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"
34 #include "cabac_functions.h"
35 #include "error_resilience.h"
40 #include "h264chroma.h"
41 #include "h264_mvpred.h"
45 #include "mpegutils.h"
46 #include "mpegvideo.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 if (h->avctx->hwaccel) {
200 const AVHWAccel *hwaccel = h->avctx->hwaccel;
201 av_assert0(!pic->hwaccel_picture_private);
202 if (hwaccel->frame_priv_data_size) {
203 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
204 if (!pic->hwaccel_priv_buf)
205 return AVERROR(ENOMEM);
206 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
209 if (CONFIG_GRAY && !h->avctx->hwaccel && h->flags & AV_CODEC_FLAG_GRAY && pic->f->data[2]) {
210 int h_chroma_shift, v_chroma_shift;
211 av_pix_fmt_get_chroma_sub_sample(pic->f->format,
212 &h_chroma_shift, &v_chroma_shift);
214 for(i=0; i<AV_CEIL_RSHIFT(pic->f->height, v_chroma_shift); i++) {
215 memset(pic->f->data[1] + pic->f->linesize[1]*i,
216 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
217 memset(pic->f->data[2] + pic->f->linesize[2]*i,
218 0x80, AV_CEIL_RSHIFT(pic->f->width, h_chroma_shift));
222 if (!h->qscale_table_pool) {
223 ret = init_table_pools(h);
228 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
229 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
230 if (!pic->qscale_table_buf || !pic->mb_type_buf)
233 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
234 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
236 for (i = 0; i < 2; i++) {
237 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
238 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
239 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
242 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
243 pic->ref_index[i] = pic->ref_index_buf[i]->data;
248 ff_h264_unref_picture(h, pic);
249 return (ret < 0) ? ret : AVERROR(ENOMEM);
252 static int find_unused_picture(H264Context *h)
256 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
257 if (!h->DPB[i].f->buf[0])
260 return AVERROR_INVALIDDATA;
264 #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
266 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
267 (((pic) && (pic) >= (old_ctx)->DPB && \
268 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
269 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
271 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
272 H264Context *new_base,
273 H264Context *old_base)
277 for (i = 0; i < count; i++) {
278 av_assert1(!from[i] ||
279 IN_RANGE(from[i], old_base, 1) ||
280 IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
281 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
285 static int h264_slice_header_init(H264Context *h);
287 int ff_h264_update_thread_context(AVCodecContext *dst,
288 const AVCodecContext *src)
290 H264Context *h = dst->priv_data, *h1 = src->priv_data;
291 int inited = h->context_initialized, err = 0;
298 // We can't fail if SPS isn't set at it breaks current skip_frame code
300 // return AVERROR_INVALIDDATA;
303 (h->width != h1->width ||
304 h->height != h1->height ||
305 h->mb_width != h1->mb_width ||
306 h->mb_height != h1->mb_height ||
308 h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
309 h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
310 h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
314 /* copy block_offset since frame_start may not be called */
315 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
318 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
319 av_buffer_unref(&h->ps.sps_list[i]);
320 if (h1->ps.sps_list[i]) {
321 h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
322 if (!h->ps.sps_list[i])
323 return AVERROR(ENOMEM);
326 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
327 av_buffer_unref(&h->ps.pps_list[i]);
328 if (h1->ps.pps_list[i]) {
329 h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
330 if (!h->ps.pps_list[i])
331 return AVERROR(ENOMEM);
335 av_buffer_unref(&h->ps.pps_ref);
338 if (h1->ps.pps_ref) {
339 h->ps.pps_ref = av_buffer_ref(h1->ps.pps_ref);
341 return AVERROR(ENOMEM);
342 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
343 h->ps.sps = h->ps.pps->sps;
346 if (need_reinit || !inited) {
347 h->width = h1->width;
348 h->height = h1->height;
349 h->mb_height = h1->mb_height;
350 h->mb_width = h1->mb_width;
351 h->mb_num = h1->mb_num;
352 h->mb_stride = h1->mb_stride;
353 h->b_stride = h1->b_stride;
354 h->x264_build = h1->x264_build;
356 if (h->context_initialized || h1->context_initialized) {
357 if ((err = h264_slice_header_init(h)) < 0) {
358 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
363 /* copy block_offset since frame_start may not be called */
364 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
367 h->avctx->coded_height = h1->avctx->coded_height;
368 h->avctx->coded_width = h1->avctx->coded_width;
369 h->avctx->width = h1->avctx->width;
370 h->avctx->height = h1->avctx->height;
371 h->width_from_caller = h1->width_from_caller;
372 h->height_from_caller = h1->height_from_caller;
373 h->coded_picture_number = h1->coded_picture_number;
374 h->first_field = h1->first_field;
375 h->picture_structure = h1->picture_structure;
376 h->mb_aff_frame = h1->mb_aff_frame;
377 h->droppable = h1->droppable;
379 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
380 ff_h264_unref_picture(h, &h->DPB[i]);
381 if (h1->DPB[i].f->buf[0] &&
382 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
386 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
387 ff_h264_unref_picture(h, &h->cur_pic);
388 if (h1->cur_pic.f->buf[0]) {
389 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
394 h->enable_er = h1->enable_er;
395 h->workaround_bugs = h1->workaround_bugs;
396 h->droppable = h1->droppable;
398 // extradata/NAL handling
399 h->is_avc = h1->is_avc;
400 h->nal_length_size = h1->nal_length_size;
402 memcpy(&h->poc, &h1->poc, sizeof(h->poc));
404 memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
405 memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
406 memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
407 memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
409 h->next_output_pic = h1->next_output_pic;
410 h->next_outputed_poc = h1->next_outputed_poc;
412 memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
413 h->nb_mmco = h1->nb_mmco;
414 h->mmco_reset = h1->mmco_reset;
415 h->explicit_ref_marking = h1->explicit_ref_marking;
416 h->long_ref_count = h1->long_ref_count;
417 h->short_ref_count = h1->short_ref_count;
419 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
420 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
421 copy_picture_range(h->delayed_pic, h1->delayed_pic,
422 MAX_DELAYED_PIC_COUNT + 2, h, h1);
424 h->frame_recovered = h1->frame_recovered;
426 av_buffer_unref(&h->sei.a53_caption.buf_ref);
427 if (h1->sei.a53_caption.buf_ref) {
428 h->sei.a53_caption.buf_ref = av_buffer_ref(h1->sei.a53_caption.buf_ref);
429 if (!h->sei.a53_caption.buf_ref)
430 return AVERROR(ENOMEM);
437 err = ff_h264_execute_ref_pic_marking(h);
438 h->poc.prev_poc_msb = h->poc.poc_msb;
439 h->poc.prev_poc_lsb = h->poc.poc_lsb;
441 h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
442 h->poc.prev_frame_num = h->poc.frame_num;
444 h->recovery_frame = h1->recovery_frame;
449 static int h264_frame_start(H264Context *h)
453 const int pixel_shift = h->pixel_shift;
455 if (!ff_thread_can_start_frame(h->avctx)) {
456 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
460 release_unused_pictures(h, 1);
461 h->cur_pic_ptr = NULL;
463 i = find_unused_picture(h);
465 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
470 pic->reference = h->droppable ? 0 : h->picture_structure;
471 pic->f->coded_picture_number = h->coded_picture_number++;
472 pic->field_picture = h->picture_structure != PICT_FRAME;
473 pic->frame_num = h->poc.frame_num;
475 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
477 * See decode_nal_units().
479 pic->f->key_frame = 0;
482 pic->invalid_gap = 0;
483 pic->sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
485 pic->f->pict_type = h->slice_ctx[0].slice_type;
487 pic->f->crop_left = h->crop_left;
488 pic->f->crop_right = h->crop_right;
489 pic->f->crop_top = h->crop_top;
490 pic->f->crop_bottom = h->crop_bottom;
492 if ((ret = alloc_picture(h, pic)) < 0)
495 h->cur_pic_ptr = pic;
496 ff_h264_unref_picture(h, &h->cur_pic);
497 if (CONFIG_ERROR_RESILIENCE) {
498 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
501 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
504 for (i = 0; i < h->nb_slice_ctx; i++) {
505 h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
506 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
509 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
510 ff_er_frame_start(&h->slice_ctx[0].er);
511 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
512 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
515 for (i = 0; i < 16; i++) {
516 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
517 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
519 for (i = 0; i < 16; i++) {
520 h->block_offset[16 + i] =
521 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
522 h->block_offset[48 + 16 + i] =
523 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
526 /* We mark the current picture as non-reference after allocating it, so
527 * that if we break out due to an error it can be released automatically
528 * in the next ff_mpv_frame_start().
530 h->cur_pic_ptr->reference = 0;
532 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
534 h->next_output_pic = NULL;
536 h->postpone_filter = 0;
538 h->mb_aff_frame = h->ps.sps->mb_aff && (h->picture_structure == PICT_FRAME);
540 if (h->sei.unregistered.x264_build >= 0)
541 h->x264_build = h->sei.unregistered.x264_build;
543 assert(h->cur_pic_ptr->long_ref == 0);
548 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
550 uint8_t *src_cb, uint8_t *src_cr,
551 int linesize, int uvlinesize,
556 const int pixel_shift = h->pixel_shift;
557 int chroma444 = CHROMA444(h);
558 int chroma422 = CHROMA422(h);
561 src_cb -= uvlinesize;
562 src_cr -= uvlinesize;
564 if (!simple && FRAME_MBAFF(h)) {
567 top_border = sl->top_borders[0][sl->mb_x];
568 AV_COPY128(top_border, src_y + 15 * linesize);
570 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
571 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
574 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
575 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
576 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
577 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
579 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
580 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
582 } else if (chroma422) {
584 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
585 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
587 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
588 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
592 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
593 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
595 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
596 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
601 } else if (MB_MBAFF(sl)) {
607 top_border = sl->top_borders[top_idx][sl->mb_x];
608 /* There are two lines saved, the line above the top macroblock
609 * of a pair, and the line above the bottom macroblock. */
610 AV_COPY128(top_border, src_y + 16 * linesize);
612 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
614 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
617 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
618 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
619 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
620 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
622 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
623 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
625 } else if (chroma422) {
627 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
628 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
630 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
631 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
635 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
636 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
638 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
639 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
646 * Initialize implicit_weight table.
647 * @param field 0/1 initialize the weight for interlaced MBAFF
648 * -1 initializes the rest
650 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
652 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
654 for (i = 0; i < 2; i++) {
655 sl->pwt.luma_weight_flag[i] = 0;
656 sl->pwt.chroma_weight_flag[i] = 0;
660 if (h->picture_structure == PICT_FRAME) {
661 cur_poc = h->cur_pic_ptr->poc;
663 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
665 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
666 sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2LL * cur_poc) {
667 sl->pwt.use_weight = 0;
668 sl->pwt.use_weight_chroma = 0;
672 ref_count0 = sl->ref_count[0];
673 ref_count1 = sl->ref_count[1];
675 cur_poc = h->cur_pic_ptr->field_poc[field];
677 ref_count0 = 16 + 2 * sl->ref_count[0];
678 ref_count1 = 16 + 2 * sl->ref_count[1];
681 sl->pwt.use_weight = 2;
682 sl->pwt.use_weight_chroma = 2;
683 sl->pwt.luma_log2_weight_denom = 5;
684 sl->pwt.chroma_log2_weight_denom = 5;
686 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
687 int64_t poc0 = sl->ref_list[0][ref0].poc;
688 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
690 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
691 int poc1 = sl->ref_list[1][ref1].poc;
692 int td = av_clip_int8(poc1 - poc0);
694 int tb = av_clip_int8(cur_poc - poc0);
695 int tx = (16384 + (FFABS(td) >> 1)) / td;
696 int dist_scale_factor = (tb * tx + 32) >> 8;
697 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
698 w = 64 - dist_scale_factor;
702 sl->pwt.implicit_weight[ref0][ref1][0] =
703 sl->pwt.implicit_weight[ref0][ref1][1] = w;
705 sl->pwt.implicit_weight[ref0][ref1][field] = w;
712 * initialize scan tables
714 static void init_scan_tables(H264Context *h)
717 for (i = 0; i < 16; i++) {
718 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
719 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
720 h->field_scan[i] = TRANSPOSE(field_scan[i]);
723 for (i = 0; i < 64; i++) {
724 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
725 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
726 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
727 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
728 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
731 if (h->ps.sps->transform_bypass) { // FIXME same ugly
732 memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
733 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
734 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
735 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
736 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
737 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
739 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
740 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
741 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
742 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
743 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
744 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
748 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
750 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
751 (CONFIG_H264_D3D11VA_HWACCEL * 2) + \
752 CONFIG_H264_NVDEC_HWACCEL + \
753 CONFIG_H264_VAAPI_HWACCEL + \
754 CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
755 CONFIG_H264_VDPAU_HWACCEL)
756 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
757 const enum AVPixelFormat *choices = pix_fmts;
760 switch (h->ps.sps->bit_depth_luma) {
763 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
764 *fmt++ = AV_PIX_FMT_GBRP9;
766 *fmt++ = AV_PIX_FMT_YUV444P9;
767 } else if (CHROMA422(h))
768 *fmt++ = AV_PIX_FMT_YUV422P9;
770 *fmt++ = AV_PIX_FMT_YUV420P9;
774 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
775 *fmt++ = AV_PIX_FMT_GBRP10;
777 *fmt++ = AV_PIX_FMT_YUV444P10;
778 } else if (CHROMA422(h))
779 *fmt++ = AV_PIX_FMT_YUV422P10;
781 *fmt++ = AV_PIX_FMT_YUV420P10;
785 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
786 *fmt++ = AV_PIX_FMT_GBRP12;
788 *fmt++ = AV_PIX_FMT_YUV444P12;
789 } else if (CHROMA422(h))
790 *fmt++ = AV_PIX_FMT_YUV422P12;
792 *fmt++ = AV_PIX_FMT_YUV420P12;
796 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
797 *fmt++ = AV_PIX_FMT_GBRP14;
799 *fmt++ = AV_PIX_FMT_YUV444P14;
800 } else if (CHROMA422(h))
801 *fmt++ = AV_PIX_FMT_YUV422P14;
803 *fmt++ = AV_PIX_FMT_YUV420P14;
806 #if CONFIG_H264_VDPAU_HWACCEL
807 *fmt++ = AV_PIX_FMT_VDPAU;
809 #if CONFIG_H264_NVDEC_HWACCEL
810 *fmt++ = AV_PIX_FMT_CUDA;
813 if (h->avctx->colorspace == AVCOL_SPC_RGB)
814 *fmt++ = AV_PIX_FMT_GBRP;
815 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
816 *fmt++ = AV_PIX_FMT_YUVJ444P;
818 *fmt++ = AV_PIX_FMT_YUV444P;
819 } else if (CHROMA422(h)) {
820 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
821 *fmt++ = AV_PIX_FMT_YUVJ422P;
823 *fmt++ = AV_PIX_FMT_YUV422P;
825 #if CONFIG_H264_DXVA2_HWACCEL
826 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
828 #if CONFIG_H264_D3D11VA_HWACCEL
829 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
830 *fmt++ = AV_PIX_FMT_D3D11;
832 #if CONFIG_H264_VAAPI_HWACCEL
833 *fmt++ = AV_PIX_FMT_VAAPI;
835 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
836 *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
838 if (h->avctx->codec->pix_fmts)
839 choices = h->avctx->codec->pix_fmts;
840 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
841 *fmt++ = AV_PIX_FMT_YUVJ420P;
843 *fmt++ = AV_PIX_FMT_YUV420P;
847 av_log(h->avctx, AV_LOG_ERROR,
848 "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
849 return AVERROR_INVALIDDATA;
852 *fmt = AV_PIX_FMT_NONE;
854 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
855 if (choices[i] == h->avctx->pix_fmt && !force_callback)
857 return ff_thread_get_format(h->avctx, choices);
860 /* export coded and cropped frame dimensions to AVCodecContext */
861 static void init_dimensions(H264Context *h)
863 const SPS *sps = (const SPS*)h->ps.sps;
864 int cr = sps->crop_right;
865 int cl = sps->crop_left;
866 int ct = sps->crop_top;
867 int cb = sps->crop_bottom;
868 int width = h->width - (cr + cl);
869 int height = h->height - (ct + cb);
870 av_assert0(sps->crop_right + sps->crop_left < (unsigned)h->width);
871 av_assert0(sps->crop_top + sps->crop_bottom < (unsigned)h->height);
873 /* handle container cropping */
874 if (h->width_from_caller > 0 && h->height_from_caller > 0 &&
875 !sps->crop_top && !sps->crop_left &&
876 FFALIGN(h->width_from_caller, 16) == FFALIGN(width, 16) &&
877 FFALIGN(h->height_from_caller, 16) == FFALIGN(height, 16) &&
878 h->width_from_caller <= width &&
879 h->height_from_caller <= height) {
880 width = h->width_from_caller;
881 height = h->height_from_caller;
884 cr = h->width - width;
885 cb = h->height - height;
887 h->width_from_caller = 0;
888 h->height_from_caller = 0;
891 h->avctx->coded_width = h->width;
892 h->avctx->coded_height = h->height;
893 h->avctx->width = width;
894 h->avctx->height = height;
901 static int h264_slice_header_init(H264Context *h)
903 const SPS *sps = h->ps.sps;
906 ff_set_sar(h->avctx, sps->sar);
907 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
908 &h->chroma_x_shift, &h->chroma_y_shift);
910 if (sps->timing_info_present_flag) {
911 int64_t den = sps->time_scale;
912 if (h->x264_build < 44U)
914 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
915 sps->num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
918 ff_h264_free_tables(h);
921 h->prev_interlaced_frame = 1;
924 ret = ff_h264_alloc_tables(h);
926 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
930 if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 14 ||
931 sps->bit_depth_luma == 11 || sps->bit_depth_luma == 13
933 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
934 sps->bit_depth_luma);
935 ret = AVERROR_INVALIDDATA;
939 h->cur_bit_depth_luma =
940 h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
941 h->cur_chroma_format_idc = sps->chroma_format_idc;
942 h->pixel_shift = sps->bit_depth_luma > 8;
943 h->chroma_format_idc = sps->chroma_format_idc;
944 h->bit_depth_luma = sps->bit_depth_luma;
946 ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
947 sps->chroma_format_idc);
948 ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
949 ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
950 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
951 sps->chroma_format_idc);
952 ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
954 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
955 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
957 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
961 for (i = 0; i < h->nb_slice_ctx; i++) {
962 H264SliceContext *sl = &h->slice_ctx[i];
965 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
966 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
967 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
969 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
970 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
976 h->context_initialized = 1;
980 ff_h264_free_tables(h);
981 h->context_initialized = 0;
985 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
988 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
989 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
990 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
996 static int h264_init_ps(H264Context *h, const H264SliceContext *sl, int first_slice)
999 int needs_reinit = 0, must_reinit, ret;
1002 av_buffer_unref(&h->ps.pps_ref);
1004 h->ps.pps_ref = av_buffer_ref(h->ps.pps_list[sl->pps_id]);
1006 return AVERROR(ENOMEM);
1007 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
1010 if (h->ps.sps != h->ps.pps->sps) {
1011 h->ps.sps = (const SPS*)h->ps.pps->sps;
1013 if (h->mb_width != h->ps.sps->mb_width ||
1014 h->mb_height != h->ps.sps->mb_height ||
1015 h->cur_bit_depth_luma != h->ps.sps->bit_depth_luma ||
1016 h->cur_chroma_format_idc != h->ps.sps->chroma_format_idc
1020 if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
1021 h->chroma_format_idc != h->ps.sps->chroma_format_idc)
1026 must_reinit = (h->context_initialized &&
1027 ( 16*sps->mb_width != h->avctx->coded_width
1028 || 16*sps->mb_height != h->avctx->coded_height
1029 || h->cur_bit_depth_luma != sps->bit_depth_luma
1030 || h->cur_chroma_format_idc != sps->chroma_format_idc
1031 || h->mb_width != sps->mb_width
1032 || h->mb_height != sps->mb_height
1034 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1035 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1038 if (first_slice && av_cmp_q(sps->sar, h->avctx->sample_aspect_ratio))
1041 if (!h->setup_finished) {
1042 h->avctx->profile = ff_h264_get_profile(sps);
1043 h->avctx->level = sps->level_idc;
1044 h->avctx->refs = sps->ref_frame_count;
1046 h->mb_width = sps->mb_width;
1047 h->mb_height = sps->mb_height;
1048 h->mb_num = h->mb_width * h->mb_height;
1049 h->mb_stride = h->mb_width + 1;
1051 h->b_stride = h->mb_width * 4;
1053 h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
1055 h->width = 16 * h->mb_width;
1056 h->height = 16 * h->mb_height;
1060 if (sps->video_signal_type_present_flag) {
1061 h->avctx->color_range = sps->full_range > 0 ? AVCOL_RANGE_JPEG
1063 if (sps->colour_description_present_flag) {
1064 if (h->avctx->colorspace != sps->colorspace)
1066 h->avctx->color_primaries = sps->color_primaries;
1067 h->avctx->color_trc = sps->color_trc;
1068 h->avctx->colorspace = sps->colorspace;
1072 if (h->sei.alternative_transfer.present &&
1073 av_color_transfer_name(h->sei.alternative_transfer.preferred_transfer_characteristics) &&
1074 h->sei.alternative_transfer.preferred_transfer_characteristics != AVCOL_TRC_UNSPECIFIED) {
1075 h->avctx->color_trc = h->sei.alternative_transfer.preferred_transfer_characteristics;
1078 h->avctx->chroma_sample_location = sps->chroma_location;
1080 if (!h->context_initialized || must_reinit || needs_reinit) {
1081 int flush_changes = h->context_initialized;
1082 h->context_initialized = 0;
1083 if (sl != h->slice_ctx) {
1084 av_log(h->avctx, AV_LOG_ERROR,
1085 "changing width %d -> %d / height %d -> %d on "
1087 h->width, h->avctx->coded_width,
1088 h->height, h->avctx->coded_height,
1089 h->current_slice + 1);
1090 return AVERROR_INVALIDDATA;
1093 av_assert1(first_slice);
1096 ff_h264_flush_change(h);
1098 if ((ret = get_pixel_format(h, 1)) < 0)
1100 h->avctx->pix_fmt = ret;
1102 av_log(h->avctx, AV_LOG_VERBOSE, "Reinit context to %dx%d, "
1103 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1105 if ((ret = h264_slice_header_init(h)) < 0) {
1106 av_log(h->avctx, AV_LOG_ERROR,
1107 "h264_slice_header_init() failed\n");
1115 static int h264_export_frame_props(H264Context *h)
1117 const SPS *sps = h->ps.sps;
1118 H264Picture *cur = h->cur_pic_ptr;
1120 cur->f->interlaced_frame = 0;
1121 cur->f->repeat_pict = 0;
1123 /* Signal interlacing information externally. */
1124 /* Prioritize picture timing SEI information over used
1125 * decoding process if it exists. */
1126 if (h->sei.picture_timing.present) {
1127 int ret = ff_h264_sei_process_picture_timing(&h->sei.picture_timing, sps,
1130 av_log(h->avctx, AV_LOG_ERROR, "Error processing a picture timing SEI\n");
1131 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1133 h->sei.picture_timing.present = 0;
1137 if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
1138 H264SEIPictureTiming *pt = &h->sei.picture_timing;
1139 switch (pt->pic_struct) {
1140 case H264_SEI_PIC_STRUCT_FRAME:
1142 case H264_SEI_PIC_STRUCT_TOP_FIELD:
1143 case H264_SEI_PIC_STRUCT_BOTTOM_FIELD:
1144 cur->f->interlaced_frame = 1;
1146 case H264_SEI_PIC_STRUCT_TOP_BOTTOM:
1147 case H264_SEI_PIC_STRUCT_BOTTOM_TOP:
1148 if (FIELD_OR_MBAFF_PICTURE(h))
1149 cur->f->interlaced_frame = 1;
1151 // try to flag soft telecine progressive
1152 cur->f->interlaced_frame = h->prev_interlaced_frame;
1154 case H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1155 case H264_SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1156 /* Signal the possibility of telecined film externally
1157 * (pic_struct 5,6). From these hints, let the applications
1158 * decide if they apply deinterlacing. */
1159 cur->f->repeat_pict = 1;
1161 case H264_SEI_PIC_STRUCT_FRAME_DOUBLING:
1162 cur->f->repeat_pict = 2;
1164 case H264_SEI_PIC_STRUCT_FRAME_TRIPLING:
1165 cur->f->repeat_pict = 4;
1169 if ((pt->ct_type & 3) &&
1170 pt->pic_struct <= H264_SEI_PIC_STRUCT_BOTTOM_TOP)
1171 cur->f->interlaced_frame = (pt->ct_type & (1 << 1)) != 0;
1173 /* Derive interlacing flag from used decoding process. */
1174 cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
1176 h->prev_interlaced_frame = cur->f->interlaced_frame;
1178 if (cur->field_poc[0] != cur->field_poc[1]) {
1179 /* Derive top_field_first from field pocs. */
1180 cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
1182 if (sps->pic_struct_present_flag && h->sei.picture_timing.present) {
1183 /* Use picture timing SEI information. Even if it is a
1184 * information of a past frame, better than nothing. */
1185 if (h->sei.picture_timing.pic_struct == H264_SEI_PIC_STRUCT_TOP_BOTTOM ||
1186 h->sei.picture_timing.pic_struct == H264_SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1187 cur->f->top_field_first = 1;
1189 cur->f->top_field_first = 0;
1190 } else if (cur->f->interlaced_frame) {
1191 /* Default to top field first when pic_struct_present_flag
1192 * is not set but interlaced frame detected */
1193 cur->f->top_field_first = 1;
1195 /* Most likely progressive */
1196 cur->f->top_field_first = 0;
1200 if (h->sei.frame_packing.present &&
1201 h->sei.frame_packing.arrangement_type <= 6 &&
1202 h->sei.frame_packing.content_interpretation_type > 0 &&
1203 h->sei.frame_packing.content_interpretation_type < 3) {
1204 H264SEIFramePacking *fp = &h->sei.frame_packing;
1205 AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
1207 switch (fp->arrangement_type) {
1208 case H264_SEI_FPA_TYPE_CHECKERBOARD:
1209 stereo->type = AV_STEREO3D_CHECKERBOARD;
1211 case H264_SEI_FPA_TYPE_INTERLEAVE_COLUMN:
1212 stereo->type = AV_STEREO3D_COLUMNS;
1214 case H264_SEI_FPA_TYPE_INTERLEAVE_ROW:
1215 stereo->type = AV_STEREO3D_LINES;
1217 case H264_SEI_FPA_TYPE_SIDE_BY_SIDE:
1218 if (fp->quincunx_sampling_flag)
1219 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
1221 stereo->type = AV_STEREO3D_SIDEBYSIDE;
1223 case H264_SEI_FPA_TYPE_TOP_BOTTOM:
1224 stereo->type = AV_STEREO3D_TOPBOTTOM;
1226 case H264_SEI_FPA_TYPE_INTERLEAVE_TEMPORAL:
1227 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
1229 case H264_SEI_FPA_TYPE_2D:
1230 stereo->type = AV_STEREO3D_2D;
1234 if (fp->content_interpretation_type == 2)
1235 stereo->flags = AV_STEREO3D_FLAG_INVERT;
1237 if (fp->arrangement_type == H264_SEI_FPA_TYPE_INTERLEAVE_TEMPORAL) {
1238 if (fp->current_frame_is_frame0_flag)
1239 stereo->view = AV_STEREO3D_VIEW_LEFT;
1241 stereo->view = AV_STEREO3D_VIEW_RIGHT;
1246 if (h->sei.display_orientation.present &&
1247 (h->sei.display_orientation.anticlockwise_rotation ||
1248 h->sei.display_orientation.hflip ||
1249 h->sei.display_orientation.vflip)) {
1250 H264SEIDisplayOrientation *o = &h->sei.display_orientation;
1251 double angle = o->anticlockwise_rotation * 360 / (double) (1 << 16);
1252 AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
1253 AV_FRAME_DATA_DISPLAYMATRIX,
1254 sizeof(int32_t) * 9);
1256 av_display_rotation_set((int32_t *)rotation->data, angle);
1257 av_display_matrix_flip((int32_t *)rotation->data,
1258 o->hflip, o->vflip);
1262 if (h->sei.afd.present) {
1263 AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
1267 *sd->data = h->sei.afd.active_format_description;
1268 h->sei.afd.present = 0;
1272 if (h->sei.a53_caption.buf_ref) {
1273 H264SEIA53Caption *a53 = &h->sei.a53_caption;
1275 AVFrameSideData *sd = av_frame_new_side_data_from_buf(cur->f, AV_FRAME_DATA_A53_CC, a53->buf_ref);
1277 av_buffer_unref(&a53->buf_ref);
1278 a53->buf_ref = NULL;
1280 h->avctx->properties |= FF_CODEC_PROPERTY_CLOSED_CAPTIONS;
1283 if (h->sei.picture_timing.timecode_cnt > 0) {
1287 AVFrameSideData *tcside = av_frame_new_side_data(cur->f,
1288 AV_FRAME_DATA_S12M_TIMECODE,
1289 sizeof(uint32_t)*4);
1291 return AVERROR(ENOMEM);
1293 tc_sd = (uint32_t*)tcside->data;
1294 tc_sd[0] = h->sei.picture_timing.timecode_cnt;
1296 for (int i = 0; i < tc_sd[0]; i++) {
1299 /* For SMPTE 12-M timecodes, frame count is a special case if > 30 FPS.
1300 See SMPTE ST 12-1:2014 Sec 12.1 for more info. */
1301 if (av_cmp_q(h->avctx->framerate, (AVRational) {30, 1}) == 1) {
1302 frames = h->sei.picture_timing.timecode[i].frame / 2;
1303 if (h->sei.picture_timing.timecode[i].frame % 2 == 1) {
1304 if (av_cmp_q(h->avctx->framerate, (AVRational) {50, 1}) == 0)
1310 frames = h->sei.picture_timing.timecode[i].frame;
1313 tc |= h->sei.picture_timing.timecode[i].dropframe << 30;
1314 tc |= (frames / 10) << 28;
1315 tc |= (frames % 10) << 24;
1316 tc |= (h->sei.picture_timing.timecode[i].seconds / 10) << 20;
1317 tc |= (h->sei.picture_timing.timecode[i].seconds % 10) << 16;
1318 tc |= (h->sei.picture_timing.timecode[i].minutes / 10) << 12;
1319 tc |= (h->sei.picture_timing.timecode[i].minutes % 10) << 8;
1320 tc |= (h->sei.picture_timing.timecode[i].hours / 10) << 4;
1321 tc |= (h->sei.picture_timing.timecode[i].hours % 10);
1325 h->sei.picture_timing.timecode_cnt = 0;
1331 static int h264_select_output_frame(H264Context *h)
1333 const SPS *sps = h->ps.sps;
1334 H264Picture *out = h->cur_pic_ptr;
1335 H264Picture *cur = h->cur_pic_ptr;
1336 int i, pics, out_of_order, out_idx;
1338 cur->mmco_reset = h->mmco_reset;
1341 if (sps->bitstream_restriction_flag ||
1342 h->avctx->strict_std_compliance >= FF_COMPLIANCE_STRICT) {
1343 h->avctx->has_b_frames = FFMAX(h->avctx->has_b_frames, sps->num_reorder_frames);
1346 for (i = 0; 1; i++) {
1347 if(i == MAX_DELAYED_PIC_COUNT || cur->poc < h->last_pocs[i]){
1349 h->last_pocs[i-1] = cur->poc;
1352 h->last_pocs[i-1]= h->last_pocs[i];
1355 out_of_order = MAX_DELAYED_PIC_COUNT - i;
1356 if( cur->f->pict_type == AV_PICTURE_TYPE_B
1357 || (h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > INT_MIN && h->last_pocs[MAX_DELAYED_PIC_COUNT-1] - (int64_t)h->last_pocs[MAX_DELAYED_PIC_COUNT-2] > 2))
1358 out_of_order = FFMAX(out_of_order, 1);
1359 if (out_of_order == MAX_DELAYED_PIC_COUNT) {
1360 av_log(h->avctx, AV_LOG_VERBOSE, "Invalid POC %d<%d\n", cur->poc, h->last_pocs[0]);
1361 for (i = 1; i < MAX_DELAYED_PIC_COUNT; i++)
1362 h->last_pocs[i] = INT_MIN;
1363 h->last_pocs[0] = cur->poc;
1364 cur->mmco_reset = 1;
1365 } else if(h->avctx->has_b_frames < out_of_order && !sps->bitstream_restriction_flag){
1366 int loglevel = h->avctx->frame_number > 1 ? AV_LOG_WARNING : AV_LOG_VERBOSE;
1367 av_log(h->avctx, loglevel, "Increasing reorder buffer to %d\n", out_of_order);
1368 h->avctx->has_b_frames = out_of_order;
1372 while (h->delayed_pic[pics])
1375 av_assert0(pics <= MAX_DELAYED_PIC_COUNT);
1377 h->delayed_pic[pics++] = cur;
1378 if (cur->reference == 0)
1379 cur->reference = DELAYED_PIC_REF;
1381 out = h->delayed_pic[0];
1383 for (i = 1; h->delayed_pic[i] &&
1384 !h->delayed_pic[i]->f->key_frame &&
1385 !h->delayed_pic[i]->mmco_reset;
1387 if (h->delayed_pic[i]->poc < out->poc) {
1388 out = h->delayed_pic[i];
1391 if (h->avctx->has_b_frames == 0 &&
1392 (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset))
1393 h->next_outputed_poc = INT_MIN;
1394 out_of_order = out->poc < h->next_outputed_poc;
1396 if (out_of_order || pics > h->avctx->has_b_frames) {
1397 out->reference &= ~DELAYED_PIC_REF;
1398 for (i = out_idx; h->delayed_pic[i]; i++)
1399 h->delayed_pic[i] = h->delayed_pic[i + 1];
1401 if (!out_of_order && pics > h->avctx->has_b_frames) {
1402 h->next_output_pic = out;
1403 if (out_idx == 0 && h->delayed_pic[0] && (h->delayed_pic[0]->f->key_frame || h->delayed_pic[0]->mmco_reset)) {
1404 h->next_outputed_poc = INT_MIN;
1406 h->next_outputed_poc = out->poc;
1408 if (out->recovered) {
1409 // We have reached an recovery point and all frames after it in
1410 // display order are "recovered".
1411 h->frame_recovered |= FRAME_RECOVERED_SEI;
1413 out->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_SEI);
1415 if (!out->recovered) {
1416 if (!(h->avctx->flags & AV_CODEC_FLAG_OUTPUT_CORRUPT) &&
1417 !(h->avctx->flags2 & AV_CODEC_FLAG2_SHOW_ALL)) {
1418 h->next_output_pic = NULL;
1420 out->f->flags |= AV_FRAME_FLAG_CORRUPT;
1424 av_log(h->avctx, AV_LOG_DEBUG, "no picture %s\n", out_of_order ? "ooo" : "");
1430 /* This function is called right after decoding the slice header for a first
1431 * slice in a field (or a frame). It decides whether we are decoding a new frame
1432 * or a second field in a pair and does the necessary setup.
1434 static int h264_field_start(H264Context *h, const H264SliceContext *sl,
1435 const H2645NAL *nal, int first_slice)
1440 int last_pic_structure, last_pic_droppable, ret;
1442 ret = h264_init_ps(h, sl, first_slice);
1448 if (sps && sps->bitstream_restriction_flag &&
1449 h->avctx->has_b_frames < sps->num_reorder_frames) {
1450 h->avctx->has_b_frames = sps->num_reorder_frames;
1453 last_pic_droppable = h->droppable;
1454 last_pic_structure = h->picture_structure;
1455 h->droppable = (nal->ref_idc == 0);
1456 h->picture_structure = sl->picture_structure;
1458 h->poc.frame_num = sl->frame_num;
1459 h->poc.poc_lsb = sl->poc_lsb;
1460 h->poc.delta_poc_bottom = sl->delta_poc_bottom;
1461 h->poc.delta_poc[0] = sl->delta_poc[0];
1462 h->poc.delta_poc[1] = sl->delta_poc[1];
1464 /* Shorten frame num gaps so we don't have to allocate reference
1465 * frames just to throw them away */
1466 if (h->poc.frame_num != h->poc.prev_frame_num) {
1467 int unwrap_prev_frame_num = h->poc.prev_frame_num;
1468 int max_frame_num = 1 << sps->log2_max_frame_num;
1470 if (unwrap_prev_frame_num > h->poc.frame_num)
1471 unwrap_prev_frame_num -= max_frame_num;
1473 if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
1474 unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
1475 if (unwrap_prev_frame_num < 0)
1476 unwrap_prev_frame_num += max_frame_num;
1478 h->poc.prev_frame_num = unwrap_prev_frame_num;
1482 /* See if we have a decoded first field looking for a pair...
1483 * Here, we're using that to see if we should mark previously
1484 * decode frames as "finished".
1485 * We have to do that before the "dummy" in-between frame allocation,
1486 * since that can modify h->cur_pic_ptr. */
1487 if (h->first_field) {
1488 int last_field = last_pic_structure == PICT_BOTTOM_FIELD;
1489 av_assert0(h->cur_pic_ptr);
1490 av_assert0(h->cur_pic_ptr->f->buf[0]);
1491 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1493 /* Mark old field/frame as completed */
1494 if (h->cur_pic_ptr->tf.owner[last_field] == h->avctx) {
1495 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, last_field);
1498 /* figure out if we have a complementary field pair */
1499 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1500 /* Previous field is unmatched. Don't display it, but let it
1501 * remain for reference if marked as such. */
1502 if (last_pic_structure != PICT_FRAME) {
1503 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1504 last_pic_structure == PICT_TOP_FIELD);
1507 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1508 /* This and previous field were reference, but had
1509 * different frame_nums. Consider this field first in
1510 * pair. Throw away previous field except for reference
1512 if (last_pic_structure != PICT_FRAME) {
1513 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1514 last_pic_structure == PICT_TOP_FIELD);
1517 /* Second field in complementary pair */
1518 if (!((last_pic_structure == PICT_TOP_FIELD &&
1519 h->picture_structure == PICT_BOTTOM_FIELD) ||
1520 (last_pic_structure == PICT_BOTTOM_FIELD &&
1521 h->picture_structure == PICT_TOP_FIELD))) {
1522 av_log(h->avctx, AV_LOG_ERROR,
1523 "Invalid field mode combination %d/%d\n",
1524 last_pic_structure, h->picture_structure);
1525 h->picture_structure = last_pic_structure;
1526 h->droppable = last_pic_droppable;
1527 return AVERROR_INVALIDDATA;
1528 } else if (last_pic_droppable != h->droppable) {
1529 avpriv_request_sample(h->avctx,
1530 "Found reference and non-reference fields in the same frame, which");
1531 h->picture_structure = last_pic_structure;
1532 h->droppable = last_pic_droppable;
1533 return AVERROR_PATCHWELCOME;
1539 while (h->poc.frame_num != h->poc.prev_frame_num && !h->first_field &&
1540 h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
1541 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1542 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1543 h->poc.frame_num, h->poc.prev_frame_num);
1544 if (!sps->gaps_in_frame_num_allowed_flag)
1545 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1546 h->last_pocs[i] = INT_MIN;
1547 ret = h264_frame_start(h);
1553 h->poc.prev_frame_num++;
1554 h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
1555 h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
1556 h->cur_pic_ptr->invalid_gap = !sps->gaps_in_frame_num_allowed_flag;
1557 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1558 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1560 h->explicit_ref_marking = 0;
1561 ret = ff_h264_execute_ref_pic_marking(h);
1562 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1564 /* Error concealment: If a ref is missing, copy the previous ref
1566 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1567 * many assumptions about there being no actual duplicates.
1568 * FIXME: This does not copy padding for out-of-frame motion
1569 * vectors. Given we are concealing a lost frame, this probably
1570 * is not noticeable by comparison, but it should be fixed. */
1571 if (h->short_ref_count) {
1573 1<<(h->ps.sps->bit_depth_luma-1),
1574 1<<(h->ps.sps->bit_depth_chroma-1),
1575 1<<(h->ps.sps->bit_depth_chroma-1),
1580 h->short_ref[0]->f->width == prev->f->width &&
1581 h->short_ref[0]->f->height == prev->f->height &&
1582 h->short_ref[0]->f->format == prev->f->format) {
1583 ff_thread_await_progress(&prev->tf, INT_MAX, 0);
1584 if (prev->field_picture)
1585 ff_thread_await_progress(&prev->tf, INT_MAX, 1);
1586 av_image_copy(h->short_ref[0]->f->data,
1587 h->short_ref[0]->f->linesize,
1588 (const uint8_t **)prev->f->data,
1593 h->short_ref[0]->poc = prev->poc + 2;
1594 } else if (!h->frame_recovered && !h->avctx->hwaccel)
1595 ff_color_frame(h->short_ref[0]->f, c);
1596 h->short_ref[0]->frame_num = h->poc.prev_frame_num;
1600 /* See if we have a decoded first field looking for a pair...
1601 * We're using that to see whether to continue decoding in that
1602 * frame, or to allocate a new one. */
1603 if (h->first_field) {
1604 av_assert0(h->cur_pic_ptr);
1605 av_assert0(h->cur_pic_ptr->f->buf[0]);
1606 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1608 /* figure out if we have a complementary field pair */
1609 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1610 /* Previous field is unmatched. Don't display it, but let it
1611 * remain for reference if marked as such. */
1612 h->missing_fields ++;
1613 h->cur_pic_ptr = NULL;
1614 h->first_field = FIELD_PICTURE(h);
1616 h->missing_fields = 0;
1617 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1618 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1619 h->picture_structure==PICT_BOTTOM_FIELD);
1620 /* This and the previous field had different frame_nums.
1621 * Consider this field first in pair. Throw away previous
1622 * one except for reference purposes. */
1624 h->cur_pic_ptr = NULL;
1625 } else if (h->cur_pic_ptr->reference & DELAYED_PIC_REF) {
1626 /* This frame was already output, we cannot draw into it
1630 h->cur_pic_ptr = NULL;
1632 /* Second field in complementary pair */
1637 /* Frame or first field in a potentially complementary pair */
1638 h->first_field = FIELD_PICTURE(h);
1641 if (!FIELD_PICTURE(h) || h->first_field) {
1642 if (h264_frame_start(h) < 0) {
1644 return AVERROR_INVALIDDATA;
1647 int field = h->picture_structure == PICT_BOTTOM_FIELD;
1648 release_unused_pictures(h, 0);
1649 h->cur_pic_ptr->tf.owner[field] = h->avctx;
1651 /* Some macroblocks can be accessed before they're available in case
1652 * of lost slices, MBAFF or threading. */
1653 if (FIELD_PICTURE(h)) {
1654 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1655 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1657 memset(h->slice_table, -1,
1658 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1661 ret = ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
1662 h->ps.sps, &h->poc, h->picture_structure, nal->ref_idc);
1666 memcpy(h->mmco, sl->mmco, sl->nb_mmco * sizeof(*h->mmco));
1667 h->nb_mmco = sl->nb_mmco;
1668 h->explicit_ref_marking = sl->explicit_ref_marking;
1670 h->picture_idr = nal->type == H264_NAL_IDR_SLICE;
1672 if (h->sei.recovery_point.recovery_frame_cnt >= 0) {
1673 const int sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
1675 if (h->poc.frame_num != sei_recovery_frame_cnt || sl->slice_type_nos != AV_PICTURE_TYPE_I)
1676 h->valid_recovery_point = 1;
1678 if ( h->recovery_frame < 0
1679 || av_mod_uintp2(h->recovery_frame - h->poc.frame_num, h->ps.sps->log2_max_frame_num) > sei_recovery_frame_cnt) {
1680 h->recovery_frame = av_mod_uintp2(h->poc.frame_num + sei_recovery_frame_cnt, h->ps.sps->log2_max_frame_num);
1682 if (!h->valid_recovery_point)
1683 h->recovery_frame = h->poc.frame_num;
1687 h->cur_pic_ptr->f->key_frame |= (nal->type == H264_NAL_IDR_SLICE);
1689 if (nal->type == H264_NAL_IDR_SLICE ||
1690 (h->recovery_frame == h->poc.frame_num && nal->ref_idc)) {
1691 h->recovery_frame = -1;
1692 h->cur_pic_ptr->recovered = 1;
1694 // If we have an IDR, all frames after it in decoded order are
1696 if (nal->type == H264_NAL_IDR_SLICE)
1697 h->frame_recovered |= FRAME_RECOVERED_IDR;
1699 h->cur_pic_ptr->recovered |= h->frame_recovered;
1701 h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
1704 /* Set the frame properties/side data. Only done for the second field in
1705 * field coded frames, since some SEI information is present for each field
1706 * and is merged by the SEI parsing code. */
1707 if (!FIELD_PICTURE(h) || !h->first_field || h->missing_fields > 1) {
1708 ret = h264_export_frame_props(h);
1712 ret = h264_select_output_frame(h);
1720 static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl,
1721 const H2645NAL *nal)
1726 unsigned int slice_type, tmp, i;
1727 int field_pic_flag, bottom_field_flag;
1728 int first_slice = sl == h->slice_ctx && !h->current_slice;
1729 int picture_structure;
1732 av_assert0(!h->setup_finished);
1734 sl->first_mb_addr = get_ue_golomb_long(&sl->gb);
1736 slice_type = get_ue_golomb_31(&sl->gb);
1737 if (slice_type > 9) {
1738 av_log(h->avctx, AV_LOG_ERROR,
1739 "slice type %d too large at %d\n",
1740 slice_type, sl->first_mb_addr);
1741 return AVERROR_INVALIDDATA;
1743 if (slice_type > 4) {
1745 sl->slice_type_fixed = 1;
1747 sl->slice_type_fixed = 0;
1749 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1750 sl->slice_type = slice_type;
1751 sl->slice_type_nos = slice_type & 3;
1753 if (nal->type == H264_NAL_IDR_SLICE &&
1754 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1755 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1756 return AVERROR_INVALIDDATA;
1759 sl->pps_id = get_ue_golomb(&sl->gb);
1760 if (sl->pps_id >= MAX_PPS_COUNT) {
1761 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", sl->pps_id);
1762 return AVERROR_INVALIDDATA;
1764 if (!h->ps.pps_list[sl->pps_id]) {
1765 av_log(h->avctx, AV_LOG_ERROR,
1766 "non-existing PPS %u referenced\n",
1768 return AVERROR_INVALIDDATA;
1770 pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
1773 sl->frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
1775 if (h->poc.frame_num != sl->frame_num) {
1776 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1777 h->poc.frame_num, sl->frame_num);
1778 return AVERROR_INVALIDDATA;
1784 if (sps->frame_mbs_only_flag) {
1785 picture_structure = PICT_FRAME;
1787 if (!sps->direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1788 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1791 field_pic_flag = get_bits1(&sl->gb);
1792 if (field_pic_flag) {
1793 bottom_field_flag = get_bits1(&sl->gb);
1794 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1796 picture_structure = PICT_FRAME;
1799 sl->picture_structure = picture_structure;
1800 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1802 if (picture_structure == PICT_FRAME) {
1803 sl->curr_pic_num = sl->frame_num;
1804 sl->max_pic_num = 1 << sps->log2_max_frame_num;
1806 sl->curr_pic_num = 2 * sl->frame_num + 1;
1807 sl->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
1810 if (nal->type == H264_NAL_IDR_SLICE)
1811 get_ue_golomb_long(&sl->gb); /* idr_pic_id */
1813 if (sps->poc_type == 0) {
1814 sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
1816 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1817 sl->delta_poc_bottom = get_se_golomb(&sl->gb);
1820 if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
1821 sl->delta_poc[0] = get_se_golomb(&sl->gb);
1823 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1824 sl->delta_poc[1] = get_se_golomb(&sl->gb);
1827 sl->redundant_pic_count = 0;
1828 if (pps->redundant_pic_cnt_present)
1829 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1831 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1832 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1834 ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
1835 &sl->gb, pps, sl->slice_type_nos,
1836 picture_structure, h->avctx);
1840 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1841 ret = ff_h264_decode_ref_pic_list_reordering(sl, h->avctx);
1843 sl->ref_count[1] = sl->ref_count[0] = 0;
1848 sl->pwt.use_weight = 0;
1849 for (i = 0; i < 2; i++) {
1850 sl->pwt.luma_weight_flag[i] = 0;
1851 sl->pwt.chroma_weight_flag[i] = 0;
1853 if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1854 (pps->weighted_bipred_idc == 1 &&
1855 sl->slice_type_nos == AV_PICTURE_TYPE_B)) {
1856 ret = ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
1857 sl->slice_type_nos, &sl->pwt,
1858 picture_structure, h->avctx);
1863 sl->explicit_ref_marking = 0;
1865 ret = ff_h264_decode_ref_pic_marking(sl, &sl->gb, nal, h->avctx);
1866 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1867 return AVERROR_INVALIDDATA;
1870 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
1871 tmp = get_ue_golomb_31(&sl->gb);
1873 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1874 return AVERROR_INVALIDDATA;
1876 sl->cabac_init_idc = tmp;
1879 sl->last_qscale_diff = 0;
1880 tmp = pps->init_qp + (unsigned)get_se_golomb(&sl->gb);
1881 if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
1882 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1883 return AVERROR_INVALIDDATA;
1886 sl->chroma_qp[0] = get_chroma_qp(pps, 0, sl->qscale);
1887 sl->chroma_qp[1] = get_chroma_qp(pps, 1, sl->qscale);
1888 // FIXME qscale / qp ... stuff
1889 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1890 get_bits1(&sl->gb); /* sp_for_switch_flag */
1891 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1892 sl->slice_type == AV_PICTURE_TYPE_SI)
1893 get_se_golomb(&sl->gb); /* slice_qs_delta */
1895 sl->deblocking_filter = 1;
1896 sl->slice_alpha_c0_offset = 0;
1897 sl->slice_beta_offset = 0;
1898 if (pps->deblocking_filter_parameters_present) {
1899 tmp = get_ue_golomb_31(&sl->gb);
1901 av_log(h->avctx, AV_LOG_ERROR,
1902 "deblocking_filter_idc %u out of range\n", tmp);
1903 return AVERROR_INVALIDDATA;
1905 sl->deblocking_filter = tmp;
1906 if (sl->deblocking_filter < 2)
1907 sl->deblocking_filter ^= 1; // 1<->0
1909 if (sl->deblocking_filter) {
1910 int slice_alpha_c0_offset_div2 = get_se_golomb(&sl->gb);
1911 int slice_beta_offset_div2 = get_se_golomb(&sl->gb);
1912 if (slice_alpha_c0_offset_div2 > 6 ||
1913 slice_alpha_c0_offset_div2 < -6 ||
1914 slice_beta_offset_div2 > 6 ||
1915 slice_beta_offset_div2 < -6) {
1916 av_log(h->avctx, AV_LOG_ERROR,
1917 "deblocking filter parameters %d %d out of range\n",
1918 slice_alpha_c0_offset_div2, slice_beta_offset_div2);
1919 return AVERROR_INVALIDDATA;
1921 sl->slice_alpha_c0_offset = slice_alpha_c0_offset_div2 * 2;
1922 sl->slice_beta_offset = slice_beta_offset_div2 * 2;
1929 /* do all the per-slice initialization needed before we can start decoding the
1931 static int h264_slice_init(H264Context *h, H264SliceContext *sl,
1932 const H2645NAL *nal)
1936 if (h->picture_idr && nal->type != H264_NAL_IDR_SLICE) {
1937 av_log(h->avctx, AV_LOG_ERROR, "Invalid mix of IDR and non-IDR slices\n");
1938 return AVERROR_INVALIDDATA;
1941 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1942 if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1943 sl->first_mb_addr >= h->mb_num) {
1944 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1945 return AVERROR_INVALIDDATA;
1947 sl->resync_mb_x = sl->mb_x = sl->first_mb_addr % h->mb_width;
1948 sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
1949 FIELD_OR_MBAFF_PICTURE(h);
1950 if (h->picture_structure == PICT_BOTTOM_FIELD)
1951 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1952 av_assert1(sl->mb_y < h->mb_height);
1954 ret = ff_h264_build_ref_list(h, sl);
1958 if (h->ps.pps->weighted_bipred_idc == 2 &&
1959 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1960 implicit_weight_table(h, sl, -1);
1961 if (FRAME_MBAFF(h)) {
1962 implicit_weight_table(h, sl, 0);
1963 implicit_weight_table(h, sl, 1);
1967 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1968 ff_h264_direct_dist_scale_factor(h, sl);
1969 if (!h->setup_finished)
1970 ff_h264_direct_ref_list_init(h, sl);
1972 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1973 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1974 h->nal_unit_type != H264_NAL_IDR_SLICE) ||
1975 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1976 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1977 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1978 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1979 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1981 sl->deblocking_filter = 0;
1983 if (sl->deblocking_filter == 1 && h->nb_slice_ctx > 1) {
1984 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1985 /* Cheat slightly for speed:
1986 * Do not bother to deblock across slices. */
1987 sl->deblocking_filter = 2;
1989 h->postpone_filter = 1;
1992 sl->qp_thresh = 15 -
1993 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1995 h->ps.pps->chroma_qp_index_offset[0],
1996 h->ps.pps->chroma_qp_index_offset[1]) +
1997 6 * (h->ps.sps->bit_depth_luma - 8);
1999 sl->slice_num = ++h->current_slice;
2002 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
2003 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
2004 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
2005 && sl->slice_num >= MAX_SLICES) {
2006 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
2007 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);
2010 for (j = 0; j < 2; j++) {
2012 int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
2013 for (i = 0; i < 16; i++) {
2015 if (j < sl->list_count && i < sl->ref_count[j] &&
2016 sl->ref_list[j][i].parent->f->buf[0]) {
2018 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
2019 for (k = 0; k < h->short_ref_count; k++)
2020 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
2024 for (k = 0; k < h->long_ref_count; k++)
2025 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
2026 id_list[i] = h->short_ref_count + k;
2034 for (i = 0; i < 16; i++)
2035 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
2037 ref2frm[18 + 1] = -1;
2038 for (i = 16; i < 48; i++)
2039 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
2040 (sl->ref_list[j][i].reference & 3);
2043 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
2044 av_log(h->avctx, AV_LOG_DEBUG,
2045 "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",
2047 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
2048 sl->mb_y * h->mb_width + sl->mb_x,
2049 av_get_picture_type_char(sl->slice_type),
2050 sl->slice_type_fixed ? " fix" : "",
2051 nal->type == H264_NAL_IDR_SLICE ? " IDR" : "",
2053 h->cur_pic_ptr->field_poc[0],
2054 h->cur_pic_ptr->field_poc[1],
2055 sl->ref_count[0], sl->ref_count[1],
2057 sl->deblocking_filter,
2058 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
2060 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
2061 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2067 int ff_h264_queue_decode_slice(H264Context *h, const H2645NAL *nal)
2069 H264SliceContext *sl = h->slice_ctx + h->nb_slice_ctx_queued;
2070 int first_slice = sl == h->slice_ctx && !h->current_slice;
2075 ret = h264_slice_header_parse(h, sl, nal);
2079 // discard redundant pictures
2080 if (sl->redundant_pic_count > 0) {
2081 sl->ref_count[0] = sl->ref_count[1] = 0;
2085 if (sl->first_mb_addr == 0 || !h->current_slice) {
2086 if (h->setup_finished) {
2087 av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
2088 return AVERROR_INVALIDDATA;
2092 if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
2093 if (h->current_slice) {
2094 // this slice starts a new field
2095 // first decode any pending queued slices
2096 if (h->nb_slice_ctx_queued) {
2097 H264SliceContext tmp_ctx;
2099 ret = ff_h264_execute_decode_slices(h);
2100 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
2103 memcpy(&tmp_ctx, h->slice_ctx, sizeof(tmp_ctx));
2104 memcpy(h->slice_ctx, sl, sizeof(tmp_ctx));
2105 memcpy(sl, &tmp_ctx, sizeof(tmp_ctx));
2109 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
2110 ret = ff_h264_field_end(h, h->slice_ctx, 1);
2113 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == H264_NAL_IDR_SLICE) {
2114 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
2115 ret = ff_h264_field_end(h, h->slice_ctx, 1);
2116 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
2117 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
2118 h->cur_pic_ptr = NULL;
2122 return AVERROR_INVALIDDATA;
2125 if (!h->first_field) {
2126 if (h->cur_pic_ptr && !h->droppable) {
2127 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
2128 h->picture_structure == PICT_BOTTOM_FIELD);
2130 h->cur_pic_ptr = NULL;
2134 if (!h->current_slice)
2135 av_assert0(sl == h->slice_ctx);
2137 if (h->current_slice == 0 && !h->first_field) {
2139 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
2140 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
2141 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
2142 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != H264_NAL_IDR_SLICE && h->sei.recovery_point.recovery_frame_cnt < 0) ||
2143 h->avctx->skip_frame >= AVDISCARD_ALL) {
2149 const PPS *pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
2151 if (h->ps.pps->sps_id != pps->sps_id ||
2152 h->ps.pps->transform_8x8_mode != pps->transform_8x8_mode /*||
2153 (h->setup_finished && h->ps.pps != pps)*/) {
2154 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
2155 return AVERROR_INVALIDDATA;
2157 if (h->ps.sps != pps->sps) {
2158 av_log(h->avctx, AV_LOG_ERROR,
2159 "SPS changed in the middle of the frame\n");
2160 return AVERROR_INVALIDDATA;
2164 if (h->current_slice == 0) {
2165 ret = h264_field_start(h, sl, nal, first_slice);
2169 if (h->picture_structure != sl->picture_structure ||
2170 h->droppable != (nal->ref_idc == 0)) {
2171 av_log(h->avctx, AV_LOG_ERROR,
2172 "Changing field mode (%d -> %d) between slices is not allowed\n",
2173 h->picture_structure, sl->picture_structure);
2174 return AVERROR_INVALIDDATA;
2175 } else if (!h->cur_pic_ptr) {
2176 av_log(h->avctx, AV_LOG_ERROR,
2177 "unset cur_pic_ptr on slice %d\n",
2178 h->current_slice + 1);
2179 return AVERROR_INVALIDDATA;
2183 ret = h264_slice_init(h, sl, nal);
2187 h->nb_slice_ctx_queued++;
2192 int ff_h264_get_slice_type(const H264SliceContext *sl)
2194 switch (sl->slice_type) {
2195 case AV_PICTURE_TYPE_P:
2197 case AV_PICTURE_TYPE_B:
2199 case AV_PICTURE_TYPE_I:
2201 case AV_PICTURE_TYPE_SP:
2203 case AV_PICTURE_TYPE_SI:
2206 return AVERROR_INVALIDDATA;
2210 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
2211 H264SliceContext *sl,
2212 int mb_type, int top_xy,
2213 int left_xy[LEFT_MBS],
2215 int left_type[LEFT_MBS],
2216 int mb_xy, int list)
2218 int b_stride = h->b_stride;
2219 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
2220 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
2221 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2222 if (USES_LIST(top_type, list)) {
2223 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2224 const int b8_xy = 4 * top_xy + 2;
2225 const int *ref2frm = &h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2226 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2227 ref_cache[0 - 1 * 8] =
2228 ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
2229 ref_cache[2 - 1 * 8] =
2230 ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
2232 AV_ZERO128(mv_dst - 1 * 8);
2233 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2236 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2237 if (USES_LIST(left_type[LTOP], list)) {
2238 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2239 const int b8_xy = 4 * left_xy[LTOP] + 1;
2240 const int *ref2frm = &h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2241 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2242 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2243 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2244 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2246 ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2247 ref_cache[-1 + 16] =
2248 ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2250 AV_ZERO32(mv_dst - 1 + 0);
2251 AV_ZERO32(mv_dst - 1 + 8);
2252 AV_ZERO32(mv_dst - 1 + 16);
2253 AV_ZERO32(mv_dst - 1 + 24);
2256 ref_cache[-1 + 16] =
2257 ref_cache[-1 + 24] = LIST_NOT_USED;
2262 if (!USES_LIST(mb_type, list)) {
2263 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2264 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2265 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2266 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2267 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2272 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2273 const int *ref2frm = &h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
2274 uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
2275 uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
2276 AV_WN32A(&ref_cache[0 * 8], ref01);
2277 AV_WN32A(&ref_cache[1 * 8], ref01);
2278 AV_WN32A(&ref_cache[2 * 8], ref23);
2279 AV_WN32A(&ref_cache[3 * 8], ref23);
2283 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
2284 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2285 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2286 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2287 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2292 * @return non zero if the loop filter can be skipped
2294 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
2296 const int mb_xy = sl->mb_xy;
2297 int top_xy, left_xy[LEFT_MBS];
2298 int top_type, left_type[LEFT_MBS];
2302 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
2304 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2305 if (FRAME_MBAFF(h)) {
2306 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2307 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2309 if (left_mb_field_flag != curr_mb_field_flag)
2310 left_xy[LTOP] -= h->mb_stride;
2312 if (curr_mb_field_flag)
2313 top_xy += h->mb_stride &
2314 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2315 if (left_mb_field_flag != curr_mb_field_flag)
2316 left_xy[LBOT] += h->mb_stride;
2320 sl->top_mb_xy = top_xy;
2321 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2322 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2324 /* For sufficiently low qp, filtering wouldn't do anything.
2325 * This is a conservative estimate: could also check beta_offset
2326 * and more accurate chroma_qp. */
2327 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2328 int qp = h->cur_pic.qscale_table[mb_xy];
2329 if (qp <= qp_thresh &&
2330 (left_xy[LTOP] < 0 ||
2331 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2333 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2334 if (!FRAME_MBAFF(h))
2336 if ((left_xy[LTOP] < 0 ||
2337 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2338 (top_xy < h->mb_stride ||
2339 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2344 top_type = h->cur_pic.mb_type[top_xy];
2345 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2346 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2347 if (sl->deblocking_filter == 2) {
2348 if (h->slice_table[top_xy] != sl->slice_num)
2350 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2351 left_type[LTOP] = left_type[LBOT] = 0;
2353 if (h->slice_table[top_xy] == 0xFFFF)
2355 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2356 left_type[LTOP] = left_type[LBOT] = 0;
2358 sl->top_type = top_type;
2359 sl->left_type[LTOP] = left_type[LTOP];
2360 sl->left_type[LBOT] = left_type[LBOT];
2362 if (IS_INTRA(mb_type))
2365 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2366 top_type, left_type, mb_xy, 0);
2367 if (sl->list_count == 2)
2368 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2369 top_type, left_type, mb_xy, 1);
2371 nnz = h->non_zero_count[mb_xy];
2372 nnz_cache = sl->non_zero_count_cache;
2373 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2374 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2375 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2376 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2377 sl->cbp = h->cbp_table[mb_xy];
2380 nnz = h->non_zero_count[top_xy];
2381 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2384 if (left_type[LTOP]) {
2385 nnz = h->non_zero_count[left_xy[LTOP]];
2386 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2387 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2388 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2389 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2392 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2393 * from what the loop filter needs */
2394 if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
2395 if (IS_8x8DCT(top_type)) {
2396 nnz_cache[4 + 8 * 0] =
2397 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2398 nnz_cache[6 + 8 * 0] =
2399 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2401 if (IS_8x8DCT(left_type[LTOP])) {
2402 nnz_cache[3 + 8 * 1] =
2403 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2405 if (IS_8x8DCT(left_type[LBOT])) {
2406 nnz_cache[3 + 8 * 3] =
2407 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2410 if (IS_8x8DCT(mb_type)) {
2411 nnz_cache[scan8[0]] =
2412 nnz_cache[scan8[1]] =
2413 nnz_cache[scan8[2]] =
2414 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2416 nnz_cache[scan8[0 + 4]] =
2417 nnz_cache[scan8[1 + 4]] =
2418 nnz_cache[scan8[2 + 4]] =
2419 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2421 nnz_cache[scan8[0 + 8]] =
2422 nnz_cache[scan8[1 + 8]] =
2423 nnz_cache[scan8[2 + 8]] =
2424 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2426 nnz_cache[scan8[0 + 12]] =
2427 nnz_cache[scan8[1 + 12]] =
2428 nnz_cache[scan8[2 + 12]] =
2429 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2436 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2438 uint8_t *dest_y, *dest_cb, *dest_cr;
2439 int linesize, uvlinesize, mb_x, mb_y;
2440 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2441 const int old_slice_type = sl->slice_type;
2442 const int pixel_shift = h->pixel_shift;
2443 const int block_h = 16 >> h->chroma_y_shift;
2445 if (h->postpone_filter)
2448 if (sl->deblocking_filter) {
2449 for (mb_x = start_x; mb_x < end_x; mb_x++)
2450 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2452 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2453 mb_type = h->cur_pic.mb_type[mb_xy];
2457 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2461 dest_y = h->cur_pic.f->data[0] +
2462 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2463 dest_cb = h->cur_pic.f->data[1] +
2464 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2465 mb_y * sl->uvlinesize * block_h;
2466 dest_cr = h->cur_pic.f->data[2] +
2467 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2468 mb_y * sl->uvlinesize * block_h;
2469 // FIXME simplify above
2472 linesize = sl->mb_linesize = sl->linesize * 2;
2473 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2474 if (mb_y & 1) { // FIXME move out of this function?
2475 dest_y -= sl->linesize * 15;
2476 dest_cb -= sl->uvlinesize * (block_h - 1);
2477 dest_cr -= sl->uvlinesize * (block_h - 1);
2480 linesize = sl->mb_linesize = sl->linesize;
2481 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2483 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2485 if (fill_filter_caches(h, sl, mb_type))
2487 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mb_xy]);
2488 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mb_xy]);
2490 if (FRAME_MBAFF(h)) {
2491 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2492 linesize, uvlinesize);
2494 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2495 dest_cr, linesize, uvlinesize);
2499 sl->slice_type = old_slice_type;
2501 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2502 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
2503 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
2506 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2508 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2509 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2510 h->cur_pic.mb_type[mb_xy - 1] :
2511 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2512 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2513 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2517 * Draw edges and report progress for the last MB row.
2519 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2521 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2522 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2523 int height = 16 << FRAME_MBAFF(h);
2524 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2526 if (sl->deblocking_filter) {
2527 if ((top + height) >= pic_height)
2528 height += deblock_border;
2529 top -= deblock_border;
2532 if (top >= pic_height || (top + height) < 0)
2535 height = FFMIN(height, pic_height - top);
2537 height = top + height;
2541 ff_h264_draw_horiz_band(h, sl, top, height);
2543 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2546 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2547 h->picture_structure == PICT_BOTTOM_FIELD);
2550 static void er_add_slice(H264SliceContext *sl,
2551 int startx, int starty,
2552 int endx, int endy, int status)
2554 if (!sl->h264->enable_er)
2557 if (CONFIG_ERROR_RESILIENCE) {
2558 ERContext *er = &sl->h264->slice_ctx[0].er;
2560 ff_er_add_slice(er, startx, starty, endx, endy, status);
2564 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2566 H264SliceContext *sl = arg;
2567 const H264Context *h = sl->h264;
2568 int lf_x_start = sl->mb_x;
2569 int orig_deblock = sl->deblocking_filter;
2572 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2573 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2575 ret = alloc_scratch_buffers(sl, sl->linesize);
2579 sl->mb_skip_run = -1;
2581 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2583 if (h->postpone_filter)
2584 sl->deblocking_filter = 0;
2586 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2587 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2589 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2590 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2592 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2593 prev_status &= ~ VP_START;
2594 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2595 h->slice_ctx[0].er.error_occurred = 1;
2599 if (h->ps.pps->cabac) {
2601 align_get_bits(&sl->gb);
2604 ret = ff_init_cabac_decoder(&sl->cabac,
2605 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2606 (get_bits_left(&sl->gb) + 7) / 8);
2610 ff_h264_init_cabac_states(h, sl);
2614 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2615 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2616 sl->next_slice_idx);
2617 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2618 sl->mb_y, ER_MB_ERROR);
2619 return AVERROR_INVALIDDATA;
2622 ret = ff_h264_decode_mb_cabac(h, sl);
2625 ff_h264_hl_decode_mb(h, sl);
2627 // FIXME optimal? or let mb_decode decode 16x32 ?
2628 if (ret >= 0 && FRAME_MBAFF(h)) {
2631 ret = ff_h264_decode_mb_cabac(h, sl);
2634 ff_h264_hl_decode_mb(h, sl);
2637 eos = get_cabac_terminate(&sl->cabac);
2639 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2640 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2641 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2642 sl->mb_y, ER_MB_END);
2643 if (sl->mb_x >= lf_x_start)
2644 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2647 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2648 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2649 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2650 av_log(h->avctx, AV_LOG_ERROR,
2651 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2653 sl->cabac.bytestream_end - sl->cabac.bytestream);
2654 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2655 sl->mb_y, ER_MB_ERROR);
2656 return AVERROR_INVALIDDATA;
2659 if (++sl->mb_x >= h->mb_width) {
2660 loop_filter(h, sl, lf_x_start, sl->mb_x);
2661 sl->mb_x = lf_x_start = 0;
2662 decode_finish_row(h, sl);
2664 if (FIELD_OR_MBAFF_PICTURE(h)) {
2666 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2667 predict_field_decoding_flag(h, sl);
2671 if (eos || sl->mb_y >= h->mb_height) {
2672 ff_tlog(h->avctx, "slice end %d %d\n",
2673 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2674 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2675 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);
2685 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2686 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2687 sl->next_slice_idx);
2688 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2689 sl->mb_y, ER_MB_ERROR);
2690 return AVERROR_INVALIDDATA;
2693 ret = ff_h264_decode_mb_cavlc(h, sl);
2696 ff_h264_hl_decode_mb(h, sl);
2698 // FIXME optimal? or let mb_decode decode 16x32 ?
2699 if (ret >= 0 && FRAME_MBAFF(h)) {
2701 ret = ff_h264_decode_mb_cavlc(h, sl);
2704 ff_h264_hl_decode_mb(h, sl);
2709 av_log(h->avctx, AV_LOG_ERROR,
2710 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2711 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2712 sl->mb_y, ER_MB_ERROR);
2716 if (++sl->mb_x >= h->mb_width) {
2717 loop_filter(h, sl, lf_x_start, sl->mb_x);
2718 sl->mb_x = lf_x_start = 0;
2719 decode_finish_row(h, sl);
2721 if (FIELD_OR_MBAFF_PICTURE(h)) {
2723 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2724 predict_field_decoding_flag(h, sl);
2726 if (sl->mb_y >= h->mb_height) {
2727 ff_tlog(h->avctx, "slice end %d %d\n",
2728 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2730 if ( get_bits_left(&sl->gb) == 0
2731 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2732 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2733 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2737 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2738 sl->mb_x, sl->mb_y, ER_MB_END);
2740 return AVERROR_INVALIDDATA;
2745 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2746 ff_tlog(h->avctx, "slice end %d %d\n",
2747 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2749 if (get_bits_left(&sl->gb) == 0) {
2750 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2751 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2752 if (sl->mb_x > lf_x_start)
2753 loop_filter(h, sl, lf_x_start, sl->mb_x);
2757 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2758 sl->mb_y, ER_MB_ERROR);
2760 return AVERROR_INVALIDDATA;
2767 sl->deblocking_filter = orig_deblock;
2772 * Call decode_slice() for each context.
2774 * @param h h264 master context
2776 int ff_h264_execute_decode_slices(H264Context *h)
2778 AVCodecContext *const avctx = h->avctx;
2779 H264SliceContext *sl;
2780 int context_count = h->nb_slice_ctx_queued;
2784 h->slice_ctx[0].next_slice_idx = INT_MAX;
2786 if (h->avctx->hwaccel || context_count < 1)
2789 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2791 if (context_count == 1) {
2793 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2794 h->postpone_filter = 0;
2796 ret = decode_slice(avctx, &h->slice_ctx[0]);
2797 h->mb_y = h->slice_ctx[0].mb_y;
2801 av_assert0(context_count > 0);
2802 for (i = 0; i < context_count; i++) {
2803 int next_slice_idx = h->mb_width * h->mb_height;
2806 sl = &h->slice_ctx[i];
2807 if (CONFIG_ERROR_RESILIENCE) {
2808 sl->er.error_count = 0;
2811 /* make sure none of those slices overlap */
2812 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2813 for (j = 0; j < context_count; j++) {
2814 H264SliceContext *sl2 = &h->slice_ctx[j];
2815 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2817 if (i == j || slice_idx2 < slice_idx)
2819 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2821 sl->next_slice_idx = next_slice_idx;
2824 avctx->execute(avctx, decode_slice, h->slice_ctx,
2825 NULL, context_count, sizeof(h->slice_ctx[0]));
2827 /* pull back stuff from slices to master context */
2828 sl = &h->slice_ctx[context_count - 1];
2830 if (CONFIG_ERROR_RESILIENCE) {
2831 for (i = 1; i < context_count; i++)
2832 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
2835 if (h->postpone_filter) {
2836 h->postpone_filter = 0;
2838 for (i = 0; i < context_count; i++) {
2841 sl = &h->slice_ctx[i];
2842 y_end = FFMIN(sl->mb_y + 1, h->mb_height);
2843 x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
2845 for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
2847 loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
2848 j == y_end - 1 ? x_end : h->mb_width);
2855 h->nb_slice_ctx_queued = 0;