2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/avassert.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/timer.h"
33 #include "cabac_functions.h"
34 #include "error_resilience.h"
38 #include "h264chroma.h"
39 #include "h264_mvpred.h"
42 #include "mpegutils.h"
43 #include "rectangle.h"
46 static const uint8_t field_scan[16+1] = {
47 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
48 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
49 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
50 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
53 static const uint8_t field_scan8x8[64+1] = {
54 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
55 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
56 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
57 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
58 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
59 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
60 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
61 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
62 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
63 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
64 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
65 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
66 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
67 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
68 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
69 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
72 static const uint8_t field_scan8x8_cavlc[64+1] = {
73 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
74 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
75 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
76 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
77 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
78 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
79 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
80 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
81 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
82 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
83 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
84 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
85 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
86 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
87 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
88 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
91 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
92 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
93 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
94 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
95 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
96 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
97 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
98 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
99 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
100 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
101 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
102 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
103 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
104 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
105 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
106 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
107 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
108 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
111 static void release_unused_pictures(H264Context *h, int remove_current)
115 /* release non reference frames */
116 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
117 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
118 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
119 ff_h264_unref_picture(h, &h->DPB[i]);
124 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
126 const H264Context *h = sl->h264;
127 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
129 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
130 // edge emu needs blocksize + filter length - 1
131 // (= 21x21 for h264)
132 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
134 av_fast_mallocz(&sl->top_borders[0], &sl->top_borders_allocated[0],
135 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
136 av_fast_mallocz(&sl->top_borders[1], &sl->top_borders_allocated[1],
137 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
139 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
140 !sl->top_borders[0] || !sl->top_borders[1]) {
141 av_freep(&sl->bipred_scratchpad);
142 av_freep(&sl->edge_emu_buffer);
143 av_freep(&sl->top_borders[0]);
144 av_freep(&sl->top_borders[1]);
146 sl->bipred_scratchpad_allocated = 0;
147 sl->edge_emu_buffer_allocated = 0;
148 sl->top_borders_allocated[0] = 0;
149 sl->top_borders_allocated[1] = 0;
150 return AVERROR(ENOMEM);
156 static int init_table_pools(H264Context *h)
158 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
159 const int mb_array_size = h->mb_stride * h->mb_height;
160 const int b4_stride = h->mb_width * 4 + 1;
161 const int b4_array_size = b4_stride * h->mb_height * 4;
163 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
165 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
166 sizeof(uint32_t), av_buffer_allocz);
167 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
168 sizeof(int16_t), av_buffer_allocz);
169 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
171 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
172 !h->ref_index_pool) {
173 av_buffer_pool_uninit(&h->qscale_table_pool);
174 av_buffer_pool_uninit(&h->mb_type_pool);
175 av_buffer_pool_uninit(&h->motion_val_pool);
176 av_buffer_pool_uninit(&h->ref_index_pool);
177 return AVERROR(ENOMEM);
183 static int alloc_picture(H264Context *h, H264Picture *pic)
187 av_assert0(!pic->f->data[0]);
190 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
191 AV_GET_BUFFER_FLAG_REF : 0);
195 pic->crop = h->ps.sps->crop;
196 pic->crop_top = h->ps.sps->crop_top;
197 pic->crop_left= h->ps.sps->crop_left;
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 inline int pic_is_unused(H264Context *h, H264Picture *pic)
259 static int find_unused_picture(H264Context *h)
263 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
264 if (pic_is_unused(h, &h->DPB[i]))
267 if (i == H264_MAX_PICTURE_COUNT)
268 return AVERROR_INVALIDDATA;
274 #define IN_RANGE(a, b, size) (((void*)(a) >= (void*)(b)) && ((void*)(a) < (void*)((b) + (size))))
276 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
277 (((pic) && (pic) >= (old_ctx)->DPB && \
278 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
279 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
281 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
282 H264Context *new_base,
283 H264Context *old_base)
287 for (i = 0; i < count; i++) {
288 av_assert1(!from[i] ||
289 IN_RANGE(from[i], old_base, 1) ||
290 IN_RANGE(from[i], old_base->DPB, H264_MAX_PICTURE_COUNT));
291 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
295 #define copy_fields(to, from, start_field, end_field) \
296 memcpy(&(to)->start_field, &(from)->start_field, \
297 (char *)&(to)->end_field - (char *)&(to)->start_field)
299 static int h264_slice_header_init(H264Context *h);
301 int ff_h264_update_thread_context(AVCodecContext *dst,
302 const AVCodecContext *src)
304 H264Context *h = dst->priv_data, *h1 = src->priv_data;
305 int inited = h->context_initialized, err = 0;
312 // We can't fail if SPS isn't set at it breaks current skip_frame code
314 // return AVERROR_INVALIDDATA;
317 (h->width != h1->width ||
318 h->height != h1->height ||
319 h->mb_width != h1->mb_width ||
320 h->mb_height != h1->mb_height ||
322 h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
323 h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
324 h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
328 /* copy block_offset since frame_start may not be called */
329 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
332 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
333 av_buffer_unref(&h->ps.sps_list[i]);
334 if (h1->ps.sps_list[i]) {
335 h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
336 if (!h->ps.sps_list[i])
337 return AVERROR(ENOMEM);
340 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
341 av_buffer_unref(&h->ps.pps_list[i]);
342 if (h1->ps.pps_list[i]) {
343 h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
344 if (!h->ps.pps_list[i])
345 return AVERROR(ENOMEM);
349 av_buffer_unref(&h->ps.pps_ref);
350 av_buffer_unref(&h->ps.sps_ref);
353 if (h1->ps.pps_ref) {
354 h->ps.pps_ref = av_buffer_ref(h1->ps.pps_ref);
356 return AVERROR(ENOMEM);
357 h->ps.pps = h->ps.pps_ref->data;
359 if (h1->ps.sps_ref) {
360 h->ps.sps_ref = av_buffer_ref(h1->ps.sps_ref);
362 return AVERROR(ENOMEM);
363 h->ps.sps = h->ps.sps_ref->data;
366 if (need_reinit || !inited) {
367 h->width = h1->width;
368 h->height = h1->height;
369 h->mb_height = h1->mb_height;
370 h->mb_width = h1->mb_width;
371 h->mb_num = h1->mb_num;
372 h->mb_stride = h1->mb_stride;
373 h->b_stride = h1->b_stride;
375 if (h->context_initialized || h1->context_initialized) {
376 if ((err = h264_slice_header_init(h)) < 0) {
377 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
381 /* copy block_offset since frame_start may not be called */
382 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
385 h->avctx->coded_height = h1->avctx->coded_height;
386 h->avctx->coded_width = h1->avctx->coded_width;
387 h->avctx->width = h1->avctx->width;
388 h->avctx->height = h1->avctx->height;
389 h->coded_picture_number = h1->coded_picture_number;
390 h->first_field = h1->first_field;
391 h->picture_structure = h1->picture_structure;
392 h->droppable = h1->droppable;
393 h->low_delay = h1->low_delay;
394 h->backup_width = h1->backup_width;
395 h->backup_height = h1->backup_height;
396 h->backup_pix_fmt = h1->backup_pix_fmt;
398 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
399 ff_h264_unref_picture(h, &h->DPB[i]);
400 if (h1->DPB[i].f->buf[0] &&
401 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
405 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
406 ff_h264_unref_picture(h, &h->cur_pic);
407 if (h1->cur_pic.f->buf[0]) {
408 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
413 h->enable_er = h1->enable_er;
414 h->workaround_bugs = h1->workaround_bugs;
415 h->low_delay = h1->low_delay;
416 h->droppable = h1->droppable;
418 // extradata/NAL handling
419 h->is_avc = h1->is_avc;
420 h->nal_length_size = h1->nal_length_size;
421 h->x264_build = h1->x264_build;
424 copy_fields(h, h1, poc, current_slice);
426 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
427 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
428 copy_picture_range(h->delayed_pic, h1->delayed_pic,
429 MAX_DELAYED_PIC_COUNT + 2, h, h1);
431 h->frame_recovered = h1->frame_recovered;
437 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
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 1<<(h->ps.sps->bit_depth_luma-1),
456 1<<(h->ps.sps->bit_depth_chroma-1),
457 1<<(h->ps.sps->bit_depth_chroma-1),
461 if (!ff_thread_can_start_frame(h->avctx)) {
462 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
466 release_unused_pictures(h, 1);
467 h->cur_pic_ptr = NULL;
469 i = find_unused_picture(h);
471 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
476 pic->reference = h->droppable ? 0 : h->picture_structure;
477 pic->f->coded_picture_number = h->coded_picture_number++;
478 pic->field_picture = h->picture_structure != PICT_FRAME;
479 pic->frame_num = h->poc.frame_num;
481 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
483 * See decode_nal_units().
485 pic->f->key_frame = 0;
488 pic->invalid_gap = 0;
489 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
491 if ((ret = alloc_picture(h, pic)) < 0)
493 if(!h->frame_recovered && !h->avctx->hwaccel
495 && !(h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
498 ff_color_frame(pic->f, c);
500 h->cur_pic_ptr = pic;
501 ff_h264_unref_picture(h, &h->cur_pic);
502 if (CONFIG_ERROR_RESILIENCE) {
503 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
506 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
509 for (i = 0; i < h->nb_slice_ctx; i++) {
510 h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
511 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
514 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
515 ff_er_frame_start(&h->slice_ctx[0].er);
516 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
517 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
520 for (i = 0; i < 16; i++) {
521 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
522 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
524 for (i = 0; i < 16; i++) {
525 h->block_offset[16 + i] =
526 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
527 h->block_offset[48 + 16 + i] =
528 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
531 /* We mark the current picture as non-reference after allocating it, so
532 * that if we break out due to an error it can be released automatically
533 * in the next ff_mpv_frame_start().
535 h->cur_pic_ptr->reference = 0;
537 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
539 h->next_output_pic = NULL;
541 assert(h->cur_pic_ptr->long_ref == 0);
546 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
548 uint8_t *src_cb, uint8_t *src_cr,
549 int linesize, int uvlinesize,
554 const int pixel_shift = h->pixel_shift;
555 int chroma444 = CHROMA444(h);
556 int chroma422 = CHROMA422(h);
559 src_cb -= uvlinesize;
560 src_cr -= uvlinesize;
562 if (!simple && FRAME_MBAFF(h)) {
565 top_border = sl->top_borders[0][sl->mb_x];
566 AV_COPY128(top_border, src_y + 15 * linesize);
568 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
569 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
572 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
573 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
574 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
575 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
577 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
578 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
580 } else if (chroma422) {
582 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
583 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
585 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
586 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
590 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
591 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
593 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
594 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
599 } else if (MB_MBAFF(sl)) {
605 top_border = sl->top_borders[top_idx][sl->mb_x];
606 /* There are two lines saved, the line above the top macroblock
607 * of a pair, and the line above the bottom macroblock. */
608 AV_COPY128(top_border, src_y + 16 * linesize);
610 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
612 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
615 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
616 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
617 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
618 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
620 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
621 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
623 } else if (chroma422) {
625 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
626 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
628 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
629 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
633 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
634 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
636 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
637 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
644 * Initialize implicit_weight table.
645 * @param field 0/1 initialize the weight for interlaced MBAFF
646 * -1 initializes the rest
648 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
650 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
652 for (i = 0; i < 2; i++) {
653 sl->pwt.luma_weight_flag[i] = 0;
654 sl->pwt.chroma_weight_flag[i] = 0;
658 if (h->picture_structure == PICT_FRAME) {
659 cur_poc = h->cur_pic_ptr->poc;
661 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
663 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
664 sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2 * cur_poc) {
665 sl->pwt.use_weight = 0;
666 sl->pwt.use_weight_chroma = 0;
670 ref_count0 = sl->ref_count[0];
671 ref_count1 = sl->ref_count[1];
673 cur_poc = h->cur_pic_ptr->field_poc[field];
675 ref_count0 = 16 + 2 * sl->ref_count[0];
676 ref_count1 = 16 + 2 * sl->ref_count[1];
679 sl->pwt.use_weight = 2;
680 sl->pwt.use_weight_chroma = 2;
681 sl->pwt.luma_log2_weight_denom = 5;
682 sl->pwt.chroma_log2_weight_denom = 5;
684 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
685 int64_t poc0 = sl->ref_list[0][ref0].poc;
686 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
688 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
689 int poc1 = sl->ref_list[1][ref1].poc;
690 int td = av_clip_int8(poc1 - poc0);
692 int tb = av_clip_int8(cur_poc - poc0);
693 int tx = (16384 + (FFABS(td) >> 1)) / td;
694 int dist_scale_factor = (tb * tx + 32) >> 8;
695 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
696 w = 64 - dist_scale_factor;
700 sl->pwt.implicit_weight[ref0][ref1][0] =
701 sl->pwt.implicit_weight[ref0][ref1][1] = w;
703 sl->pwt.implicit_weight[ref0][ref1][field] = w;
710 * initialize scan tables
712 static void init_scan_tables(H264Context *h)
715 for (i = 0; i < 16; i++) {
716 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
717 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
718 h->field_scan[i] = TRANSPOSE(field_scan[i]);
721 for (i = 0; i < 64; i++) {
722 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
723 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
724 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
725 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
726 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
729 if (h->ps.sps->transform_bypass) { // FIXME same ugly
730 memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
731 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
732 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
733 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
734 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
735 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
737 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
738 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
739 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
740 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
741 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
742 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
746 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
748 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
749 CONFIG_H264_D3D11VA_HWACCEL + \
750 CONFIG_H264_VAAPI_HWACCEL + \
751 (CONFIG_H264_VDA_HWACCEL * 2) + \
752 CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
753 CONFIG_H264_VDPAU_HWACCEL)
754 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
755 const enum AVPixelFormat *choices = pix_fmts;
758 switch (h->ps.sps->bit_depth_luma) {
761 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
762 *fmt++ = AV_PIX_FMT_GBRP9;
764 *fmt++ = AV_PIX_FMT_YUV444P9;
765 } else if (CHROMA422(h))
766 *fmt++ = AV_PIX_FMT_YUV422P9;
768 *fmt++ = AV_PIX_FMT_YUV420P9;
772 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
773 *fmt++ = AV_PIX_FMT_GBRP10;
775 *fmt++ = AV_PIX_FMT_YUV444P10;
776 } else if (CHROMA422(h))
777 *fmt++ = AV_PIX_FMT_YUV422P10;
779 *fmt++ = AV_PIX_FMT_YUV420P10;
783 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
784 *fmt++ = AV_PIX_FMT_GBRP12;
786 *fmt++ = AV_PIX_FMT_YUV444P12;
787 } else if (CHROMA422(h))
788 *fmt++ = AV_PIX_FMT_YUV422P12;
790 *fmt++ = AV_PIX_FMT_YUV420P12;
794 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
795 *fmt++ = AV_PIX_FMT_GBRP14;
797 *fmt++ = AV_PIX_FMT_YUV444P14;
798 } else if (CHROMA422(h))
799 *fmt++ = AV_PIX_FMT_YUV422P14;
801 *fmt++ = AV_PIX_FMT_YUV420P14;
804 #if CONFIG_H264_VDPAU_HWACCEL
805 *fmt++ = AV_PIX_FMT_VDPAU;
808 if (h->avctx->colorspace == AVCOL_SPC_RGB)
809 *fmt++ = AV_PIX_FMT_GBRP;
810 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
811 *fmt++ = AV_PIX_FMT_YUVJ444P;
813 *fmt++ = AV_PIX_FMT_YUV444P;
814 } else if (CHROMA422(h)) {
815 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
816 *fmt++ = AV_PIX_FMT_YUVJ422P;
818 *fmt++ = AV_PIX_FMT_YUV422P;
820 #if CONFIG_H264_DXVA2_HWACCEL
821 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
823 #if CONFIG_H264_D3D11VA_HWACCEL
824 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
826 #if CONFIG_H264_VAAPI_HWACCEL
827 *fmt++ = AV_PIX_FMT_VAAPI;
829 #if CONFIG_H264_VDA_HWACCEL
830 *fmt++ = AV_PIX_FMT_VDA_VLD;
831 *fmt++ = AV_PIX_FMT_VDA;
833 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
834 *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
836 if (h->avctx->codec->pix_fmts)
837 choices = h->avctx->codec->pix_fmts;
838 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
839 *fmt++ = AV_PIX_FMT_YUVJ420P;
841 *fmt++ = AV_PIX_FMT_YUV420P;
845 av_log(h->avctx, AV_LOG_ERROR,
846 "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
847 return AVERROR_INVALIDDATA;
850 *fmt = AV_PIX_FMT_NONE;
852 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
853 if (choices[i] == h->avctx->pix_fmt && !force_callback)
855 return ff_thread_get_format(h->avctx, choices);
858 /* export coded and cropped frame dimensions to AVCodecContext */
859 static int init_dimensions(H264Context *h)
861 SPS *sps = h->ps.sps;
862 int width = h->width - (sps->crop_right + sps->crop_left);
863 int height = h->height - (sps->crop_top + sps->crop_bottom);
864 av_assert0(sps->crop_right + sps->crop_left < (unsigned)h->width);
865 av_assert0(sps->crop_top + sps->crop_bottom < (unsigned)h->height);
867 /* handle container cropping */
868 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
869 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16) &&
870 h->avctx->width <= width &&
871 h->avctx->height <= height
873 width = h->avctx->width;
874 height = h->avctx->height;
877 if (width <= 0 || height <= 0) {
878 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
880 if (h->avctx->err_recognition & AV_EF_EXPLODE)
881 return AVERROR_INVALIDDATA;
883 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
894 h->avctx->coded_width = h->width;
895 h->avctx->coded_height = h->height;
896 h->avctx->width = width;
897 h->avctx->height = height;
902 static int h264_slice_header_init(H264Context *h)
904 const SPS *sps = h->ps.sps;
905 int nb_slices = (HAVE_THREADS &&
906 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
907 h->avctx->thread_count : 1;
910 ff_set_sar(h->avctx, sps->sar);
911 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
912 &h->chroma_x_shift, &h->chroma_y_shift);
914 if (sps->timing_info_present_flag) {
915 int64_t den = sps->time_scale;
916 if (h->x264_build < 44U)
918 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
919 sps->num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
922 ff_h264_free_tables(h);
925 h->prev_interlaced_frame = 1;
928 ret = ff_h264_alloc_tables(h);
930 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
935 if (h->avctx->codec &&
936 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU &&
937 (sps->bit_depth_luma != 8 || sps->chroma_format_idc > 1)) {
938 av_log(h->avctx, AV_LOG_ERROR,
939 "VDPAU decoding does not support video colorspace.\n");
940 ret = AVERROR_INVALIDDATA;
945 if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 14 ||
946 sps->bit_depth_luma == 11 || sps->bit_depth_luma == 13
948 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
949 sps->bit_depth_luma);
950 ret = AVERROR_INVALIDDATA;
954 h->cur_bit_depth_luma =
955 h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
956 h->cur_chroma_format_idc = sps->chroma_format_idc;
957 h->pixel_shift = sps->bit_depth_luma > 8;
958 h->chroma_format_idc = sps->chroma_format_idc;
959 h->bit_depth_luma = sps->bit_depth_luma;
961 ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
962 sps->chroma_format_idc);
963 ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
964 ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
965 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
966 sps->chroma_format_idc);
967 ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
969 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
972 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
974 max_slices = H264_MAX_THREADS;
975 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
976 " reducing to %d\n", nb_slices, max_slices);
977 nb_slices = max_slices;
979 h->slice_context_count = nb_slices;
980 h->max_contexts = FFMIN(h->max_contexts, nb_slices);
982 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
983 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
985 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
989 for (i = 0; i < h->slice_context_count; i++) {
990 H264SliceContext *sl = &h->slice_ctx[i];
993 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
994 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
995 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
997 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
998 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1004 h->context_initialized = 1;
1008 ff_h264_free_tables(h);
1009 h->context_initialized = 0;
1013 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1016 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1017 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1018 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1025 * Decode a slice header.
1026 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1028 * @param h h264context
1030 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1032 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1036 unsigned int first_mb_in_slice;
1037 unsigned int pps_id;
1039 unsigned int slice_type, tmp, i, j;
1040 int last_pic_structure, last_pic_droppable;
1042 int needs_reinit = 0;
1043 int field_pic_flag, bottom_field_flag;
1044 int first_slice = sl == h->slice_ctx && !h->current_slice;
1045 int frame_num, droppable, picture_structure;
1046 int mb_aff_frame, last_mb_aff_frame;
1049 av_assert0(!h->setup_finished);
1051 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1052 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1054 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1056 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1057 if (h->current_slice) {
1058 if (h->setup_finished) {
1059 av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
1060 return AVERROR_INVALIDDATA;
1062 if (h->max_contexts > 1) {
1063 if (!h->single_decode_warning) {
1064 av_log(h->avctx, AV_LOG_WARNING, "Cannot decode multiple access units as slice threads\n");
1065 h->single_decode_warning = 1;
1067 h->max_contexts = 1;
1068 return SLICE_SINGLETHREAD;
1071 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
1072 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1073 h->current_slice = 0;
1076 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == NAL_IDR_SLICE) {
1077 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
1078 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1079 h->current_slice = 0;
1080 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1081 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1082 h->cur_pic_ptr = NULL;
1086 return AVERROR_INVALIDDATA;
1089 if (!h->first_field) {
1090 if (h->cur_pic_ptr && !h->droppable) {
1091 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1092 h->picture_structure == PICT_BOTTOM_FIELD);
1094 h->cur_pic_ptr = NULL;
1098 if (!h->current_slice)
1099 av_assert0(sl == h->slice_ctx);
1101 slice_type = get_ue_golomb_31(&sl->gb);
1102 if (slice_type > 9) {
1103 av_log(h->avctx, AV_LOG_ERROR,
1104 "slice type %d too large at %d\n",
1105 slice_type, first_mb_in_slice);
1106 return AVERROR_INVALIDDATA;
1108 if (slice_type > 4) {
1110 sl->slice_type_fixed = 1;
1112 sl->slice_type_fixed = 0;
1114 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1115 sl->slice_type = slice_type;
1116 sl->slice_type_nos = slice_type & 3;
1118 if (h->nal_unit_type == NAL_IDR_SLICE &&
1119 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1120 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1121 return AVERROR_INVALIDDATA;
1124 if (h->current_slice == 0 && !h->first_field) {
1126 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1127 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1128 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1129 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE && h->sei_recovery_frame_cnt < 0) ||
1130 h->avctx->skip_frame >= AVDISCARD_ALL) {
1131 return SLICE_SKIPED;
1135 // to make a few old functions happy, it's wrong though
1136 if (!h->setup_finished)
1137 h->pict_type = sl->slice_type;
1139 pps_id = get_ue_golomb(&sl->gb);
1140 if (pps_id >= MAX_PPS_COUNT) {
1141 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1142 return AVERROR_INVALIDDATA;
1144 if (!h->ps.pps_list[pps_id]) {
1145 av_log(h->avctx, AV_LOG_ERROR,
1146 "non-existing PPS %u referenced\n",
1148 return AVERROR_INVALIDDATA;
1150 if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
1151 av_log(h->avctx, AV_LOG_ERROR,
1152 "PPS change from %d to %d forbidden\n",
1153 h->au_pps_id, pps_id);
1154 return AVERROR_INVALIDDATA;
1157 pps = (const PPS*)h->ps.pps_list[pps_id]->data;
1159 if (!h->ps.sps_list[pps->sps_id]) {
1160 av_log(h->avctx, AV_LOG_ERROR,
1161 "non-existing SPS %u referenced\n",
1163 return AVERROR_INVALIDDATA;
1167 av_buffer_unref(&h->ps.pps_ref);
1169 h->ps.pps_ref = av_buffer_ref(h->ps.pps_list[pps_id]);
1171 return AVERROR(ENOMEM);
1172 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
1174 if (h->ps.pps->sps_id != pps->sps_id ||
1175 h->ps.pps->transform_8x8_mode != pps->transform_8x8_mode /*||
1176 (h->setup_finished && h->ps.pps != pps)*/) {
1177 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1178 return AVERROR_INVALIDDATA;
1182 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data ||
1183 pps->sps_id != h->current_sps_id) {
1186 av_log(h->avctx, AV_LOG_ERROR,
1187 "SPS changed in the middle of the frame\n");
1188 return AVERROR_INVALIDDATA;
1191 av_buffer_unref(&h->ps.sps_ref);
1193 h->ps.sps_ref = av_buffer_ref(h->ps.sps_list[h->ps.pps->sps_id]);
1195 return AVERROR(ENOMEM);
1196 h->ps.sps = (const SPS*)h->ps.sps_ref->data;
1198 if (h->mb_width != h->ps.sps->mb_width ||
1199 h->mb_height != h->ps.sps->mb_height * (2 - h->ps.sps->frame_mbs_only_flag) ||
1200 h->cur_bit_depth_luma != h->ps.sps->bit_depth_luma ||
1201 h->cur_chroma_format_idc != h->ps.sps->chroma_format_idc
1205 if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
1206 h->chroma_format_idc != h->ps.sps->chroma_format_idc)
1209 if (h->flags & AV_CODEC_FLAG_LOW_DELAY ||
1210 (h->ps.sps->bitstream_restriction_flag &&
1211 !h->ps.sps->num_reorder_frames)) {
1212 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1213 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1214 "Reenabling low delay requires a codec flush.\n");
1219 if (h->avctx->has_b_frames < 2)
1220 h->avctx->has_b_frames = !h->low_delay;
1227 must_reinit = (h->context_initialized &&
1228 ( 16*sps->mb_width != h->avctx->coded_width
1229 || 16*sps->mb_height * (2 - sps->frame_mbs_only_flag) != h->avctx->coded_height
1230 || h->cur_bit_depth_luma != sps->bit_depth_luma
1231 || h->cur_chroma_format_idc != sps->chroma_format_idc
1232 || h->mb_width != sps->mb_width
1233 || h->mb_height != sps->mb_height * (2 - sps->frame_mbs_only_flag)
1235 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1236 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1239 if (first_slice && av_cmp_q(sps->sar, h->avctx->sample_aspect_ratio))
1242 if (!h->setup_finished) {
1243 h->avctx->profile = ff_h264_get_profile(sps);
1244 h->avctx->level = sps->level_idc;
1245 h->avctx->refs = sps->ref_frame_count;
1247 h->mb_width = sps->mb_width;
1248 h->mb_height = sps->mb_height * (2 - sps->frame_mbs_only_flag);
1249 h->mb_num = h->mb_width * h->mb_height;
1250 h->mb_stride = h->mb_width + 1;
1252 h->b_stride = h->mb_width * 4;
1254 h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
1256 h->width = 16 * h->mb_width;
1257 h->height = 16 * h->mb_height;
1259 ret = init_dimensions(h);
1263 if (sps->video_signal_type_present_flag) {
1264 h->avctx->color_range = sps->full_range > 0 ? AVCOL_RANGE_JPEG
1266 if (sps->colour_description_present_flag) {
1267 if (h->avctx->colorspace != sps->colorspace)
1269 h->avctx->color_primaries = sps->color_primaries;
1270 h->avctx->color_trc = sps->color_trc;
1271 h->avctx->colorspace = sps->colorspace;
1276 if (h->context_initialized &&
1277 (must_reinit || needs_reinit)) {
1278 h->context_initialized = 0;
1279 if (sl != h->slice_ctx) {
1280 av_log(h->avctx, AV_LOG_ERROR,
1281 "changing width %d -> %d / height %d -> %d on "
1283 h->width, h->avctx->coded_width,
1284 h->height, h->avctx->coded_height,
1285 h->current_slice + 1);
1286 return AVERROR_INVALIDDATA;
1289 av_assert1(first_slice);
1291 ff_h264_flush_change(h);
1293 if ((ret = get_pixel_format(h, 1)) < 0)
1295 h->avctx->pix_fmt = ret;
1297 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1298 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1300 if ((ret = h264_slice_header_init(h)) < 0) {
1301 av_log(h->avctx, AV_LOG_ERROR,
1302 "h264_slice_header_init() failed\n");
1306 if (!h->context_initialized) {
1307 if (sl != h->slice_ctx) {
1308 av_log(h->avctx, AV_LOG_ERROR,
1309 "Cannot (re-)initialize context during parallel decoding.\n");
1310 return AVERROR_PATCHWELCOME;
1313 if ((ret = get_pixel_format(h, 1)) < 0)
1315 h->avctx->pix_fmt = ret;
1317 if ((ret = h264_slice_header_init(h)) < 0) {
1318 av_log(h->avctx, AV_LOG_ERROR,
1319 "h264_slice_header_init() failed\n");
1324 frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
1326 if (h->poc.frame_num != frame_num) {
1327 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1328 h->poc.frame_num, frame_num);
1329 return AVERROR_INVALIDDATA;
1333 if (!h->setup_finished)
1334 h->poc.frame_num = frame_num;
1338 last_mb_aff_frame = h->mb_aff_frame;
1339 last_pic_structure = h->picture_structure;
1340 last_pic_droppable = h->droppable;
1342 droppable = h->nal_ref_idc == 0;
1343 if (sps->frame_mbs_only_flag) {
1344 picture_structure = PICT_FRAME;
1346 if (!h->ps.sps->direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1347 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1350 field_pic_flag = get_bits1(&sl->gb);
1352 if (field_pic_flag) {
1353 bottom_field_flag = get_bits1(&sl->gb);
1354 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1356 picture_structure = PICT_FRAME;
1357 mb_aff_frame = sps->mb_aff;
1361 if (h->current_slice) {
1362 if (last_pic_structure != picture_structure ||
1363 last_pic_droppable != droppable ||
1364 last_mb_aff_frame != mb_aff_frame) {
1365 av_log(h->avctx, AV_LOG_ERROR,
1366 "Changing field mode (%d -> %d) between slices is not allowed\n",
1367 last_pic_structure, h->picture_structure);
1368 return AVERROR_INVALIDDATA;
1369 } else if (!h->cur_pic_ptr) {
1370 av_log(h->avctx, AV_LOG_ERROR,
1371 "unset cur_pic_ptr on slice %d\n",
1372 h->current_slice + 1);
1373 return AVERROR_INVALIDDATA;
1377 if (!h->setup_finished) {
1378 h->droppable = droppable;
1379 h->picture_structure = picture_structure;
1380 h->mb_aff_frame = mb_aff_frame;
1382 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1384 if (h->current_slice == 0) {
1385 /* Shorten frame num gaps so we don't have to allocate reference
1386 * frames just to throw them away */
1387 if (h->poc.frame_num != h->poc.prev_frame_num) {
1388 int unwrap_prev_frame_num = h->poc.prev_frame_num;
1389 int max_frame_num = 1 << sps->log2_max_frame_num;
1391 if (unwrap_prev_frame_num > h->poc.frame_num)
1392 unwrap_prev_frame_num -= max_frame_num;
1394 if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
1395 unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
1396 if (unwrap_prev_frame_num < 0)
1397 unwrap_prev_frame_num += max_frame_num;
1399 h->poc.prev_frame_num = unwrap_prev_frame_num;
1403 /* See if we have a decoded first field looking for a pair...
1404 * Here, we're using that to see if we should mark previously
1405 * decode frames as "finished".
1406 * We have to do that before the "dummy" in-between frame allocation,
1407 * since that can modify h->cur_pic_ptr. */
1408 if (h->first_field) {
1409 av_assert0(h->cur_pic_ptr);
1410 av_assert0(h->cur_pic_ptr->f->buf[0]);
1411 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1413 /* Mark old field/frame as completed */
1414 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1415 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1416 last_pic_structure == PICT_BOTTOM_FIELD);
1419 /* figure out if we have a complementary field pair */
1420 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1421 /* Previous field is unmatched. Don't display it, but let it
1422 * remain for reference if marked as such. */
1423 if (last_pic_structure != PICT_FRAME) {
1424 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1425 last_pic_structure == PICT_TOP_FIELD);
1428 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1429 /* This and previous field were reference, but had
1430 * different frame_nums. Consider this field first in
1431 * pair. Throw away previous field except for reference
1433 if (last_pic_structure != PICT_FRAME) {
1434 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1435 last_pic_structure == PICT_TOP_FIELD);
1438 /* Second field in complementary pair */
1439 if (!((last_pic_structure == PICT_TOP_FIELD &&
1440 h->picture_structure == PICT_BOTTOM_FIELD) ||
1441 (last_pic_structure == PICT_BOTTOM_FIELD &&
1442 h->picture_structure == PICT_TOP_FIELD))) {
1443 av_log(h->avctx, AV_LOG_ERROR,
1444 "Invalid field mode combination %d/%d\n",
1445 last_pic_structure, h->picture_structure);
1446 h->picture_structure = last_pic_structure;
1447 h->droppable = last_pic_droppable;
1448 return AVERROR_INVALIDDATA;
1449 } else if (last_pic_droppable != h->droppable) {
1450 avpriv_request_sample(h->avctx,
1451 "Found reference and non-reference fields in the same frame, which");
1452 h->picture_structure = last_pic_structure;
1453 h->droppable = last_pic_droppable;
1454 return AVERROR_PATCHWELCOME;
1460 while (h->poc.frame_num != h->poc.prev_frame_num && !h->first_field &&
1461 h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
1462 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1463 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1464 h->poc.frame_num, h->poc.prev_frame_num);
1465 if (!sps->gaps_in_frame_num_allowed_flag)
1466 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1467 h->last_pocs[i] = INT_MIN;
1468 ret = h264_frame_start(h);
1474 h->poc.prev_frame_num++;
1475 h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
1476 h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
1477 h->cur_pic_ptr->invalid_gap = !sps->gaps_in_frame_num_allowed_flag;
1478 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1479 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1480 ret = ff_generate_sliding_window_mmcos(h, 1);
1481 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1483 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1484 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1486 /* Error concealment: If a ref is missing, copy the previous ref
1488 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1489 * many assumptions about there being no actual duplicates.
1490 * FIXME: This does not copy padding for out-of-frame motion
1491 * vectors. Given we are concealing a lost frame, this probably
1492 * is not noticeable by comparison, but it should be fixed. */
1493 if (h->short_ref_count) {
1495 h->short_ref[0]->f->width == prev->f->width &&
1496 h->short_ref[0]->f->height == prev->f->height &&
1497 h->short_ref[0]->f->format == prev->f->format) {
1498 av_image_copy(h->short_ref[0]->f->data,
1499 h->short_ref[0]->f->linesize,
1500 (const uint8_t **)prev->f->data,
1505 h->short_ref[0]->poc = prev->poc + 2;
1507 h->short_ref[0]->frame_num = h->poc.prev_frame_num;
1511 /* See if we have a decoded first field looking for a pair...
1512 * We're using that to see whether to continue decoding in that
1513 * frame, or to allocate a new one. */
1514 if (h->first_field) {
1515 av_assert0(h->cur_pic_ptr);
1516 av_assert0(h->cur_pic_ptr->f->buf[0]);
1517 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1519 /* figure out if we have a complementary field pair */
1520 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1521 /* Previous field is unmatched. Don't display it, but let it
1522 * remain for reference if marked as such. */
1523 h->missing_fields ++;
1524 h->cur_pic_ptr = NULL;
1525 h->first_field = FIELD_PICTURE(h);
1527 h->missing_fields = 0;
1528 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1529 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1530 h->picture_structure==PICT_BOTTOM_FIELD);
1531 /* This and the previous field had different frame_nums.
1532 * Consider this field first in pair. Throw away previous
1533 * one except for reference purposes. */
1535 h->cur_pic_ptr = NULL;
1537 /* Second field in complementary pair */
1542 /* Frame or first field in a potentially complementary pair */
1543 h->first_field = FIELD_PICTURE(h);
1546 if (!FIELD_PICTURE(h) || h->first_field) {
1547 if (h264_frame_start(h) < 0) {
1549 return AVERROR_INVALIDDATA;
1552 release_unused_pictures(h, 0);
1554 /* Some macroblocks can be accessed before they're available in case
1555 * of lost slices, MBAFF or threading. */
1556 if (FIELD_PICTURE(h)) {
1557 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1558 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1560 memset(h->slice_table, -1,
1561 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1565 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1566 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1567 first_mb_in_slice >= h->mb_num) {
1568 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1569 return AVERROR_INVALIDDATA;
1571 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1572 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1573 FIELD_OR_MBAFF_PICTURE(h);
1574 if (h->picture_structure == PICT_BOTTOM_FIELD)
1575 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1576 av_assert1(sl->mb_y < h->mb_height);
1578 if (h->picture_structure == PICT_FRAME) {
1579 h->curr_pic_num = h->poc.frame_num;
1580 h->max_pic_num = 1 << sps->log2_max_frame_num;
1582 h->curr_pic_num = 2 * h->poc.frame_num + 1;
1583 h->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
1586 if (h->nal_unit_type == NAL_IDR_SLICE)
1587 get_ue_golomb_long(&sl->gb); /* idr_pic_id */
1589 if (sps->poc_type == 0) {
1590 int poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
1592 if (!h->setup_finished)
1593 h->poc.poc_lsb = poc_lsb;
1595 if (pps->pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1596 int delta_poc_bottom = get_se_golomb(&sl->gb);
1597 if (!h->setup_finished)
1598 h->poc.delta_poc_bottom = delta_poc_bottom;
1602 if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
1603 int delta_poc = get_se_golomb(&sl->gb);
1605 if (!h->setup_finished)
1606 h->poc.delta_poc[0] = delta_poc;
1608 if (pps->pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1609 delta_poc = get_se_golomb(&sl->gb);
1611 if (!h->setup_finished)
1612 h->poc.delta_poc[1] = delta_poc;
1616 if (!h->setup_finished)
1617 ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
1618 sps, &h->poc, h->picture_structure, h->nal_ref_idc);
1620 if (pps->redundant_pic_cnt_present)
1621 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1623 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1624 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1626 ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
1627 &sl->gb, pps, sl->slice_type_nos,
1628 h->picture_structure, h->avctx);
1632 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1633 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1635 sl->ref_count[1] = sl->ref_count[0] = 0;
1640 if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1641 (pps->weighted_bipred_idc == 1 &&
1642 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1643 ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
1644 sl->slice_type_nos, &sl->pwt);
1645 else if (pps->weighted_bipred_idc == 2 &&
1646 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1647 implicit_weight_table(h, sl, -1);
1649 sl->pwt.use_weight = 0;
1650 for (i = 0; i < 2; i++) {
1651 sl->pwt.luma_weight_flag[i] = 0;
1652 sl->pwt.chroma_weight_flag[i] = 0;
1656 // If frame-mt is enabled, only update mmco tables for the first slice
1657 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1658 // or h->mmco, which will cause ref list mix-ups and decoding errors
1659 // further down the line. This may break decoding if the first slice is
1660 // corrupt, thus we only do this if frame-mt is enabled.
1661 if (h->nal_ref_idc) {
1662 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1663 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1664 h->current_slice == 0);
1665 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1666 return AVERROR_INVALIDDATA;
1669 if (FRAME_MBAFF(h)) {
1670 ff_h264_fill_mbaff_ref_list(h, sl);
1672 if (pps->weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1673 implicit_weight_table(h, sl, 0);
1674 implicit_weight_table(h, sl, 1);
1678 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1679 ff_h264_direct_dist_scale_factor(h, sl);
1680 ff_h264_direct_ref_list_init(h, sl);
1682 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
1683 tmp = get_ue_golomb_31(&sl->gb);
1685 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1686 return AVERROR_INVALIDDATA;
1688 sl->cabac_init_idc = tmp;
1691 sl->last_qscale_diff = 0;
1692 tmp = pps->init_qp + get_se_golomb(&sl->gb);
1693 if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
1694 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1695 return AVERROR_INVALIDDATA;
1698 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1699 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1700 // FIXME qscale / qp ... stuff
1701 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1702 get_bits1(&sl->gb); /* sp_for_switch_flag */
1703 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1704 sl->slice_type == AV_PICTURE_TYPE_SI)
1705 get_se_golomb(&sl->gb); /* slice_qs_delta */
1707 sl->deblocking_filter = 1;
1708 sl->slice_alpha_c0_offset = 0;
1709 sl->slice_beta_offset = 0;
1710 if (pps->deblocking_filter_parameters_present) {
1711 tmp = get_ue_golomb_31(&sl->gb);
1713 av_log(h->avctx, AV_LOG_ERROR,
1714 "deblocking_filter_idc %u out of range\n", tmp);
1715 return AVERROR_INVALIDDATA;
1717 sl->deblocking_filter = tmp;
1718 if (sl->deblocking_filter < 2)
1719 sl->deblocking_filter ^= 1; // 1<->0
1721 if (sl->deblocking_filter) {
1722 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1723 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1724 if (sl->slice_alpha_c0_offset > 12 ||
1725 sl->slice_alpha_c0_offset < -12 ||
1726 sl->slice_beta_offset > 12 ||
1727 sl->slice_beta_offset < -12) {
1728 av_log(h->avctx, AV_LOG_ERROR,
1729 "deblocking filter parameters %d %d out of range\n",
1730 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1731 return AVERROR_INVALIDDATA;
1736 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1737 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1738 h->nal_unit_type != NAL_IDR_SLICE) ||
1739 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1740 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1741 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1742 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1743 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1744 h->nal_ref_idc == 0))
1745 sl->deblocking_filter = 0;
1747 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1748 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1749 /* Cheat slightly for speed:
1750 * Do not bother to deblock across slices. */
1751 sl->deblocking_filter = 2;
1753 h->max_contexts = 1;
1754 if (!h->single_decode_warning) {
1755 av_log(h->avctx, AV_LOG_INFO,
1756 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1757 "To parallelize slice decoding you need video encoded with disable_deblocking_filter_idc set to 2 (deblock only edges that do not cross slices).\n"
1758 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1759 "but will generate non-standard-compliant output.\n");
1760 h->single_decode_warning = 1;
1762 if (sl != h->slice_ctx) {
1763 av_log(h->avctx, AV_LOG_ERROR,
1764 "Deblocking switched inside frame.\n");
1765 return SLICE_SINGLETHREAD;
1769 sl->qp_thresh = 15 -
1770 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1772 pps->chroma_qp_index_offset[0],
1773 pps->chroma_qp_index_offset[1]) +
1774 6 * (sps->bit_depth_luma - 8);
1776 sl->slice_num = ++h->current_slice;
1779 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1780 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1781 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1782 && sl->slice_num >= MAX_SLICES) {
1783 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1784 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);
1787 for (j = 0; j < 2; j++) {
1789 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1790 for (i = 0; i < 16; i++) {
1792 if (j < sl->list_count && i < sl->ref_count[j] &&
1793 sl->ref_list[j][i].parent->f->buf[0]) {
1795 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1796 for (k = 0; k < h->short_ref_count; k++)
1797 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1801 for (k = 0; k < h->long_ref_count; k++)
1802 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1803 id_list[i] = h->short_ref_count + k;
1811 for (i = 0; i < 16; i++)
1812 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1814 ref2frm[18 + 1] = -1;
1815 for (i = 16; i < 48; i++)
1816 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1817 (sl->ref_list[j][i].reference & 3);
1820 h->au_pps_id = pps_id;
1821 h->current_sps_id = h->ps.pps->sps_id;
1823 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1824 av_log(h->avctx, AV_LOG_DEBUG,
1825 "slice:%d %s mb:%d %c%s%s pps:%u frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
1827 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1829 av_get_picture_type_char(sl->slice_type),
1830 sl->slice_type_fixed ? " fix" : "",
1831 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1832 pps_id, h->poc.frame_num,
1833 h->cur_pic_ptr->field_poc[0],
1834 h->cur_pic_ptr->field_poc[1],
1835 sl->ref_count[0], sl->ref_count[1],
1837 sl->deblocking_filter,
1838 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1840 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1841 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1847 int ff_h264_get_slice_type(const H264SliceContext *sl)
1849 switch (sl->slice_type) {
1850 case AV_PICTURE_TYPE_P:
1852 case AV_PICTURE_TYPE_B:
1854 case AV_PICTURE_TYPE_I:
1856 case AV_PICTURE_TYPE_SP:
1858 case AV_PICTURE_TYPE_SI:
1861 return AVERROR_INVALIDDATA;
1865 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1866 H264SliceContext *sl,
1867 int mb_type, int top_xy,
1868 int left_xy[LEFT_MBS],
1870 int left_type[LEFT_MBS],
1871 int mb_xy, int list)
1873 int b_stride = h->b_stride;
1874 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1875 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1876 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1877 if (USES_LIST(top_type, list)) {
1878 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1879 const int b8_xy = 4 * top_xy + 2;
1880 int *ref2frm = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1881 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1882 ref_cache[0 - 1 * 8] =
1883 ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
1884 ref_cache[2 - 1 * 8] =
1885 ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
1887 AV_ZERO128(mv_dst - 1 * 8);
1888 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1891 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1892 if (USES_LIST(left_type[LTOP], list)) {
1893 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1894 const int b8_xy = 4 * left_xy[LTOP] + 1;
1895 int *ref2frm = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1896 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1897 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1898 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1899 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1901 ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1902 ref_cache[-1 + 16] =
1903 ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1905 AV_ZERO32(mv_dst - 1 + 0);
1906 AV_ZERO32(mv_dst - 1 + 8);
1907 AV_ZERO32(mv_dst - 1 + 16);
1908 AV_ZERO32(mv_dst - 1 + 24);
1911 ref_cache[-1 + 16] =
1912 ref_cache[-1 + 24] = LIST_NOT_USED;
1917 if (!USES_LIST(mb_type, list)) {
1918 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1919 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1920 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1921 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1922 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1927 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1928 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1929 uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
1930 uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
1931 AV_WN32A(&ref_cache[0 * 8], ref01);
1932 AV_WN32A(&ref_cache[1 * 8], ref01);
1933 AV_WN32A(&ref_cache[2 * 8], ref23);
1934 AV_WN32A(&ref_cache[3 * 8], ref23);
1938 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1939 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1940 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1941 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1942 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1947 * @return non zero if the loop filter can be skipped
1949 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
1951 const int mb_xy = sl->mb_xy;
1952 int top_xy, left_xy[LEFT_MBS];
1953 int top_type, left_type[LEFT_MBS];
1957 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1959 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1960 * stuff, I can't imagine that these complex rules are worth it. */
1962 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1963 if (FRAME_MBAFF(h)) {
1964 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1965 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1967 if (left_mb_field_flag != curr_mb_field_flag)
1968 left_xy[LTOP] -= h->mb_stride;
1970 if (curr_mb_field_flag)
1971 top_xy += h->mb_stride &
1972 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1973 if (left_mb_field_flag != curr_mb_field_flag)
1974 left_xy[LBOT] += h->mb_stride;
1978 sl->top_mb_xy = top_xy;
1979 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1980 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1982 /* For sufficiently low qp, filtering wouldn't do anything.
1983 * This is a conservative estimate: could also check beta_offset
1984 * and more accurate chroma_qp. */
1985 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1986 int qp = h->cur_pic.qscale_table[mb_xy];
1987 if (qp <= qp_thresh &&
1988 (left_xy[LTOP] < 0 ||
1989 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1991 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1992 if (!FRAME_MBAFF(h))
1994 if ((left_xy[LTOP] < 0 ||
1995 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1996 (top_xy < h->mb_stride ||
1997 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2002 top_type = h->cur_pic.mb_type[top_xy];
2003 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2004 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2005 if (sl->deblocking_filter == 2) {
2006 if (h->slice_table[top_xy] != sl->slice_num)
2008 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2009 left_type[LTOP] = left_type[LBOT] = 0;
2011 if (h->slice_table[top_xy] == 0xFFFF)
2013 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2014 left_type[LTOP] = left_type[LBOT] = 0;
2016 sl->top_type = top_type;
2017 sl->left_type[LTOP] = left_type[LTOP];
2018 sl->left_type[LBOT] = left_type[LBOT];
2020 if (IS_INTRA(mb_type))
2023 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2024 top_type, left_type, mb_xy, 0);
2025 if (sl->list_count == 2)
2026 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2027 top_type, left_type, mb_xy, 1);
2029 nnz = h->non_zero_count[mb_xy];
2030 nnz_cache = sl->non_zero_count_cache;
2031 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2032 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2033 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2034 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2035 sl->cbp = h->cbp_table[mb_xy];
2038 nnz = h->non_zero_count[top_xy];
2039 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2042 if (left_type[LTOP]) {
2043 nnz = h->non_zero_count[left_xy[LTOP]];
2044 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2045 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2046 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2047 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2050 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2051 * from what the loop filter needs */
2052 if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
2053 if (IS_8x8DCT(top_type)) {
2054 nnz_cache[4 + 8 * 0] =
2055 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2056 nnz_cache[6 + 8 * 0] =
2057 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2059 if (IS_8x8DCT(left_type[LTOP])) {
2060 nnz_cache[3 + 8 * 1] =
2061 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2063 if (IS_8x8DCT(left_type[LBOT])) {
2064 nnz_cache[3 + 8 * 3] =
2065 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2068 if (IS_8x8DCT(mb_type)) {
2069 nnz_cache[scan8[0]] =
2070 nnz_cache[scan8[1]] =
2071 nnz_cache[scan8[2]] =
2072 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2074 nnz_cache[scan8[0 + 4]] =
2075 nnz_cache[scan8[1 + 4]] =
2076 nnz_cache[scan8[2 + 4]] =
2077 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2079 nnz_cache[scan8[0 + 8]] =
2080 nnz_cache[scan8[1 + 8]] =
2081 nnz_cache[scan8[2 + 8]] =
2082 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2084 nnz_cache[scan8[0 + 12]] =
2085 nnz_cache[scan8[1 + 12]] =
2086 nnz_cache[scan8[2 + 12]] =
2087 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2094 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2096 uint8_t *dest_y, *dest_cb, *dest_cr;
2097 int linesize, uvlinesize, mb_x, mb_y;
2098 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2099 const int old_slice_type = sl->slice_type;
2100 const int pixel_shift = h->pixel_shift;
2101 const int block_h = 16 >> h->chroma_y_shift;
2103 if (sl->deblocking_filter) {
2104 for (mb_x = start_x; mb_x < end_x; mb_x++)
2105 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2107 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2108 sl->slice_num = h->slice_table[mb_xy];
2109 mb_type = h->cur_pic.mb_type[mb_xy];
2110 sl->list_count = h->list_counts[mb_xy];
2114 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2118 dest_y = h->cur_pic.f->data[0] +
2119 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2120 dest_cb = h->cur_pic.f->data[1] +
2121 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2122 mb_y * sl->uvlinesize * block_h;
2123 dest_cr = h->cur_pic.f->data[2] +
2124 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2125 mb_y * sl->uvlinesize * block_h;
2126 // FIXME simplify above
2129 linesize = sl->mb_linesize = sl->linesize * 2;
2130 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2131 if (mb_y & 1) { // FIXME move out of this function?
2132 dest_y -= sl->linesize * 15;
2133 dest_cb -= sl->uvlinesize * (block_h - 1);
2134 dest_cr -= sl->uvlinesize * (block_h - 1);
2137 linesize = sl->mb_linesize = sl->linesize;
2138 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2140 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2142 if (fill_filter_caches(h, sl, mb_type))
2144 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2145 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2147 if (FRAME_MBAFF(h)) {
2148 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2149 linesize, uvlinesize);
2151 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2152 dest_cr, linesize, uvlinesize);
2156 sl->slice_type = old_slice_type;
2158 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2159 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2160 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2163 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2165 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2166 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2167 h->cur_pic.mb_type[mb_xy - 1] :
2168 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2169 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2170 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2174 * Draw edges and report progress for the last MB row.
2176 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2178 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2179 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2180 int height = 16 << FRAME_MBAFF(h);
2181 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2183 if (sl->deblocking_filter) {
2184 if ((top + height) >= pic_height)
2185 height += deblock_border;
2186 top -= deblock_border;
2189 if (top >= pic_height || (top + height) < 0)
2192 height = FFMIN(height, pic_height - top);
2194 height = top + height;
2198 ff_h264_draw_horiz_band(h, sl, top, height);
2200 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2203 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2204 h->picture_structure == PICT_BOTTOM_FIELD);
2207 static void er_add_slice(H264SliceContext *sl,
2208 int startx, int starty,
2209 int endx, int endy, int status)
2211 if (!sl->h264->enable_er)
2214 if (CONFIG_ERROR_RESILIENCE) {
2215 ERContext *er = &sl->h264->slice_ctx[0].er;
2217 ff_er_add_slice(er, startx, starty, endx, endy, status);
2221 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2223 H264SliceContext *sl = arg;
2224 const H264Context *h = sl->h264;
2225 int lf_x_start = sl->mb_x;
2228 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2229 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2231 ret = alloc_scratch_buffers(sl, sl->linesize);
2235 sl->mb_skip_run = -1;
2237 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2239 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2240 avctx->codec_id != AV_CODEC_ID_H264 ||
2241 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2243 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2244 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2246 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2247 prev_status &= ~ VP_START;
2248 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2249 h->slice_ctx[0].er.error_occurred = 1;
2253 if (h->ps.pps->cabac) {
2255 align_get_bits(&sl->gb);
2258 ret = ff_init_cabac_decoder(&sl->cabac,
2259 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2260 (get_bits_left(&sl->gb) + 7) / 8);
2264 ff_h264_init_cabac_states(h, sl);
2269 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2270 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2271 sl->next_slice_idx);
2272 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2273 sl->mb_y, ER_MB_ERROR);
2274 return AVERROR_INVALIDDATA;
2277 ret = ff_h264_decode_mb_cabac(h, sl);
2278 // STOP_TIMER("decode_mb_cabac")
2281 ff_h264_hl_decode_mb(h, sl);
2283 // FIXME optimal? or let mb_decode decode 16x32 ?
2284 if (ret >= 0 && FRAME_MBAFF(h)) {
2287 ret = ff_h264_decode_mb_cabac(h, sl);
2290 ff_h264_hl_decode_mb(h, sl);
2293 eos = get_cabac_terminate(&sl->cabac);
2295 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2296 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2297 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2298 sl->mb_y, ER_MB_END);
2299 if (sl->mb_x >= lf_x_start)
2300 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2303 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2304 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2305 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2306 av_log(h->avctx, AV_LOG_ERROR,
2307 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2309 sl->cabac.bytestream_end - sl->cabac.bytestream);
2310 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2311 sl->mb_y, ER_MB_ERROR);
2312 return AVERROR_INVALIDDATA;
2315 if (++sl->mb_x >= h->mb_width) {
2316 loop_filter(h, sl, lf_x_start, sl->mb_x);
2317 sl->mb_x = lf_x_start = 0;
2318 decode_finish_row(h, sl);
2320 if (FIELD_OR_MBAFF_PICTURE(h)) {
2322 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2323 predict_field_decoding_flag(h, sl);
2327 if (eos || sl->mb_y >= h->mb_height) {
2328 ff_tlog(h->avctx, "slice end %d %d\n",
2329 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2330 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2331 sl->mb_y, ER_MB_END);
2332 if (sl->mb_x > lf_x_start)
2333 loop_filter(h, sl, lf_x_start, sl->mb_x);
2341 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2342 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2343 sl->next_slice_idx);
2344 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2345 sl->mb_y, ER_MB_ERROR);
2346 return AVERROR_INVALIDDATA;
2349 ret = ff_h264_decode_mb_cavlc(h, sl);
2352 ff_h264_hl_decode_mb(h, sl);
2354 // FIXME optimal? or let mb_decode decode 16x32 ?
2355 if (ret >= 0 && FRAME_MBAFF(h)) {
2357 ret = ff_h264_decode_mb_cavlc(h, sl);
2360 ff_h264_hl_decode_mb(h, sl);
2365 av_log(h->avctx, AV_LOG_ERROR,
2366 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2367 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2368 sl->mb_y, ER_MB_ERROR);
2372 if (++sl->mb_x >= h->mb_width) {
2373 loop_filter(h, sl, lf_x_start, sl->mb_x);
2374 sl->mb_x = lf_x_start = 0;
2375 decode_finish_row(h, sl);
2377 if (FIELD_OR_MBAFF_PICTURE(h)) {
2379 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2380 predict_field_decoding_flag(h, sl);
2382 if (sl->mb_y >= h->mb_height) {
2383 ff_tlog(h->avctx, "slice end %d %d\n",
2384 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2386 if ( get_bits_left(&sl->gb) == 0
2387 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2388 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2389 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2393 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2394 sl->mb_x, sl->mb_y, ER_MB_END);
2396 return AVERROR_INVALIDDATA;
2401 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2402 ff_tlog(h->avctx, "slice end %d %d\n",
2403 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2405 if (get_bits_left(&sl->gb) == 0) {
2406 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2407 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2408 if (sl->mb_x > lf_x_start)
2409 loop_filter(h, sl, lf_x_start, sl->mb_x);
2413 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2414 sl->mb_y, ER_MB_ERROR);
2416 return AVERROR_INVALIDDATA;
2424 * Call decode_slice() for each context.
2426 * @param h h264 master context
2427 * @param context_count number of contexts to execute
2429 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2431 AVCodecContext *const avctx = h->avctx;
2432 H264SliceContext *sl;
2435 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2437 h->slice_ctx[0].next_slice_idx = INT_MAX;
2439 if (h->avctx->hwaccel
2440 #if FF_API_CAP_VDPAU
2441 || h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU
2445 if (context_count == 1) {
2448 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2450 ret = decode_slice(avctx, &h->slice_ctx[0]);
2451 h->mb_y = h->slice_ctx[0].mb_y;
2454 av_assert0(context_count > 0);
2455 for (i = 0; i < context_count; i++) {
2456 int next_slice_idx = h->mb_width * h->mb_height;
2459 sl = &h->slice_ctx[i];
2460 if (CONFIG_ERROR_RESILIENCE) {
2461 sl->er.error_count = 0;
2464 /* make sure none of those slices overlap */
2465 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2466 for (j = 0; j < context_count; j++) {
2467 H264SliceContext *sl2 = &h->slice_ctx[j];
2468 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2470 if (i == j || slice_idx2 < slice_idx)
2472 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2474 sl->next_slice_idx = next_slice_idx;
2477 avctx->execute(avctx, decode_slice, h->slice_ctx,
2478 NULL, context_count, sizeof(h->slice_ctx[0]));
2480 /* pull back stuff from slices to master context */
2481 sl = &h->slice_ctx[context_count - 1];
2483 if (CONFIG_ERROR_RESILIENCE) {
2484 for (i = 1; i < context_count; i++)
2485 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;