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/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 H.264)
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->mb_aff_frame = h1->mb_aff_frame;
381 h->droppable = h1->droppable;
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->droppable = h1->droppable;
402 // extradata/NAL handling
403 h->is_avc = h1->is_avc;
404 h->nal_length_size = h1->nal_length_size;
406 memcpy(&h->poc, &h1->poc, sizeof(h->poc));
408 memcpy(h->short_ref, h1->short_ref, sizeof(h->short_ref));
409 memcpy(h->long_ref, h1->long_ref, sizeof(h->long_ref));
410 memcpy(h->delayed_pic, h1->delayed_pic, sizeof(h->delayed_pic));
411 memcpy(h->last_pocs, h1->last_pocs, sizeof(h->last_pocs));
413 h->next_output_pic = h1->next_output_pic;
414 h->next_outputed_poc = h1->next_outputed_poc;
416 memcpy(h->mmco, h1->mmco, sizeof(h->mmco));
417 h->nb_mmco = h1->nb_mmco;
418 h->mmco_reset = h1->mmco_reset;
419 h->explicit_ref_marking = h1->explicit_ref_marking;
420 h->long_ref_count = h1->long_ref_count;
421 h->short_ref_count = h1->short_ref_count;
423 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
424 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
425 copy_picture_range(h->delayed_pic, h1->delayed_pic,
426 MAX_DELAYED_PIC_COUNT + 2, h, h1);
432 err = ff_h264_execute_ref_pic_marking(h);
433 h->poc.prev_poc_msb = h->poc.poc_msb;
434 h->poc.prev_poc_lsb = h->poc.poc_lsb;
436 h->poc.prev_frame_num_offset = h->poc.frame_num_offset;
437 h->poc.prev_frame_num = h->poc.frame_num;
439 h->recovery_frame = h1->recovery_frame;
440 h->frame_recovered = h1->frame_recovered;
445 static int h264_frame_start(H264Context *h)
449 const int pixel_shift = h->pixel_shift;
451 ret = initialize_cur_frame(h);
455 pic = h->cur_pic_ptr;
456 pic->reference = h->droppable ? 0 : h->picture_structure;
457 pic->f->coded_picture_number = h->coded_picture_number++;
458 pic->field_picture = h->picture_structure != PICT_FRAME;
459 pic->frame_num = h->poc.frame_num;
461 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
463 * See decode_nal_units().
465 pic->f->key_frame = 0;
469 pic->f->pict_type = h->slice_ctx[0].slice_type;
471 if (CONFIG_ERROR_RESILIENCE && h->enable_er)
472 ff_er_frame_start(&h->slice_ctx[0].er);
474 for (i = 0; i < 16; i++) {
475 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
476 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
478 for (i = 0; i < 16; i++) {
479 h->block_offset[16 + i] =
480 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
481 h->block_offset[48 + 16 + i] =
482 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
485 /* Some macroblocks can be accessed before they're available in case
486 * of lost slices, MBAFF or threading. */
487 memset(h->slice_table, -1,
488 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
490 /* We mark the current picture as non-reference after allocating it, so
491 * that if we break out due to an error it can be released automatically
492 * in the next ff_mpv_frame_start().
494 h->cur_pic_ptr->reference = 0;
496 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
498 h->next_output_pic = NULL;
500 h->postpone_filter = 0;
502 h->mb_aff_frame = h->ps.sps->mb_aff && (h->picture_structure == PICT_FRAME);
504 assert(h->cur_pic_ptr->long_ref == 0);
509 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
511 uint8_t *src_cb, uint8_t *src_cr,
512 int linesize, int uvlinesize,
517 const int pixel_shift = h->pixel_shift;
518 int chroma444 = CHROMA444(h);
519 int chroma422 = CHROMA422(h);
522 src_cb -= uvlinesize;
523 src_cr -= uvlinesize;
525 if (!simple && FRAME_MBAFF(h)) {
528 top_border = sl->top_borders[0][sl->mb_x];
529 AV_COPY128(top_border, src_y + 15 * linesize);
531 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
532 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
535 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
536 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
537 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
538 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
540 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
541 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
543 } else if (chroma422) {
545 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
546 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
548 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
549 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
553 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
554 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
556 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
557 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
562 } else if (MB_MBAFF(sl)) {
568 top_border = sl->top_borders[top_idx][sl->mb_x];
569 /* There are two lines saved, the line above the top macroblock
570 * of a pair, and the line above the bottom macroblock. */
571 AV_COPY128(top_border, src_y + 16 * linesize);
573 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
575 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
578 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
579 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
580 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
581 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
583 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
584 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
586 } else if (chroma422) {
588 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
589 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
591 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
592 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
596 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
597 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
599 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
600 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
607 * Initialize implicit_weight table.
608 * @param field 0/1 initialize the weight for interlaced MBAFF
609 * -1 initializes the rest
611 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
613 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
615 for (i = 0; i < 2; i++) {
616 sl->pwt.luma_weight_flag[i] = 0;
617 sl->pwt.chroma_weight_flag[i] = 0;
621 if (h->picture_structure == PICT_FRAME) {
622 cur_poc = h->cur_pic_ptr->poc;
624 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
626 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
627 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
628 sl->pwt.use_weight = 0;
629 sl->pwt.use_weight_chroma = 0;
633 ref_count0 = sl->ref_count[0];
634 ref_count1 = sl->ref_count[1];
636 cur_poc = h->cur_pic_ptr->field_poc[field];
638 ref_count0 = 16 + 2 * sl->ref_count[0];
639 ref_count1 = 16 + 2 * sl->ref_count[1];
642 sl->pwt.use_weight = 2;
643 sl->pwt.use_weight_chroma = 2;
644 sl->pwt.luma_log2_weight_denom = 5;
645 sl->pwt.chroma_log2_weight_denom = 5;
647 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
648 int poc0 = sl->ref_list[0][ref0].poc;
649 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
651 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
652 int poc1 = sl->ref_list[1][ref1].poc;
653 int td = av_clip_int8(poc1 - poc0);
655 int tb = av_clip_int8(cur_poc - poc0);
656 int tx = (16384 + (FFABS(td) >> 1)) / td;
657 int dist_scale_factor = (tb * tx + 32) >> 8;
658 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
659 w = 64 - dist_scale_factor;
663 sl->pwt.implicit_weight[ref0][ref1][0] =
664 sl->pwt.implicit_weight[ref0][ref1][1] = w;
666 sl->pwt.implicit_weight[ref0][ref1][field] = w;
673 * initialize scan tables
675 static void init_scan_tables(H264Context *h)
678 for (i = 0; i < 16; i++) {
679 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
680 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
681 h->field_scan[i] = TRANSPOSE(field_scan[i]);
684 for (i = 0; i < 64; i++) {
685 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
686 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
687 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
688 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
689 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
692 if (h->ps.sps->transform_bypass) { // FIXME same ugly
693 h->zigzag_scan_q0 = ff_zigzag_scan;
694 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
695 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
696 h->field_scan_q0 = field_scan;
697 h->field_scan8x8_q0 = field_scan8x8;
698 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
700 h->zigzag_scan_q0 = h->zigzag_scan;
701 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
702 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
703 h->field_scan_q0 = h->field_scan;
704 h->field_scan8x8_q0 = h->field_scan8x8;
705 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
709 static enum AVPixelFormat get_pixel_format(H264Context *h)
711 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
712 CONFIG_H264_D3D11VA_HWACCEL + \
713 CONFIG_H264_VAAPI_HWACCEL + \
714 (CONFIG_H264_VDA_HWACCEL * 2) + \
715 CONFIG_H264_VDPAU_HWACCEL)
716 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
717 const enum AVPixelFormat *choices = pix_fmts;
719 switch (h->ps.sps->bit_depth_luma) {
722 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
723 *fmt++ = AV_PIX_FMT_GBRP9;
725 *fmt++ = AV_PIX_FMT_YUV444P9;
726 } else if (CHROMA422(h))
727 *fmt++ = AV_PIX_FMT_YUV422P9;
729 *fmt++ = AV_PIX_FMT_YUV420P9;
733 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
734 *fmt++ = AV_PIX_FMT_GBRP10;
736 *fmt++ = AV_PIX_FMT_YUV444P10;
737 } else if (CHROMA422(h))
738 *fmt++ = AV_PIX_FMT_YUV422P10;
740 *fmt++ = AV_PIX_FMT_YUV420P10;
743 #if CONFIG_H264_VDPAU_HWACCEL
744 *fmt++ = AV_PIX_FMT_VDPAU;
747 if (h->avctx->colorspace == AVCOL_SPC_RGB)
748 *fmt++ = AV_PIX_FMT_GBRP;
749 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
750 *fmt++ = AV_PIX_FMT_YUVJ444P;
752 *fmt++ = AV_PIX_FMT_YUV444P;
753 } else if (CHROMA422(h)) {
754 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
755 *fmt++ = AV_PIX_FMT_YUVJ422P;
757 *fmt++ = AV_PIX_FMT_YUV422P;
759 #if CONFIG_H264_DXVA2_HWACCEL
760 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
762 #if CONFIG_H264_D3D11VA_HWACCEL
763 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
765 #if CONFIG_H264_VAAPI_HWACCEL
766 *fmt++ = AV_PIX_FMT_VAAPI;
768 #if CONFIG_H264_VDA_HWACCEL
769 *fmt++ = AV_PIX_FMT_VDA_VLD;
770 *fmt++ = AV_PIX_FMT_VDA;
772 if (h->avctx->codec->pix_fmts)
773 choices = h->avctx->codec->pix_fmts;
774 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
775 *fmt++ = AV_PIX_FMT_YUVJ420P;
777 *fmt++ = AV_PIX_FMT_YUV420P;
781 av_log(h->avctx, AV_LOG_ERROR,
782 "Unsupported bit depth %d\n", h->ps.sps->bit_depth_luma);
783 return AVERROR_INVALIDDATA;
786 *fmt = AV_PIX_FMT_NONE;
788 return ff_get_format(h->avctx, choices);
791 /* export coded and cropped frame dimensions to AVCodecContext */
792 static int init_dimensions(H264Context *h)
794 SPS *sps = h->ps.sps;
795 int width = h->width - (sps->crop_right + sps->crop_left);
796 int height = h->height - (sps->crop_top + sps->crop_bottom);
798 /* handle container cropping */
799 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
800 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16)) {
801 width = h->avctx->width;
802 height = h->avctx->height;
805 if (width <= 0 || height <= 0) {
806 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
808 if (h->avctx->err_recognition & AV_EF_EXPLODE)
809 return AVERROR_INVALIDDATA;
811 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
822 h->avctx->coded_width = h->width;
823 h->avctx->coded_height = h->height;
824 h->avctx->width = width;
825 h->avctx->height = height;
830 static int h264_slice_header_init(H264Context *h)
832 const SPS *sps = h->ps.sps;
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 (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
879 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
881 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
885 for (i = 0; i < h->nb_slice_ctx; i++) {
886 H264SliceContext *sl = &h->slice_ctx[i];
889 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
890 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
891 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
893 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
894 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
900 h->context_initialized = 1;
905 static int h264_init_ps(H264Context *h, const H264SliceContext *sl)
908 int needs_reinit = 0, ret;
910 h->ps.pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
911 if (h->ps.sps != (const SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data) {
912 h->ps.sps = (SPS*)h->ps.sps_list[h->ps.pps->sps_id]->data;
914 if (h->bit_depth_luma != h->ps.sps->bit_depth_luma ||
915 h->chroma_format_idc != h->ps.sps->chroma_format_idc)
920 h->avctx->profile = ff_h264_get_profile(sps);
921 h->avctx->level = sps->level_idc;
922 h->avctx->refs = sps->ref_frame_count;
924 if (h->mb_width != sps->mb_width ||
925 h->mb_height != sps->mb_height * (2 - sps->frame_mbs_only_flag))
928 h->mb_width = sps->mb_width;
929 h->mb_height = sps->mb_height * (2 - sps->frame_mbs_only_flag);
930 h->mb_num = h->mb_width * h->mb_height;
931 h->mb_stride = h->mb_width + 1;
933 h->b_stride = h->mb_width * 4;
935 h->chroma_y_shift = sps->chroma_format_idc <= 1; // 400 uses yuv420p
937 h->width = 16 * h->mb_width;
938 h->height = 16 * h->mb_height;
940 ret = init_dimensions(h);
944 if (sps->video_signal_type_present_flag) {
945 h->avctx->color_range = sps->full_range ? AVCOL_RANGE_JPEG
947 if (sps->colour_description_present_flag) {
948 if (h->avctx->colorspace != sps->colorspace)
950 h->avctx->color_primaries = sps->color_primaries;
951 h->avctx->color_trc = sps->color_trc;
952 h->avctx->colorspace = sps->colorspace;
956 if (!h->context_initialized || needs_reinit) {
957 h->context_initialized = 0;
958 if (sl != h->slice_ctx) {
959 av_log(h->avctx, AV_LOG_ERROR,
960 "changing width %d -> %d / height %d -> %d on "
962 h->width, h->avctx->coded_width,
963 h->height, h->avctx->coded_height,
964 h->current_slice + 1);
965 return AVERROR_INVALIDDATA;
968 ff_h264_flush_change(h);
970 if ((ret = get_pixel_format(h)) < 0)
972 h->avctx->pix_fmt = ret;
974 av_log(h->avctx, AV_LOG_VERBOSE, "Reinit context to %dx%d, "
975 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
977 if ((ret = h264_slice_header_init(h)) < 0) {
978 av_log(h->avctx, AV_LOG_ERROR,
979 "h264_slice_header_init() failed\n");
987 /* This function is called right after decoding the slice header for a first
988 * slice in a field (or a frame). It decides whether we are decoding a new frame
989 * or a second field in a pair and does the necessary setup.
991 static int h264_field_start(H264Context *h, const H264SliceContext *sl,
996 int last_pic_structure, last_pic_droppable, ret;
998 ret = h264_init_ps(h, sl);
1004 last_pic_droppable = h->droppable;
1005 last_pic_structure = h->picture_structure;
1006 h->droppable = (nal->ref_idc == 0);
1007 h->picture_structure = sl->picture_structure;
1009 h->poc.frame_num = sl->frame_num;
1010 h->poc.poc_lsb = sl->poc_lsb;
1011 h->poc.delta_poc_bottom = sl->delta_poc_bottom;
1012 h->poc.delta_poc[0] = sl->delta_poc[0];
1013 h->poc.delta_poc[1] = sl->delta_poc[1];
1015 /* Shorten frame num gaps so we don't have to allocate reference
1016 * frames just to throw them away */
1017 if (h->poc.frame_num != h->poc.prev_frame_num) {
1018 int unwrap_prev_frame_num = h->poc.prev_frame_num;
1019 int max_frame_num = 1 << sps->log2_max_frame_num;
1021 if (unwrap_prev_frame_num > h->poc.frame_num)
1022 unwrap_prev_frame_num -= max_frame_num;
1024 if ((h->poc.frame_num - unwrap_prev_frame_num) > sps->ref_frame_count) {
1025 unwrap_prev_frame_num = (h->poc.frame_num - sps->ref_frame_count) - 1;
1026 if (unwrap_prev_frame_num < 0)
1027 unwrap_prev_frame_num += max_frame_num;
1029 h->poc.prev_frame_num = unwrap_prev_frame_num;
1033 /* See if we have a decoded first field looking for a pair...
1034 * Here, we're using that to see if we should mark previously
1035 * decode frames as "finished".
1036 * We have to do that before the "dummy" in-between frame allocation,
1037 * since that can modify s->current_picture_ptr. */
1038 if (h->first_field) {
1039 assert(h->cur_pic_ptr);
1040 assert(h->cur_pic_ptr->f->buf[0]);
1041 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1043 /* figure out if we have a complementary field pair */
1044 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1045 /* Previous field is unmatched. Don't display it, but let it
1046 * remain for reference if marked as such. */
1047 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1048 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1049 last_pic_structure == PICT_TOP_FIELD);
1052 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1053 /* This and previous field were reference, but had
1054 * different frame_nums. Consider this field first in
1055 * pair. Throw away previous field except for reference
1057 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1058 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1059 last_pic_structure == PICT_TOP_FIELD);
1062 /* Second field in complementary pair */
1063 if (!((last_pic_structure == PICT_TOP_FIELD &&
1064 h->picture_structure == PICT_BOTTOM_FIELD) ||
1065 (last_pic_structure == PICT_BOTTOM_FIELD &&
1066 h->picture_structure == PICT_TOP_FIELD))) {
1067 av_log(h->avctx, AV_LOG_ERROR,
1068 "Invalid field mode combination %d/%d\n",
1069 last_pic_structure, h->picture_structure);
1070 h->picture_structure = last_pic_structure;
1071 h->droppable = last_pic_droppable;
1072 return AVERROR_INVALIDDATA;
1073 } else if (last_pic_droppable != h->droppable) {
1074 avpriv_request_sample(h->avctx,
1075 "Found reference and non-reference fields in the same frame, which");
1076 h->picture_structure = last_pic_structure;
1077 h->droppable = last_pic_droppable;
1078 return AVERROR_PATCHWELCOME;
1084 while (h->poc.frame_num != h->poc.prev_frame_num &&
1085 h->poc.frame_num != (h->poc.prev_frame_num + 1) % (1 << sps->log2_max_frame_num)) {
1086 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1087 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1088 h->poc.frame_num, h->poc.prev_frame_num);
1089 ret = initialize_cur_frame(h);
1095 h->poc.prev_frame_num++;
1096 h->poc.prev_frame_num %= 1 << sps->log2_max_frame_num;
1097 h->cur_pic_ptr->frame_num = h->poc.prev_frame_num;
1098 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1099 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1101 h->explicit_ref_marking = 0;
1102 ret = ff_h264_execute_ref_pic_marking(h);
1103 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1105 /* Error concealment: If a ref is missing, copy the previous ref
1107 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1108 * many assumptions about there being no actual duplicates.
1109 * FIXME: This does not copy padding for out-of-frame motion
1110 * vectors. Given we are concealing a lost frame, this probably
1111 * is not noticeable by comparison, but it should be fixed. */
1112 if (h->short_ref_count) {
1114 h->short_ref[0]->f->width == prev->f->width &&
1115 h->short_ref[0]->f->height == prev->f->height &&
1116 h->short_ref[0]->f->format == prev->f->format) {
1117 av_image_copy(h->short_ref[0]->f->data,
1118 h->short_ref[0]->f->linesize,
1119 (const uint8_t **)prev->f->data,
1124 h->short_ref[0]->poc = prev->poc + 2;
1126 h->short_ref[0]->frame_num = h->poc.prev_frame_num;
1130 /* See if we have a decoded first field looking for a pair...
1131 * We're using that to see whether to continue decoding in that
1132 * frame, or to allocate a new one. */
1133 if (h->first_field) {
1134 assert(h->cur_pic_ptr);
1135 assert(h->cur_pic_ptr->f->buf[0]);
1136 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1138 /* figure out if we have a complementary field pair */
1139 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1140 /* Previous field is unmatched. Don't display it, but let it
1141 * remain for reference if marked as such. */
1142 h->cur_pic_ptr = NULL;
1143 h->first_field = FIELD_PICTURE(h);
1145 if (h->cur_pic_ptr->frame_num != h->poc.frame_num) {
1146 /* This and the previous field had different frame_nums.
1147 * Consider this field first in pair. Throw away previous
1148 * one except for reference purposes. */
1150 h->cur_pic_ptr = NULL;
1152 /* Second field in complementary pair */
1157 /* Frame or first field in a potentially complementary pair */
1158 h->first_field = FIELD_PICTURE(h);
1161 if (!FIELD_PICTURE(h) || h->first_field) {
1162 if (h264_frame_start(h) < 0) {
1164 return AVERROR_INVALIDDATA;
1167 release_unused_pictures(h, 0);
1173 static int h264_slice_header_parse(const H264Context *h, H264SliceContext *sl,
1174 const H2645NAL *nal)
1179 unsigned int slice_type, tmp, i;
1180 int field_pic_flag, bottom_field_flag;
1181 int droppable, picture_structure;
1183 sl->first_mb_addr = get_ue_golomb(&sl->gb);
1185 slice_type = get_ue_golomb_31(&sl->gb);
1186 if (slice_type > 9) {
1187 av_log(h->avctx, AV_LOG_ERROR,
1188 "slice type %d too large at %d\n",
1189 slice_type, sl->first_mb_addr);
1190 return AVERROR_INVALIDDATA;
1192 if (slice_type > 4) {
1194 sl->slice_type_fixed = 1;
1196 sl->slice_type_fixed = 0;
1198 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1199 sl->slice_type = slice_type;
1200 sl->slice_type_nos = slice_type & 3;
1202 if (nal->type == NAL_IDR_SLICE &&
1203 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1204 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1205 return AVERROR_INVALIDDATA;
1208 sl->pps_id = get_ue_golomb(&sl->gb);
1209 if (sl->pps_id >= MAX_PPS_COUNT) {
1210 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", sl->pps_id);
1211 return AVERROR_INVALIDDATA;
1213 if (!h->ps.pps_list[sl->pps_id]) {
1214 av_log(h->avctx, AV_LOG_ERROR,
1215 "non-existing PPS %u referenced\n",
1217 return AVERROR_INVALIDDATA;
1219 pps = (const PPS*)h->ps.pps_list[sl->pps_id]->data;
1221 if (!h->ps.sps_list[pps->sps_id]) {
1222 av_log(h->avctx, AV_LOG_ERROR,
1223 "non-existing SPS %u referenced\n", pps->sps_id);
1224 return AVERROR_INVALIDDATA;
1226 sps = (const SPS*)h->ps.sps_list[pps->sps_id]->data;
1228 sl->frame_num = get_bits(&sl->gb, sps->log2_max_frame_num);
1232 droppable = nal->ref_idc == 0;
1233 if (sps->frame_mbs_only_flag) {
1234 picture_structure = PICT_FRAME;
1236 field_pic_flag = get_bits1(&sl->gb);
1237 if (field_pic_flag) {
1238 bottom_field_flag = get_bits1(&sl->gb);
1239 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1241 picture_structure = PICT_FRAME;
1244 sl->picture_structure = picture_structure;
1245 sl->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1247 if (picture_structure == PICT_FRAME) {
1248 sl->curr_pic_num = sl->frame_num;
1249 sl->max_pic_num = 1 << sps->log2_max_frame_num;
1251 sl->curr_pic_num = 2 * sl->frame_num + 1;
1252 sl->max_pic_num = 1 << (sps->log2_max_frame_num + 1);
1255 if (nal->type == NAL_IDR_SLICE)
1256 get_ue_golomb(&sl->gb); /* idr_pic_id */
1258 if (sps->poc_type == 0) {
1259 sl->poc_lsb = get_bits(&sl->gb, sps->log2_max_poc_lsb);
1261 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1262 sl->delta_poc_bottom = get_se_golomb(&sl->gb);
1265 if (sps->poc_type == 1 && !sps->delta_pic_order_always_zero_flag) {
1266 sl->delta_poc[0] = get_se_golomb(&sl->gb);
1268 if (pps->pic_order_present == 1 && picture_structure == PICT_FRAME)
1269 sl->delta_poc[1] = get_se_golomb(&sl->gb);
1272 if (pps->redundant_pic_cnt_present)
1273 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1275 if (sl->slice_type_nos == AV_PICTURE_TYPE_B)
1276 sl->direct_spatial_mv_pred = get_bits1(&sl->gb);
1278 ret = ff_h264_parse_ref_count(&sl->list_count, sl->ref_count,
1279 &sl->gb, pps, sl->slice_type_nos,
1284 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1285 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1287 sl->ref_count[1] = sl->ref_count[0] = 0;
1292 sl->pwt.use_weight = 0;
1293 for (i = 0; i < 2; i++) {
1294 sl->pwt.luma_weight_flag[i] = 0;
1295 sl->pwt.chroma_weight_flag[i] = 0;
1297 if ((pps->weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1298 (pps->weighted_bipred_idc == 1 &&
1299 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1300 ff_h264_pred_weight_table(&sl->gb, sps, sl->ref_count,
1301 sl->slice_type_nos, &sl->pwt);
1303 sl->explicit_ref_marking = 0;
1305 ret = ff_h264_decode_ref_pic_marking(h, sl, &sl->gb);
1306 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1307 return AVERROR_INVALIDDATA;
1310 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && pps->cabac) {
1311 tmp = get_ue_golomb_31(&sl->gb);
1313 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1314 return AVERROR_INVALIDDATA;
1316 sl->cabac_init_idc = tmp;
1319 sl->last_qscale_diff = 0;
1320 tmp = pps->init_qp + get_se_golomb(&sl->gb);
1321 if (tmp > 51 + 6 * (sps->bit_depth_luma - 8)) {
1322 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1323 return AVERROR_INVALIDDATA;
1326 sl->chroma_qp[0] = get_chroma_qp(pps, 0, sl->qscale);
1327 sl->chroma_qp[1] = get_chroma_qp(pps, 1, sl->qscale);
1328 // FIXME qscale / qp ... stuff
1329 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1330 get_bits1(&sl->gb); /* sp_for_switch_flag */
1331 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1332 sl->slice_type == AV_PICTURE_TYPE_SI)
1333 get_se_golomb(&sl->gb); /* slice_qs_delta */
1335 sl->deblocking_filter = 1;
1336 sl->slice_alpha_c0_offset = 0;
1337 sl->slice_beta_offset = 0;
1338 if (pps->deblocking_filter_parameters_present) {
1339 tmp = get_ue_golomb_31(&sl->gb);
1341 av_log(h->avctx, AV_LOG_ERROR,
1342 "deblocking_filter_idc %u out of range\n", tmp);
1343 return AVERROR_INVALIDDATA;
1345 sl->deblocking_filter = tmp;
1346 if (sl->deblocking_filter < 2)
1347 sl->deblocking_filter ^= 1; // 1<->0
1349 if (sl->deblocking_filter) {
1350 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1351 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1352 if (sl->slice_alpha_c0_offset > 12 ||
1353 sl->slice_alpha_c0_offset < -12 ||
1354 sl->slice_beta_offset > 12 ||
1355 sl->slice_beta_offset < -12) {
1356 av_log(h->avctx, AV_LOG_ERROR,
1357 "deblocking filter parameters %d %d out of range\n",
1358 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1359 return AVERROR_INVALIDDATA;
1368 * Decode a slice header.
1369 * This will (re)initialize the decoder and call h264_frame_start() as needed.
1371 * @param h h264context
1373 * @return 0 if okay, <0 if an error occurred
1375 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl,
1376 const H2645NAL *nal)
1380 ret = h264_slice_header_parse(h, sl, nal);
1384 if (sl->first_mb_addr == 0) { // FIXME better field boundary detection
1385 if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1386 ff_h264_field_end(h, sl, 1);
1389 h->current_slice = 0;
1390 if (!h->first_field) {
1391 if (h->cur_pic_ptr && !h->droppable) {
1392 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1393 h->picture_structure == PICT_BOTTOM_FIELD);
1395 h->cur_pic_ptr = NULL;
1399 if (h->current_slice == 0) {
1400 ret = h264_field_start(h, sl, nal);
1404 if (h->ps.pps != (const PPS*)h->ps.pps_list[sl->pps_id]->data) {
1405 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1406 return AVERROR_INVALIDDATA;
1409 if (h->picture_structure != sl->picture_structure ||
1410 h->droppable != (nal->ref_idc == 0)) {
1411 av_log(h->avctx, AV_LOG_ERROR,
1412 "Changing field mode (%d -> %d) between slices is not allowed\n",
1413 h->picture_structure, sl->picture_structure);
1414 return AVERROR_INVALIDDATA;
1415 } else if (!h->cur_pic_ptr) {
1416 av_log(h->avctx, AV_LOG_ERROR,
1417 "unset cur_pic_ptr on slice %d\n",
1418 h->current_slice + 1);
1419 return AVERROR_INVALIDDATA;
1423 assert(h->mb_num == h->mb_width * h->mb_height);
1424 if (sl->first_mb_addr << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1425 sl->first_mb_addr >= h->mb_num) {
1426 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1427 return AVERROR_INVALIDDATA;
1429 sl->resync_mb_x = sl->mb_x = sl->first_mb_addr % h->mb_width;
1430 sl->resync_mb_y = sl->mb_y = (sl->first_mb_addr / h->mb_width) <<
1431 FIELD_OR_MBAFF_PICTURE(h);
1432 if (h->picture_structure == PICT_BOTTOM_FIELD)
1433 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1434 assert(sl->mb_y < h->mb_height);
1436 if (!h->setup_finished) {
1437 ff_h264_init_poc(h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc,
1438 h->ps.sps, &h->poc, h->picture_structure, nal->ref_idc);
1440 memcpy(h->mmco, sl->mmco, sl->nb_mmco * sizeof(*h->mmco));
1441 h->nb_mmco = sl->nb_mmco;
1442 h->explicit_ref_marking = sl->explicit_ref_marking;
1445 ret = ff_h264_build_ref_list(h, sl);
1449 if (h->ps.pps->weighted_bipred_idc == 2 &&
1450 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1451 implicit_weight_table(h, sl, -1);
1452 if (FRAME_MBAFF(h)) {
1453 implicit_weight_table(h, sl, 0);
1454 implicit_weight_table(h, sl, 1);
1458 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1459 ff_h264_direct_dist_scale_factor(h, sl);
1460 ff_h264_direct_ref_list_init(h, sl);
1462 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1463 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1464 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1465 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1466 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1467 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1469 sl->deblocking_filter = 0;
1471 if (sl->deblocking_filter == 1 && h->nb_slice_ctx > 1) {
1472 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1473 /* Cheat slightly for speed:
1474 * Do not bother to deblock across slices. */
1475 sl->deblocking_filter = 2;
1477 h->postpone_filter = 1;
1480 sl->qp_thresh = 15 -
1481 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1483 h->ps.pps->chroma_qp_index_offset[0],
1484 h->ps.pps->chroma_qp_index_offset[1]) +
1485 6 * (h->ps.sps->bit_depth_luma - 8);
1487 sl->slice_num = ++h->current_slice;
1488 if (sl->slice_num >= MAX_SLICES) {
1489 av_log(h->avctx, AV_LOG_ERROR,
1490 "Too many slices, increase MAX_SLICES and recompile\n");
1493 for (j = 0; j < 2; j++) {
1495 int *ref2frm = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1496 for (i = 0; i < 16; i++) {
1498 if (j < sl->list_count && i < sl->ref_count[j] &&
1499 sl->ref_list[j][i].parent->f->buf[0]) {
1501 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1502 for (k = 0; k < h->short_ref_count; k++)
1503 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1507 for (k = 0; k < h->long_ref_count; k++)
1508 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1509 id_list[i] = h->short_ref_count + k;
1517 for (i = 0; i < 16; i++)
1518 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1520 ref2frm[18 + 1] = -1;
1521 for (i = 16; i < 48; i++)
1522 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1523 (sl->ref_list[j][i].reference & 3);
1526 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1527 av_log(h->avctx, AV_LOG_DEBUG,
1528 "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",
1530 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1531 sl->mb_y * h->mb_width + sl->mb_x,
1532 av_get_picture_type_char(sl->slice_type),
1533 sl->slice_type_fixed ? " fix" : "",
1534 nal->type == NAL_IDR_SLICE ? " IDR" : "",
1536 h->cur_pic_ptr->field_poc[0],
1537 h->cur_pic_ptr->field_poc[1],
1538 sl->ref_count[0], sl->ref_count[1],
1540 sl->deblocking_filter,
1541 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1543 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1544 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1550 int ff_h264_get_slice_type(const H264SliceContext *sl)
1552 switch (sl->slice_type) {
1553 case AV_PICTURE_TYPE_P:
1555 case AV_PICTURE_TYPE_B:
1557 case AV_PICTURE_TYPE_I:
1559 case AV_PICTURE_TYPE_SP:
1561 case AV_PICTURE_TYPE_SI:
1564 return AVERROR_INVALIDDATA;
1568 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1569 H264SliceContext *sl,
1570 int mb_type, int top_xy,
1571 int left_xy[LEFT_MBS],
1573 int left_type[LEFT_MBS],
1574 int mb_xy, int list)
1576 int b_stride = h->b_stride;
1577 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1578 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1579 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1580 if (USES_LIST(top_type, list)) {
1581 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1582 const int b8_xy = 4 * top_xy + 2;
1583 int (*ref2frm)[64] = h->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1584 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1585 ref_cache[0 - 1 * 8] =
1586 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1587 ref_cache[2 - 1 * 8] =
1588 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1590 AV_ZERO128(mv_dst - 1 * 8);
1591 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1594 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1595 if (USES_LIST(left_type[LTOP], list)) {
1596 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1597 const int b8_xy = 4 * left_xy[LTOP] + 1;
1598 int (*ref2frm)[64] = h->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1599 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1600 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1601 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1602 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1604 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1605 ref_cache[-1 + 16] =
1606 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1608 AV_ZERO32(mv_dst - 1 + 0);
1609 AV_ZERO32(mv_dst - 1 + 8);
1610 AV_ZERO32(mv_dst - 1 + 16);
1611 AV_ZERO32(mv_dst - 1 + 24);
1614 ref_cache[-1 + 16] =
1615 ref_cache[-1 + 24] = LIST_NOT_USED;
1620 if (!USES_LIST(mb_type, list)) {
1621 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1622 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1623 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1624 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1625 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1630 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1631 int (*ref2frm)[64] = h->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1632 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1633 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1634 AV_WN32A(&ref_cache[0 * 8], ref01);
1635 AV_WN32A(&ref_cache[1 * 8], ref01);
1636 AV_WN32A(&ref_cache[2 * 8], ref23);
1637 AV_WN32A(&ref_cache[3 * 8], ref23);
1641 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1642 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1643 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1644 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1645 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1650 * @return non zero if the loop filter can be skipped
1652 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
1654 const int mb_xy = sl->mb_xy;
1655 int top_xy, left_xy[LEFT_MBS];
1656 int top_type, left_type[LEFT_MBS];
1660 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1662 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1663 if (FRAME_MBAFF(h)) {
1664 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1665 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1667 if (left_mb_field_flag != curr_mb_field_flag)
1668 left_xy[LTOP] -= h->mb_stride;
1670 if (curr_mb_field_flag)
1671 top_xy += h->mb_stride &
1672 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1673 if (left_mb_field_flag != curr_mb_field_flag)
1674 left_xy[LBOT] += h->mb_stride;
1678 sl->top_mb_xy = top_xy;
1679 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1680 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1682 /* For sufficiently low qp, filtering wouldn't do anything.
1683 * This is a conservative estimate: could also check beta_offset
1684 * and more accurate chroma_qp. */
1685 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1686 int qp = h->cur_pic.qscale_table[mb_xy];
1687 if (qp <= qp_thresh &&
1688 (left_xy[LTOP] < 0 ||
1689 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1691 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1692 if (!FRAME_MBAFF(h))
1694 if ((left_xy[LTOP] < 0 ||
1695 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1696 (top_xy < h->mb_stride ||
1697 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1702 top_type = h->cur_pic.mb_type[top_xy];
1703 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1704 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1705 if (sl->deblocking_filter == 2) {
1706 if (h->slice_table[top_xy] != sl->slice_num)
1708 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
1709 left_type[LTOP] = left_type[LBOT] = 0;
1711 if (h->slice_table[top_xy] == 0xFFFF)
1713 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1714 left_type[LTOP] = left_type[LBOT] = 0;
1716 sl->top_type = top_type;
1717 sl->left_type[LTOP] = left_type[LTOP];
1718 sl->left_type[LBOT] = left_type[LBOT];
1720 if (IS_INTRA(mb_type))
1723 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1724 top_type, left_type, mb_xy, 0);
1725 if (sl->list_count == 2)
1726 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1727 top_type, left_type, mb_xy, 1);
1729 nnz = h->non_zero_count[mb_xy];
1730 nnz_cache = sl->non_zero_count_cache;
1731 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1732 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1733 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1734 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1735 sl->cbp = h->cbp_table[mb_xy];
1738 nnz = h->non_zero_count[top_xy];
1739 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1742 if (left_type[LTOP]) {
1743 nnz = h->non_zero_count[left_xy[LTOP]];
1744 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
1745 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
1746 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
1747 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
1750 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
1751 * from what the loop filter needs */
1752 if (!CABAC(h) && h->ps.pps->transform_8x8_mode) {
1753 if (IS_8x8DCT(top_type)) {
1754 nnz_cache[4 + 8 * 0] =
1755 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
1756 nnz_cache[6 + 8 * 0] =
1757 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
1759 if (IS_8x8DCT(left_type[LTOP])) {
1760 nnz_cache[3 + 8 * 1] =
1761 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
1763 if (IS_8x8DCT(left_type[LBOT])) {
1764 nnz_cache[3 + 8 * 3] =
1765 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
1768 if (IS_8x8DCT(mb_type)) {
1769 nnz_cache[scan8[0]] =
1770 nnz_cache[scan8[1]] =
1771 nnz_cache[scan8[2]] =
1772 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
1774 nnz_cache[scan8[0 + 4]] =
1775 nnz_cache[scan8[1 + 4]] =
1776 nnz_cache[scan8[2 + 4]] =
1777 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
1779 nnz_cache[scan8[0 + 8]] =
1780 nnz_cache[scan8[1 + 8]] =
1781 nnz_cache[scan8[2 + 8]] =
1782 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
1784 nnz_cache[scan8[0 + 12]] =
1785 nnz_cache[scan8[1 + 12]] =
1786 nnz_cache[scan8[2 + 12]] =
1787 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
1794 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
1796 uint8_t *dest_y, *dest_cb, *dest_cr;
1797 int linesize, uvlinesize, mb_x, mb_y;
1798 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
1799 const int old_slice_type = sl->slice_type;
1800 const int pixel_shift = h->pixel_shift;
1801 const int block_h = 16 >> h->chroma_y_shift;
1803 if (h->postpone_filter)
1806 if (sl->deblocking_filter) {
1807 for (mb_x = start_x; mb_x < end_x; mb_x++)
1808 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
1810 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
1811 mb_type = h->cur_pic.mb_type[mb_xy];
1815 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
1819 dest_y = h->cur_pic.f->data[0] +
1820 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
1821 dest_cb = h->cur_pic.f->data[1] +
1822 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1823 mb_y * sl->uvlinesize * block_h;
1824 dest_cr = h->cur_pic.f->data[2] +
1825 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1826 mb_y * sl->uvlinesize * block_h;
1827 // FIXME simplify above
1830 linesize = sl->mb_linesize = sl->linesize * 2;
1831 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
1832 if (mb_y & 1) { // FIXME move out of this function?
1833 dest_y -= sl->linesize * 15;
1834 dest_cb -= sl->uvlinesize * (block_h - 1);
1835 dest_cr -= sl->uvlinesize * (block_h - 1);
1838 linesize = sl->mb_linesize = sl->linesize;
1839 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
1841 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
1843 if (fill_filter_caches(h, sl, mb_type))
1845 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, h->cur_pic.qscale_table[mb_xy]);
1846 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, h->cur_pic.qscale_table[mb_xy]);
1848 if (FRAME_MBAFF(h)) {
1849 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
1850 linesize, uvlinesize);
1852 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
1853 dest_cr, linesize, uvlinesize);
1857 sl->slice_type = old_slice_type;
1859 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
1860 sl->chroma_qp[0] = get_chroma_qp(h->ps.pps, 0, sl->qscale);
1861 sl->chroma_qp[1] = get_chroma_qp(h->ps.pps, 1, sl->qscale);
1864 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
1866 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
1867 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
1868 h->cur_pic.mb_type[mb_xy - 1] :
1869 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
1870 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
1871 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1875 * Draw edges and report progress for the last MB row.
1877 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
1879 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
1880 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
1881 int height = 16 << FRAME_MBAFF(h);
1882 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
1884 if (sl->deblocking_filter) {
1885 if ((top + height) >= pic_height)
1886 height += deblock_border;
1887 top -= deblock_border;
1890 if (top >= pic_height || (top + height) < 0)
1893 height = FFMIN(height, pic_height - top);
1895 height = top + height;
1899 ff_h264_draw_horiz_band(h, sl, top, height);
1904 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
1905 h->picture_structure == PICT_BOTTOM_FIELD);
1908 static void er_add_slice(H264SliceContext *sl,
1909 int startx, int starty,
1910 int endx, int endy, int status)
1912 #if CONFIG_ERROR_RESILIENCE
1913 ERContext *er = &sl->er;
1915 if (!sl->h264->enable_er)
1918 er->ref_count = sl->ref_count[0];
1919 ff_er_add_slice(er, startx, starty, endx, endy, status);
1923 static int decode_slice(struct AVCodecContext *avctx, void *arg)
1925 H264SliceContext *sl = arg;
1926 const H264Context *h = sl->h264;
1927 int lf_x_start = sl->mb_x;
1928 int orig_deblock = sl->deblocking_filter;
1931 sl->linesize = h->cur_pic_ptr->f->linesize[0];
1932 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
1934 ret = alloc_scratch_buffers(sl, sl->linesize);
1938 sl->mb_skip_run = -1;
1940 if (h->postpone_filter)
1941 sl->deblocking_filter = 0;
1943 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
1944 avctx->codec_id != AV_CODEC_ID_H264 ||
1945 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
1947 if (h->ps.pps->cabac) {
1949 align_get_bits(&sl->gb);
1952 ff_init_cabac_decoder(&sl->cabac,
1953 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
1954 (get_bits_left(&sl->gb) + 7) / 8);
1956 ff_h264_init_cabac_states(h, sl);
1962 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
1963 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
1964 sl->next_slice_idx);
1965 return AVERROR_INVALIDDATA;
1968 ret = ff_h264_decode_mb_cabac(h, sl);
1969 // STOP_TIMER("decode_mb_cabac")
1972 ff_h264_hl_decode_mb(h, sl);
1974 // FIXME optimal? or let mb_decode decode 16x32 ?
1975 if (ret >= 0 && FRAME_MBAFF(h)) {
1978 ret = ff_h264_decode_mb_cabac(h, sl);
1981 ff_h264_hl_decode_mb(h, sl);
1984 eos = get_cabac_terminate(&sl->cabac);
1986 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
1987 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
1988 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
1989 sl->mb_y, ER_MB_END);
1990 if (sl->mb_x >= lf_x_start)
1991 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
1994 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
1995 av_log(h->avctx, AV_LOG_ERROR,
1996 "error while decoding MB %d %d, bytestream %td\n",
1998 sl->cabac.bytestream_end - sl->cabac.bytestream);
1999 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2000 sl->mb_y, ER_MB_ERROR);
2001 return AVERROR_INVALIDDATA;
2004 if (++sl->mb_x >= h->mb_width) {
2005 loop_filter(h, sl, lf_x_start, sl->mb_x);
2006 sl->mb_x = lf_x_start = 0;
2007 decode_finish_row(h, sl);
2009 if (FIELD_OR_MBAFF_PICTURE(h)) {
2011 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2012 predict_field_decoding_flag(h, sl);
2016 if (eos || sl->mb_y >= h->mb_height) {
2017 ff_tlog(h->avctx, "slice end %d %d\n",
2018 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2019 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2020 sl->mb_y, ER_MB_END);
2021 if (sl->mb_x > lf_x_start)
2022 loop_filter(h, sl, lf_x_start, sl->mb_x);
2030 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2031 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2032 sl->next_slice_idx);
2033 return AVERROR_INVALIDDATA;
2036 ret = ff_h264_decode_mb_cavlc(h, sl);
2039 ff_h264_hl_decode_mb(h, sl);
2041 // FIXME optimal? or let mb_decode decode 16x32 ?
2042 if (ret >= 0 && FRAME_MBAFF(h)) {
2044 ret = ff_h264_decode_mb_cavlc(h, sl);
2047 ff_h264_hl_decode_mb(h, sl);
2052 av_log(h->avctx, AV_LOG_ERROR,
2053 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2054 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2055 sl->mb_y, ER_MB_ERROR);
2059 if (++sl->mb_x >= h->mb_width) {
2060 loop_filter(h, sl, lf_x_start, sl->mb_x);
2061 sl->mb_x = lf_x_start = 0;
2062 decode_finish_row(h, sl);
2064 if (FIELD_OR_MBAFF_PICTURE(h)) {
2066 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2067 predict_field_decoding_flag(h, sl);
2069 if (sl->mb_y >= h->mb_height) {
2070 ff_tlog(h->avctx, "slice end %d %d\n",
2071 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2073 if (get_bits_left(&sl->gb) == 0) {
2074 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2075 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2079 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2080 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2082 return AVERROR_INVALIDDATA;
2087 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2088 ff_tlog(h->avctx, "slice end %d %d\n",
2089 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2091 if (get_bits_left(&sl->gb) == 0) {
2092 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2093 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2094 if (sl->mb_x > lf_x_start)
2095 loop_filter(h, sl, lf_x_start, sl->mb_x);
2099 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2100 sl->mb_y, ER_MB_ERROR);
2102 return AVERROR_INVALIDDATA;
2109 sl->deblocking_filter = orig_deblock;
2114 * Call decode_slice() for each context.
2116 * @param h h264 master context
2117 * @param context_count number of contexts to execute
2119 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2121 AVCodecContext *const avctx = h->avctx;
2122 H264SliceContext *sl;
2125 if (h->avctx->hwaccel)
2127 if (context_count == 1) {
2130 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2131 h->postpone_filter = 0;
2133 ret = decode_slice(avctx, &h->slice_ctx[0]);
2134 h->mb_y = h->slice_ctx[0].mb_y;
2137 for (i = 0; i < context_count; i++) {
2138 int next_slice_idx = h->mb_width * h->mb_height;
2141 sl = &h->slice_ctx[i];
2142 sl->er.error_count = 0;
2144 /* make sure none of those slices overlap */
2145 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2146 for (j = 0; j < context_count; j++) {
2147 H264SliceContext *sl2 = &h->slice_ctx[j];
2148 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2150 if (i == j || slice_idx2 < slice_idx)
2152 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2154 sl->next_slice_idx = next_slice_idx;
2157 avctx->execute(avctx, decode_slice, h->slice_ctx,
2158 NULL, context_count, sizeof(h->slice_ctx[0]));
2160 /* pull back stuff from slices to master context */
2161 sl = &h->slice_ctx[context_count - 1];
2163 for (i = 1; i < context_count; i++)
2164 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;
2166 if (h->postpone_filter) {
2167 h->postpone_filter = 0;
2169 for (i = 0; i < context_count; i++) {
2172 sl = &h->slice_ctx[i];
2173 y_end = FFMIN(sl->mb_y + 1, h->mb_height);
2174 x_end = (sl->mb_y >= h->mb_height) ? h->mb_width : sl->mb_x;
2176 for (j = sl->resync_mb_y; j < y_end; j += 1 + FIELD_OR_MBAFF_PICTURE(h)) {
2178 loop_filter(h, sl, j > sl->resync_mb_y ? 0 : sl->resync_mb_x,
2179 j == y_end - 1 ? x_end : h->mb_width);