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 / 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] = {
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] = {
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] = {
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] = {
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_malloc(&sl->top_borders[0], &sl->top_borders_allocated[0],
135 h->mb_width * 16 * 3 * sizeof(uint8_t) * 2);
136 av_fast_malloc(&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 if (h->avctx->hwaccel) {
196 const AVHWAccel *hwaccel = h->avctx->hwaccel;
197 av_assert0(!pic->hwaccel_picture_private);
198 if (hwaccel->frame_priv_data_size) {
199 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
200 if (!pic->hwaccel_priv_buf)
201 return AVERROR(ENOMEM);
202 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
206 if (!h->qscale_table_pool) {
207 ret = init_table_pools(h);
212 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
213 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
214 if (!pic->qscale_table_buf || !pic->mb_type_buf)
217 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
218 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
220 for (i = 0; i < 2; i++) {
221 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
222 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
223 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
226 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
227 pic->ref_index[i] = pic->ref_index_buf[i]->data;
232 ff_h264_unref_picture(h, pic);
233 return (ret < 0) ? ret : AVERROR(ENOMEM);
236 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
243 static int find_unused_picture(H264Context *h)
247 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
248 if (pic_is_unused(h, &h->DPB[i]))
251 if (i == H264_MAX_PICTURE_COUNT)
252 return AVERROR_INVALIDDATA;
257 static int initialize_cur_frame(H264Context *h)
262 release_unused_pictures(h, 1);
263 ff_h264_unref_picture(h, &h->cur_pic);
264 h->cur_pic_ptr = NULL;
266 ret = find_unused_picture(h);
268 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
273 ret = alloc_picture(h, cur);
277 ret = ff_h264_ref_picture(h, &h->cur_pic, cur);
280 h->cur_pic_ptr = cur;
285 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
287 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
288 ((pic && pic >= old_ctx->DPB && \
289 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
290 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
292 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
293 H264Context *new_base,
294 H264Context *old_base)
298 for (i = 0; i < count; i++) {
299 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
300 IN_RANGE(from[i], old_base->DPB,
301 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
303 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
307 static int h264_slice_header_init(H264Context *h);
309 int ff_h264_update_thread_context(AVCodecContext *dst,
310 const AVCodecContext *src)
312 H264Context *h = dst->priv_data, *h1 = src->priv_data;
313 int inited = h->context_initialized, err = 0;
317 if (dst == src || !h1->context_initialized)
321 return AVERROR_INVALIDDATA;
324 (h->width != h1->width ||
325 h->height != h1->height ||
326 h->mb_width != h1->mb_width ||
327 h->mb_height != h1->mb_height ||
329 h->ps.sps->bit_depth_luma != h1->ps.sps->bit_depth_luma ||
330 h->ps.sps->chroma_format_idc != h1->ps.sps->chroma_format_idc ||
331 h->ps.sps->colorspace != h1->ps.sps->colorspace)) {
336 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.sps_list); i++) {
337 av_buffer_unref(&h->ps.sps_list[i]);
338 if (h1->ps.sps_list[i]) {
339 h->ps.sps_list[i] = av_buffer_ref(h1->ps.sps_list[i]);
340 if (!h->ps.sps_list[i])
341 return AVERROR(ENOMEM);
344 for (i = 0; i < FF_ARRAY_ELEMS(h->ps.pps_list); i++) {
345 av_buffer_unref(&h->ps.pps_list[i]);
346 if (h1->ps.pps_list[i]) {
347 h->ps.pps_list[i] = av_buffer_ref(h1->ps.pps_list[i]);
348 if (!h->ps.pps_list[i])
349 return AVERROR(ENOMEM);
353 h->ps.sps = h1->ps.sps;
355 if (need_reinit || !inited) {
356 h->width = h1->width;
357 h->height = h1->height;
358 h->mb_height = h1->mb_height;
359 h->mb_width = h1->mb_width;
360 h->mb_num = h1->mb_num;
361 h->mb_stride = h1->mb_stride;
362 h->b_stride = h1->b_stride;
364 if ((err = h264_slice_header_init(h)) < 0) {
365 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
369 /* copy block_offset since frame_start may not be called */
370 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
373 h->avctx->coded_height = h1->avctx->coded_height;
374 h->avctx->coded_width = h1->avctx->coded_width;
375 h->avctx->width = h1->avctx->width;
376 h->avctx->height = h1->avctx->height;
377 h->coded_picture_number = h1->coded_picture_number;
378 h->first_field = h1->first_field;
379 h->picture_structure = h1->picture_structure;
380 h->droppable = h1->droppable;
381 h->low_delay = h1->low_delay;
383 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
384 ff_h264_unref_picture(h, &h->DPB[i]);
385 if (h1->DPB[i].f->buf[0] &&
386 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
390 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
391 ff_h264_unref_picture(h, &h->cur_pic);
392 if (h1->cur_pic.f->buf[0]) {
393 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
398 h->enable_er = h1->enable_er;
399 h->workaround_bugs = h1->workaround_bugs;
400 h->low_delay = h1->low_delay;
401 h->droppable = h1->droppable;
403 // extradata/NAL handling
404 h->is_avc = h1->is_avc;
405 h->nal_length_size = h1->nal_length_size;
407 memcpy(&h->poc, &h1->poc, sizeof(h->poc));
409 h->curr_pic_num = h1->curr_pic_num;
410 h->max_pic_num = h1->max_pic_num;
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->mmco_index = h1->mmco_index;
422 h->mmco_reset = h1->mmco_reset;
423 h->long_ref_count = h1->long_ref_count;
424 h->short_ref_count = h1->short_ref_count;
426 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
427 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
428 copy_picture_range(h->delayed_pic, h1->delayed_pic,
429 MAX_DELAYED_PIC_COUNT + 2, h, h1);
435 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
436 h->poc.prev_poc_msb = h->poc.poc_msb;
437 h->poc.prev_poc_lsb = h->poc.poc_lsb;
439 h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
440 h->poc.prev_frame_num = h->poc.frame_num;
442 h->recovery_frame = h1->recovery_frame;
443 h->frame_recovered = h1->frame_recovered;
448 static int h264_frame_start(H264Context *h)
452 const int pixel_shift = h->pixel_shift;
454 ret = initialize_cur_frame(h);
458 pic = h->cur_pic_ptr;
459 pic->reference = h->droppable ? 0 : h->picture_structure;
460 pic->f->coded_picture_number = h->coded_picture_number++;
461 pic->field_picture = h->picture_structure != PICT_FRAME;
462 pic->frame_num = h->poc.frame_num;
464 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
466 * See decode_nal_units().
468 pic->f->key_frame = 0;
472 if (CONFIG_ERROR_RESILIENCE && h->enable_er)
473 ff_er_frame_start(&h->slice_ctx[0].er);
475 for (i = 0; i < 16; i++) {
476 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
477 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
479 for (i = 0; i < 16; i++) {
480 h->block_offset[16 + i] =
481 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
482 h->block_offset[48 + 16 + i] =
483 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
486 /* Some macroblocks can be accessed before they're available in case
487 * of lost slices, MBAFF or threading. */
488 memset(h->slice_table, -1,
489 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
491 /* We mark the current picture as non-reference after allocating it, so
492 * that if we break out due to an error it can be released automatically
493 * in the next ff_mpv_frame_start().
495 h->cur_pic_ptr->reference = 0;
497 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
499 h->next_output_pic = NULL;
501 assert(h->cur_pic_ptr->long_ref == 0);
506 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
508 uint8_t *src_cb, uint8_t *src_cr,
509 int linesize, int uvlinesize,
514 const int pixel_shift = h->pixel_shift;
515 int chroma444 = CHROMA444(h);
516 int chroma422 = CHROMA422(h);
519 src_cb -= uvlinesize;
520 src_cr -= uvlinesize;
522 if (!simple && FRAME_MBAFF(h)) {
525 top_border = sl->top_borders[0][sl->mb_x];
526 AV_COPY128(top_border, src_y + 15 * linesize);
528 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
529 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
532 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
533 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
534 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
535 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
537 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
538 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
540 } else if (chroma422) {
542 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
543 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
545 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
546 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
550 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
551 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
553 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
554 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
559 } else if (MB_MBAFF(sl)) {
565 top_border = sl->top_borders[top_idx][sl->mb_x];
566 /* There are two lines saved, the line above the top macroblock
567 * of a pair, and the line above the bottom macroblock. */
568 AV_COPY128(top_border, src_y + 16 * linesize);
570 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
572 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
575 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
576 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
577 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
578 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
580 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
581 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
583 } else if (chroma422) {
585 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
586 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
588 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
589 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
593 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
594 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
596 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
597 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
604 * Initialize implicit_weight table.
605 * @param field 0/1 initialize the weight for interlaced MBAFF
606 * -1 initializes the rest
608 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
610 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
612 for (i = 0; i < 2; i++) {
613 sl->pwt.luma_weight_flag[i] = 0;
614 sl->pwt.chroma_weight_flag[i] = 0;
618 if (h->picture_structure == PICT_FRAME) {
619 cur_poc = h->cur_pic_ptr->poc;
621 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
623 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
624 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
625 sl->pwt.use_weight = 0;
626 sl->pwt.use_weight_chroma = 0;
630 ref_count0 = sl->ref_count[0];
631 ref_count1 = sl->ref_count[1];
633 cur_poc = h->cur_pic_ptr->field_poc[field];
635 ref_count0 = 16 + 2 * sl->ref_count[0];
636 ref_count1 = 16 + 2 * sl->ref_count[1];
639 sl->pwt.use_weight = 2;
640 sl->pwt.use_weight_chroma = 2;
641 sl->pwt.luma_log2_weight_denom = 5;
642 sl->pwt.chroma_log2_weight_denom = 5;
644 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
645 int poc0 = sl->ref_list[0][ref0].poc;
646 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
648 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
649 int poc1 = sl->ref_list[1][ref1].poc;
650 int td = av_clip_int8(poc1 - poc0);
652 int tb = av_clip_int8(cur_poc - poc0);
653 int tx = (16384 + (FFABS(td) >> 1)) / td;
654 int dist_scale_factor = (tb * tx + 32) >> 8;
655 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
656 w = 64 - dist_scale_factor;
660 sl->pwt.implicit_weight[ref0][ref1][0] =
661 sl->pwt.implicit_weight[ref0][ref1][1] = w;
663 sl->pwt.implicit_weight[ref0][ref1][field] = w;
670 * initialize scan tables
672 static void init_scan_tables(H264Context *h)
675 for (i = 0; i < 16; i++) {
676 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
677 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
678 h->field_scan[i] = TRANSPOSE(field_scan[i]);
681 for (i = 0; i < 64; i++) {
682 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
683 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
684 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
685 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
686 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
689 if (h->ps.sps->transform_bypass) { // FIXME same ugly
690 h->zigzag_scan_q0 = ff_zigzag_scan;
691 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
692 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
693 h->field_scan_q0 = field_scan;
694 h->field_scan8x8_q0 = field_scan8x8;
695 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
697 h->zigzag_scan_q0 = h->zigzag_scan;
698 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
699 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
700 h->field_scan_q0 = h->field_scan;
701 h->field_scan8x8_q0 = h->field_scan8x8;
702 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
706 static enum AVPixelFormat get_pixel_format(H264Context *h)
708 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
709 CONFIG_H264_D3D11VA_HWACCEL + \
710 CONFIG_H264_VAAPI_HWACCEL + \
711 (CONFIG_H264_VDA_HWACCEL * 2) + \
712 CONFIG_H264_VDPAU_HWACCEL)
713 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
714 const enum AVPixelFormat *choices = pix_fmts;
716 switch (h->ps.sps->bit_depth_luma) {
719 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
720 *fmt++ = AV_PIX_FMT_GBRP9;
722 *fmt++ = AV_PIX_FMT_YUV444P9;
723 } else if (CHROMA422(h))
724 *fmt++ = AV_PIX_FMT_YUV422P9;
726 *fmt++ = AV_PIX_FMT_YUV420P9;
730 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
731 *fmt++ = AV_PIX_FMT_GBRP10;
733 *fmt++ = AV_PIX_FMT_YUV444P10;
734 } else if (CHROMA422(h))
735 *fmt++ = AV_PIX_FMT_YUV422P10;
737 *fmt++ = AV_PIX_FMT_YUV420P10;
740 #if CONFIG_H264_VDPAU_HWACCEL
741 *fmt++ = AV_PIX_FMT_VDPAU;
744 if (h->avctx->colorspace == AVCOL_SPC_RGB)
745 *fmt++ = AV_PIX_FMT_GBRP;
746 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
747 *fmt++ = AV_PIX_FMT_YUVJ444P;
749 *fmt++ = AV_PIX_FMT_YUV444P;
750 } else if (CHROMA422(h)) {
751 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
752 *fmt++ = AV_PIX_FMT_YUVJ422P;
754 *fmt++ = AV_PIX_FMT_YUV422P;
756 #if CONFIG_H264_DXVA2_HWACCEL
757 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
759 #if CONFIG_H264_D3D11VA_HWACCEL
760 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
762 #if CONFIG_H264_VAAPI_HWACCEL
763 *fmt++ = AV_PIX_FMT_VAAPI;
765 #if CONFIG_H264_VDA_HWACCEL
766 *fmt++ = AV_PIX_FMT_VDA_VLD;
767 *fmt++ = AV_PIX_FMT_VDA;
769 if (h->avctx->codec->pix_fmts)
770 choices = h->avctx->codec->pix_fmts;
771 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
772 *fmt++ = AV_PIX_FMT_YUVJ420P;
774 *fmt++ = AV_PIX_FMT_YUV420P;
778 av_log(h->avctx, AV_LOG_ERROR,
779 "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
780 return AVERROR_INVALIDDATA;
783 *fmt = AV_PIX_FMT_NONE;
785 return ff_get_format(h->avctx, choices);
788 /* export coded and cropped frame dimensions to AVCodecContext */
789 static int init_dimensions(H264Context *h)
791 SPS *sps = h->ps.sps;
792 int width = h->width - (sps->crop_right + sps->crop_left);
793 int height = h->height - (sps->crop_top + sps->crop_bottom);
795 /* handle container cropping */
796 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
797 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16)) {
798 width = h->avctx->width;
799 height = h->avctx->height;
802 if (width <= 0 || height <= 0) {
803 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
805 if (h->avctx->err_recognition & AV_EF_EXPLODE)
806 return AVERROR_INVALIDDATA;
808 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
819 h->avctx->coded_width = h->width;
820 h->avctx->coded_height = h->height;
821 h->avctx->width = width;
822 h->avctx->height = height;
827 static int h264_slice_header_init(H264Context *h)
829 const SPS *sps = h->ps.sps;
830 int nb_slices = (HAVE_THREADS &&
831 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
832 h->avctx->thread_count : 1;
835 ff_set_sar(h->avctx, sps->sar);
836 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
837 &h->chroma_x_shift, &h->chroma_y_shift);
839 if (sps->timing_info_present_flag) {
840 int64_t den = sps->time_scale;
841 if (h->sei.unregistered.x264_build < 44U)
843 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
844 sps->num_units_in_tick, den, 1 << 30);
847 ff_h264_free_tables(h);
850 h->prev_interlaced_frame = 1;
853 ret = ff_h264_alloc_tables(h);
855 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
859 if (sps->bit_depth_luma < 8 || sps->bit_depth_luma > 10) {
860 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
861 sps->bit_depth_luma);
862 return AVERROR_INVALIDDATA;
865 h->avctx->bits_per_raw_sample = sps->bit_depth_luma;
866 h->pixel_shift = sps->bit_depth_luma > 8;
867 h->chroma_format_idc = sps->chroma_format_idc;
868 h->bit_depth_luma = sps->bit_depth_luma;
870 ff_h264dsp_init(&h->h264dsp, sps->bit_depth_luma,
871 sps->chroma_format_idc);
872 ff_h264chroma_init(&h->h264chroma, sps->bit_depth_chroma);
873 ff_h264qpel_init(&h->h264qpel, sps->bit_depth_luma);
874 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, sps->bit_depth_luma,
875 sps->chroma_format_idc);
876 ff_videodsp_init(&h->vdsp, sps->bit_depth_luma);
878 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
881 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
883 max_slices = H264_MAX_THREADS;
884 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
885 " reducing to %d\n", nb_slices, max_slices);
886 nb_slices = max_slices;
888 h->slice_context_count = nb_slices;
890 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
891 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
893 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
897 for (i = 0; i < h->slice_context_count; i++) {
898 H264SliceContext *sl = &h->slice_ctx[i];
901 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
902 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
903 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
905 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
906 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
912 h->context_initialized = 1;
918 * Decode a slice header.
919 * This will (re)intialize the decoder and call h264_frame_start() as needed.
921 * @param h h264context
923 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
925 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
929 unsigned int first_mb_in_slice;
932 unsigned int slice_type, tmp, i, j;
933 int last_pic_structure, last_pic_droppable;
934 int needs_reinit = 0;
935 int field_pic_flag, bottom_field_flag;
936 int frame_num, droppable, picture_structure;
937 int mb_aff_frame = 0;
939 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
940 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
942 first_mb_in_slice = get_ue_golomb(&sl->gb);
944 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
945 if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
946 ff_h264_field_end(h, sl, 1);
949 h->current_slice = 0;
950 if (!h->first_field) {
951 if (h->cur_pic_ptr && !h->droppable) {
952 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
953 h->picture_structure == PICT_BOTTOM_FIELD);
955 h->cur_pic_ptr = NULL;
959 slice_type = get_ue_golomb_31(&sl->gb);
960 if (slice_type > 9) {
961 av_log(h->avctx, AV_LOG_ERROR,
962 "slice type %d too large at %d\n",
963 slice_type, first_mb_in_slice);
964 return AVERROR_INVALIDDATA;
966 if (slice_type > 4) {
968 sl->slice_type_fixed = 1;
970 sl->slice_type_fixed = 0;
972 slice_type = ff_h264_golomb_to_pict_type[slice_type];
973 sl->slice_type = slice_type;
974 sl->slice_type_nos = slice_type & 3;
976 if (h->nal_unit_type == NAL_IDR_SLICE &&
977 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
978 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
979 return AVERROR_INVALIDDATA;
982 // to make a few old functions happy, it's wrong though
983 if (!h->setup_finished)
984 h->pict_type = sl->slice_type;
986 pps_id = get_ue_golomb(&sl->gb);
987 if (pps_id >= MAX_PPS_COUNT) {
988 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
989 return AVERROR_INVALIDDATA;
991 if (!h->ps.pps_list[pps_id]) {
992 av_log(h->avctx, AV_LOG_ERROR,
993 "non-existing PPS %u referenced\n",
995 return AVERROR_INVALIDDATA;
997 if (!h->setup_finished) {
998 h->ps.pps = (const PPS*)h->ps.pps_list[pps_id]->data;
999 } else if (h->ps.pps != (const PPS*)h->ps.pps_list[pps_id]->data) {
1000 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1001 return AVERROR_INVALIDDATA;
1004 if (!h->ps.sps_list[h->ps.pps->sps_id]) {
1005 av_log(h->avctx, AV_LOG_ERROR,
1006 "non-existing SPS %u referenced\n",
1008 return AVERROR_INVALIDDATA;
1011 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
1012 h->ps.sps = (SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data;
1014 if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
1015 h->chroma_format_idc != h->ps.sps->chroma_format_idc)
1018 if (h->flags & AV_CODEC_FLAG_LOW_DELAY ||
1019 (h->ps.sps->bitstream_restriction_flag &&
1020 !h->ps.sps->num_reorder_frames)) {
1021 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1022 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1023 "Reenabling low delay requires a codec flush.\n");
1028 if (h->avctx->has_b_frames < 2)
1029 h->avctx->has_b_frames = !h->low_delay;
1036 if (!h->setup_finished) {
1037 h->avctx->profile = ff_h264_get_profile(sps);
1038 h->avctx->level = sps->level_idc;
1039 h->avctx->refs = sps->ref_frame_count;
1041 if (h->mb_width != sps->mb_width ||
1042 h->mb_height != sps->mb_height * (2 - sps->frame_mbs_only_flag))
1045 h->mb_width = sps->mb_width;
1046 h->mb_height = sps->mb_height * (2 - sps->frame_mbs_only_flag);
1047 h->mb_num = h->mb_width * h->mb_height;
1048 h->mb_stride = h->mb_width + 1;
1050 h->b_stride = h->mb_width * 4;
1052 h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
1054 h->width = 16 * h->mb_width;
1055 h->height = 16 * h->mb_height;
1057 ret = init_dimensions(h);
1061 if (sps->video_signal_type_present_flag) {
1062 h->avctx->color_range = sps->full_range ? AVCOL_RANGE_JPEG
1064 if (sps->colour_description_present_flag) {
1065 if (h->avctx->colorspace != sps->colorspace)
1067 h->avctx->color_primaries = sps->color_primaries;
1068 h->avctx->color_trc = sps->color_trc;
1069 h->avctx->colorspace = sps->colorspace;
1074 if (h->context_initialized && needs_reinit) {
1075 h->context_initialized = 0;
1076 if (sl != h->slice_ctx) {
1077 av_log(h->avctx, AV_LOG_ERROR,
1078 "changing width %d -> %d / height %d -> %d on "
1080 h->width, h->avctx->coded_width,
1081 h->height, h->avctx->coded_height,
1082 h->current_slice + 1);
1083 return AVERROR_INVALIDDATA;
1086 ff_h264_flush_change(h);
1088 if ((ret = get_pixel_format(h)) < 0)
1090 h->avctx->pix_fmt = ret;
1092 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1093 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1095 if ((ret = h264_slice_header_init(h)) < 0) {
1096 av_log(h->avctx, AV_LOG_ERROR,
1097 "h264_slice_header_init() failed\n");
1101 if (!h->context_initialized) {
1102 if (sl != h->slice_ctx) {
1103 av_log(h->avctx, AV_LOG_ERROR,
1104 "Cannot (re-)initialize context during parallel decoding.\n");
1105 return AVERROR_PATCHWELCOME;
1108 if ((ret = get_pixel_format(h)) < 0)
1110 h->avctx->pix_fmt = ret;
1112 if ((ret = h264_slice_header_init(h)) < 0) {
1113 av_log(h->avctx, AV_LOG_ERROR,
1114 "h264_slice_header_init() failed\n");
1119 frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
1120 if (!h->setup_finished)
1121 h->poc.frame_num = frame_num;
1125 last_pic_structure = h->picture_structure;
1126 last_pic_droppable = h->droppable;
1128 droppable = h->nal_ref_idc == 0;
1129 if (sps->frame_mbs_only_flag) {
1130 picture_structure = PICT_FRAME;
1132 field_pic_flag = get_bits1(&sl->gb);
1133 if (field_pic_flag) {
1134 bottom_field_flag = get_bits1(&sl->gb);
1135 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1137 picture_structure = PICT_FRAME;
1138 mb_aff_frame = sps->mb_aff;
1141 if (!h->setup_finished) {
1142 h->droppable = droppable;
1143 h->picture_structure = picture_structure;
1144 h->mb_aff_frame = mb_aff_frame;
1146 sl->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1148 if (h->current_slice != 0) {
1149 if (last_pic_structure != picture_structure ||
1150 last_pic_droppable != droppable) {
1151 av_log(h->avctx, AV_LOG_ERROR,
1152 "Changing field mode (%d -> %d) between slices is not allowed\n",
1153 last_pic_structure, h->picture_structure);
1154 return AVERROR_INVALIDDATA;
1155 } else if (!h->cur_pic_ptr) {
1156 av_log(h->avctx, AV_LOG_ERROR,
1157 "unset cur_pic_ptr on slice %d\n",
1158 h->current_slice + 1);
1159 return AVERROR_INVALIDDATA;
1162 /* Shorten frame num gaps so we don't have to allocate reference
1163 * frames just to throw them away */
1164 if (h->poc.frame_num != h->poc.prev_frame_num) {
1165 int unwrap_prev_frame_num = h->poc.prev_frame_num;
1166 int max_frame_num = 1 << sps->log2_max_frame_num;
1168 if (unwrap_prev_frame_num > h->poc.frame_num)
1169 unwrap_prev_frame_num -= max_frame_num;
1171 if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
1172 unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
1173 if (unwrap_prev_frame_num < 0)
1174 unwrap_prev_frame_num += max_frame_num;
1176 h->poc.prev_frame_num = unwrap_prev_frame_num;
1180 /* See if we have a decoded first field looking for a pair...
1181 * Here, we're using that to see if we should mark previously
1182 * decode frames as "finished".
1183 * We have to do that before the "dummy" in-between frame allocation,
1184 * since that can modify s->current_picture_ptr. */
1185 if (h->first_field) {
1186 assert(h->cur_pic_ptr);
1187 assert(h->cur_pic_ptr->f->buf[0]);
1188 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1190 /* figure out if we have a complementary field pair */
1191 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1192 /* Previous field is unmatched. Don't display it, but let it
1193 * remain for reference if marked as such. */
1194 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1195 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1196 last_pic_structure == PICT_TOP_FIELD);
1199 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1200 /* This and previous field were reference, but had
1201 * different frame_nums. Consider this field first in
1202 * pair. Throw away previous field except for reference
1204 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1205 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1206 last_pic_structure == PICT_TOP_FIELD);
1209 /* Second field in complementary pair */
1210 if (!((last_pic_structure == PICT_TOP_FIELD &&
1211 h->picture_structure == PICT_BOTTOM_FIELD) ||
1212 (last_pic_structure == PICT_BOTTOM_FIELD &&
1213 h->picture_structure == PICT_TOP_FIELD))) {
1214 av_log(h->avctx, AV_LOG_ERROR,
1215 "Invalid field mode combination %d/%d\n",
1216 last_pic_structure, h->picture_structure);
1217 h->picture_structure = last_pic_structure;
1218 h->droppable = last_pic_droppable;
1219 return AVERROR_INVALIDDATA;
1220 } else if (last_pic_droppable != h->droppable) {
1221 avpriv_request_sample(h->avctx,
1222 "Found reference and non-reference fields in the same frame, which");
1223 h->picture_structure = last_pic_structure;
1224 h->droppable = last_pic_droppable;
1225 return AVERROR_PATCHWELCOME;
1231 while (h->poc.frame_num != h->poc.prev_frame_num &&
1232 h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
1233 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1234 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1235 h->poc.frame_num, h->poc.prev_frame_num);
1236 ret = initialize_cur_frame(h);
1242 h->poc.prev_frame_num++;
1243 h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
1244 h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
1245 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1246 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1247 ret = ff_generate_sliding_window_mmcos(h, 1);
1248 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1250 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1251 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1253 /* Error concealment: If a ref is missing, copy the previous ref
1255 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1256 * many assumptions about there being no actual duplicates.
1257 * FIXME: This does not copy padding for out-of-frame motion
1258 * vectors. Given we are concealing a lost frame, this probably
1259 * is not noticeable by comparison, but it should be fixed. */
1260 if (h->short_ref_count) {
1262 h->short_ref[0]->f->width == prev->f->width &&
1263 h->short_ref[0]->f->height == prev->f->height &&
1264 h->short_ref[0]->f->format == prev->f->format) {
1265 av_image_copy(h->short_ref[0]->f->data,
1266 h->short_ref[0]->f->linesize,
1267 (const uint8_t **)prev->f->data,
1272 h->short_ref[0]->poc = prev->poc + 2;
1274 h->short_ref[0]->frame_num = h->poc.prev_frame_num;
1278 /* See if we have a decoded first field looking for a pair...
1279 * We're using that to see whether to continue decoding in that
1280 * frame, or to allocate a new one. */
1281 if (h->first_field) {
1282 assert(h->cur_pic_ptr);
1283 assert(h->cur_pic_ptr->f->buf[0]);
1284 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1286 /* figure out if we have a complementary field pair */
1287 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1288 /* Previous field is unmatched. Don't display it, but let it
1289 * remain for reference if marked as such. */
1290 h->cur_pic_ptr = NULL;
1291 h->first_field = FIELD_PICTURE(h);
1293 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1294 /* This and the previous field had different frame_nums.
1295 * Consider this field first in pair. Throw away previous
1296 * one except for reference purposes. */
1298 h->cur_pic_ptr = NULL;
1300 /* Second field in complementary pair */
1305 /* Frame or first field in a potentially complementary pair */
1306 h->first_field = FIELD_PICTURE(h);
1309 if (!FIELD_PICTURE(h) || h->first_field) {
1310 if (h264_frame_start(h) < 0) {
1312 return AVERROR_INVALIDDATA;
1315 release_unused_pictures(h, 0);
1319 assert(h->mb_num == h->mb_width * h->mb_height);
1320 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1321 first_mb_in_slice >= h->mb_num) {
1322 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1323 return AVERROR_INVALIDDATA;
1325 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1326 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1327 FIELD_OR_MBAFF_PICTURE(h);
1328 if (h->picture_structure == PICT_BOTTOM_FIELD)
1329 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1330 assert(sl->mb_y < h->mb_height);
1332 if (h->picture_structure == PICT_FRAME) {
1333 h->curr_pic_num = h->poc.frame_num;
1334 h->max_pic_num = 1 << sps->log2_max_frame_num;
1336 h->curr_pic_num = 2 * h->poc.frame_num + 1;
1337 h->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
1340 if (h->nal_unit_type == NAL_IDR_SLICE)
1341 get_ue_golomb(&sl->gb); /* idr_pic_id */
1343 if (sps->poc_type == 0) {
1344 int poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
1346 if (!h->setup_finished)
1347 h->poc.poc_lsb = poc_lsb;
1349 if (pps->pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1350 int delta_poc_bottom = get_se_golomb(&sl->gb);
1351 if (!h->setup_finished)
1352 h->poc.delta_poc_bottom = delta_poc_bottom;
1356 if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
1357 int delta_poc = get_se_golomb(&sl->gb);
1359 if (!h->setup_finished)
1360 h->poc.delta_poc[0] = delta_poc;
1362 if (pps->pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1363 delta_poc = get_se_golomb(&sl->gb);
1365 if (!h->setup_finished)
1366 h->poc.delta_poc[1] = delta_poc;
1370 if (!h->setup_finished)
1371 ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
1372 sps, &h->poc, h->picture_structure, h->nal_ref_idc);
1374 if (pps->redundant_pic_cnt_present)
1375 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1377 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1378 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1380 ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
1381 &sl->gb, pps, sl->slice_type_nos,
1382 h->picture_structure);
1386 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1387 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1389 sl->ref_count[1] = sl->ref_count[0] = 0;
1394 if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1395 (pps->weighted_bipred_idc == 1 &&
1396 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1397 ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
1398 sl->slice_type_nos, &sl->pwt);
1399 else if (pps->weighted_bipred_idc == 2 &&
1400 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1401 implicit_weight_table(h, sl, -1);
1403 sl->pwt.use_weight = 0;
1404 for (i = 0; i < 2; i++) {
1405 sl->pwt.luma_weight_flag[i] = 0;
1406 sl->pwt.chroma_weight_flag[i] = 0;
1410 // If frame-mt is enabled, only update mmco tables for the first slice
1411 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1412 // or h->mmco, which will cause ref list mix-ups and decoding errors
1413 // further down the line. This may break decoding if the first slice is
1414 // corrupt, thus we only do this if frame-mt is enabled.
1415 if (h->nal_ref_idc) {
1416 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1417 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1418 h->current_slice == 0);
1419 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1420 return AVERROR_INVALIDDATA;
1423 if (FRAME_MBAFF(h)) {
1424 ff_h264_fill_mbaff_ref_list(h, sl);
1426 if (pps->weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1427 implicit_weight_table(h, sl, 0);
1428 implicit_weight_table(h, sl, 1);
1432 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1433 ff_h264_direct_dist_scale_factor(h, sl);
1434 ff_h264_direct_ref_list_init(h, sl);
1436 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
1437 tmp = get_ue_golomb_31(&sl->gb);
1439 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1440 return AVERROR_INVALIDDATA;
1442 sl->cabac_init_idc = tmp;
1445 sl->last_qscale_diff = 0;
1446 tmp = pps->init_qp + get_se_golomb(&sl->gb);
1447 if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
1448 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1449 return AVERROR_INVALIDDATA;
1452 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1453 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1454 // FIXME qscale / qp ... stuff
1455 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1456 get_bits1(&sl->gb); /* sp_for_switch_flag */
1457 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1458 sl->slice_type == AV_PICTURE_TYPE_SI)
1459 get_se_golomb(&sl->gb); /* slice_qs_delta */
1461 sl->deblocking_filter = 1;
1462 sl->slice_alpha_c0_offset = 0;
1463 sl->slice_beta_offset = 0;
1464 if (pps->deblocking_filter_parameters_present) {
1465 tmp = get_ue_golomb_31(&sl->gb);
1467 av_log(h->avctx, AV_LOG_ERROR,
1468 "deblocking_filter_idc %u out of range\n", tmp);
1469 return AVERROR_INVALIDDATA;
1471 sl->deblocking_filter = tmp;
1472 if (sl->deblocking_filter < 2)
1473 sl->deblocking_filter ^= 1; // 1<->0
1475 if (sl->deblocking_filter) {
1476 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1477 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1478 if (sl->slice_alpha_c0_offset > 12 ||
1479 sl->slice_alpha_c0_offset < -12 ||
1480 sl->slice_beta_offset > 12 ||
1481 sl->slice_beta_offset < -12) {
1482 av_log(h->avctx, AV_LOG_ERROR,
1483 "deblocking filter parameters %d %d out of range\n",
1484 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1485 return AVERROR_INVALIDDATA;
1490 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1491 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1492 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1493 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1494 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1495 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1496 h->nal_ref_idc == 0))
1497 sl->deblocking_filter = 0;
1499 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1500 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1501 /* Cheat slightly for speed:
1502 * Do not bother to deblock across slices. */
1503 sl->deblocking_filter = 2;
1505 h->max_contexts = 1;
1506 if (!h->single_decode_warning) {
1507 av_log(h->avctx, AV_LOG_INFO,
1508 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1509 h->single_decode_warning = 1;
1511 if (sl != h->slice_ctx) {
1512 av_log(h->avctx, AV_LOG_ERROR,
1513 "Deblocking switched inside frame.\n");
1518 sl->qp_thresh = 15 -
1519 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1521 pps->chroma_qp_index_offset[0],
1522 pps->chroma_qp_index_offset[1]) +
1523 6 * (sps->bit_depth_luma - 8);
1525 sl->slice_num = ++h->current_slice;
1526 if (sl->slice_num >= MAX_SLICES) {
1527 av_log(h->avctx, AV_LOG_ERROR,
1528 "Too many slices, increase MAX_SLICES and recompile\n");
1531 for (j = 0; j < 2; j++) {
1533 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1534 for (i = 0; i < 16; i++) {
1536 if (j < sl->list_count && i < sl->ref_count[j] &&
1537 sl->ref_list[j][i].parent->f->buf[0]) {
1539 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1540 for (k = 0; k < h->short_ref_count; k++)
1541 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1545 for (k = 0; k < h->long_ref_count; k++)
1546 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1547 id_list[i] = h->short_ref_count + k;
1555 for (i = 0; i < 16; i++)
1556 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1558 ref2frm[18 + 1] = -1;
1559 for (i = 16; i < 48; i++)
1560 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1561 (sl->ref_list[j][i].reference & 3);
1564 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1565 av_log(h->avctx, AV_LOG_DEBUG,
1566 "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",
1568 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1570 av_get_picture_type_char(sl->slice_type),
1571 sl->slice_type_fixed ? " fix" : "",
1572 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1573 pps_id, h->poc.frame_num,
1574 h->cur_pic_ptr->field_poc[0],
1575 h->cur_pic_ptr->field_poc[1],
1576 sl->ref_count[0], sl->ref_count[1],
1578 sl->deblocking_filter,
1579 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1581 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1582 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1588 int ff_h264_get_slice_type(const H264SliceContext *sl)
1590 switch (sl->slice_type) {
1591 case AV_PICTURE_TYPE_P:
1593 case AV_PICTURE_TYPE_B:
1595 case AV_PICTURE_TYPE_I:
1597 case AV_PICTURE_TYPE_SP:
1599 case AV_PICTURE_TYPE_SI:
1602 return AVERROR_INVALIDDATA;
1606 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1607 H264SliceContext *sl,
1608 int mb_type, int top_xy,
1609 int left_xy[LEFT_MBS],
1611 int left_type[LEFT_MBS],
1612 int mb_xy, int list)
1614 int b_stride = h->b_stride;
1615 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1616 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1617 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1618 if (USES_LIST(top_type, list)) {
1619 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1620 const int b8_xy = 4 * top_xy + 2;
1621 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1622 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1623 ref_cache[0 - 1 * 8] =
1624 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1625 ref_cache[2 - 1 * 8] =
1626 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1628 AV_ZERO128(mv_dst - 1 * 8);
1629 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1632 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1633 if (USES_LIST(left_type[LTOP], list)) {
1634 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1635 const int b8_xy = 4 * left_xy[LTOP] + 1;
1636 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1637 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1638 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1639 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1640 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1642 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1643 ref_cache[-1 + 16] =
1644 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1646 AV_ZERO32(mv_dst - 1 + 0);
1647 AV_ZERO32(mv_dst - 1 + 8);
1648 AV_ZERO32(mv_dst - 1 + 16);
1649 AV_ZERO32(mv_dst - 1 + 24);
1652 ref_cache[-1 + 16] =
1653 ref_cache[-1 + 24] = LIST_NOT_USED;
1658 if (!USES_LIST(mb_type, list)) {
1659 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1660 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1661 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1662 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1663 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1668 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1669 int (*ref2frm)[64] = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1670 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1671 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1672 AV_WN32A(&ref_cache[0 * 8], ref01);
1673 AV_WN32A(&ref_cache[1 * 8], ref01);
1674 AV_WN32A(&ref_cache[2 * 8], ref23);
1675 AV_WN32A(&ref_cache[3 * 8], ref23);
1679 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1680 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1681 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1682 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1683 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1688 * @return non zero if the loop filter can be skipped
1690 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
1692 const int mb_xy = sl->mb_xy;
1693 int top_xy, left_xy[LEFT_MBS];
1694 int top_type, left_type[LEFT_MBS];
1698 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1700 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1701 * stuff, I can't imagine that these complex rules are worth it. */
1703 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1704 if (FRAME_MBAFF(h)) {
1705 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1706 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1708 if (left_mb_field_flag != curr_mb_field_flag)
1709 left_xy[LTOP] -= h->mb_stride;
1711 if (curr_mb_field_flag)
1712 top_xy += h->mb_stride &
1713 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1714 if (left_mb_field_flag != curr_mb_field_flag)
1715 left_xy[LBOT] += h->mb_stride;
1719 sl->top_mb_xy = top_xy;
1720 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1721 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1723 /* For sufficiently low qp, filtering wouldn't do anything.
1724 * This is a conservative estimate: could also check beta_offset
1725 * and more accurate chroma_qp. */
1726 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1727 int qp = h->cur_pic.qscale_table[mb_xy];
1728 if (qp <= qp_thresh &&
1729 (left_xy[LTOP] < 0 ||
1730 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1732 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1733 if (!FRAME_MBAFF(h))
1735 if ((left_xy[LTOP] < 0 ||
1736 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1737 (top_xy < h->mb_stride ||
1738 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1743 top_type = h->cur_pic.mb_type[top_xy];
1744 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1745 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1746 if (sl->deblocking_filter == 2) {
1747 if (h->slice_table[top_xy] != sl->slice_num)
1749 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
1750 left_type[LTOP] = left_type[LBOT] = 0;
1752 if (h->slice_table[top_xy] == 0xFFFF)
1754 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1755 left_type[LTOP] = left_type[LBOT] = 0;
1757 sl->top_type = top_type;
1758 sl->left_type[LTOP] = left_type[LTOP];
1759 sl->left_type[LBOT] = left_type[LBOT];
1761 if (IS_INTRA(mb_type))
1764 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1765 top_type, left_type, mb_xy, 0);
1766 if (sl->list_count == 2)
1767 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1768 top_type, left_type, mb_xy, 1);
1770 nnz = h->non_zero_count[mb_xy];
1771 nnz_cache = sl->non_zero_count_cache;
1772 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1773 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1774 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1775 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1776 sl->cbp = h->cbp_table[mb_xy];
1779 nnz = h->non_zero_count[top_xy];
1780 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1783 if (left_type[LTOP]) {
1784 nnz = h->non_zero_count[left_xy[LTOP]];
1785 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
1786 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
1787 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
1788 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
1791 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
1792 * from what the loop filter needs */
1793 if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
1794 if (IS_8x8DCT(top_type)) {
1795 nnz_cache[4 + 8 * 0] =
1796 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
1797 nnz_cache[6 + 8 * 0] =
1798 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
1800 if (IS_8x8DCT(left_type[LTOP])) {
1801 nnz_cache[3 + 8 * 1] =
1802 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
1804 if (IS_8x8DCT(left_type[LBOT])) {
1805 nnz_cache[3 + 8 * 3] =
1806 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
1809 if (IS_8x8DCT(mb_type)) {
1810 nnz_cache[scan8[0]] =
1811 nnz_cache[scan8[1]] =
1812 nnz_cache[scan8[2]] =
1813 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
1815 nnz_cache[scan8[0 + 4]] =
1816 nnz_cache[scan8[1 + 4]] =
1817 nnz_cache[scan8[2 + 4]] =
1818 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
1820 nnz_cache[scan8[0 + 8]] =
1821 nnz_cache[scan8[1 + 8]] =
1822 nnz_cache[scan8[2 + 8]] =
1823 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
1825 nnz_cache[scan8[0 + 12]] =
1826 nnz_cache[scan8[1 + 12]] =
1827 nnz_cache[scan8[2 + 12]] =
1828 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
1835 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
1837 uint8_t *dest_y, *dest_cb, *dest_cr;
1838 int linesize, uvlinesize, mb_x, mb_y;
1839 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
1840 const int old_slice_type = sl->slice_type;
1841 const int pixel_shift = h->pixel_shift;
1842 const int block_h = 16 >> h->chroma_y_shift;
1844 if (sl->deblocking_filter) {
1845 for (mb_x = start_x; mb_x < end_x; mb_x++)
1846 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
1848 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
1849 sl->slice_num = h->slice_table[mb_xy];
1850 mb_type = h->cur_pic.mb_type[mb_xy];
1851 sl->list_count = h->list_counts[mb_xy];
1855 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
1859 dest_y = h->cur_pic.f->data[0] +
1860 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
1861 dest_cb = h->cur_pic.f->data[1] +
1862 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1863 mb_y * sl->uvlinesize * block_h;
1864 dest_cr = h->cur_pic.f->data[2] +
1865 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1866 mb_y * sl->uvlinesize * block_h;
1867 // FIXME simplify above
1870 linesize = sl->mb_linesize = sl->linesize * 2;
1871 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
1872 if (mb_y & 1) { // FIXME move out of this function?
1873 dest_y -= sl->linesize * 15;
1874 dest_cb -= sl->uvlinesize * (block_h - 1);
1875 dest_cr -= sl->uvlinesize * (block_h - 1);
1878 linesize = sl->mb_linesize = sl->linesize;
1879 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
1881 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
1883 if (fill_filter_caches(h, sl, mb_type))
1885 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
1886 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
1888 if (FRAME_MBAFF(h)) {
1889 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
1890 linesize, uvlinesize);
1892 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
1893 dest_cr, linesize, uvlinesize);
1897 sl->slice_type = old_slice_type;
1899 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
1900 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1901 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1904 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
1906 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
1907 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
1908 h->cur_pic.mb_type[mb_xy - 1] :
1909 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
1910 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
1911 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1915 * Draw edges and report progress for the last MB row.
1917 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
1919 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
1920 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
1921 int height = 16 << FRAME_MBAFF(h);
1922 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
1924 if (sl->deblocking_filter) {
1925 if ((top + height) >= pic_height)
1926 height += deblock_border;
1927 top -= deblock_border;
1930 if (top >= pic_height || (top + height) < 0)
1933 height = FFMIN(height, pic_height - top);
1935 height = top + height;
1939 ff_h264_draw_horiz_band(h, sl, top, height);
1944 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
1945 h->picture_structure == PICT_BOTTOM_FIELD);
1948 static void er_add_slice(H264SliceContext *sl,
1949 int startx, int starty,
1950 int endx, int endy, int status)
1952 #if CONFIG_ERROR_RESILIENCE
1953 ERContext *er = &sl->er;
1955 if (!sl->h264->enable_er)
1958 er->ref_count = sl->ref_count[0];
1959 ff_er_add_slice(er, startx, starty, endx, endy, status);
1963 static int decode_slice(struct AVCodecContext *avctx, void *arg)
1965 H264SliceContext *sl = arg;
1966 const H264Context *h = sl->h264;
1967 int lf_x_start = sl->mb_x;
1970 sl->linesize = h->cur_pic_ptr->f->linesize[0];
1971 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
1973 ret = alloc_scratch_buffers(sl, sl->linesize);
1977 sl->mb_skip_run = -1;
1979 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
1980 avctx->codec_id != AV_CODEC_ID_H264 ||
1981 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
1983 if (h->ps.pps->cabac) {
1985 align_get_bits(&sl->gb);
1988 ff_init_cabac_decoder(&sl->cabac,
1989 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
1990 (get_bits_left(&sl->gb) + 7) / 8);
1992 ff_h264_init_cabac_states(h, sl);
1998 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
1999 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2000 sl->next_slice_idx);
2001 return AVERROR_INVALIDDATA;
2004 ret = ff_h264_decode_mb_cabac(h, sl);
2005 // STOP_TIMER("decode_mb_cabac")
2008 ff_h264_hl_decode_mb(h, sl);
2010 // FIXME optimal? or let mb_decode decode 16x32 ?
2011 if (ret >= 0 && FRAME_MBAFF(h)) {
2014 ret = ff_h264_decode_mb_cabac(h, sl);
2017 ff_h264_hl_decode_mb(h, sl);
2020 eos = get_cabac_terminate(&sl->cabac);
2022 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2023 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2024 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2025 sl->mb_y, ER_MB_END);
2026 if (sl->mb_x >= lf_x_start)
2027 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2030 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2031 av_log(h->avctx, AV_LOG_ERROR,
2032 "error while decoding MB %d %d, bytestream %td\n",
2034 sl->cabac.bytestream_end - sl->cabac.bytestream);
2035 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2036 sl->mb_y, ER_MB_ERROR);
2037 return AVERROR_INVALIDDATA;
2040 if (++sl->mb_x >= h->mb_width) {
2041 loop_filter(h, sl, lf_x_start, sl->mb_x);
2042 sl->mb_x = lf_x_start = 0;
2043 decode_finish_row(h, sl);
2045 if (FIELD_OR_MBAFF_PICTURE(h)) {
2047 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2048 predict_field_decoding_flag(h, sl);
2052 if (eos || sl->mb_y >= h->mb_height) {
2053 ff_tlog(h->avctx, "slice end %d %d\n",
2054 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2055 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2056 sl->mb_y, ER_MB_END);
2057 if (sl->mb_x > lf_x_start)
2058 loop_filter(h, sl, lf_x_start, sl->mb_x);
2066 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2067 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2068 sl->next_slice_idx);
2069 return AVERROR_INVALIDDATA;
2072 ret = ff_h264_decode_mb_cavlc(h, sl);
2075 ff_h264_hl_decode_mb(h, sl);
2077 // FIXME optimal? or let mb_decode decode 16x32 ?
2078 if (ret >= 0 && FRAME_MBAFF(h)) {
2080 ret = ff_h264_decode_mb_cavlc(h, sl);
2083 ff_h264_hl_decode_mb(h, sl);
2088 av_log(h->avctx, AV_LOG_ERROR,
2089 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2090 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2091 sl->mb_y, ER_MB_ERROR);
2095 if (++sl->mb_x >= h->mb_width) {
2096 loop_filter(h, sl, lf_x_start, sl->mb_x);
2097 sl->mb_x = lf_x_start = 0;
2098 decode_finish_row(h, sl);
2100 if (FIELD_OR_MBAFF_PICTURE(h)) {
2102 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2103 predict_field_decoding_flag(h, sl);
2105 if (sl->mb_y >= h->mb_height) {
2106 ff_tlog(h->avctx, "slice end %d %d\n",
2107 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2109 if (get_bits_left(&sl->gb) == 0) {
2110 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2111 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2115 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2116 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2118 return AVERROR_INVALIDDATA;
2123 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2124 ff_tlog(h->avctx, "slice end %d %d\n",
2125 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2127 if (get_bits_left(&sl->gb) == 0) {
2128 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2129 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2130 if (sl->mb_x > lf_x_start)
2131 loop_filter(h, sl, lf_x_start, sl->mb_x);
2135 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2136 sl->mb_y, ER_MB_ERROR);
2138 return AVERROR_INVALIDDATA;
2146 * Call decode_slice() for each context.
2148 * @param h h264 master context
2149 * @param context_count number of contexts to execute
2151 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2153 AVCodecContext *const avctx = h->avctx;
2154 H264SliceContext *sl;
2157 if (h->avctx->hwaccel)
2159 if (context_count == 1) {
2162 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2164 ret = decode_slice(avctx, &h->slice_ctx[0]);
2165 h->mb_y = h->slice_ctx[0].mb_y;
2168 for (i = 0; i < context_count; i++) {
2169 int next_slice_idx = h->mb_width * h->mb_height;
2172 sl = &h->slice_ctx[i];
2173 sl->er.error_count = 0;
2175 /* make sure none of those slices overlap */
2176 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2177 for (j = 0; j < context_count; j++) {
2178 H264SliceContext *sl2 = &h->slice_ctx[j];
2179 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2181 if (i == j || slice_idx2 < slice_idx)
2183 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2185 sl->next_slice_idx = next_slice_idx;
2188 avctx->execute(avctx, decode_slice, h->slice_ctx,
2189 NULL, context_count, sizeof(h->slice_ctx[0]));
2191 /* pull back stuff from slices to master context */
2192 sl = &h->slice_ctx[context_count - 1];
2194 for (i = 1; i < context_count; i++)
2195 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;