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 Libav.
7 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG-4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/avassert.h"
29 #include "libavutil/display.h"
30 #include "libavutil/imgutils.h"
31 #include "libavutil/stereo3d.h"
32 #include "libavutil/timer.h"
35 #include "cabac_functions.h"
36 #include "error_resilience.h"
41 #include "h264chroma.h"
42 #include "h264_mvpred.h"
46 #include "mpegutils.h"
47 #include "rectangle.h"
50 static const uint8_t field_scan[16] = {
51 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
52 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
53 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
54 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
57 static const uint8_t field_scan8x8[64] = {
58 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
59 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
60 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
61 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
62 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
63 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
64 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
65 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
66 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
67 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
68 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
69 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
70 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
71 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
72 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
73 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
76 static const uint8_t field_scan8x8_cavlc[64] = {
77 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
78 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
79 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
80 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
81 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
82 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
83 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
84 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
85 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
86 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
87 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
88 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
89 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
90 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
91 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
92 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
95 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
96 static const uint8_t zigzag_scan8x8_cavlc[64] = {
97 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
98 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
99 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
100 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
101 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
102 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
103 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
104 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
105 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
106 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
107 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
108 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
109 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
110 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
111 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
112 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
115 static void release_unused_pictures(H264Context *h, int remove_current)
119 /* release non reference frames */
120 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
121 if (h->DPB[i].f->buf[0] && !h->DPB[i].reference &&
122 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
123 ff_h264_unref_picture(h, &h->DPB[i]);
128 static int alloc_scratch_buffers(H264SliceContext *sl, int linesize)
130 const H264Context *h = sl->h264;
131 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
133 av_fast_malloc(&sl->bipred_scratchpad, &sl->bipred_scratchpad_allocated, 16 * 6 * alloc_size);
134 // edge emu needs blocksize + filter length - 1
135 // (= 21x21 for H.264)
136 av_fast_malloc(&sl->edge_emu_buffer, &sl->edge_emu_buffer_allocated, alloc_size * 2 * 21);
138 av_fast_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
139 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
140 av_fast_malloc(&sl->top_borders[1], &sl->top_borders_allocated[1],
141 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
143 if (!sl->bipred_scratchpad || !sl->edge_emu_buffer ||
144 !sl->top_borders[0] || !sl->top_borders[1]) {
145 av_freep(&sl->bipred_scratchpad);
146 av_freep(&sl->edge_emu_buffer);
147 av_freep(&sl->top_borders[0]);
148 av_freep(&sl->top_borders[1]);
150 sl->bipred_scratchpad_allocated = 0;
151 sl->edge_emu_buffer_allocated = 0;
152 sl->top_borders_allocated[0] = 0;
153 sl->top_borders_allocated[1] = 0;
154 return AVERROR(ENOMEM);
160 static int init_table_pools(H264Context *h)
162 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
163 const int mb_array_size = h->mb_stride * h->mb_height;
164 const int b4_stride = h->mb_width * 4 + 1;
165 const int b4_array_size = b4_stride * h->mb_height * 4;
167 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
169 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
170 sizeof(uint32_t), av_buffer_allocz);
171 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
172 sizeof(int16_t), av_buffer_allocz);
173 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
175 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
176 !h->ref_index_pool) {
177 av_buffer_pool_uninit(&h->qscale_table_pool);
178 av_buffer_pool_uninit(&h->mb_type_pool);
179 av_buffer_pool_uninit(&h->motion_val_pool);
180 av_buffer_pool_uninit(&h->ref_index_pool);
181 return AVERROR(ENOMEM);
187 static int alloc_picture(H264Context *h, H264Picture *pic)
191 av_assert0(!pic->f->data[0]);
194 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
195 AV_GET_BUFFER_FLAG_REF : 0);
199 if (h->avctx->hwaccel) {
200 const AVHWAccel *hwaccel = h->avctx->hwaccel;
201 av_assert0(!pic->hwaccel_picture_private);
202 if (hwaccel->frame_priv_data_size) {
203 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
204 if (!pic->hwaccel_priv_buf)
205 return AVERROR(ENOMEM);
206 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
210 if (!h->qscale_table_pool) {
211 ret = init_table_pools(h);
216 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
217 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
218 if (!pic->qscale_table_buf || !pic->mb_type_buf)
221 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
222 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
224 for (i = 0; i < 2; i++) {
225 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
226 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
227 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
230 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
231 pic->ref_index[i] = pic->ref_index_buf[i]->data;
236 ff_h264_unref_picture(h, pic);
237 return (ret < 0) ? ret : AVERROR(ENOMEM);
240 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
247 static int find_unused_picture(H264Context *h)
251 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
252 if (pic_is_unused(h, &h->DPB[i]))
255 if (i == H264_MAX_PICTURE_COUNT)
256 return AVERROR_INVALIDDATA;
261 static int initialize_cur_frame(H264Context *h)
266 release_unused_pictures(h, 1);
267 ff_h264_unref_picture(h, &h->cur_pic);
268 h->cur_pic_ptr = NULL;
270 ret = find_unused_picture(h);
272 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
277 ret = alloc_picture(h, cur);
281 ret = ff_h264_ref_picture(h, &h->cur_pic, cur);
284 h->cur_pic_ptr = cur;
289 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
291 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
292 ((pic && pic >= old_ctx->DPB && \
293 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
294 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
296 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
297 H264Context *new_base,
298 H264Context *old_base)
302 for (i = 0; i < count; i++) {
303 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
304 IN_RANGE(from[i], old_base->DPB,
305 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
307 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
311 static int h264_slice_header_init(H264Context *h);
313 int ff_h264_update_thread_context(AVCodecContext *dst,
314 const AVCodecContext *src)
316 H264Context *h = dst->priv_data, *h1 = src->priv_data;
317 int inited = h->context_initialized, err = 0;
321 if (dst == src || !h1->context_initialized)
325 return AVERROR_INVALIDDATA;
328 (h->width != h1->width ||
329 h->height != h1->height ||
330 h->mb_width != h1->mb_width ||
331 h->mb_height != h1->mb_height ||
333 h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
334 h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
335 h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
340 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
341 av_buffer_unref(&h->ps.sps_list[i]);
342 if (h1->ps.sps_list[i]) {
343 h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
344 if (!h->ps.sps_list[i])
345 return AVERROR(ENOMEM);
348 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
349 av_buffer_unref(&h->ps.pps_list[i]);
350 if (h1->ps.pps_list[i]) {
351 h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
352 if (!h->ps.pps_list[i])
353 return AVERROR(ENOMEM);
357 h->ps.sps = h1->ps.sps;
359 if (need_reinit || !inited) {
360 h->width = h1->width;
361 h->height = h1->height;
362 h->mb_height = h1->mb_height;
363 h->mb_width = h1->mb_width;
364 h->mb_num = h1->mb_num;
365 h->mb_stride = h1->mb_stride;
366 h->b_stride = h1->b_stride;
368 if ((err = h264_slice_header_init(h)) < 0) {
369 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
373 /* copy block_offset since frame_start may not be called */
374 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
377 h->avctx->coded_height = h1->avctx->coded_height;
378 h->avctx->coded_width = h1->avctx->coded_width;
379 h->avctx->width = h1->avctx->width;
380 h->avctx->height = h1->avctx->height;
381 h->coded_picture_number = h1->coded_picture_number;
382 h->first_field = h1->first_field;
383 h->picture_structure = h1->picture_structure;
384 h->mb_aff_frame = h1->mb_aff_frame;
385 h->droppable = h1->droppable;
387 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
388 ff_h264_unref_picture(h, &h->DPB[i]);
389 if (h1->DPB[i].f->buf[0] &&
390 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
394 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
395 ff_h264_unref_picture(h, &h->cur_pic);
396 if (h1->cur_pic.f->buf[0]) {
397 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
402 h->enable_er = h1->enable_er;
403 h->workaround_bugs = h1->workaround_bugs;
404 h->droppable = h1->droppable;
406 // extradata/NAL handling
407 h->is_avc = h1->is_avc;
408 h->nal_length_size = h1->nal_length_size;
410 memcpy(&h->poc, &h1->poc, sizeof(h->poc));
412 memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
413 memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
414 memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
415 memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
417 h->next_output_pic = h1->next_output_pic;
418 h->next_outputed_poc = h1->next_outputed_poc;
420 memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
421 h->nb_mmco = h1->nb_mmco;
422 h->mmco_reset = h1->mmco_reset;
423 h->explicit_ref_marking = h1->explicit_ref_marking;
424 h->long_ref_count = h1->long_ref_count;
425 h->short_ref_count = h1->short_ref_count;
427 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
428 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
429 copy_picture_range(h->delayed_pic, h1->delayed_pic,
430 MAX_DELAYED_PIC_COUNT + 2, h, h1);
436 err = ff_h264_execute_ref_pic_marking(h);
437 h->poc.prev_poc_msb = h->poc.poc_msb;
438 h->poc.prev_poc_lsb = h->poc.poc_lsb;
440 h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
441 h->poc.prev_frame_num = h->poc.frame_num;
443 h->recovery_frame = h1->recovery_frame;
444 h->frame_recovered = h1->frame_recovered;
449 static int h264_frame_start(H264Context *h)
453 const int pixel_shift = h->pixel_shift;
455 ret = initialize_cur_frame(h);
459 pic = h->cur_pic_ptr;
460 pic->reference = h->droppable ? 0 : h->picture_structure;
461 pic->f->coded_picture_number = h->coded_picture_number++;
462 pic->field_picture = h->picture_structure != PICT_FRAME;
463 pic->frame_num = h->poc.frame_num;
465 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
467 * See decode_nal_units().
469 pic->f->key_frame = 0;
473 pic->f->pict_type = h->slice_ctx[0].slice_type;
475 if (CONFIG_ERROR_RESILIENCE && h->enable_er)
476 ff_er_frame_start(&h->slice_ctx[0].er);
478 for (i = 0; i < 16; i++) {
479 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
480 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
482 for (i = 0; i < 16; i++) {
483 h->block_offset[16 + i] =
484 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
485 h->block_offset[48 + 16 + i] =
486 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
489 /* Some macroblocks can be accessed before they're available in case
490 * of lost slices, MBAFF or threading. */
491 memset(h->slice_table, -1,
492 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
494 /* We mark the current picture as non-reference after allocating it, so
495 * that if we break out due to an error it can be released automatically
496 * in the next ff_mpv_frame_start().
498 h->cur_pic_ptr->reference = 0;
500 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
502 h->next_output_pic = NULL;
504 h->postpone_filter = 0;
506 h->mb_aff_frame = h->ps.sps->mb_aff && (h->picture_structure == PICT_FRAME);
508 assert(h->cur_pic_ptr->long_ref == 0);
513 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
515 uint8_t *src_cb, uint8_t *src_cr,
516 int linesize, int uvlinesize,
521 const int pixel_shift = h->pixel_shift;
522 int chroma444 = CHROMA444(h);
523 int chroma422 = CHROMA422(h);
526 src_cb -= uvlinesize;
527 src_cr -= uvlinesize;
529 if (!simple && FRAME_MBAFF(h)) {
532 top_border = sl->top_borders[0][sl->mb_x];
533 AV_COPY128(top_border, src_y + 15 * linesize);
535 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
536 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
539 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
540 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
541 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
542 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
544 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
545 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
547 } else if (chroma422) {
549 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
550 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
552 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
553 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
557 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
558 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
560 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
561 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
566 } else if (MB_MBAFF(sl)) {
572 top_border = sl->top_borders[top_idx][sl->mb_x];
573 /* There are two lines saved, the line above the top macroblock
574 * of a pair, and the line above the bottom macroblock. */
575 AV_COPY128(top_border, src_y + 16 * linesize);
577 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
579 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
582 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
583 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
584 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
585 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
587 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
588 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
590 } else if (chroma422) {
592 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
593 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
595 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
596 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
600 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
601 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
603 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
604 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
611 * Initialize implicit_weight table.
612 * @param field 0/1 initialize the weight for interlaced MBAFF
613 * -1 initializes the rest
615 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
617 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
619 for (i = 0; i < 2; i++) {
620 sl->pwt.luma_weight_flag[i] = 0;
621 sl->pwt.chroma_weight_flag[i] = 0;
625 if (h->picture_structure == PICT_FRAME) {
626 cur_poc = h->cur_pic_ptr->poc;
628 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
630 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
631 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
632 sl->pwt.use_weight = 0;
633 sl->pwt.use_weight_chroma = 0;
637 ref_count0 = sl->ref_count[0];
638 ref_count1 = sl->ref_count[1];
640 cur_poc = h->cur_pic_ptr->field_poc[field];
642 ref_count0 = 16 + 2 * sl->ref_count[0];
643 ref_count1 = 16 + 2 * sl->ref_count[1];
646 sl->pwt.use_weight = 2;
647 sl->pwt.use_weight_chroma = 2;
648 sl->pwt.luma_log2_weight_denom = 5;
649 sl->pwt.chroma_log2_weight_denom = 5;
651 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
652 int poc0 = sl->ref_list[0][ref0].poc;
653 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
655 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
656 int poc1 = sl->ref_list[1][ref1].poc;
657 int td = av_clip_int8(poc1 - poc0);
659 int tb = av_clip_int8(cur_poc - poc0);
660 int tx = (16384 + (FFABS(td) >> 1)) / td;
661 int dist_scale_factor = (tb * tx + 32) >> 8;
662 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
663 w = 64 - dist_scale_factor;
667 sl->pwt.implicit_weight[ref0][ref1][0] =
668 sl->pwt.implicit_weight[ref0][ref1][1] = w;
670 sl->pwt.implicit_weight[ref0][ref1][field] = w;
677 * initialize scan tables
679 static void init_scan_tables(H264Context *h)
682 for (i = 0; i < 16; i++) {
683 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
684 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
685 h->field_scan[i] = TRANSPOSE(field_scan[i]);
688 for (i = 0; i < 64; i++) {
689 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
690 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
691 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
692 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
693 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
696 if (h->ps.sps->transform_bypass) { // FIXME same ugly
697 h->zigzag_scan_q0 = ff_zigzag_scan;
698 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
699 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
700 h->field_scan_q0 = field_scan;
701 h->field_scan8x8_q0 = field_scan8x8;
702 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
704 h->zigzag_scan_q0 = h->zigzag_scan;
705 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
706 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
707 h->field_scan_q0 = h->field_scan;
708 h->field_scan8x8_q0 = h->field_scan8x8;
709 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
713 static enum AVPixelFormat get_pixel_format(H264Context *h)
715 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
716 CONFIG_H264_D3D11VA_HWACCEL + \
717 CONFIG_H264_VAAPI_HWACCEL + \
718 (CONFIG_H264_VDA_HWACCEL * 2) + \
719 CONFIG_H264_VDPAU_HWACCEL)
720 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
721 const enum AVPixelFormat *choices = pix_fmts;
723 switch (h->ps.sps->bit_depth_luma) {
726 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
727 *fmt++ = AV_PIX_FMT_GBRP9;
729 *fmt++ = AV_PIX_FMT_YUV444P9;
730 } else if (CHROMA422(h))
731 *fmt++ = AV_PIX_FMT_YUV422P9;
733 *fmt++ = AV_PIX_FMT_YUV420P9;
737 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
738 *fmt++ = AV_PIX_FMT_GBRP10;
740 *fmt++ = AV_PIX_FMT_YUV444P10;
741 } else if (CHROMA422(h))
742 *fmt++ = AV_PIX_FMT_YUV422P10;
744 *fmt++ = AV_PIX_FMT_YUV420P10;
747 #if CONFIG_H264_VDPAU_HWACCEL
748 *fmt++ = AV_PIX_FMT_VDPAU;
751 if (h->avctx->colorspace == AVCOL_SPC_RGB)
752 *fmt++ = AV_PIX_FMT_GBRP;
753 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
754 *fmt++ = AV_PIX_FMT_YUVJ444P;
756 *fmt++ = AV_PIX_FMT_YUV444P;
757 } else if (CHROMA422(h)) {
758 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
759 *fmt++ = AV_PIX_FMT_YUVJ422P;
761 *fmt++ = AV_PIX_FMT_YUV422P;
763 #if CONFIG_H264_DXVA2_HWACCEL
764 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
766 #if CONFIG_H264_D3D11VA_HWACCEL
767 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
769 #if CONFIG_H264_VAAPI_HWACCEL
770 *fmt++ = AV_PIX_FMT_VAAPI;
772 #if CONFIG_H264_VDA_HWACCEL
773 *fmt++ = AV_PIX_FMT_VDA_VLD;
774 *fmt++ = AV_PIX_FMT_VDA;
776 if (h->avctx->codec->pix_fmts)
777 choices = h->avctx->codec->pix_fmts;
778 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
779 *fmt++ = AV_PIX_FMT_YUVJ420P;
781 *fmt++ = AV_PIX_FMT_YUV420P;
785 av_log(h->avctx, AV_LOG_ERROR,
786 "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
787 return AVERROR_INVALIDDATA;
790 *fmt = AV_PIX_FMT_NONE;
792 return ff_get_format(h->avctx, choices);
795 /* export coded and cropped frame dimensions to AVCodecContext */
796 static int init_dimensions(H264Context *h)
798 SPS *sps = h->ps.sps;
799 int width = h->width - (sps->crop_right + sps->crop_left);
800 int height = h->height - (sps->crop_top + sps->crop_bottom);
802 /* handle container cropping */
803 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
804 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16)) {
805 width = h->avctx->width;
806 height = h->avctx->height;
809 if (width <= 0 || height <= 0) {
810 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
812 if (h->avctx->err_recognition & AV_EF_EXPLODE)
813 return AVERROR_INVALIDDATA;
815 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
826 h->avctx->coded_width = h->width;
827 h->avctx->coded_height = h->height;
828 h->avctx->width = width;
829 h->avctx->height = height;
834 static int h264_slice_header_init(H264Context *h)
836 const SPS *sps = h->ps.sps;
839 ff_set_sar(h->avctx, sps->sar);
840 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
841 &h->chroma_x_shift, &h->chroma_y_shift);
843 if (sps->timing_info_present_flag) {
844 int64_t den = sps->time_scale;
845 if (h->sei.unregistered.x264_build < 44U)
847 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
848 sps->num_units_in_tick, den, 1 << 30);
851 ff_h264_free_tables(h);
854 h->prev_interlaced_frame = 1;
857 ret = ff_h264_alloc_tables(h);
859 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
863 if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 10) {
864 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
865 sps->bit_depth_luma);
866 return AVERROR_INVALIDDATA;
869 h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
870 h->pixel_shift = sps->bit_depth_luma > 8;
871 h->chroma_format_idc = sps->chroma_format_idc;
872 h->bit_depth_luma = sps->bit_depth_luma;
874 ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
875 sps->chroma_format_idc);
876 ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
877 ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
878 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
879 sps->chroma_format_idc);
880 ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
882 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
883 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
885 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
889 for (i = 0; i < h->nb_slice_ctx; i++) {
890 H264SliceContext *sl = &h->slice_ctx[i];
893 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
894 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
895 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
897 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
898 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
904 h->context_initialized = 1;
909 static int h264_init_ps(H264Context *h, const H264SliceContext *sl)
912 int needs_reinit = 0, ret;
914 h->ps.pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
915 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
916 h->ps.sps = (SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data;
918 if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
919 h->chroma_format_idc != h->ps.sps->chroma_format_idc)
924 h->avctx->profile = ff_h264_get_profile(sps);
925 h->avctx->level = sps->level_idc;
926 h->avctx->refs = sps->ref_frame_count;
928 if (h->mb_width != sps->mb_width ||
929 h->mb_height != sps->mb_height * (2 - sps->frame_mbs_only_flag))
932 h->mb_width = sps->mb_width;
933 h->mb_height = sps->mb_height * (2 - sps->frame_mbs_only_flag);
934 h->mb_num = h->mb_width * h->mb_height;
935 h->mb_stride = h->mb_width + 1;
937 h->b_stride = h->mb_width * 4;
939 h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
941 h->width = 16 * h->mb_width;
942 h->height = 16 * h->mb_height;
944 ret = init_dimensions(h);
948 if (sps->video_signal_type_present_flag) {
949 h->avctx->color_range = sps->full_range ? AVCOL_RANGE_JPEG
951 if (sps->colour_description_present_flag) {
952 if (h->avctx->colorspace != sps->colorspace)
954 h->avctx->color_primaries = sps->color_primaries;
955 h->avctx->color_trc = sps->color_trc;
956 h->avctx->colorspace = sps->colorspace;
960 if (!h->context_initialized || needs_reinit) {
961 h->context_initialized = 0;
962 if (sl != h->slice_ctx) {
963 av_log(h->avctx, AV_LOG_ERROR,
964 "changing width %d -> %d / height %d -> %d on "
966 h->width, h->avctx->coded_width,
967 h->height, h->avctx->coded_height,
968 h->current_slice + 1);
969 return AVERROR_INVALIDDATA;
972 ff_h264_flush_change(h);
974 if ((ret = get_pixel_format(h)) < 0)
976 h->avctx->pix_fmt = ret;
978 av_log(h->avctx, AV_LOG_VERBOSE, "Reinit context to %dx%d, "
979 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
981 if ((ret = h264_slice_header_init(h)) < 0) {
982 av_log(h->avctx, AV_LOG_ERROR,
983 "h264_slice_header_init() failed\n");
991 static int h264_export_frame_props(H264Context *h)
993 const SPS *sps = h->ps.sps;
994 H264Picture *cur = h->cur_pic_ptr;
996 cur->f->interlaced_frame = 0;
997 cur->f->repeat_pict = 0;
999 /* Signal interlacing information externally. */
1000 /* Prioritize picture timing SEI information over used
1001 * decoding process if it exists. */
1003 if (sps->pic_struct_present_flag) {
1004 H264SEIPictureTiming *pt = &h->sei.picture_timing;
1005 switch (pt->pic_struct) {
1006 case SEI_PIC_STRUCT_FRAME:
1008 case SEI_PIC_STRUCT_TOP_FIELD:
1009 case SEI_PIC_STRUCT_BOTTOM_FIELD:
1010 cur->f->interlaced_frame = 1;
1012 case SEI_PIC_STRUCT_TOP_BOTTOM:
1013 case SEI_PIC_STRUCT_BOTTOM_TOP:
1014 if (FIELD_OR_MBAFF_PICTURE(h))
1015 cur->f->interlaced_frame = 1;
1017 // try to flag soft telecine progressive
1018 cur->f->interlaced_frame = h->prev_interlaced_frame;
1020 case SEI_PIC_STRUCT_TOP_BOTTOM_TOP:
1021 case SEI_PIC_STRUCT_BOTTOM_TOP_BOTTOM:
1022 /* Signal the possibility of telecined film externally
1023 * (pic_struct 5,6). From these hints, let the applications
1024 * decide if they apply deinterlacing. */
1025 cur->f->repeat_pict = 1;
1027 case SEI_PIC_STRUCT_FRAME_DOUBLING:
1028 cur->f->repeat_pict = 2;
1030 case SEI_PIC_STRUCT_FRAME_TRIPLING:
1031 cur->f->repeat_pict = 4;
1035 if ((pt->ct_type & 3) &&
1036 pt->pic_struct <= SEI_PIC_STRUCT_BOTTOM_TOP)
1037 cur->f->interlaced_frame = (pt->ct_type & (1 << 1)) != 0;
1039 /* Derive interlacing flag from used decoding process. */
1040 cur->f->interlaced_frame = FIELD_OR_MBAFF_PICTURE(h);
1042 h->prev_interlaced_frame = cur->f->interlaced_frame;
1044 if (cur->field_poc[0] != cur->field_poc[1]) {
1045 /* Derive top_field_first from field pocs. */
1046 cur->f->top_field_first = cur->field_poc[0] < cur->field_poc[1];
1048 if (cur->f->interlaced_frame || sps->pic_struct_present_flag) {
1049 /* Use picture timing SEI information. Even if it is a
1050 * information of a past frame, better than nothing. */
1051 if (h->sei.picture_timing.pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM ||
1052 h->sei.picture_timing.pic_struct == SEI_PIC_STRUCT_TOP_BOTTOM_TOP)
1053 cur->f->top_field_first = 1;
1055 cur->f->top_field_first = 0;
1057 /* Most likely progressive */
1058 cur->f->top_field_first = 0;
1062 if (h->sei.frame_packing.present &&
1063 h->sei.frame_packing.arrangement_type >= 0 &&
1064 h->sei.frame_packing.arrangement_type <= 6 &&
1065 h->sei.frame_packing.content_interpretation_type > 0 &&
1066 h->sei.frame_packing.content_interpretation_type < 3) {
1067 H264SEIFramePacking *fp = &h->sei.frame_packing;
1068 AVStereo3D *stereo = av_stereo3d_create_side_data(cur->f);
1070 return AVERROR(ENOMEM);
1072 switch (fp->arrangement_type) {
1074 stereo->type = AV_STEREO3D_CHECKERBOARD;
1077 stereo->type = AV_STEREO3D_COLUMNS;
1080 stereo->type = AV_STEREO3D_LINES;
1083 if (fp->quincunx_subsampling)
1084 stereo->type = AV_STEREO3D_SIDEBYSIDE_QUINCUNX;
1086 stereo->type = AV_STEREO3D_SIDEBYSIDE;
1089 stereo->type = AV_STEREO3D_TOPBOTTOM;
1092 stereo->type = AV_STEREO3D_FRAMESEQUENCE;
1095 stereo->type = AV_STEREO3D_2D;
1099 if (fp->content_interpretation_type == 2)
1100 stereo->flags = AV_STEREO3D_FLAG_INVERT;
1103 if (h->sei.display_orientation.present &&
1104 (h->sei.display_orientation.anticlockwise_rotation ||
1105 h->sei.display_orientation.hflip ||
1106 h->sei.display_orientation.vflip)) {
1107 H264SEIDisplayOrientation *o = &h->sei.display_orientation;
1108 double angle = o->anticlockwise_rotation * 360 / (double) (1 << 16);
1109 AVFrameSideData *rotation = av_frame_new_side_data(cur->f,
1110 AV_FRAME_DATA_DISPLAYMATRIX,
1111 sizeof(int32_t) * 9);
1113 return AVERROR(ENOMEM);
1115 av_display_rotation_set((int32_t *)rotation->data, angle);
1116 av_display_matrix_flip((int32_t *)rotation->data,
1117 o->hflip, o->vflip);
1120 if (h->sei.afd.present) {
1121 AVFrameSideData *sd = av_frame_new_side_data(cur->f, AV_FRAME_DATA_AFD,
1124 return AVERROR(ENOMEM);
1126 *sd->data = h->sei.afd.active_format_description;
1127 h->sei.afd.present = 0;
1130 if (h->sei.a53_caption.a53_caption) {
1131 H264SEIA53Caption *a53 = &h->sei.a53_caption;
1132 AVFrameSideData *sd = av_frame_new_side_data(cur->f,
1133 AV_FRAME_DATA_A53_CC,
1134 a53->a53_caption_size);
1136 return AVERROR(ENOMEM);
1138 memcpy(sd->data, a53->a53_caption, a53->a53_caption_size);
1139 av_freep(&a53->a53_caption);
1140 a53->a53_caption_size = 0;
1146 /* This function is called right after decoding the slice header for a first
1147 * slice in a field (or a frame). It decides whether we are decoding a new frame
1148 * or a second field in a pair and does the necessary setup.
1150 static int h264_field_start(H264Context *h, const H264SliceContext *sl,
1151 const H2645NAL *nal)
1155 int last_pic_structure, last_pic_droppable, ret;
1157 ret = h264_init_ps(h, sl);
1163 last_pic_droppable = h->droppable;
1164 last_pic_structure = h->picture_structure;
1165 h->droppable = (nal->ref_idc == 0);
1166 h->picture_structure = sl->picture_structure;
1168 h->poc.frame_num = sl->frame_num;
1169 h->poc.poc_lsb = sl->poc_lsb;
1170 h->poc.delta_poc_bottom = sl->delta_poc_bottom;
1171 h->poc.delta_poc[0] = sl->delta_poc[0];
1172 h->poc.delta_poc[1] = sl->delta_poc[1];
1174 /* Shorten frame num gaps so we don't have to allocate reference
1175 * frames just to throw them away */
1176 if (h->poc.frame_num != h->poc.prev_frame_num) {
1177 int unwrap_prev_frame_num = h->poc.prev_frame_num;
1178 int max_frame_num = 1 << sps->log2_max_frame_num;
1180 if (unwrap_prev_frame_num > h->poc.frame_num)
1181 unwrap_prev_frame_num -= max_frame_num;
1183 if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
1184 unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
1185 if (unwrap_prev_frame_num < 0)
1186 unwrap_prev_frame_num += max_frame_num;
1188 h->poc.prev_frame_num = unwrap_prev_frame_num;
1192 /* See if we have a decoded first field looking for a pair...
1193 * Here, we're using that to see if we should mark previously
1194 * decode frames as "finished".
1195 * We have to do that before the "dummy" in-between frame allocation,
1196 * since that can modify s->current_picture_ptr. */
1197 if (h->first_field) {
1198 assert(h->cur_pic_ptr);
1199 assert(h->cur_pic_ptr->f->buf[0]);
1200 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1202 /* figure out if we have a complementary field pair */
1203 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1204 /* Previous field is unmatched. Don't display it, but let it
1205 * remain for reference if marked as such. */
1206 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1207 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1208 last_pic_structure == PICT_TOP_FIELD);
1211 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1212 /* This and previous field were reference, but had
1213 * different frame_nums. Consider this field first in
1214 * pair. Throw away previous field except for reference
1216 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1217 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1218 last_pic_structure == PICT_TOP_FIELD);
1221 /* Second field in complementary pair */
1222 if (!((last_pic_structure == PICT_TOP_FIELD &&
1223 h->picture_structure == PICT_BOTTOM_FIELD) ||
1224 (last_pic_structure == PICT_BOTTOM_FIELD &&
1225 h->picture_structure == PICT_TOP_FIELD))) {
1226 av_log(h->avctx, AV_LOG_ERROR,
1227 "Invalid field mode combination %d/%d\n",
1228 last_pic_structure, h->picture_structure);
1229 h->picture_structure = last_pic_structure;
1230 h->droppable = last_pic_droppable;
1231 return AVERROR_INVALIDDATA;
1232 } else if (last_pic_droppable != h->droppable) {
1233 avpriv_request_sample(h->avctx,
1234 "Found reference and non-reference fields in the same frame, which");
1235 h->picture_structure = last_pic_structure;
1236 h->droppable = last_pic_droppable;
1237 return AVERROR_PATCHWELCOME;
1243 while (h->poc.frame_num != h->poc.prev_frame_num &&
1244 h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
1245 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1246 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1247 h->poc.frame_num, h->poc.prev_frame_num);
1248 ret = initialize_cur_frame(h);
1254 h->poc.prev_frame_num++;
1255 h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
1256 h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
1257 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1258 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1260 h->explicit_ref_marking = 0;
1261 ret = ff_h264_execute_ref_pic_marking(h);
1262 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1264 /* Error concealment: If a ref is missing, copy the previous ref
1266 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1267 * many assumptions about there being no actual duplicates.
1268 * FIXME: This does not copy padding for out-of-frame motion
1269 * vectors. Given we are concealing a lost frame, this probably
1270 * is not noticeable by comparison, but it should be fixed. */
1271 if (h->short_ref_count) {
1273 h->short_ref[0]->f->width == prev->f->width &&
1274 h->short_ref[0]->f->height == prev->f->height &&
1275 h->short_ref[0]->f->format == prev->f->format) {
1276 av_image_copy(h->short_ref[0]->f->data,
1277 h->short_ref[0]->f->linesize,
1278 (const uint8_t **)prev->f->data,
1283 h->short_ref[0]->poc = prev->poc + 2;
1285 h->short_ref[0]->frame_num = h->poc.prev_frame_num;
1289 /* See if we have a decoded first field looking for a pair...
1290 * We're using that to see whether to continue decoding in that
1291 * frame, or to allocate a new one. */
1292 if (h->first_field) {
1293 assert(h->cur_pic_ptr);
1294 assert(h->cur_pic_ptr->f->buf[0]);
1295 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1297 /* figure out if we have a complementary field pair */
1298 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1299 /* Previous field is unmatched. Don't display it, but let it
1300 * remain for reference if marked as such. */
1301 h->cur_pic_ptr = NULL;
1302 h->first_field = FIELD_PICTURE(h);
1304 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1305 /* This and the previous field had different frame_nums.
1306 * Consider this field first in pair. Throw away previous
1307 * one except for reference purposes. */
1309 h->cur_pic_ptr = NULL;
1311 /* Second field in complementary pair */
1316 /* Frame or first field in a potentially complementary pair */
1317 h->first_field = FIELD_PICTURE(h);
1320 if (!FIELD_PICTURE(h) || h->first_field) {
1321 if (h264_frame_start(h) < 0) {
1323 return AVERROR_INVALIDDATA;
1326 release_unused_pictures(h, 0);
1329 ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
1330 h->ps.sps, &h->poc, h->picture_structure, nal->ref_idc);
1332 memcpy(h->mmco, sl->mmco, sl->nb_mmco * sizeof(*h->mmco));
1333 h->nb_mmco = sl->nb_mmco;
1334 h->explicit_ref_marking = sl->explicit_ref_marking;
1336 h->picture_idr = nal->type == H264_NAL_IDR_SLICE;
1338 if (h->sei.recovery_point.recovery_frame_cnt >= 0 && h->recovery_frame < 0) {
1339 h->recovery_frame = (h->poc.frame_num + h->sei.recovery_point.recovery_frame_cnt) &
1340 ((1 << h->ps.sps->log2_max_frame_num) - 1);
1343 h->cur_pic_ptr->f->key_frame |= (nal->type == H264_NAL_IDR_SLICE) ||
1344 (h->sei.recovery_point.recovery_frame_cnt >= 0);
1346 if (nal->type == H264_NAL_IDR_SLICE || h->recovery_frame == h->poc.frame_num) {
1347 h->recovery_frame = -1;
1348 h->cur_pic_ptr->recovered = 1;
1350 // If we have an IDR, all frames after it in decoded order are
1352 if (nal->type == H264_NAL_IDR_SLICE)
1353 h->frame_recovered |= FRAME_RECOVERED_IDR;
1354 h->cur_pic_ptr->recovered |= !!(h->frame_recovered & FRAME_RECOVERED_IDR);
1356 /* Set the frame properties/side data. Only done for the second field in
1357 * field coded frames, since some SEI information is present for each field
1358 * and is merged by the SEI parsing code. */
1359 if (!FIELD_PICTURE(h) || !h->first_field) {
1360 ret = h264_export_frame_props(h);
1365 if (h->avctx->hwaccel) {
1366 ret = h->avctx->hwaccel->start_frame(h->avctx, NULL, 0);
1374 static int h264_slice_header_parse(H264SliceContext *sl, const H2645NAL *nal,
1375 const H264ParamSets *ps, AVCodecContext *avctx)
1380 unsigned int slice_type, tmp, i;
1381 int field_pic_flag, bottom_field_flag, picture_structure;
1383 sl->first_mb_addr = get_ue_golomb(&sl->gb);
1385 slice_type = get_ue_golomb_31(&sl->gb);
1386 if (slice_type > 9) {
1387 av_log(avctx, AV_LOG_ERROR,
1388 "slice type %d too large at %d\n",
1389 slice_type, sl->first_mb_addr);
1390 return AVERROR_INVALIDDATA;
1392 if (slice_type > 4) {
1394 sl->slice_type_fixed = 1;
1396 sl->slice_type_fixed = 0;
1398 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1399 sl->slice_type = slice_type;
1400 sl->slice_type_nos = slice_type & 3;
1402 if (nal->type == H264_NAL_IDR_SLICE &&
1403 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1404 av_log(avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1405 return AVERROR_INVALIDDATA;
1408 sl->pps_id = get_ue_golomb(&sl->gb);
1409 if (sl->pps_id >= MAX_PPS_COUNT) {
1410 av_log(avctx, AV_LOG_ERROR, "pps_id %u out of range\n", sl->pps_id);
1411 return AVERROR_INVALIDDATA;
1413 if (!ps->pps_list[sl->pps_id]) {
1414 av_log(avctx, AV_LOG_ERROR,
1415 "non-existing PPS %u referenced\n",
1417 return AVERROR_INVALIDDATA;
1419 pps = (const PPS*)ps->pps_list[sl->pps_id]->data;
1421 if (!ps->sps_list[pps->sps_id]) {
1422 av_log(avctx, AV_LOG_ERROR,
1423 "non-existing SPS %u referenced\n", pps->sps_id);
1424 return AVERROR_INVALIDDATA;
1426 sps = (const SPS*)ps->sps_list[pps->sps_id]->data;
1428 sl->frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
1432 if (sps->frame_mbs_only_flag) {
1433 picture_structure = PICT_FRAME;
1435 field_pic_flag = get_bits1(&sl->gb);
1436 if (field_pic_flag) {
1437 bottom_field_flag = get_bits1(&sl->gb);
1438 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1440 picture_structure = PICT_FRAME;
1443 sl->picture_structure = picture_structure;
1444 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1446 if (picture_structure == PICT_FRAME) {
1447 sl->curr_pic_num = sl->frame_num;
1448 sl->max_pic_num = 1 << sps->log2_max_frame_num;
1450 sl->curr_pic_num = 2 * sl->frame_num + 1;
1451 sl->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
1454 if (nal->type == H264_NAL_IDR_SLICE)
1455 get_ue_golomb(&sl->gb); /* idr_pic_id */
1457 if (sps->poc_type == 0) {
1458 sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
1460 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1461 sl->delta_poc_bottom = get_se_golomb(&sl->gb);
1464 if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
1465 sl->delta_poc[0] = get_se_golomb(&sl->gb);
1467 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1468 sl->delta_poc[1] = get_se_golomb(&sl->gb);
1471 sl->redundant_pic_count = 0;
1472 if (pps->redundant_pic_cnt_present)
1473 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1475 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1476 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1478 ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
1479 &sl->gb, pps, sl->slice_type_nos,
1484 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1485 ret = ff_h264_decode_ref_pic_list_reordering(sl, avctx);
1487 sl->ref_count[1] = sl->ref_count[0] = 0;
1492 sl->pwt.use_weight = 0;
1493 for (i = 0; i < 2; i++) {
1494 sl->pwt.luma_weight_flag[i] = 0;
1495 sl->pwt.chroma_weight_flag[i] = 0;
1497 if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1498 (pps->weighted_bipred_idc == 1 &&
1499 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1500 ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
1501 sl->slice_type_nos, &sl->pwt);
1503 sl->explicit_ref_marking = 0;
1505 ret = ff_h264_decode_ref_pic_marking(sl, &sl->gb, nal, avctx);
1506 if (ret < 0 && (avctx->err_recognition & AV_EF_EXPLODE))
1507 return AVERROR_INVALIDDATA;
1510 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
1511 tmp = get_ue_golomb_31(&sl->gb);
1513 av_log(avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1514 return AVERROR_INVALIDDATA;
1516 sl->cabac_init_idc = tmp;
1519 sl->last_qscale_diff = 0;
1520 tmp = pps->init_qp + get_se_golomb(&sl->gb);
1521 if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
1522 av_log(avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1523 return AVERROR_INVALIDDATA;
1526 sl->chroma_qp[0] = get_chroma_qp(pps, 0, sl->qscale);
1527 sl->chroma_qp[1] = get_chroma_qp(pps, 1, sl->qscale);
1528 // FIXME qscale / qp ... stuff
1529 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1530 get_bits1(&sl->gb); /* sp_for_switch_flag */
1531 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1532 sl->slice_type == AV_PICTURE_TYPE_SI)
1533 get_se_golomb(&sl->gb); /* slice_qs_delta */
1535 sl->deblocking_filter = 1;
1536 sl->slice_alpha_c0_offset = 0;
1537 sl->slice_beta_offset = 0;
1538 if (pps->deblocking_filter_parameters_present) {
1539 tmp = get_ue_golomb_31(&sl->gb);
1541 av_log(avctx, AV_LOG_ERROR,
1542 "deblocking_filter_idc %u out of range\n", tmp);
1543 return AVERROR_INVALIDDATA;
1545 sl->deblocking_filter = tmp;
1546 if (sl->deblocking_filter < 2)
1547 sl->deblocking_filter ^= 1; // 1<->0
1549 if (sl->deblocking_filter) {
1550 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1551 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1552 if (sl->slice_alpha_c0_offset > 12 ||
1553 sl->slice_alpha_c0_offset < -12 ||
1554 sl->slice_beta_offset > 12 ||
1555 sl->slice_beta_offset < -12) {
1556 av_log(avctx, AV_LOG_ERROR,
1557 "deblocking filter parameters %d %d out of range\n",
1558 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1559 return AVERROR_INVALIDDATA;
1568 * Decode a slice header.
1569 * This will (re)initialize the decoder and call h264_frame_start() as needed.
1571 * @param h h264context
1573 * @return 0 if okay, <0 if an error occurred
1575 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl,
1576 const H2645NAL *nal)
1580 ret = h264_slice_header_parse(sl, nal, &h->ps, h->avctx);
1584 // discard redundant pictures
1585 if (sl->redundant_pic_count > 0)
1588 if (!h->setup_finished) {
1589 if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
1590 if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1591 ff_h264_field_end(h, sl, 1);
1594 h->current_slice = 0;
1595 if (!h->first_field) {
1596 if (h->cur_pic_ptr && !h->droppable) {
1597 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1598 h->picture_structure == PICT_BOTTOM_FIELD);
1600 h->cur_pic_ptr = NULL;
1604 if (h->current_slice == 0) {
1605 ret = h264_field_start(h, sl, nal);
1611 if (h->current_slice > 0) {
1612 if (h->ps.pps != (const PPS*)h->ps.pps_list[sl->pps_id]->data) {
1613 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1614 return AVERROR_INVALIDDATA;
1617 if (h->picture_structure != sl->picture_structure ||
1618 h->droppable != (nal->ref_idc == 0)) {
1619 av_log(h->avctx, AV_LOG_ERROR,
1620 "Changing field mode (%d -> %d) between slices is not allowed\n",
1621 h->picture_structure, sl->picture_structure);
1622 return AVERROR_INVALIDDATA;
1623 } else if (!h->cur_pic_ptr) {
1624 av_log(h->avctx, AV_LOG_ERROR,
1625 "unset cur_pic_ptr on slice %d\n",
1626 h->current_slice + 1);
1627 return AVERROR_INVALIDDATA;
1631 if (h->picture_idr && nal->type != H264_NAL_IDR_SLICE) {
1632 av_log(h->avctx, AV_LOG_ERROR, "Invalid mix of IDR and non-IDR slices\n");
1633 return AVERROR_INVALIDDATA;
1636 assert(h->mb_num == h->mb_width * h->mb_height);
1637 if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1638 sl->first_mb_addr >= h->mb_num) {
1639 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1640 return AVERROR_INVALIDDATA;
1642 sl->resync_mb_x = sl->mb_x = sl->first_mb_addr % h->mb_width;
1643 sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
1644 FIELD_OR_MBAFF_PICTURE(h);
1645 if (h->picture_structure == PICT_BOTTOM_FIELD)
1646 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1647 assert(sl->mb_y < h->mb_height);
1649 ret = ff_h264_build_ref_list(h, sl);
1653 if (h->ps.pps->weighted_bipred_idc == 2 &&
1654 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1655 implicit_weight_table(h, sl, -1);
1656 if (FRAME_MBAFF(h)) {
1657 implicit_weight_table(h, sl, 0);
1658 implicit_weight_table(h, sl, 1);
1662 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1663 ff_h264_direct_dist_scale_factor(h, sl);
1664 ff_h264_direct_ref_list_init(h, sl);
1666 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1667 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1668 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1669 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1670 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1671 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1673 sl->deblocking_filter = 0;
1675 if (sl->deblocking_filter == 1 && h->nb_slice_ctx > 1) {
1676 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1677 /* Cheat slightly for speed:
1678 * Do not bother to deblock across slices. */
1679 sl->deblocking_filter = 2;
1681 h->postpone_filter = 1;
1684 sl->qp_thresh = 15 -
1685 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1687 h->ps.pps->chroma_qp_index_offset[0],
1688 h->ps.pps->chroma_qp_index_offset[1]) +
1689 6 * (h->ps.sps->bit_depth_luma - 8);
1691 sl->slice_num = ++h->current_slice;
1692 if (sl->slice_num >= MAX_SLICES) {
1693 av_log(h->avctx, AV_LOG_ERROR,
1694 "Too many slices, increase MAX_SLICES and recompile\n");
1697 for (j = 0; j < 2; j++) {
1699 int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1700 for (i = 0; i < 16; i++) {
1702 if (j < sl->list_count && i < sl->ref_count[j] &&
1703 sl->ref_list[j][i].parent->f->buf[0]) {
1705 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1706 for (k = 0; k < h->short_ref_count; k++)
1707 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1711 for (k = 0; k < h->long_ref_count; k++)
1712 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1713 id_list[i] = h->short_ref_count + k;
1721 for (i = 0; i < 16; i++)
1722 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1724 ref2frm[18 + 1] = -1;
1725 for (i = 16; i < 48; i++)
1726 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1727 (sl->ref_list[j][i].reference & 3);
1730 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1731 av_log(h->avctx, AV_LOG_DEBUG,
1732 "slice:%d %s mb:%d %c%s%s frame:%d poc:%d/%d ref:%d/%d qp:%d loop:%d:%d:%d weight:%d%s %s\n",
1734 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1735 sl->mb_y * h->mb_width + sl->mb_x,
1736 av_get_picture_type_char(sl->slice_type),
1737 sl->slice_type_fixed ? " fix" : "",
1738 nal->type == H264_NAL_IDR_SLICE ? " IDR" : "",
1740 h->cur_pic_ptr->field_poc[0],
1741 h->cur_pic_ptr->field_poc[1],
1742 sl->ref_count[0], sl->ref_count[1],
1744 sl->deblocking_filter,
1745 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1747 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1748 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1754 int ff_h264_get_slice_type(const H264SliceContext *sl)
1756 switch (sl->slice_type) {
1757 case AV_PICTURE_TYPE_P:
1759 case AV_PICTURE_TYPE_B:
1761 case AV_PICTURE_TYPE_I:
1763 case AV_PICTURE_TYPE_SP:
1765 case AV_PICTURE_TYPE_SI:
1768 return AVERROR_INVALIDDATA;
1772 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1773 H264SliceContext *sl,
1774 int mb_type, int top_xy,
1775 int left_xy[LEFT_MBS],
1777 int left_type[LEFT_MBS],
1778 int mb_xy, int list)
1780 int b_stride = h->b_stride;
1781 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1782 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1783 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1784 if (USES_LIST(top_type, list)) {
1785 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1786 const int b8_xy = 4 * top_xy + 2;
1787 const int *ref2frm = &h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
1788 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1789 ref_cache[0 - 1 * 8] =
1790 ref_cache[1 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 0]];
1791 ref_cache[2 - 1 * 8] =
1792 ref_cache[3 - 1 * 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 1]];
1794 AV_ZERO128(mv_dst - 1 * 8);
1795 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1798 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1799 if (USES_LIST(left_type[LTOP], list)) {
1800 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1801 const int b8_xy = 4 * left_xy[LTOP] + 1;
1802 const int *ref2frm = &h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
1803 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1804 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1805 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1806 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1808 ref_cache[-1 + 8] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1809 ref_cache[-1 + 16] =
1810 ref_cache[-1 + 24] = ref2frm[h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1812 AV_ZERO32(mv_dst - 1 + 0);
1813 AV_ZERO32(mv_dst - 1 + 8);
1814 AV_ZERO32(mv_dst - 1 + 16);
1815 AV_ZERO32(mv_dst - 1 + 24);
1818 ref_cache[-1 + 16] =
1819 ref_cache[-1 + 24] = LIST_NOT_USED;
1824 if (!USES_LIST(mb_type, list)) {
1825 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1826 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1827 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1828 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1829 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1834 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1835 const int *ref2frm = &h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][list][(MB_MBAFF(sl) ? 20 : 2)];
1836 uint32_t ref01 = (pack16to32(ref2frm[ref[0]], ref2frm[ref[1]]) & 0x00FF00FF) * 0x0101;
1837 uint32_t ref23 = (pack16to32(ref2frm[ref[2]], ref2frm[ref[3]]) & 0x00FF00FF) * 0x0101;
1838 AV_WN32A(&ref_cache[0 * 8], ref01);
1839 AV_WN32A(&ref_cache[1 * 8], ref01);
1840 AV_WN32A(&ref_cache[2 * 8], ref23);
1841 AV_WN32A(&ref_cache[3 * 8], ref23);
1845 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1846 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1847 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1848 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1849 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1854 * @return non zero if the loop filter can be skipped
1856 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
1858 const int mb_xy = sl->mb_xy;
1859 int top_xy, left_xy[LEFT_MBS];
1860 int top_type, left_type[LEFT_MBS];
1864 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1866 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1867 if (FRAME_MBAFF(h)) {
1868 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1869 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1871 if (left_mb_field_flag != curr_mb_field_flag)
1872 left_xy[LTOP] -= h->mb_stride;
1874 if (curr_mb_field_flag)
1875 top_xy += h->mb_stride &
1876 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1877 if (left_mb_field_flag != curr_mb_field_flag)
1878 left_xy[LBOT] += h->mb_stride;
1882 sl->top_mb_xy = top_xy;
1883 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1884 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1886 /* For sufficiently low qp, filtering wouldn't do anything.
1887 * This is a conservative estimate: could also check beta_offset
1888 * and more accurate chroma_qp. */
1889 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1890 int qp = h->cur_pic.qscale_table[mb_xy];
1891 if (qp <= qp_thresh &&
1892 (left_xy[LTOP] < 0 ||
1893 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1895 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1896 if (!FRAME_MBAFF(h))
1898 if ((left_xy[LTOP] < 0 ||
1899 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1900 (top_xy < h->mb_stride ||
1901 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1906 top_type = h->cur_pic.mb_type[top_xy];
1907 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1908 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1909 if (sl->deblocking_filter == 2) {
1910 if (h->slice_table[top_xy] != sl->slice_num)
1912 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
1913 left_type[LTOP] = left_type[LBOT] = 0;
1915 if (h->slice_table[top_xy] == 0xFFFF)
1917 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1918 left_type[LTOP] = left_type[LBOT] = 0;
1920 sl->top_type = top_type;
1921 sl->left_type[LTOP] = left_type[LTOP];
1922 sl->left_type[LBOT] = left_type[LBOT];
1924 if (IS_INTRA(mb_type))
1927 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1928 top_type, left_type, mb_xy, 0);
1929 if (sl->list_count == 2)
1930 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1931 top_type, left_type, mb_xy, 1);
1933 nnz = h->non_zero_count[mb_xy];
1934 nnz_cache = sl->non_zero_count_cache;
1935 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1936 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1937 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1938 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1939 sl->cbp = h->cbp_table[mb_xy];
1942 nnz = h->non_zero_count[top_xy];
1943 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1946 if (left_type[LTOP]) {
1947 nnz = h->non_zero_count[left_xy[LTOP]];
1948 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
1949 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
1950 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
1951 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
1954 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
1955 * from what the loop filter needs */
1956 if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
1957 if (IS_8x8DCT(top_type)) {
1958 nnz_cache[4 + 8 * 0] =
1959 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
1960 nnz_cache[6 + 8 * 0] =
1961 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
1963 if (IS_8x8DCT(left_type[LTOP])) {
1964 nnz_cache[3 + 8 * 1] =
1965 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
1967 if (IS_8x8DCT(left_type[LBOT])) {
1968 nnz_cache[3 + 8 * 3] =
1969 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
1972 if (IS_8x8DCT(mb_type)) {
1973 nnz_cache[scan8[0]] =
1974 nnz_cache[scan8[1]] =
1975 nnz_cache[scan8[2]] =
1976 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
1978 nnz_cache[scan8[0 + 4]] =
1979 nnz_cache[scan8[1 + 4]] =
1980 nnz_cache[scan8[2 + 4]] =
1981 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
1983 nnz_cache[scan8[0 + 8]] =
1984 nnz_cache[scan8[1 + 8]] =
1985 nnz_cache[scan8[2 + 8]] =
1986 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
1988 nnz_cache[scan8[0 + 12]] =
1989 nnz_cache[scan8[1 + 12]] =
1990 nnz_cache[scan8[2 + 12]] =
1991 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
1998 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2000 uint8_t *dest_y, *dest_cb, *dest_cr;
2001 int linesize, uvlinesize, mb_x, mb_y;
2002 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
2003 const int old_slice_type = sl->slice_type;
2004 const int pixel_shift = h->pixel_shift;
2005 const int block_h = 16 >> h->chroma_y_shift;
2007 if (h->postpone_filter)
2010 if (sl->deblocking_filter) {
2011 for (mb_x = start_x; mb_x < end_x; mb_x++)
2012 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2014 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
2015 mb_type = h->cur_pic.mb_type[mb_xy];
2019 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2023 dest_y = h->cur_pic.f->data[0] +
2024 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
2025 dest_cb = h->cur_pic.f->data[1] +
2026 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2027 mb_y * sl->uvlinesize * block_h;
2028 dest_cr = h->cur_pic.f->data[2] +
2029 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2030 mb_y * sl->uvlinesize * block_h;
2031 // FIXME simplify above
2034 linesize = sl->mb_linesize = sl->linesize * 2;
2035 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
2036 if (mb_y & 1) { // FIXME move out of this function?
2037 dest_y -= sl->linesize * 15;
2038 dest_cb -= sl->uvlinesize * (block_h - 1);
2039 dest_cr -= sl->uvlinesize * (block_h - 1);
2042 linesize = sl->mb_linesize = sl->linesize;
2043 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
2045 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
2047 if (fill_filter_caches(h, sl, mb_type))
2049 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mb_xy]);
2050 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mb_xy]);
2052 if (FRAME_MBAFF(h)) {
2053 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2054 linesize, uvlinesize);
2056 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2057 dest_cr, linesize, uvlinesize);
2061 sl->slice_type = old_slice_type;
2063 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
2064 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
2065 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
2068 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
2070 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
2071 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2072 h->cur_pic.mb_type[mb_xy - 1] :
2073 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2074 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2075 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2079 * Draw edges and report progress for the last MB row.
2081 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
2083 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
2084 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2085 int height = 16 << FRAME_MBAFF(h);
2086 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2088 if (sl->deblocking_filter) {
2089 if ((top + height) >= pic_height)
2090 height += deblock_border;
2091 top -= deblock_border;
2094 if (top >= pic_height || (top + height) < 0)
2097 height = FFMIN(height, pic_height - top);
2099 height = top + height;
2103 ff_h264_draw_horiz_band(h, sl, top, height);
2108 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2109 h->picture_structure == PICT_BOTTOM_FIELD);
2112 static void er_add_slice(H264SliceContext *sl,
2113 int startx, int starty,
2114 int endx, int endy, int status)
2116 #if CONFIG_ERROR_RESILIENCE
2117 ERContext *er = &sl->er;
2119 if (!sl->h264->enable_er)
2122 er->ref_count = sl->ref_count[0];
2123 ff_er_add_slice(er, startx, starty, endx, endy, status);
2127 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2129 H264SliceContext *sl = arg;
2130 const H264Context *h = sl->h264;
2131 int lf_x_start = sl->mb_x;
2132 int orig_deblock = sl->deblocking_filter;
2135 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2136 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2138 ret = alloc_scratch_buffers(sl, sl->linesize);
2142 sl->mb_skip_run = -1;
2144 if (h->postpone_filter)
2145 sl->deblocking_filter = 0;
2147 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2148 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2150 if (h->ps.pps->cabac) {
2152 align_get_bits(&sl->gb);
2155 ff_init_cabac_decoder(&sl->cabac,
2156 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2157 (get_bits_left(&sl->gb) + 7) / 8);
2159 ff_h264_init_cabac_states(h, sl);
2165 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2166 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2167 sl->next_slice_idx);
2168 return AVERROR_INVALIDDATA;
2171 ret = ff_h264_decode_mb_cabac(h, sl);
2172 // STOP_TIMER("decode_mb_cabac")
2175 ff_h264_hl_decode_mb(h, sl);
2177 // FIXME optimal? or let mb_decode decode 16x32 ?
2178 if (ret >= 0 && FRAME_MBAFF(h)) {
2181 ret = ff_h264_decode_mb_cabac(h, sl);
2184 ff_h264_hl_decode_mb(h, sl);
2187 eos = get_cabac_terminate(&sl->cabac);
2189 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2190 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2191 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2192 sl->mb_y, ER_MB_END);
2193 if (sl->mb_x >= lf_x_start)
2194 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2197 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2198 av_log(h->avctx, AV_LOG_ERROR,
2199 "error while decoding MB %d %d, bytestream %td\n",
2201 sl->cabac.bytestream_end - sl->cabac.bytestream);
2202 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2203 sl->mb_y, ER_MB_ERROR);
2204 return AVERROR_INVALIDDATA;
2207 if (++sl->mb_x >= h->mb_width) {
2208 loop_filter(h, sl, lf_x_start, sl->mb_x);
2209 sl->mb_x = lf_x_start = 0;
2210 decode_finish_row(h, sl);
2212 if (FIELD_OR_MBAFF_PICTURE(h)) {
2214 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2215 predict_field_decoding_flag(h, sl);
2219 if (eos || sl->mb_y >= h->mb_height) {
2220 ff_tlog(h->avctx, "slice end %d %d\n",
2221 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2222 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2223 sl->mb_y, ER_MB_END);
2224 if (sl->mb_x > lf_x_start)
2225 loop_filter(h, sl, lf_x_start, sl->mb_x);
2233 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2234 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2235 sl->next_slice_idx);
2236 return AVERROR_INVALIDDATA;
2239 ret = ff_h264_decode_mb_cavlc(h, sl);
2242 ff_h264_hl_decode_mb(h, sl);
2244 // FIXME optimal? or let mb_decode decode 16x32 ?
2245 if (ret >= 0 && FRAME_MBAFF(h)) {
2247 ret = ff_h264_decode_mb_cavlc(h, sl);
2250 ff_h264_hl_decode_mb(h, sl);
2255 av_log(h->avctx, AV_LOG_ERROR,
2256 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2257 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2258 sl->mb_y, ER_MB_ERROR);
2262 if (++sl->mb_x >= h->mb_width) {
2263 loop_filter(h, sl, lf_x_start, sl->mb_x);
2264 sl->mb_x = lf_x_start = 0;
2265 decode_finish_row(h, sl);
2267 if (FIELD_OR_MBAFF_PICTURE(h)) {
2269 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2270 predict_field_decoding_flag(h, sl);
2272 if (sl->mb_y >= h->mb_height) {
2273 ff_tlog(h->avctx, "slice end %d %d\n",
2274 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2276 if (get_bits_left(&sl->gb) == 0) {
2277 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2278 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2282 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2283 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2285 return AVERROR_INVALIDDATA;
2290 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2291 ff_tlog(h->avctx, "slice end %d %d\n",
2292 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2294 if (get_bits_left(&sl->gb) == 0) {
2295 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2296 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2297 if (sl->mb_x > lf_x_start)
2298 loop_filter(h, sl, lf_x_start, sl->mb_x);
2302 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2303 sl->mb_y, ER_MB_ERROR);
2305 return AVERROR_INVALIDDATA;
2312 sl->deblocking_filter = orig_deblock;
2317 * Call decode_slice() for each context.
2319 * @param h h264 master context
2320 * @param context_count number of contexts to execute
2322 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2324 AVCodecContext *const avctx = h->avctx;
2325 H264SliceContext *sl;
2328 if (h->avctx->hwaccel)
2330 if (context_count == 1) {
2333 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2334 h->postpone_filter = 0;
2336 ret = decode_slice(avctx, &h->slice_ctx[0]);
2337 h->mb_y = h->slice_ctx[0].mb_y;
2340 for (i = 0; i < context_count; i++) {
2341 int next_slice_idx = h->mb_width * h->mb_height;
2344 sl = &h->slice_ctx[i];
2345 sl->er.error_count = 0;
2347 /* make sure none of those slices overlap */
2348 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2349 for (j = 0; j < context_count; j++) {
2350 H264SliceContext *sl2 = &h->slice_ctx[j];
2351 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2353 if (i == j || slice_idx2 < slice_idx)
2355 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2357 sl->next_slice_idx = next_slice_idx;
2360 avctx->execute(avctx, decode_slice, h->slice_ctx,
2361 NULL, context_count, sizeof(h->slice_ctx[0]));
2363 /* pull back stuff from slices to master context */
2364 sl = &h->slice_ctx[context_count - 1];
2366 for (i = 1; i < context_count; i++)
2367 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
2369 if (h->postpone_filter) {
2370 h->postpone_filter = 0;
2372 for (i = 0; i < context_count; i++) {
2375 sl = &h->slice_ctx[i];
2376 y_end = FFMIN(sl->mb_y + 1, h->mb_height);
2377 x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
2379 for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
2381 loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
2382 j == y_end - 1 ? x_end : h->mb_width);