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 static int h264_slice_header_init(H264Context *h);
297 int ff_h264_update_thread_context(AVCodecContext *dst,
298 const AVCodecContext *src)
300 H264Context *h = dst->priv_data, *h1 = src->priv_data;
301 int inited = h->context_initialized, err = 0;
308 // We can't fail if SPS isn't set at it breaks current skip_frame code
310 // return AVERROR_INVALIDDATA;
313 (h->width != h1->width ||
314 h->height != h1->height ||
315 h->mb_width != h1->mb_width ||
316 h->mb_height != h1->mb_height ||
318 h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
319 h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
320 h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
324 /* copy block_offset since frame_start may not be called */
325 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
328 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
329 av_buffer_unref(&h->ps.sps_list[i]);
330 if (h1->ps.sps_list[i]) {
331 h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
332 if (!h->ps.sps_list[i])
333 return AVERROR(ENOMEM);
336 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
337 av_buffer_unref(&h->ps.pps_list[i]);
338 if (h1->ps.pps_list[i]) {
339 h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
340 if (!h->ps.pps_list[i])
341 return AVERROR(ENOMEM);
345 av_buffer_unref(&h->ps.pps_ref);
346 av_buffer_unref(&h->ps.sps_ref);
349 if (h1->ps.pps_ref) {
350 h->ps.pps_ref = av_buffer_ref(h1->ps.pps_ref);
352 return AVERROR(ENOMEM);
353 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
355 if (h1->ps.sps_ref) {
356 h->ps.sps_ref = av_buffer_ref(h1->ps.sps_ref);
358 return AVERROR(ENOMEM);
359 h->ps.sps = (SPS*)h->ps.sps_ref->data;
362 if (need_reinit || !inited) {
363 h->width = h1->width;
364 h->height = h1->height;
365 h->mb_height = h1->mb_height;
366 h->mb_width = h1->mb_width;
367 h->mb_num = h1->mb_num;
368 h->mb_stride = h1->mb_stride;
369 h->b_stride = h1->b_stride;
371 if (h->context_initialized || h1->context_initialized) {
372 if ((err = h264_slice_header_init(h)) < 0) {
373 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
377 /* copy block_offset since frame_start may not be called */
378 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
381 h->avctx->coded_height = h1->avctx->coded_height;
382 h->avctx->coded_width = h1->avctx->coded_width;
383 h->avctx->width = h1->avctx->width;
384 h->avctx->height = h1->avctx->height;
385 h->coded_picture_number = h1->coded_picture_number;
386 h->first_field = h1->first_field;
387 h->picture_structure = h1->picture_structure;
388 h->droppable = h1->droppable;
389 h->low_delay = h1->low_delay;
390 h->backup_width = h1->backup_width;
391 h->backup_height = h1->backup_height;
392 h->backup_pix_fmt = h1->backup_pix_fmt;
394 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
395 ff_h264_unref_picture(h, &h->DPB[i]);
396 if (h1->DPB[i].f->buf[0] &&
397 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
401 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
402 ff_h264_unref_picture(h, &h->cur_pic);
403 if (h1->cur_pic.f->buf[0]) {
404 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
409 h->enable_er = h1->enable_er;
410 h->workaround_bugs = h1->workaround_bugs;
411 h->low_delay = h1->low_delay;
412 h->droppable = h1->droppable;
414 // extradata/NAL handling
415 h->is_avc = h1->is_avc;
416 h->nal_length_size = h1->nal_length_size;
417 h->sei.unregistered.x264_build = h1->sei.unregistered.x264_build;
419 memcpy(&h->poc, &h1->poc, sizeof(h->poc));
421 h->curr_pic_num = h1->curr_pic_num;
422 h->max_pic_num = h1->max_pic_num;
424 memcpy(h->default_ref, h1->default_ref, sizeof(h->default_ref));
425 memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
426 memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
427 memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
428 memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
430 h->next_output_pic = h1->next_output_pic;
431 h->next_outputed_poc = h1->next_outputed_poc;
433 memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
434 h->mmco_index = h1->mmco_index;
435 h->mmco_reset = h1->mmco_reset;
436 h->long_ref_count = h1->long_ref_count;
437 h->short_ref_count = h1->short_ref_count;
439 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
440 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
441 copy_picture_range(h->delayed_pic, h1->delayed_pic,
442 MAX_DELAYED_PIC_COUNT + 2, h, h1);
444 h->frame_recovered = h1->frame_recovered;
450 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
451 h->poc.prev_poc_msb = h->poc.poc_msb;
452 h->poc.prev_poc_lsb = h->poc.poc_lsb;
454 h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
455 h->poc.prev_frame_num = h->poc.frame_num;
457 h->recovery_frame = h1->recovery_frame;
462 static int h264_frame_start(H264Context *h)
466 const int pixel_shift = h->pixel_shift;
468 1<<(h->ps.sps->bit_depth_luma-1),
469 1<<(h->ps.sps->bit_depth_chroma-1),
470 1<<(h->ps.sps->bit_depth_chroma-1),
474 if (!ff_thread_can_start_frame(h->avctx)) {
475 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
479 release_unused_pictures(h, 1);
480 h->cur_pic_ptr = NULL;
482 i = find_unused_picture(h);
484 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
489 pic->reference = h->droppable ? 0 : h->picture_structure;
490 pic->f->coded_picture_number = h->coded_picture_number++;
491 pic->field_picture = h->picture_structure != PICT_FRAME;
492 pic->frame_num = h->poc.frame_num;
494 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
496 * See decode_nal_units().
498 pic->f->key_frame = 0;
501 pic->invalid_gap = 0;
502 pic->sei_recovery_frame_cnt = h->sei.recovery_point.recovery_frame_cnt;
504 if ((ret = alloc_picture(h, pic)) < 0)
506 if(!h->frame_recovered && !h->avctx->hwaccel
508 && !(h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU)
511 ff_color_frame(pic->f, c);
513 h->cur_pic_ptr = pic;
514 ff_h264_unref_picture(h, &h->cur_pic);
515 if (CONFIG_ERROR_RESILIENCE) {
516 ff_h264_set_erpic(&h->slice_ctx[0].er.cur_pic, NULL);
519 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
522 for (i = 0; i < h->nb_slice_ctx; i++) {
523 h->slice_ctx[i].linesize = h->cur_pic_ptr->f->linesize[0];
524 h->slice_ctx[i].uvlinesize = h->cur_pic_ptr->f->linesize[1];
527 if (CONFIG_ERROR_RESILIENCE && h->enable_er) {
528 ff_er_frame_start(&h->slice_ctx[0].er);
529 ff_h264_set_erpic(&h->slice_ctx[0].er.last_pic, NULL);
530 ff_h264_set_erpic(&h->slice_ctx[0].er.next_pic, NULL);
533 for (i = 0; i < 16; i++) {
534 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
535 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
537 for (i = 0; i < 16; i++) {
538 h->block_offset[16 + i] =
539 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
540 h->block_offset[48 + 16 + i] =
541 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
544 /* We mark the current picture as non-reference after allocating it, so
545 * that if we break out due to an error it can be released automatically
546 * in the next ff_mpv_frame_start().
548 h->cur_pic_ptr->reference = 0;
550 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
552 h->next_output_pic = NULL;
554 assert(h->cur_pic_ptr->long_ref == 0);
559 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
561 uint8_t *src_cb, uint8_t *src_cr,
562 int linesize, int uvlinesize,
567 const int pixel_shift = h->pixel_shift;
568 int chroma444 = CHROMA444(h);
569 int chroma422 = CHROMA422(h);
572 src_cb -= uvlinesize;
573 src_cr -= uvlinesize;
575 if (!simple && FRAME_MBAFF(h)) {
578 top_border = sl->top_borders[0][sl->mb_x];
579 AV_COPY128(top_border, src_y + 15 * linesize);
581 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
582 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
585 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
586 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
587 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
588 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
590 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
591 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
593 } else if (chroma422) {
595 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
596 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
598 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
599 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
603 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
604 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
606 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
607 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
612 } else if (MB_MBAFF(sl)) {
618 top_border = sl->top_borders[top_idx][sl->mb_x];
619 /* There are two lines saved, the line above the top macroblock
620 * of a pair, and the line above the bottom macroblock. */
621 AV_COPY128(top_border, src_y + 16 * linesize);
623 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
625 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
628 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
629 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
630 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
631 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
633 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
634 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
636 } else if (chroma422) {
638 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
639 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
641 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
642 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
646 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
647 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
649 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
650 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
657 * Initialize implicit_weight table.
658 * @param field 0/1 initialize the weight for interlaced MBAFF
659 * -1 initializes the rest
661 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
663 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
665 for (i = 0; i < 2; i++) {
666 sl->pwt.luma_weight_flag[i] = 0;
667 sl->pwt.chroma_weight_flag[i] = 0;
671 if (h->picture_structure == PICT_FRAME) {
672 cur_poc = h->cur_pic_ptr->poc;
674 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
676 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
677 sl->ref_list[0][0].poc + (int64_t)sl->ref_list[1][0].poc == 2 * cur_poc) {
678 sl->pwt.use_weight = 0;
679 sl->pwt.use_weight_chroma = 0;
683 ref_count0 = sl->ref_count[0];
684 ref_count1 = sl->ref_count[1];
686 cur_poc = h->cur_pic_ptr->field_poc[field];
688 ref_count0 = 16 + 2 * sl->ref_count[0];
689 ref_count1 = 16 + 2 * sl->ref_count[1];
692 sl->pwt.use_weight = 2;
693 sl->pwt.use_weight_chroma = 2;
694 sl->pwt.luma_log2_weight_denom = 5;
695 sl->pwt.chroma_log2_weight_denom = 5;
697 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
698 int64_t poc0 = sl->ref_list[0][ref0].poc;
699 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
701 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
702 int poc1 = sl->ref_list[1][ref1].poc;
703 int td = av_clip_int8(poc1 - poc0);
705 int tb = av_clip_int8(cur_poc - poc0);
706 int tx = (16384 + (FFABS(td) >> 1)) / td;
707 int dist_scale_factor = (tb * tx + 32) >> 8;
708 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
709 w = 64 - dist_scale_factor;
713 sl->pwt.implicit_weight[ref0][ref1][0] =
714 sl->pwt.implicit_weight[ref0][ref1][1] = w;
716 sl->pwt.implicit_weight[ref0][ref1][field] = w;
723 * initialize scan tables
725 static void init_scan_tables(H264Context *h)
728 for (i = 0; i < 16; i++) {
729 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
730 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
731 h->field_scan[i] = TRANSPOSE(field_scan[i]);
734 for (i = 0; i < 64; i++) {
735 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
736 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
737 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
738 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
739 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
742 if (h->ps.sps->transform_bypass) { // FIXME same ugly
743 memcpy(h->zigzag_scan_q0 , ff_zigzag_scan , sizeof(h->zigzag_scan_q0 ));
744 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
745 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
746 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
747 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
748 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
750 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
751 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
752 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
753 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
754 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
755 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
759 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
761 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
762 CONFIG_H264_D3D11VA_HWACCEL + \
763 CONFIG_H264_VAAPI_HWACCEL + \
764 (CONFIG_H264_VDA_HWACCEL * 2) + \
765 CONFIG_H264_VIDEOTOOLBOX_HWACCEL + \
766 CONFIG_H264_VDPAU_HWACCEL)
767 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
768 const enum AVPixelFormat *choices = pix_fmts;
771 switch (h->ps.sps->bit_depth_luma) {
774 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
775 *fmt++ = AV_PIX_FMT_GBRP9;
777 *fmt++ = AV_PIX_FMT_YUV444P9;
778 } else if (CHROMA422(h))
779 *fmt++ = AV_PIX_FMT_YUV422P9;
781 *fmt++ = AV_PIX_FMT_YUV420P9;
785 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
786 *fmt++ = AV_PIX_FMT_GBRP10;
788 *fmt++ = AV_PIX_FMT_YUV444P10;
789 } else if (CHROMA422(h))
790 *fmt++ = AV_PIX_FMT_YUV422P10;
792 *fmt++ = AV_PIX_FMT_YUV420P10;
796 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
797 *fmt++ = AV_PIX_FMT_GBRP12;
799 *fmt++ = AV_PIX_FMT_YUV444P12;
800 } else if (CHROMA422(h))
801 *fmt++ = AV_PIX_FMT_YUV422P12;
803 *fmt++ = AV_PIX_FMT_YUV420P12;
807 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
808 *fmt++ = AV_PIX_FMT_GBRP14;
810 *fmt++ = AV_PIX_FMT_YUV444P14;
811 } else if (CHROMA422(h))
812 *fmt++ = AV_PIX_FMT_YUV422P14;
814 *fmt++ = AV_PIX_FMT_YUV420P14;
817 #if CONFIG_H264_VDPAU_HWACCEL
818 *fmt++ = AV_PIX_FMT_VDPAU;
821 if (h->avctx->colorspace == AVCOL_SPC_RGB)
822 *fmt++ = AV_PIX_FMT_GBRP;
823 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
824 *fmt++ = AV_PIX_FMT_YUVJ444P;
826 *fmt++ = AV_PIX_FMT_YUV444P;
827 } else if (CHROMA422(h)) {
828 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
829 *fmt++ = AV_PIX_FMT_YUVJ422P;
831 *fmt++ = AV_PIX_FMT_YUV422P;
833 #if CONFIG_H264_DXVA2_HWACCEL
834 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
836 #if CONFIG_H264_D3D11VA_HWACCEL
837 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
839 #if CONFIG_H264_VAAPI_HWACCEL
840 *fmt++ = AV_PIX_FMT_VAAPI;
842 #if CONFIG_H264_VDA_HWACCEL
843 *fmt++ = AV_PIX_FMT_VDA_VLD;
844 *fmt++ = AV_PIX_FMT_VDA;
846 #if CONFIG_H264_VIDEOTOOLBOX_HWACCEL
847 *fmt++ = AV_PIX_FMT_VIDEOTOOLBOX;
849 if (h->avctx->codec->pix_fmts)
850 choices = h->avctx->codec->pix_fmts;
851 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
852 *fmt++ = AV_PIX_FMT_YUVJ420P;
854 *fmt++ = AV_PIX_FMT_YUV420P;
858 av_log(h->avctx, AV_LOG_ERROR,
859 "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
860 return AVERROR_INVALIDDATA;
863 *fmt = AV_PIX_FMT_NONE;
865 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
866 if (choices[i] == h->avctx->pix_fmt && !force_callback)
868 return ff_thread_get_format(h->avctx, choices);
871 /* export coded and cropped frame dimensions to AVCodecContext */
872 static int init_dimensions(H264Context *h)
874 SPS *sps = h->ps.sps;
875 int width = h->width - (sps->crop_right + sps->crop_left);
876 int height = h->height - (sps->crop_top + sps->crop_bottom);
877 av_assert0(sps->crop_right + sps->crop_left < (unsigned)h->width);
878 av_assert0(sps->crop_top + sps->crop_bottom < (unsigned)h->height);
880 /* handle container cropping */
881 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
882 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16) &&
883 h->avctx->width <= width &&
884 h->avctx->height <= height
886 width = h->avctx->width;
887 height = h->avctx->height;
890 if (width <= 0 || height <= 0) {
891 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
893 if (h->avctx->err_recognition & AV_EF_EXPLODE)
894 return AVERROR_INVALIDDATA;
896 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
907 h->avctx->coded_width = h->width;
908 h->avctx->coded_height = h->height;
909 h->avctx->width = width;
910 h->avctx->height = height;
915 static int h264_slice_header_init(H264Context *h)
917 const SPS *sps = h->ps.sps;
918 int nb_slices = (HAVE_THREADS &&
919 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
920 h->avctx->thread_count : 1;
923 ff_set_sar(h->avctx, sps->sar);
924 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
925 &h->chroma_x_shift, &h->chroma_y_shift);
927 if (sps->timing_info_present_flag) {
928 int64_t den = sps->time_scale;
929 if (h->sei.unregistered.x264_build < 44U)
931 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
932 sps->num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
935 ff_h264_free_tables(h);
938 h->prev_interlaced_frame = 1;
941 ret = ff_h264_alloc_tables(h);
943 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
948 if (h->avctx->codec &&
949 h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU &&
950 (sps->bit_depth_luma != 8 || sps->chroma_format_idc > 1)) {
951 av_log(h->avctx, AV_LOG_ERROR,
952 "VDPAU decoding does not support video colorspace.\n");
953 ret = AVERROR_INVALIDDATA;
958 if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 14 ||
959 sps->bit_depth_luma == 11 || sps->bit_depth_luma == 13
961 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
962 sps->bit_depth_luma);
963 ret = AVERROR_INVALIDDATA;
967 h->cur_bit_depth_luma =
968 h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
969 h->cur_chroma_format_idc = sps->chroma_format_idc;
970 h->pixel_shift = sps->bit_depth_luma > 8;
971 h->chroma_format_idc = sps->chroma_format_idc;
972 h->bit_depth_luma = sps->bit_depth_luma;
974 ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
975 sps->chroma_format_idc);
976 ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
977 ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
978 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
979 sps->chroma_format_idc);
980 ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
982 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
985 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
987 max_slices = H264_MAX_THREADS;
988 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
989 " reducing to %d\n", nb_slices, max_slices);
990 nb_slices = max_slices;
992 h->slice_context_count = nb_slices;
993 h->max_contexts = FFMIN(h->max_contexts, nb_slices);
995 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
996 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
998 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1002 for (i = 0; i < h->slice_context_count; i++) {
1003 H264SliceContext *sl = &h->slice_ctx[i];
1006 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
1007 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
1008 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
1010 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
1011 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1017 h->context_initialized = 1;
1021 ff_h264_free_tables(h);
1022 h->context_initialized = 0;
1026 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1029 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1030 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1031 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1038 * Decode a slice header.
1039 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1041 * @param h h264context
1043 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1045 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1049 unsigned int first_mb_in_slice;
1050 unsigned int pps_id;
1052 unsigned int slice_type, tmp, i, j;
1053 int last_pic_structure, last_pic_droppable;
1055 int needs_reinit = 0;
1056 int field_pic_flag, bottom_field_flag;
1057 int first_slice = sl == h->slice_ctx && !h->current_slice;
1058 int frame_num, droppable, picture_structure;
1059 int mb_aff_frame, last_mb_aff_frame;
1062 av_assert0(!h->setup_finished);
1064 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1065 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1067 first_mb_in_slice = get_ue_golomb_long(&sl->gb);
1069 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1070 if (h->current_slice) {
1071 if (h->setup_finished) {
1072 av_log(h->avctx, AV_LOG_ERROR, "Too many fields\n");
1073 return AVERROR_INVALIDDATA;
1075 if (h->max_contexts > 1) {
1076 if (!h->single_decode_warning) {
1077 av_log(h->avctx, AV_LOG_WARNING, "Cannot decode multiple access units as slice threads\n");
1078 h->single_decode_warning = 1;
1080 h->max_contexts = 1;
1081 return SLICE_SINGLETHREAD;
1084 if (h->cur_pic_ptr && FIELD_PICTURE(h) && h->first_field) {
1085 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1086 h->current_slice = 0;
1089 } else if (h->cur_pic_ptr && !FIELD_PICTURE(h) && !h->first_field && h->nal_unit_type == NAL_IDR_SLICE) {
1090 av_log(h, AV_LOG_WARNING, "Broken frame packetizing\n");
1091 ret = ff_h264_field_end(h, h->slice_ctx, 1);
1092 h->current_slice = 0;
1093 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1094 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1095 h->cur_pic_ptr = NULL;
1099 return AVERROR_INVALIDDATA;
1102 if (!h->first_field) {
1103 if (h->cur_pic_ptr && !h->droppable) {
1104 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1105 h->picture_structure == PICT_BOTTOM_FIELD);
1107 h->cur_pic_ptr = NULL;
1111 if (!h->current_slice)
1112 av_assert0(sl == h->slice_ctx);
1114 slice_type = get_ue_golomb_31(&sl->gb);
1115 if (slice_type > 9) {
1116 av_log(h->avctx, AV_LOG_ERROR,
1117 "slice type %d too large at %d\n",
1118 slice_type, first_mb_in_slice);
1119 return AVERROR_INVALIDDATA;
1121 if (slice_type > 4) {
1123 sl->slice_type_fixed = 1;
1125 sl->slice_type_fixed = 0;
1127 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1128 sl->slice_type = slice_type;
1129 sl->slice_type_nos = slice_type & 3;
1131 if (h->nal_unit_type == NAL_IDR_SLICE &&
1132 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1133 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1134 return AVERROR_INVALIDDATA;
1137 if (h->current_slice == 0 && !h->first_field) {
1139 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1140 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1141 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1142 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE && h->sei.recovery_point.recovery_frame_cnt < 0) ||
1143 h->avctx->skip_frame >= AVDISCARD_ALL) {
1144 return SLICE_SKIPED;
1148 // to make a few old functions happy, it's wrong though
1149 if (!h->setup_finished)
1150 h->pict_type = sl->slice_type;
1152 pps_id = get_ue_golomb(&sl->gb);
1153 if (pps_id >= MAX_PPS_COUNT) {
1154 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1155 return AVERROR_INVALIDDATA;
1157 if (!h->ps.pps_list[pps_id]) {
1158 av_log(h->avctx, AV_LOG_ERROR,
1159 "non-existing PPS %u referenced\n",
1161 return AVERROR_INVALIDDATA;
1163 if (h->au_pps_id >= 0 && pps_id != h->au_pps_id) {
1164 av_log(h->avctx, AV_LOG_ERROR,
1165 "PPS change from %d to %d forbidden\n",
1166 h->au_pps_id, pps_id);
1167 return AVERROR_INVALIDDATA;
1170 pps = (const PPS*)h->ps.pps_list[pps_id]->data;
1172 if (!h->ps.sps_list[pps->sps_id]) {
1173 av_log(h->avctx, AV_LOG_ERROR,
1174 "non-existing SPS %u referenced\n",
1176 return AVERROR_INVALIDDATA;
1180 av_buffer_unref(&h->ps.pps_ref);
1182 h->ps.pps_ref = av_buffer_ref(h->ps.pps_list[pps_id]);
1184 return AVERROR(ENOMEM);
1185 h->ps.pps = (const PPS*)h->ps.pps_ref->data;
1187 if (h->ps.pps->sps_id != pps->sps_id ||
1188 h->ps.pps->transform_8x8_mode != pps->transform_8x8_mode /*||
1189 (h->setup_finished && h->ps.pps != pps)*/) {
1190 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1191 return AVERROR_INVALIDDATA;
1195 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data ||
1196 pps->sps_id != h->current_sps_id) {
1199 av_log(h->avctx, AV_LOG_ERROR,
1200 "SPS changed in the middle of the frame\n");
1201 return AVERROR_INVALIDDATA;
1204 av_buffer_unref(&h->ps.sps_ref);
1206 h->ps.sps_ref = av_buffer_ref(h->ps.sps_list[h->ps.pps->sps_id]);
1208 return AVERROR(ENOMEM);
1209 h->ps.sps = (const SPS*)h->ps.sps_ref->data;
1211 if (h->mb_width != h->ps.sps->mb_width ||
1212 h->mb_height != h->ps.sps->mb_height * (2 - h->ps.sps->frame_mbs_only_flag) ||
1213 h->cur_bit_depth_luma != h->ps.sps->bit_depth_luma ||
1214 h->cur_chroma_format_idc != h->ps.sps->chroma_format_idc
1218 if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
1219 h->chroma_format_idc != h->ps.sps->chroma_format_idc)
1222 if (h->flags & AV_CODEC_FLAG_LOW_DELAY ||
1223 (h->ps.sps->bitstream_restriction_flag &&
1224 !h->ps.sps->num_reorder_frames)) {
1225 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1226 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1227 "Reenabling low delay requires a codec flush.\n");
1232 if (h->avctx->has_b_frames < 2)
1233 h->avctx->has_b_frames = !h->low_delay;
1240 must_reinit = (h->context_initialized &&
1241 ( 16*sps->mb_width != h->avctx->coded_width
1242 || 16*sps->mb_height * (2 - sps->frame_mbs_only_flag) != h->avctx->coded_height
1243 || h->cur_bit_depth_luma != sps->bit_depth_luma
1244 || h->cur_chroma_format_idc != sps->chroma_format_idc
1245 || h->mb_width != sps->mb_width
1246 || h->mb_height != sps->mb_height * (2 - sps->frame_mbs_only_flag)
1248 if (h->avctx->pix_fmt == AV_PIX_FMT_NONE
1249 || (non_j_pixfmt(h->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h, 0))))
1252 if (first_slice && av_cmp_q(sps->sar, h->avctx->sample_aspect_ratio))
1255 if (!h->setup_finished) {
1256 h->avctx->profile = ff_h264_get_profile(sps);
1257 h->avctx->level = sps->level_idc;
1258 h->avctx->refs = sps->ref_frame_count;
1260 h->mb_width = sps->mb_width;
1261 h->mb_height = sps->mb_height * (2 - sps->frame_mbs_only_flag);
1262 h->mb_num = h->mb_width * h->mb_height;
1263 h->mb_stride = h->mb_width + 1;
1265 h->b_stride = h->mb_width * 4;
1267 h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
1269 h->width = 16 * h->mb_width;
1270 h->height = 16 * h->mb_height;
1272 ret = init_dimensions(h);
1276 if (sps->video_signal_type_present_flag) {
1277 h->avctx->color_range = sps->full_range > 0 ? AVCOL_RANGE_JPEG
1279 if (sps->colour_description_present_flag) {
1280 if (h->avctx->colorspace != sps->colorspace)
1282 h->avctx->color_primaries = sps->color_primaries;
1283 h->avctx->color_trc = sps->color_trc;
1284 h->avctx->colorspace = sps->colorspace;
1289 if (h->context_initialized &&
1290 (must_reinit || needs_reinit)) {
1291 h->context_initialized = 0;
1292 if (sl != h->slice_ctx) {
1293 av_log(h->avctx, AV_LOG_ERROR,
1294 "changing width %d -> %d / height %d -> %d on "
1296 h->width, h->avctx->coded_width,
1297 h->height, h->avctx->coded_height,
1298 h->current_slice + 1);
1299 return AVERROR_INVALIDDATA;
1302 av_assert1(first_slice);
1304 ff_h264_flush_change(h);
1306 if ((ret = get_pixel_format(h, 1)) < 0)
1308 h->avctx->pix_fmt = ret;
1310 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1311 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1313 if ((ret = h264_slice_header_init(h)) < 0) {
1314 av_log(h->avctx, AV_LOG_ERROR,
1315 "h264_slice_header_init() failed\n");
1319 if (!h->context_initialized) {
1320 if (sl != h->slice_ctx) {
1321 av_log(h->avctx, AV_LOG_ERROR,
1322 "Cannot (re-)initialize context during parallel decoding.\n");
1323 return AVERROR_PATCHWELCOME;
1326 if ((ret = get_pixel_format(h, 1)) < 0)
1328 h->avctx->pix_fmt = ret;
1330 if ((ret = h264_slice_header_init(h)) < 0) {
1331 av_log(h->avctx, AV_LOG_ERROR,
1332 "h264_slice_header_init() failed\n");
1337 frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
1339 if (h->poc.frame_num != frame_num) {
1340 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1341 h->poc.frame_num, frame_num);
1342 return AVERROR_INVALIDDATA;
1346 if (!h->setup_finished)
1347 h->poc.frame_num = frame_num;
1351 last_mb_aff_frame = h->mb_aff_frame;
1352 last_pic_structure = h->picture_structure;
1353 last_pic_droppable = h->droppable;
1355 droppable = h->nal_ref_idc == 0;
1356 if (sps->frame_mbs_only_flag) {
1357 picture_structure = PICT_FRAME;
1359 if (!h->ps.sps->direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1360 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1363 field_pic_flag = get_bits1(&sl->gb);
1365 if (field_pic_flag) {
1366 bottom_field_flag = get_bits1(&sl->gb);
1367 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1369 picture_structure = PICT_FRAME;
1370 mb_aff_frame = sps->mb_aff;
1374 if (h->current_slice) {
1375 if (last_pic_structure != picture_structure ||
1376 last_pic_droppable != droppable ||
1377 last_mb_aff_frame != mb_aff_frame) {
1378 av_log(h->avctx, AV_LOG_ERROR,
1379 "Changing field mode (%d -> %d) between slices is not allowed\n",
1380 last_pic_structure, h->picture_structure);
1381 return AVERROR_INVALIDDATA;
1382 } else if (!h->cur_pic_ptr) {
1383 av_log(h->avctx, AV_LOG_ERROR,
1384 "unset cur_pic_ptr on slice %d\n",
1385 h->current_slice + 1);
1386 return AVERROR_INVALIDDATA;
1390 if (!h->setup_finished) {
1391 h->droppable = droppable;
1392 h->picture_structure = picture_structure;
1393 h->mb_aff_frame = mb_aff_frame;
1395 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1397 if (h->current_slice == 0) {
1398 /* Shorten frame num gaps so we don't have to allocate reference
1399 * frames just to throw them away */
1400 if (h->poc.frame_num != h->poc.prev_frame_num) {
1401 int unwrap_prev_frame_num = h->poc.prev_frame_num;
1402 int max_frame_num = 1 << sps->log2_max_frame_num;
1404 if (unwrap_prev_frame_num > h->poc.frame_num)
1405 unwrap_prev_frame_num -= max_frame_num;
1407 if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
1408 unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
1409 if (unwrap_prev_frame_num < 0)
1410 unwrap_prev_frame_num += max_frame_num;
1412 h->poc.prev_frame_num = unwrap_prev_frame_num;
1416 /* See if we have a decoded first field looking for a pair...
1417 * Here, we're using that to see if we should mark previously
1418 * decode frames as "finished".
1419 * We have to do that before the "dummy" in-between frame allocation,
1420 * since that can modify h->cur_pic_ptr. */
1421 if (h->first_field) {
1422 av_assert0(h->cur_pic_ptr);
1423 av_assert0(h->cur_pic_ptr->f->buf[0]);
1424 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1426 /* Mark old field/frame as completed */
1427 if (h->cur_pic_ptr->tf.owner == h->avctx) {
1428 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1429 last_pic_structure == PICT_BOTTOM_FIELD);
1432 /* figure out if we have a complementary field pair */
1433 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1434 /* Previous field is unmatched. Don't display it, but let it
1435 * remain for reference if marked as such. */
1436 if (last_pic_structure != PICT_FRAME) {
1437 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1438 last_pic_structure == PICT_TOP_FIELD);
1441 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1442 /* This and previous field were reference, but had
1443 * different frame_nums. Consider this field first in
1444 * pair. Throw away previous field except for reference
1446 if (last_pic_structure != PICT_FRAME) {
1447 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1448 last_pic_structure == PICT_TOP_FIELD);
1451 /* Second field in complementary pair */
1452 if (!((last_pic_structure == PICT_TOP_FIELD &&
1453 h->picture_structure == PICT_BOTTOM_FIELD) ||
1454 (last_pic_structure == PICT_BOTTOM_FIELD &&
1455 h->picture_structure == PICT_TOP_FIELD))) {
1456 av_log(h->avctx, AV_LOG_ERROR,
1457 "Invalid field mode combination %d/%d\n",
1458 last_pic_structure, h->picture_structure);
1459 h->picture_structure = last_pic_structure;
1460 h->droppable = last_pic_droppable;
1461 return AVERROR_INVALIDDATA;
1462 } else if (last_pic_droppable != h->droppable) {
1463 avpriv_request_sample(h->avctx,
1464 "Found reference and non-reference fields in the same frame, which");
1465 h->picture_structure = last_pic_structure;
1466 h->droppable = last_pic_droppable;
1467 return AVERROR_PATCHWELCOME;
1473 while (h->poc.frame_num != h->poc.prev_frame_num && !h->first_field &&
1474 h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
1475 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1476 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1477 h->poc.frame_num, h->poc.prev_frame_num);
1478 if (!sps->gaps_in_frame_num_allowed_flag)
1479 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1480 h->last_pocs[i] = INT_MIN;
1481 ret = h264_frame_start(h);
1487 h->poc.prev_frame_num++;
1488 h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
1489 h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
1490 h->cur_pic_ptr->invalid_gap = !sps->gaps_in_frame_num_allowed_flag;
1491 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1492 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1493 ret = ff_generate_sliding_window_mmcos(h, 1);
1494 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1496 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1497 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1499 /* Error concealment: If a ref is missing, copy the previous ref
1501 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1502 * many assumptions about there being no actual duplicates.
1503 * FIXME: This does not copy padding for out-of-frame motion
1504 * vectors. Given we are concealing a lost frame, this probably
1505 * is not noticeable by comparison, but it should be fixed. */
1506 if (h->short_ref_count) {
1508 h->short_ref[0]->f->width == prev->f->width &&
1509 h->short_ref[0]->f->height == prev->f->height &&
1510 h->short_ref[0]->f->format == prev->f->format) {
1511 av_image_copy(h->short_ref[0]->f->data,
1512 h->short_ref[0]->f->linesize,
1513 (const uint8_t **)prev->f->data,
1518 h->short_ref[0]->poc = prev->poc + 2;
1520 h->short_ref[0]->frame_num = h->poc.prev_frame_num;
1524 /* See if we have a decoded first field looking for a pair...
1525 * We're using that to see whether to continue decoding in that
1526 * frame, or to allocate a new one. */
1527 if (h->first_field) {
1528 av_assert0(h->cur_pic_ptr);
1529 av_assert0(h->cur_pic_ptr->f->buf[0]);
1530 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1532 /* figure out if we have a complementary field pair */
1533 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1534 /* Previous field is unmatched. Don't display it, but let it
1535 * remain for reference if marked as such. */
1536 h->missing_fields ++;
1537 h->cur_pic_ptr = NULL;
1538 h->first_field = FIELD_PICTURE(h);
1540 h->missing_fields = 0;
1541 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1542 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1543 h->picture_structure==PICT_BOTTOM_FIELD);
1544 /* This and the previous field had different frame_nums.
1545 * Consider this field first in pair. Throw away previous
1546 * one except for reference purposes. */
1548 h->cur_pic_ptr = NULL;
1550 /* Second field in complementary pair */
1555 /* Frame or first field in a potentially complementary pair */
1556 h->first_field = FIELD_PICTURE(h);
1559 if (!FIELD_PICTURE(h) || h->first_field) {
1560 if (h264_frame_start(h) < 0) {
1562 return AVERROR_INVALIDDATA;
1565 release_unused_pictures(h, 0);
1567 /* Some macroblocks can be accessed before they're available in case
1568 * of lost slices, MBAFF or threading. */
1569 if (FIELD_PICTURE(h)) {
1570 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1571 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1573 memset(h->slice_table, -1,
1574 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1578 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1579 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1580 first_mb_in_slice >= h->mb_num) {
1581 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1582 return AVERROR_INVALIDDATA;
1584 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1585 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1586 FIELD_OR_MBAFF_PICTURE(h);
1587 if (h->picture_structure == PICT_BOTTOM_FIELD)
1588 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1589 av_assert1(sl->mb_y < h->mb_height);
1591 if (h->picture_structure == PICT_FRAME) {
1592 h->curr_pic_num = h->poc.frame_num;
1593 h->max_pic_num = 1 << sps->log2_max_frame_num;
1595 h->curr_pic_num = 2 * h->poc.frame_num + 1;
1596 h->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
1599 if (h->nal_unit_type == NAL_IDR_SLICE)
1600 get_ue_golomb_long(&sl->gb); /* idr_pic_id */
1602 if (sps->poc_type == 0) {
1603 int poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
1605 if (!h->setup_finished)
1606 h->poc.poc_lsb = poc_lsb;
1608 if (pps->pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1609 int delta_poc_bottom = get_se_golomb(&sl->gb);
1610 if (!h->setup_finished)
1611 h->poc.delta_poc_bottom = delta_poc_bottom;
1615 if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
1616 int delta_poc = get_se_golomb(&sl->gb);
1618 if (!h->setup_finished)
1619 h->poc.delta_poc[0] = delta_poc;
1621 if (pps->pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1622 delta_poc = get_se_golomb(&sl->gb);
1624 if (!h->setup_finished)
1625 h->poc.delta_poc[1] = delta_poc;
1629 if (!h->setup_finished)
1630 ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
1631 sps, &h->poc, h->picture_structure, h->nal_ref_idc);
1633 if (pps->redundant_pic_cnt_present)
1634 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1636 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1637 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1639 ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
1640 &sl->gb, pps, sl->slice_type_nos,
1641 h->picture_structure, h->avctx);
1645 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1646 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1648 sl->ref_count[1] = sl->ref_count[0] = 0;
1653 if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1654 (pps->weighted_bipred_idc == 1 &&
1655 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1656 ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
1657 sl->slice_type_nos, &sl->pwt, h->avctx);
1658 else if (pps->weighted_bipred_idc == 2 &&
1659 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1660 implicit_weight_table(h, sl, -1);
1662 sl->pwt.use_weight = 0;
1663 for (i = 0; i < 2; i++) {
1664 sl->pwt.luma_weight_flag[i] = 0;
1665 sl->pwt.chroma_weight_flag[i] = 0;
1669 // If frame-mt is enabled, only update mmco tables for the first slice
1670 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1671 // or h->mmco, which will cause ref list mix-ups and decoding errors
1672 // further down the line. This may break decoding if the first slice is
1673 // corrupt, thus we only do this if frame-mt is enabled.
1674 if (h->nal_ref_idc) {
1675 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1676 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1677 h->current_slice == 0);
1678 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1679 return AVERROR_INVALIDDATA;
1682 if (FRAME_MBAFF(h)) {
1683 ff_h264_fill_mbaff_ref_list(sl);
1685 if (pps->weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1686 implicit_weight_table(h, sl, 0);
1687 implicit_weight_table(h, sl, 1);
1691 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1692 ff_h264_direct_dist_scale_factor(h, sl);
1693 ff_h264_direct_ref_list_init(h, sl);
1695 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
1696 tmp = get_ue_golomb_31(&sl->gb);
1698 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1699 return AVERROR_INVALIDDATA;
1701 sl->cabac_init_idc = tmp;
1704 sl->last_qscale_diff = 0;
1705 tmp = pps->init_qp + get_se_golomb(&sl->gb);
1706 if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
1707 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1708 return AVERROR_INVALIDDATA;
1711 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1712 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1713 // FIXME qscale / qp ... stuff
1714 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1715 get_bits1(&sl->gb); /* sp_for_switch_flag */
1716 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1717 sl->slice_type == AV_PICTURE_TYPE_SI)
1718 get_se_golomb(&sl->gb); /* slice_qs_delta */
1720 sl->deblocking_filter = 1;
1721 sl->slice_alpha_c0_offset = 0;
1722 sl->slice_beta_offset = 0;
1723 if (pps->deblocking_filter_parameters_present) {
1724 tmp = get_ue_golomb_31(&sl->gb);
1726 av_log(h->avctx, AV_LOG_ERROR,
1727 "deblocking_filter_idc %u out of range\n", tmp);
1728 return AVERROR_INVALIDDATA;
1730 sl->deblocking_filter = tmp;
1731 if (sl->deblocking_filter < 2)
1732 sl->deblocking_filter ^= 1; // 1<->0
1734 if (sl->deblocking_filter) {
1735 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1736 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1737 if (sl->slice_alpha_c0_offset > 12 ||
1738 sl->slice_alpha_c0_offset < -12 ||
1739 sl->slice_beta_offset > 12 ||
1740 sl->slice_beta_offset < -12) {
1741 av_log(h->avctx, AV_LOG_ERROR,
1742 "deblocking filter parameters %d %d out of range\n",
1743 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1744 return AVERROR_INVALIDDATA;
1749 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1750 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1751 h->nal_unit_type != NAL_IDR_SLICE) ||
1752 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1753 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1754 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1755 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1756 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1757 h->nal_ref_idc == 0))
1758 sl->deblocking_filter = 0;
1760 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1761 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1762 /* Cheat slightly for speed:
1763 * Do not bother to deblock across slices. */
1764 sl->deblocking_filter = 2;
1766 h->max_contexts = 1;
1767 if (!h->single_decode_warning) {
1768 av_log(h->avctx, AV_LOG_INFO,
1769 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1770 "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"
1771 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1772 "but will generate non-standard-compliant output.\n");
1773 h->single_decode_warning = 1;
1775 if (sl != h->slice_ctx) {
1776 av_log(h->avctx, AV_LOG_ERROR,
1777 "Deblocking switched inside frame.\n");
1778 return SLICE_SINGLETHREAD;
1782 sl->qp_thresh = 15 -
1783 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1785 pps->chroma_qp_index_offset[0],
1786 pps->chroma_qp_index_offset[1]) +
1787 6 * (sps->bit_depth_luma - 8);
1789 sl->slice_num = ++h->current_slice;
1792 h->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= sl->resync_mb_y;
1793 if ( h->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= sl->resync_mb_y
1794 && h->slice_row[sl->slice_num&(MAX_SLICES-1)] <= sl->resync_mb_y
1795 && sl->slice_num >= MAX_SLICES) {
1796 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1797 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);
1800 for (j = 0; j < 2; j++) {
1802 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1803 for (i = 0; i < 16; i++) {
1805 if (j < sl->list_count && i < sl->ref_count[j] &&
1806 sl->ref_list[j][i].parent->f->buf[0]) {
1808 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1809 for (k = 0; k < h->short_ref_count; k++)
1810 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1814 for (k = 0; k < h->long_ref_count; k++)
1815 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1816 id_list[i] = h->short_ref_count + k;
1824 for (i = 0; i < 16; i++)
1825 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1827 ref2frm[18 + 1] = -1;
1828 for (i = 16; i < 48; i++)
1829 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1830 (sl->ref_list[j][i].reference & 3);
1833 h->au_pps_id = pps_id;
1834 h->current_sps_id = h->ps.pps->sps_id;
1836 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1837 av_log(h->avctx, AV_LOG_DEBUG,
1838 "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",
1840 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1842 av_get_picture_type_char(sl->slice_type),
1843 sl->slice_type_fixed ? " fix" : "",
1844 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1845 pps_id, h->poc.frame_num,
1846 h->cur_pic_ptr->field_poc[0],
1847 h->cur_pic_ptr->field_poc[1],
1848 sl->ref_count[0], sl->ref_count[1],
1850 sl->deblocking_filter,
1851 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1853 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1854 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1860 int ff_h264_get_slice_type(const H264SliceContext *sl)
1862 switch (sl->slice_type) {
1863 case AV_PICTURE_TYPE_P:
1865 case AV_PICTURE_TYPE_B:
1867 case AV_PICTURE_TYPE_I:
1869 case AV_PICTURE_TYPE_SP:
1871 case AV_PICTURE_TYPE_SI:
1874 return AVERROR_INVALIDDATA;
1878 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1879 H264SliceContext *sl,
1880 int mb_type, int top_xy,
1881 int left_xy[LEFT_MBS],
1883 int left_type[LEFT_MBS],
1884 int mb_xy, int list)
1886 int b_stride = h->b_stride;
1887 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1888 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1889 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1890 if (USES_LIST(top_type, list)) {
1891 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1892 const int b8_xy = 4 * top_xy + 2;
1893 int *ref2frm = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1894 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1895 ref_cache[0 - 1 * 8] =
1896 ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
1897 ref_cache[2 - 1 * 8] =
1898 ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
1900 AV_ZERO128(mv_dst - 1 * 8);
1901 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1904 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1905 if (USES_LIST(left_type[LTOP], list)) {
1906 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1907 const int b8_xy = 4 * left_xy[LTOP] + 1;
1908 int *ref2frm = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1909 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1910 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1911 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1912 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1914 ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1915 ref_cache[-1 + 16] =
1916 ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1918 AV_ZERO32(mv_dst - 1 + 0);
1919 AV_ZERO32(mv_dst - 1 + 8);
1920 AV_ZERO32(mv_dst - 1 + 16);
1921 AV_ZERO32(mv_dst - 1 + 24);
1924 ref_cache[-1 + 16] =
1925 ref_cache[-1 + 24] = LIST_NOT_USED;
1930 if (!USES_LIST(mb_type, list)) {
1931 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1932 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1933 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1934 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1935 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1940 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1941 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list] + (MB_MBAFF(sl) ? 20 : 2);
1942 uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
1943 uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
1944 AV_WN32A(&ref_cache[0 * 8], ref01);
1945 AV_WN32A(&ref_cache[1 * 8], ref01);
1946 AV_WN32A(&ref_cache[2 * 8], ref23);
1947 AV_WN32A(&ref_cache[3 * 8], ref23);
1951 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1952 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1953 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1954 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1955 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1960 * @return non zero if the loop filter can be skipped
1962 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
1964 const int mb_xy = sl->mb_xy;
1965 int top_xy, left_xy[LEFT_MBS];
1966 int top_type, left_type[LEFT_MBS];
1970 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1972 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1973 * stuff, I can't imagine that these complex rules are worth it. */
1975 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1976 if (FRAME_MBAFF(h)) {
1977 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1978 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1980 if (left_mb_field_flag != curr_mb_field_flag)
1981 left_xy[LTOP] -= h->mb_stride;
1983 if (curr_mb_field_flag)
1984 top_xy += h->mb_stride &
1985 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1986 if (left_mb_field_flag != curr_mb_field_flag)
1987 left_xy[LBOT] += h->mb_stride;
1991 sl->top_mb_xy = top_xy;
1992 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1993 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1995 /* For sufficiently low qp, filtering wouldn't do anything.
1996 * This is a conservative estimate: could also check beta_offset
1997 * and more accurate chroma_qp. */
1998 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1999 int qp = h->cur_pic.qscale_table[mb_xy];
2000 if (qp <= qp_thresh &&
2001 (left_xy[LTOP] < 0 ||
2002 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2004 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2005 if (!FRAME_MBAFF(h))
2007 if ((left_xy[LTOP] < 0 ||
2008 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2009 (top_xy < h->mb_stride ||
2010 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2015 top_type = h->cur_pic.mb_type[top_xy];
2016 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2017 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2018 if (sl->deblocking_filter == 2) {
2019 if (h->slice_table[top_xy] != sl->slice_num)
2021 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2022 left_type[LTOP] = left_type[LBOT] = 0;
2024 if (h->slice_table[top_xy] == 0xFFFF)
2026 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2027 left_type[LTOP] = left_type[LBOT] = 0;
2029 sl->top_type = top_type;
2030 sl->left_type[LTOP] = left_type[LTOP];
2031 sl->left_type[LBOT] = left_type[LBOT];
2033 if (IS_INTRA(mb_type))
2036 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2037 top_type, left_type, mb_xy, 0);
2038 if (sl->list_count == 2)
2039 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2040 top_type, left_type, mb_xy, 1);
2042 nnz = h->non_zero_count[mb_xy];
2043 nnz_cache = sl->non_zero_count_cache;
2044 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2045 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2046 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2047 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2048 sl->cbp = h->cbp_table[mb_xy];
2051 nnz = h->non_zero_count[top_xy];
2052 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2055 if (left_type[LTOP]) {
2056 nnz = h->non_zero_count[left_xy[LTOP]];
2057 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2058 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2059 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2060 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2063 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2064 * from what the loop filter needs */
2065 if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
2066 if (IS_8x8DCT(top_type)) {
2067 nnz_cache[4 + 8 * 0] =
2068 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2069 nnz_cache[6 + 8 * 0] =
2070 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2072 if (IS_8x8DCT(left_type[LTOP])) {
2073 nnz_cache[3 + 8 * 1] =
2074 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2076 if (IS_8x8DCT(left_type[LBOT])) {
2077 nnz_cache[3 + 8 * 3] =
2078 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2081 if (IS_8x8DCT(mb_type)) {
2082 nnz_cache[scan8[0]] =
2083 nnz_cache[scan8[1]] =
2084 nnz_cache[scan8[2]] =
2085 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2087 nnz_cache[scan8[0 + 4]] =
2088 nnz_cache[scan8[1 + 4]] =
2089 nnz_cache[scan8[2 + 4]] =
2090 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2092 nnz_cache[scan8[0 + 8]] =
2093 nnz_cache[scan8[1 + 8]] =
2094 nnz_cache[scan8[2 + 8]] =
2095 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2097 nnz_cache[scan8[0 + 12]] =
2098 nnz_cache[scan8[1 + 12]] =
2099 nnz_cache[scan8[2 + 12]] =
2100 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2107 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2109 uint8_t *dest_y, *dest_cb, *dest_cr;
2110 int linesize, uvlinesize, mb_x, mb_y;
2111 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2112 const int old_slice_type = sl->slice_type;
2113 const int pixel_shift = h->pixel_shift;
2114 const int block_h = 16 >> h->chroma_y_shift;
2116 if (sl->deblocking_filter) {
2117 for (mb_x = start_x; mb_x < end_x; mb_x++)
2118 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2120 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2121 sl->slice_num = h->slice_table[mb_xy];
2122 mb_type = h->cur_pic.mb_type[mb_xy];
2123 sl->list_count = h->list_counts[mb_xy];
2127 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2131 dest_y = h->cur_pic.f->data[0] +
2132 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2133 dest_cb = h->cur_pic.f->data[1] +
2134 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2135 mb_y * sl->uvlinesize * block_h;
2136 dest_cr = h->cur_pic.f->data[2] +
2137 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2138 mb_y * sl->uvlinesize * block_h;
2139 // FIXME simplify above
2142 linesize = sl->mb_linesize = sl->linesize * 2;
2143 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2144 if (mb_y & 1) { // FIXME move out of this function?
2145 dest_y -= sl->linesize * 15;
2146 dest_cb -= sl->uvlinesize * (block_h - 1);
2147 dest_cr -= sl->uvlinesize * (block_h - 1);
2150 linesize = sl->mb_linesize = sl->linesize;
2151 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2153 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2155 if (fill_filter_caches(h, sl, mb_type))
2157 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2158 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2160 if (FRAME_MBAFF(h)) {
2161 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2162 linesize, uvlinesize);
2164 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2165 dest_cr, linesize, uvlinesize);
2169 sl->slice_type = old_slice_type;
2171 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2172 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2173 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2176 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2178 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2179 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2180 h->cur_pic.mb_type[mb_xy - 1] :
2181 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2182 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2183 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2187 * Draw edges and report progress for the last MB row.
2189 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2191 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2192 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2193 int height = 16 << FRAME_MBAFF(h);
2194 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2196 if (sl->deblocking_filter) {
2197 if ((top + height) >= pic_height)
2198 height += deblock_border;
2199 top -= deblock_border;
2202 if (top >= pic_height || (top + height) < 0)
2205 height = FFMIN(height, pic_height - top);
2207 height = top + height;
2211 ff_h264_draw_horiz_band(h, sl, top, height);
2213 if (h->droppable || sl->h264->slice_ctx[0].er.error_occurred)
2216 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2217 h->picture_structure == PICT_BOTTOM_FIELD);
2220 static void er_add_slice(H264SliceContext *sl,
2221 int startx, int starty,
2222 int endx, int endy, int status)
2224 if (!sl->h264->enable_er)
2227 if (CONFIG_ERROR_RESILIENCE) {
2228 ERContext *er = &sl->h264->slice_ctx[0].er;
2230 ff_er_add_slice(er, startx, starty, endx, endy, status);
2234 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2236 H264SliceContext *sl = arg;
2237 const H264Context *h = sl->h264;
2238 int lf_x_start = sl->mb_x;
2241 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2242 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2244 ret = alloc_scratch_buffers(sl, sl->linesize);
2248 sl->mb_skip_run = -1;
2250 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * sl->linesize * ((scan8[15] - scan8[0]) >> 3));
2252 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2253 avctx->codec_id != AV_CODEC_ID_H264 ||
2254 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2256 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->slice_ctx[0].er.error_status_table) {
2257 const int start_i = av_clip(sl->resync_mb_x + sl->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2259 int prev_status = h->slice_ctx[0].er.error_status_table[h->slice_ctx[0].er.mb_index2xy[start_i - 1]];
2260 prev_status &= ~ VP_START;
2261 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2262 h->slice_ctx[0].er.error_occurred = 1;
2266 if (h->ps.pps->cabac) {
2268 align_get_bits(&sl->gb);
2271 ret = ff_init_cabac_decoder(&sl->cabac,
2272 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2273 (get_bits_left(&sl->gb) + 7) / 8);
2277 ff_h264_init_cabac_states(h, sl);
2282 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2283 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2284 sl->next_slice_idx);
2285 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2286 sl->mb_y, ER_MB_ERROR);
2287 return AVERROR_INVALIDDATA;
2290 ret = ff_h264_decode_mb_cabac(h, sl);
2291 // STOP_TIMER("decode_mb_cabac")
2294 ff_h264_hl_decode_mb(h, sl);
2296 // FIXME optimal? or let mb_decode decode 16x32 ?
2297 if (ret >= 0 && FRAME_MBAFF(h)) {
2300 ret = ff_h264_decode_mb_cabac(h, sl);
2303 ff_h264_hl_decode_mb(h, sl);
2306 eos = get_cabac_terminate(&sl->cabac);
2308 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2309 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2310 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2311 sl->mb_y, ER_MB_END);
2312 if (sl->mb_x >= lf_x_start)
2313 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2316 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2317 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2318 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2319 av_log(h->avctx, AV_LOG_ERROR,
2320 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2322 sl->cabac.bytestream_end - sl->cabac.bytestream);
2323 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2324 sl->mb_y, ER_MB_ERROR);
2325 return AVERROR_INVALIDDATA;
2328 if (++sl->mb_x >= h->mb_width) {
2329 loop_filter(h, sl, lf_x_start, sl->mb_x);
2330 sl->mb_x = lf_x_start = 0;
2331 decode_finish_row(h, sl);
2333 if (FIELD_OR_MBAFF_PICTURE(h)) {
2335 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2336 predict_field_decoding_flag(h, sl);
2340 if (eos || sl->mb_y >= h->mb_height) {
2341 ff_tlog(h->avctx, "slice end %d %d\n",
2342 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2343 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2344 sl->mb_y, ER_MB_END);
2345 if (sl->mb_x > lf_x_start)
2346 loop_filter(h, sl, lf_x_start, sl->mb_x);
2354 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2355 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2356 sl->next_slice_idx);
2357 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2358 sl->mb_y, ER_MB_ERROR);
2359 return AVERROR_INVALIDDATA;
2362 ret = ff_h264_decode_mb_cavlc(h, sl);
2365 ff_h264_hl_decode_mb(h, sl);
2367 // FIXME optimal? or let mb_decode decode 16x32 ?
2368 if (ret >= 0 && FRAME_MBAFF(h)) {
2370 ret = ff_h264_decode_mb_cavlc(h, sl);
2373 ff_h264_hl_decode_mb(h, sl);
2378 av_log(h->avctx, AV_LOG_ERROR,
2379 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2380 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2381 sl->mb_y, ER_MB_ERROR);
2385 if (++sl->mb_x >= h->mb_width) {
2386 loop_filter(h, sl, lf_x_start, sl->mb_x);
2387 sl->mb_x = lf_x_start = 0;
2388 decode_finish_row(h, sl);
2390 if (FIELD_OR_MBAFF_PICTURE(h)) {
2392 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2393 predict_field_decoding_flag(h, sl);
2395 if (sl->mb_y >= h->mb_height) {
2396 ff_tlog(h->avctx, "slice end %d %d\n",
2397 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2399 if ( get_bits_left(&sl->gb) == 0
2400 || get_bits_left(&sl->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2401 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2402 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2406 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2407 sl->mb_x, sl->mb_y, ER_MB_END);
2409 return AVERROR_INVALIDDATA;
2414 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2415 ff_tlog(h->avctx, "slice end %d %d\n",
2416 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2418 if (get_bits_left(&sl->gb) == 0) {
2419 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2420 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2421 if (sl->mb_x > lf_x_start)
2422 loop_filter(h, sl, lf_x_start, sl->mb_x);
2426 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2427 sl->mb_y, ER_MB_ERROR);
2429 return AVERROR_INVALIDDATA;
2437 * Call decode_slice() for each context.
2439 * @param h h264 master context
2440 * @param context_count number of contexts to execute
2442 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2444 AVCodecContext *const avctx = h->avctx;
2445 H264SliceContext *sl;
2448 av_assert0(context_count && h->slice_ctx[context_count - 1].mb_y < h->mb_height);
2450 h->slice_ctx[0].next_slice_idx = INT_MAX;
2452 if (h->avctx->hwaccel
2453 #if FF_API_CAP_VDPAU
2454 || h->avctx->codec->capabilities & AV_CODEC_CAP_HWACCEL_VDPAU
2458 if (context_count == 1) {
2461 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2463 ret = decode_slice(avctx, &h->slice_ctx[0]);
2464 h->mb_y = h->slice_ctx[0].mb_y;
2467 av_assert0(context_count > 0);
2468 for (i = 0; i < context_count; i++) {
2469 int next_slice_idx = h->mb_width * h->mb_height;
2472 sl = &h->slice_ctx[i];
2473 if (CONFIG_ERROR_RESILIENCE) {
2474 sl->er.error_count = 0;
2477 /* make sure none of those slices overlap */
2478 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2479 for (j = 0; j < context_count; j++) {
2480 H264SliceContext *sl2 = &h->slice_ctx[j];
2481 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2483 if (i == j || slice_idx2 < slice_idx)
2485 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2487 sl->next_slice_idx = next_slice_idx;
2490 avctx->execute(avctx, decode_slice, h->slice_ctx,
2491 NULL, context_count, sizeof(h->slice_ctx[0]));
2493 /* pull back stuff from slices to master context */
2494 sl = &h->slice_ctx[context_count - 1];
2496 if (CONFIG_ERROR_RESILIENCE) {
2497 for (i = 1; i < context_count; i++)
2498 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;