type),
1972 sl->slice_type_fixed ? " fix" : "",
1973 nal->type == H264_NAL_IDR_SLICE ? " IDR" : "",
1975 h->cur_pic_ptr->field_poc[0],
1976 h->cur_pic_ptr->field_poc[1],
1977 sl->ref_count[0], sl->ref_count[1],
1979 sl->deblocking_filter,
1980 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1982 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1983 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1989 int ff_h264_queue_decode_slice(H264Context *h, const H2645NAL *nal)
1991 H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
1992 int first_slice = sl == h->slice_ctx && !h->current_slice;
1997 ret = h264_slice_header_parse(h, sl, nal);
2001 // discard redundant pictures
2002 if (sl->redundant_pic_count > 0) {
2003 sl->ref_count[0] = sl->ref_count[1] = 0;
2007 if (sl->first_mb_addr == 0 || !h->current_slice) {
2008 if (h->setup_finished) {
2009 av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
2010 return AVERROR_INVALIDDATA;
2014 if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
2015 if (h->current_slice) {
2016 // this slice starts a new field
2017 // first decode any pending queued slices
2018 if (h->nb_slice_ctx_queued) {
2019 H264SliceContext tmp_ctx;
2021 ret = ff_h264_execute_decode_slices(h);
2022 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
2025 memcpy(&tmp_ctx, h->slice_ctx, sizeof(tmp_ctx));
2026 memcpy(h->slice_ctx, sl, sizeof(tmp_ctx));
2027 memcpy(sl, &tmp_ctx, sizeof(tmp_ctx));
2031 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
2032 ret = ff_h264_field_end(h, h->slice_ctx, 1);
2035 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == H264_NAL_IDR_SLICE) {
2036 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
2037 ret = ff_h264_field_end(h, h->slice_ctx, 1);
2038 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
2039 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
2040 h->cur_pic_ptr = NULL;
2044 return AVERROR_INVALIDDATA;
2047 if (!h->first_field) {
2048 if (h->cur_pic_ptr && !h->droppable) {
2049 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
2050 h->picture_structure == PICT_BOTTOM_FIELD);
2052 h->cur_pic_ptr = NULL;
2056 if (!h->current_slice)
2057 av_assert0(sl == h->slice_ctx);
2059 if (h->current_slice == 0 && !h->first_field) {
2061 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
2062 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
2063 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
2064 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != H264_NAL_IDR_SLICE && h->sei.recovery_point.recovery_frame_cnt < 0) ||
2065 h->avctx->skip_frame >= AVDISCARD_ALL) {
2071 const PPS *pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
2073 if (h->ps.pps->sps_id != pps->sps_id ||
2074 h->ps.pps->transform_8x8_mode != pps->transform_8x8_mode /*||
2075 (h->setup_finished && h->ps.pps != pps)*/) {
2076 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
2077 return AVERROR_INVALIDDATA;
2079 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
2080 av_log(h->avctx, AV_LOG_ERROR,
2081 "SPS changed in the middle of the frame\n");
2082 return AVERROR_INVALIDDATA;
2086 if (h->current_slice == 0) {
2087 ret = h264_field_start(h, sl, nal, first_slice);
2091 if (h->picture_structure != sl->picture_structure ||
2092 h->droppable != (nal->ref_idc == 0)) {
2093 av_log(h->avctx, AV_LOG_ERROR,
2094 "Changing field mode (%d -> %d) between slices is not allowed\n",
2095 h->picture_structure, sl->picture_structure);
2096 return AVERROR_INVALIDDATA;
2097 } else if (!h->cur_pic_ptr) {
2098 av_log(h->avctx, AV_LOG_ERROR,
2099 "unset cur_pic_ptr on slice %d\n",
2100 h->current_slice + 1);
2101 return AVERROR_INVALIDDATA;
2105 ret = h264_slice_init(h, sl, nal);
2109 h->nb_slice_ctx_queued++;
2114 int ff_h264_get_slice_type(const H264SliceContext *sl)
2116 switch (sl->slice_type) {
2117 case AV_PICTURE_TYPE_P:
2119 case AV_PICTURE_TYPE_B:
2121 case AV_PICTURE_TYPE_I:
2123 case AV_PICTURE_TYPE_SP:
2125 case AV_PICTURE_TYPE_SI:
2128 return AVERROR_INVALIDDATA;
2132 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
2133 H264SliceContext *sl,
2134 int mb_type, int top_xy,
2135 int left_xy[LEFT_MBS],
2137 int left_type[LEFT_MBS],
2138 int mb_xy, int list)
2140 int b_stride = h->b_stride;
2141 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
2142 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
2143 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2144 if (USES_LIST(top_type, list)) {
2145 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2146 const int b8_xy = 4 * top_xy + 2;
2147 const int *ref2frm = &h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2148 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2149 ref_cache[0 - 1 * 8] =
2150 ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
2151 ref_cache[2 - 1 * 8] =
2152 ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
2154 AV_ZERO128(mv_dst - 1 * 8);
2155 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2158 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2159 if (USES_LIST(left_type[LTOP], list)) {
2160 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2161 const int b8_xy = 4 * left_xy[LTOP] + 1;
2162 const int *ref2frm = &h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2163 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2164 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2165 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2166 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2168 ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2169 ref_cache[-1 + 16] =
2170 ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2172 AV_ZERO32(mv_dst - 1 + 0);
2173 AV_ZERO32(mv_dst - 1 + 8);
2174 AV_ZERO32(mv_dst - 1 + 16);
2175 AV_ZERO32(mv_dst - 1 + 24);
2178 ref_cache[-1 + 16] =
2179 ref_cache[-1 + 24] = LIST_NOT_USED;
2184 if (!USES_LIST(mb_type, list)) {
2185 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2186 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2187 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2188 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2189 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2194 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2195 const int *ref2frm = &h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2196 uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
2197 uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
2198 AV_WN32A(&ref_cache[0 * 8], ref01);
2199 AV_WN32A(&ref_cache[1 * 8], ref01);
2200 AV_WN32A(&ref_cache[2 * 8], ref23);
2201 AV_WN32A(&ref_cache[3 * 8], ref23);
2205 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2206 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2207 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2208 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2209 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2214 * @return non zero if the loop filter can be skipped
2216 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2218 const int mb_xy = sl->mb_xy;
2219 int top_xy, left_xy[LEFT_MBS];
2220 int top_type, left_type[LEFT_MBS];
2224 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2226 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2227 if (FRAME_MBAFF(h)) {
2228 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2229 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2231 if (left_mb_field_flag != curr_mb_field_flag)
2232 left_xy[LTOP] -= h->mb_stride;
2234 if (curr_mb_field_flag)
2235 top_xy += h->mb_stride &
2236 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2237 if (left_mb_field_flag != curr_mb_field_flag)
2238 left_xy[LBOT] += h->mb_stride;
2242 sl->top_mb_xy = top_xy;
2243 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2244 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2246 /* For sufficiently low qp, filtering wouldn't do anything.
2247 * This is a conservative estimate: could also check beta_offset
2248 * and more accurate chroma_qp. */
2249 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2250 int qp = h->cur_pic.qscale_table[mb_xy];
2251 if (qp <= qp_thresh &&
2252 (left_xy[LTOP] < 0 ||
2253 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2255 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2256 if (!FRAME_MBAFF(h))
2258 if ((left_xy[LTOP] < 0 ||
2259 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2260 (top_xy < h->mb_stride ||
2261 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2266 top_type = h->cur_pic.mb_type[top_xy];
2267 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2268 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2269 if (sl->deblocking_filter == 2) {
2270 if (h->slice_table[top_xy] != sl->slice_num)
2272 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2273 left_type[LTOP] = left_type[LBOT] = 0;
2275 if (h->slice_table[top_xy] == 0xFFFF)
2277 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2278 left_type[LTOP] = left_type[LBOT] = 0;
2280 sl->top_type = top_type;
2281 sl->left_type[LTOP] = left_type[LTOP];
2282 sl->left_type[LBOT] = left_type[LBOT];
2284 if (IS_INTRA(mb_type))
2287 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2288 top_type, left_type, mb_xy, 0);
2289 if (sl->list_count == 2)
2290 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2291 top_type, left_type, mb_xy, 1);
2293 nnz = h->non_zero_count[mb_xy];
2294 nnz_cache = sl->non_zero_count_cache;
2295 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2296 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2297 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2298 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2299 sl->cbp = h->cbp_table[mb_xy];
2302 nnz = h->non_zero_count[top_xy];
2303 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2306 if (left_type[LTOP]) {
2307 nnz = h->non_zero_count[left_xy[LTOP]];
2308 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2309 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2310 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2311 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2314 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2315 * from what the loop filter needs */
2316 if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
2317 if (IS_8x8DCT(top_type)) {
2318 nnz_cache[4 + 8 * 0] =
2319 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2320 nnz_cache[6 + 8 * 0] =
2321 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2323 if (IS_8x8DCT(left_type[LTOP])) {
2324 nnz_cache[3 + 8 * 1] =
2325 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2327 if (IS_8x8DCT(left_type[LBOT])) {
2328 nnz_cache[3 + 8 * 3] =
2329 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2332 if (IS_8x8DCT(mb_type)) {
2333 nnz_cache[scan8[0]] =
2334 nnz_cache[scan8[1]] =
2335 nnz_cache[scan8[2]] =
2336 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2338 nnz_cache[scan8[0 + 4]] =
2339 nnz_cache[scan8[1 + 4]] =
2340 nnz_cache[scan8[2 + 4]] =
2341 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2343 nnz_cache[scan8[0 + 8]] =
2344 nnz_cache[scan8[1 + 8]] =
2345 nnz_cache[scan8[2 + 8]] =
2346 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2348 nnz_cache[scan8[0 + 12]] =
2349 nnz_cache[scan8[1 + 12]] =
2350 nnz_cache[scan8[2 + 12]] =
2351 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2358 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2360 uint8_t *dest_y, *dest_cb, *dest_cr;
2361 int linesize, uvlinesize, mb_x, mb_y;
2362 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2363 const int old_slice_type = sl->slice_type;
2364 const int pixel_shift = h->pixel_shift;
2365 const int block_h = 16 >> h->chroma_y_shift;
2367 if (h->postpone_filter)
2370 if (sl->deblocking_filter) {
2371 for (mb_x = start_x; mb_x < end_x; mb_x++)
2372 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2374 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2375 mb_type = h->cur_pic.mb_type[mb_xy];
2379 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2383 dest_y = h->cur_pic.f->data[0] +
2384 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2385 dest_cb = h->cur_pic.f->data[1] +
2386 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2387 mb_y * sl->uvlinesize * block_h;
2388 dest_cr = h->cur_pic.f->data[2] +
2389 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2390 mb_y * sl->uvlinesize * block_h;
2391 // FIXME simplify above
2394 linesize = sl->mb_linesize = sl->linesize * 2;
2395 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2396 if (mb_y & 1) { // FIXME move out of this function?
2397 dest_y -= sl->linesize * 15;
2398 dest_cb -= sl->uvlinesize * (block_h - 1);
2399 dest_cr -= sl->uvlinesize * (block_h - 1);
2402 linesize = sl->mb_linesize = sl->linesize;
2403 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2405 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2407 if (fill_filter_caches(h, sl, mb_type))
2409 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mb_xy]);
2410 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mb_xy]);
2412 if (FRAME_MBAFF(h)) {
2413 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2414 linesize, uvlinesize);
2416 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2417 dest_cr, linesize, uvlinesize);
2421 sl->slice_type = old_slice_type;
2423 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2424 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
2425 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
2428 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2430 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2431 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2432 h->cur_pic.mb_type[mb_xy - 1] :
2433 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2434 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2435 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2439 * Draw edges and report progress for the last MB row.
2441 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2443 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2444 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2445 int height = 16 << FRAME_MBAFF(h);
2446 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2448 if (sl->deblocking_filter) {
2449 if ((top + height) >= pic_height)
2450 height += deblock_border;
2451 top -= deblock_border;
2454 if (top >= pic_height || (top + height) < 0)
2457 height = FFMIN(height, pic_height - top);
2459 height = top + height;
2463 ff_h264_draw_horiz_band(h, sl, top, height);
2465 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2468 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2469 h->picture_structure == PICT_BOTTOM_FIELD);
2472 static void er_add_slice(H264SliceContext *sl,
2473 int startx, int starty,
2474 int endx, int endy, int status)
2476 if (!sl->h264->enable_er)
2479 if (CONFIG_ERROR_RESILIENCE) {
2480 ERContext *er = &sl->h264->slice_ctx[0].er;
2482 ff_er_add_slice(er, startx, starty, endx, endy, status);
2486 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2488 H264SliceContext *sl = arg;
2489 const H264Context *h = sl->h264;
2490 int lf_x_start = sl->mb_x;
2491 int orig_deblock = sl->deblocking_filter;
2494 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2495 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2497 ret = alloc_scratch_buffers(sl, sl->linesize);
2501 sl->mb_skip_run = -1;
2503 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2505 if (h->postpone_filter)
2506 sl->deblocking_filter = 0;
2508 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2509 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2511 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2512 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2514 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2515 prev_status &= ~ VP_START;
2516 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2517 h->slice_ctx[0].er.error_occurred = 1;
2521 if (h->ps.pps->cabac) {
2523 align_get_bits(&sl->gb);
2526 ret = ff_init_cabac_decoder(&sl->cabac,
2527 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2528 (get_bits_left(&sl->gb) + 7) / 8);
2532 ff_h264_init_cabac_states(h, sl);
2537 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2538 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2539 sl->next_slice_idx);
2540 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2541 sl->mb_y, ER_MB_ERROR);
2542 return AVERROR_INVALIDDATA;
2545 ret = ff_h264_decode_mb_cabac(h, sl);
2546 // STOP_TIMER("decode_mb_cabac")
2549 ff_h264_hl_decode_mb(h, sl);
2551 // FIXME optimal? or let mb_decode decode 16x32 ?
2552 if (ret >= 0 && FRAME_MBAFF(h)) {
2555 ret = ff_h264_decode_mb_cabac(h, sl);
2558 ff_h264_hl_decode_mb(h, sl);
2561 eos = get_cabac_terminate(&sl->cabac);
2563 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2564 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2565 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2566 sl->mb_y, ER_MB_END);
2567 if (sl->mb_x >= lf_x_start)
2568 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2571 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2572 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2573 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2574 av_log(h->avctx, AV_LOG_ERROR,
2575 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2577 sl->cabac.bytestream_end - sl->cabac.bytestream);
2578 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2579 sl->mb_y, ER_MB_ERROR);
2580 return AVERROR_INVALIDDATA;
2583 if (++sl->mb_x >= h->mb_width) {
2584 loop_filter(h, sl, lf_x_start, sl->mb_x);
2585 sl->mb_x = lf_x_start = 0;
2586 decode_finish_row(h, sl);
2588 if (FIELD_OR_MBAFF_PICTURE(h)) {
2590 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2591 predict_field_decoding_flag(h, sl);
2595 if (eos || sl->mb_y >= h->mb_height) {
2596 ff_tlog(h->avctx, "slice end %d %d\n",
2597 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2598 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2599 sl->mb_y, ER_MB_END);
2600 if (sl->mb_x > lf_x_start)
2601 loop_filter(h, sl, lf_x_start, sl->mb_x);
2609 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2610 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2611 sl->next_slice_idx);
2612 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2613 sl->mb_y, ER_MB_ERROR);
2614 return AVERROR_INVALIDDATA;
2617 ret = ff_h264_decode_mb_cavlc(h, sl);
2620 ff_h264_hl_decode_mb(h, sl);
2622 // FIXME optimal? or let mb_decode decode 16x32 ?
2623 if (ret >= 0 && FRAME_MBAFF(h)) {
2625 ret = ff_h264_decode_mb_cavlc(h, sl);
2628 ff_h264_hl_decode_mb(h, sl);
2633 av_log(h->avctx, AV_LOG_ERROR,
2634 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2635 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2636 sl->mb_y, ER_MB_ERROR);
2640 if (++sl->mb_x >= h->mb_width) {
2641 loop_filter(h, sl, lf_x_start, sl->mb_x);
2642 sl->mb_x = lf_x_start = 0;
2643 decode_finish_row(h, sl);
2645 if (FIELD_OR_MBAFF_PICTURE(h)) {
2647 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2648 predict_field_decoding_flag(h, sl);
2650 if (sl->mb_y >= h->mb_height) {
2651 ff_tlog(h->avctx, "slice end %d %d\n",
2652 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2654 if ( get_bits_left(&sl->gb) == 0
2655 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2656 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2657 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2661 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2662 sl->mb_x, sl->mb_y, ER_MB_END);
2664 return AVERROR_INVALIDDATA;
2669 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2670 ff_tlog(h->avctx, "slice end %d %d\n",
2671 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2673 if (get_bits_left(&sl->gb) == 0) {
2674 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2675 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2676 if (sl->mb_x > lf_x_start)
2677 loop_filter(h, sl, lf_x_start, sl->mb_x);
2681 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2682 sl->mb_y, ER_MB_ERROR);
2684 return AVERROR_INVALIDDATA;
2691 sl->deblocking_filter = orig_deblock;
2696 * Call decode_slice() for each context.
2698 * @param h h264 master context
2700 int ff_h264_execute_decode_slices(H264Context *h)
2702 AVCodecContext *const avctx = h->avctx;
2703 H264SliceContext *sl;
2704 int context_count = h->nb_slice_ctx_queued;
2708 h->slice_ctx[0].next_slice_idx = INT_MAX;
2710 if (h->avctx->hwaccel || context_count < 1
2711 #if FF_API_CAP_VDPAU
2712 || h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU
2717 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2719 if (context_count == 1) {
2721 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2722 h->postpone_filter = 0;
2724 ret = decode_slice(avctx, &h->slice_ctx[0]);
2725 h->mb_y = h->slice_ctx[0].mb_y;
2729 av_assert0(context_count > 0);
2730 for (i = 0; i < context_count; i++) {
2731 int next_slice_idx = h->mb_width * h->mb_height;
2734 sl = &h->slice_ctx[i];
2735 if (CONFIG_ERROR_RESILIENCE) {
2736 sl->er.error_count = 0;
2739 /* make sure none of those slices overlap */
2740 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2741 for (j = 0; j < context_count; j++) {
2742 H264SliceContext *sl2 = &h->slice_ctx[j];
2743 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2745 if (i == j || slice_idx2 < slice_idx)
2747 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2749 sl->next_slice_idx = next_slice_idx;
2752 avctx->execute(avctx, decode_slice, h->slice_ctx,
2753 NULL, context_count, sizeof(h->slice_ctx[0]));
2755 /* pull back stuff from slices to master context */
2756 sl = &h->slice_ctx[context_count - 1];
2758 if (CONFIG_ERROR_RESILIENCE) {
2759 for (i = 1; i < context_count; i++)
2760 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
2763 if (h->postpone_filter) {
2764 h->postpone_filter = 0;
2766 for (i = 0; i < context_count; i++) {
2769 sl = &h->slice_ctx[i];
2770 y_end = FFMIN(sl->mb_y + 1, h->mb_height);
2771 x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
2773 for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
2775 loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
2776 j == y_end - 1 ? x_end : h->mb_width);
2783 h->nb_slice_ctx_queued = 0;