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 static void init_dequant8_coeff_table(H264Context *h)
288 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
290 for (i = 0; i < 6; i++) {
291 h->dequant8_coeff[i] = h->dequant8_buffer[i];
292 for (j = 0; j < i; j++)
293 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
294 64 * sizeof(uint8_t))) {
295 h->dequant8_coeff[i] = h->dequant8_buffer[j];
301 for (q = 0; q < max_qp + 1; q++) {
302 int shift = ff_h264_quant_div6[q];
303 int idx = ff_h264_quant_rem6[q];
304 for (x = 0; x < 64; x++)
305 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
306 ((uint32_t)ff_h264_dequant8_coeff_init[idx][ff_h264_dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
307 h->pps.scaling_matrix8[i][x]) << shift;
312 static void init_dequant4_coeff_table(H264Context *h)
315 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
316 for (i = 0; i < 6; i++) {
317 h->dequant4_coeff[i] = h->dequant4_buffer[i];
318 for (j = 0; j < i; j++)
319 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
320 16 * sizeof(uint8_t))) {
321 h->dequant4_coeff[i] = h->dequant4_buffer[j];
327 for (q = 0; q < max_qp + 1; q++) {
328 int shift = ff_h264_quant_div6[q] + 2;
329 int idx = ff_h264_quant_rem6[q];
330 for (x = 0; x < 16; x++)
331 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
332 ((uint32_t)ff_h264_dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
333 h->pps.scaling_matrix4[i][x]) << shift;
338 void h264_init_dequant_tables(H264Context *h)
341 init_dequant4_coeff_table(h);
342 if (h->pps.transform_8x8_mode)
343 init_dequant8_coeff_table(h);
344 if (h->sps.transform_bypass) {
345 for (i = 0; i < 6; i++)
346 for (x = 0; x < 16; x++)
347 h->dequant4_coeff[i][0][x] = 1 << 6;
348 if (h->pps.transform_8x8_mode)
349 for (i = 0; i < 6; i++)
350 for (x = 0; x < 64; x++)
351 h->dequant8_coeff[i][0][x] = 1 << 6;
355 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
357 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
358 ((pic && pic >= old_ctx->DPB && \
359 pic < old_ctx->DPB + H264_MAX_PICTURE_COUNT) ? \
360 &new_ctx->DPB[pic - old_ctx->DPB] : NULL)
362 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
363 H264Context *new_base,
364 H264Context *old_base)
368 for (i = 0; i < count; i++) {
369 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
370 IN_RANGE(from[i], old_base->DPB,
371 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
373 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
377 static int copy_parameter_set(void **to, void **from, int count, int size)
381 for (i = 0; i < count; i++) {
382 if (to[i] && !from[i]) {
384 } else if (from[i] && !to[i]) {
385 to[i] = av_malloc(size);
387 return AVERROR(ENOMEM);
391 memcpy(to[i], from[i], size);
397 #define copy_fields(to, from, start_field, end_field) \
398 memcpy(&to->start_field, &from->start_field, \
399 (char *)&to->end_field - (char *)&to->start_field)
401 static int h264_slice_header_init(H264Context *h);
403 int ff_h264_update_thread_context(AVCodecContext *dst,
404 const AVCodecContext *src)
406 H264Context *h = dst->priv_data, *h1 = src->priv_data;
407 int inited = h->context_initialized, err = 0;
411 if (dst == src || !h1->context_initialized)
415 (h->width != h1->width ||
416 h->height != h1->height ||
417 h->mb_width != h1->mb_width ||
418 h->mb_height != h1->mb_height ||
419 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
420 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
421 h->sps.colorspace != h1->sps.colorspace)) {
426 if ((ret = copy_parameter_set((void **)h->sps_buffers,
427 (void **)h1->sps_buffers,
428 MAX_SPS_COUNT, sizeof(SPS))) < 0)
431 if ((ret = copy_parameter_set((void **)h->pps_buffers,
432 (void **)h1->pps_buffers,
433 MAX_PPS_COUNT, sizeof(PPS))) < 0)
437 if (need_reinit || !inited) {
438 h->width = h1->width;
439 h->height = h1->height;
440 h->mb_height = h1->mb_height;
441 h->mb_width = h1->mb_width;
442 h->mb_num = h1->mb_num;
443 h->mb_stride = h1->mb_stride;
444 h->b_stride = h1->b_stride;
446 if ((err = h264_slice_header_init(h)) < 0) {
447 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed");
451 /* copy block_offset since frame_start may not be called */
452 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
455 h->avctx->coded_height = h1->avctx->coded_height;
456 h->avctx->coded_width = h1->avctx->coded_width;
457 h->avctx->width = h1->avctx->width;
458 h->avctx->height = h1->avctx->height;
459 h->coded_picture_number = h1->coded_picture_number;
460 h->first_field = h1->first_field;
461 h->picture_structure = h1->picture_structure;
462 h->droppable = h1->droppable;
463 h->low_delay = h1->low_delay;
465 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
466 ff_h264_unref_picture(h, &h->DPB[i]);
467 if (h1->DPB[i].f->buf[0] &&
468 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
472 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
473 ff_h264_unref_picture(h, &h->cur_pic);
474 if (h1->cur_pic.f->buf[0]) {
475 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
480 h->enable_er = h1->enable_er;
481 h->workaround_bugs = h1->workaround_bugs;
482 h->low_delay = h1->low_delay;
483 h->droppable = h1->droppable;
485 // extradata/NAL handling
486 h->is_avc = h1->is_avc;
487 h->nal_length_size = h1->nal_length_size;
489 // Dequantization matrices
490 // FIXME these are big - can they be only copied when PPS changes?
491 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
493 for (i = 0; i < 6; i++)
494 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
495 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
497 for (i = 0; i < 6; i++)
498 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
499 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
501 h->dequant_coeff_pps = h1->dequant_coeff_pps;
504 copy_fields(h, h1, poc_lsb, current_slice);
506 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
507 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
508 copy_picture_range(h->delayed_pic, h1->delayed_pic,
509 MAX_DELAYED_PIC_COUNT + 2, h, h1);
515 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
516 h->prev_poc_msb = h->poc_msb;
517 h->prev_poc_lsb = h->poc_lsb;
519 h->prev_frame_num_offset = h->frame_num_offset;
520 h->prev_frame_num = h->frame_num;
522 h->recovery_frame = h1->recovery_frame;
523 h->frame_recovered = h1->frame_recovered;
528 static int h264_frame_start(H264Context *h)
532 const int pixel_shift = h->pixel_shift;
534 ret = initialize_cur_frame(h);
538 pic = h->cur_pic_ptr;
539 pic->reference = h->droppable ? 0 : h->picture_structure;
540 pic->f->coded_picture_number = h->coded_picture_number++;
541 pic->field_picture = h->picture_structure != PICT_FRAME;
542 pic->frame_num = h->frame_num;
544 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
546 * See decode_nal_units().
548 pic->f->key_frame = 0;
552 if (CONFIG_ERROR_RESILIENCE && h->enable_er)
553 ff_er_frame_start(&h->slice_ctx[0].er);
555 for (i = 0; i < 16; i++) {
556 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
557 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);
559 for (i = 0; i < 16; i++) {
560 h->block_offset[16 + i] =
561 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
562 h->block_offset[48 + 16 + i] =
563 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);
566 /* Some macroblocks can be accessed before they're available in case
567 * of lost slices, MBAFF or threading. */
568 memset(h->slice_table, -1,
569 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
571 /* We mark the current picture as non-reference after allocating it, so
572 * that if we break out due to an error it can be released automatically
573 * in the next ff_mpv_frame_start().
575 h->cur_pic_ptr->reference = 0;
577 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
579 h->next_output_pic = NULL;
581 assert(h->cur_pic_ptr->long_ref == 0);
586 static av_always_inline void backup_mb_border(const H264Context *h, H264SliceContext *sl,
588 uint8_t *src_cb, uint8_t *src_cr,
589 int linesize, int uvlinesize,
594 const int pixel_shift = h->pixel_shift;
595 int chroma444 = CHROMA444(h);
596 int chroma422 = CHROMA422(h);
599 src_cb -= uvlinesize;
600 src_cr -= uvlinesize;
602 if (!simple && FRAME_MBAFF(h)) {
605 top_border = sl->top_borders[0][sl->mb_x];
606 AV_COPY128(top_border, src_y + 15 * linesize);
608 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
609 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
612 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
613 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
614 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
615 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
617 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
618 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
620 } else if (chroma422) {
622 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
623 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
625 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
626 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
630 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
631 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
633 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
634 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
639 } else if (MB_MBAFF(sl)) {
645 top_border = sl->top_borders[top_idx][sl->mb_x];
646 /* There are two lines saved, the line above the top macroblock
647 * of a pair, and the line above the bottom macroblock. */
648 AV_COPY128(top_border, src_y + 16 * linesize);
650 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
652 if (simple || !CONFIG_GRAY || !(h->flags & AV_CODEC_FLAG_GRAY)) {
655 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
656 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
657 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
658 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
660 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
661 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
663 } else if (chroma422) {
665 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
666 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
668 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
669 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
673 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
674 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
676 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
677 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
684 * Initialize implicit_weight table.
685 * @param field 0/1 initialize the weight for interlaced MBAFF
686 * -1 initializes the rest
688 static void implicit_weight_table(const H264Context *h, H264SliceContext *sl, int field)
690 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
692 for (i = 0; i < 2; i++) {
693 sl->pwt.luma_weight_flag[i] = 0;
694 sl->pwt.chroma_weight_flag[i] = 0;
698 if (h->picture_structure == PICT_FRAME) {
699 cur_poc = h->cur_pic_ptr->poc;
701 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
703 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
704 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
705 sl->pwt.use_weight = 0;
706 sl->pwt.use_weight_chroma = 0;
710 ref_count0 = sl->ref_count[0];
711 ref_count1 = sl->ref_count[1];
713 cur_poc = h->cur_pic_ptr->field_poc[field];
715 ref_count0 = 16 + 2 * sl->ref_count[0];
716 ref_count1 = 16 + 2 * sl->ref_count[1];
719 sl->pwt.use_weight = 2;
720 sl->pwt.use_weight_chroma = 2;
721 sl->pwt.luma_log2_weight_denom = 5;
722 sl->pwt.chroma_log2_weight_denom = 5;
724 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
725 int poc0 = sl->ref_list[0][ref0].poc;
726 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
728 if (!sl->ref_list[0][ref0].parent->long_ref && !sl->ref_list[1][ref1].parent->long_ref) {
729 int poc1 = sl->ref_list[1][ref1].poc;
730 int td = av_clip_int8(poc1 - poc0);
732 int tb = av_clip_int8(cur_poc - poc0);
733 int tx = (16384 + (FFABS(td) >> 1)) / td;
734 int dist_scale_factor = (tb * tx + 32) >> 8;
735 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
736 w = 64 - dist_scale_factor;
740 sl->pwt.implicit_weight[ref0][ref1][0] =
741 sl->pwt.implicit_weight[ref0][ref1][1] = w;
743 sl->pwt.implicit_weight[ref0][ref1][field] = w;
750 * initialize scan tables
752 static void init_scan_tables(H264Context *h)
755 for (i = 0; i < 16; i++) {
756 #define TRANSPOSE(x) (x >> 2) | ((x << 2) & 0xF)
757 h->zigzag_scan[i] = TRANSPOSE(ff_zigzag_scan[i]);
758 h->field_scan[i] = TRANSPOSE(field_scan[i]);
761 for (i = 0; i < 64; i++) {
762 #define TRANSPOSE(x) (x >> 3) | ((x & 7) << 3)
763 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
764 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
765 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
766 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
769 if (h->sps.transform_bypass) { // FIXME same ugly
770 h->zigzag_scan_q0 = ff_zigzag_scan;
771 h->zigzag_scan8x8_q0 = ff_zigzag_direct;
772 h->zigzag_scan8x8_cavlc_q0 = zigzag_scan8x8_cavlc;
773 h->field_scan_q0 = field_scan;
774 h->field_scan8x8_q0 = field_scan8x8;
775 h->field_scan8x8_cavlc_q0 = field_scan8x8_cavlc;
777 h->zigzag_scan_q0 = h->zigzag_scan;
778 h->zigzag_scan8x8_q0 = h->zigzag_scan8x8;
779 h->zigzag_scan8x8_cavlc_q0 = h->zigzag_scan8x8_cavlc;
780 h->field_scan_q0 = h->field_scan;
781 h->field_scan8x8_q0 = h->field_scan8x8;
782 h->field_scan8x8_cavlc_q0 = h->field_scan8x8_cavlc;
786 static enum AVPixelFormat get_pixel_format(H264Context *h)
788 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
789 CONFIG_H264_D3D11VA_HWACCEL + \
790 CONFIG_H264_VAAPI_HWACCEL + \
791 (CONFIG_H264_VDA_HWACCEL * 2) + \
792 CONFIG_H264_VDPAU_HWACCEL)
793 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
794 const enum AVPixelFormat *choices = pix_fmts;
796 switch (h->sps.bit_depth_luma) {
799 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
800 *fmt++ = AV_PIX_FMT_GBRP9;
802 *fmt++ = AV_PIX_FMT_YUV444P9;
803 } else if (CHROMA422(h))
804 *fmt++ = AV_PIX_FMT_YUV422P9;
806 *fmt++ = AV_PIX_FMT_YUV420P9;
810 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
811 *fmt++ = AV_PIX_FMT_GBRP10;
813 *fmt++ = AV_PIX_FMT_YUV444P10;
814 } else if (CHROMA422(h))
815 *fmt++ = AV_PIX_FMT_YUV422P10;
817 *fmt++ = AV_PIX_FMT_YUV420P10;
820 #if CONFIG_H264_VDPAU_HWACCEL
821 *fmt++ = AV_PIX_FMT_VDPAU;
824 if (h->avctx->colorspace == AVCOL_SPC_RGB)
825 *fmt++ = AV_PIX_FMT_GBRP;
826 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
827 *fmt++ = AV_PIX_FMT_YUVJ444P;
829 *fmt++ = AV_PIX_FMT_YUV444P;
830 } else if (CHROMA422(h)) {
831 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
832 *fmt++ = AV_PIX_FMT_YUVJ422P;
834 *fmt++ = AV_PIX_FMT_YUV422P;
836 #if CONFIG_H264_DXVA2_HWACCEL
837 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
839 #if CONFIG_H264_D3D11VA_HWACCEL
840 *fmt++ = AV_PIX_FMT_D3D11VA_VLD;
842 #if CONFIG_H264_VAAPI_HWACCEL
843 *fmt++ = AV_PIX_FMT_VAAPI;
845 #if CONFIG_H264_VDA_HWACCEL
846 *fmt++ = AV_PIX_FMT_VDA_VLD;
847 *fmt++ = AV_PIX_FMT_VDA;
849 if (h->avctx->codec->pix_fmts)
850 choices = h->avctx->codec->pix_fmts;
851 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
852 *fmt++ = AV_PIX_FMT_YUVJ420P;
854 *fmt++ = AV_PIX_FMT_YUV420P;
858 av_log(h->avctx, AV_LOG_ERROR,
859 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
860 return AVERROR_INVALIDDATA;
863 *fmt = AV_PIX_FMT_NONE;
865 return ff_get_format(h->avctx, choices);
868 /* export coded and cropped frame dimensions to AVCodecContext */
869 static int init_dimensions(H264Context *h)
871 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
872 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
874 /* handle container cropping */
875 if (FFALIGN(h->avctx->width, 16) == FFALIGN(width, 16) &&
876 FFALIGN(h->avctx->height, 16) == FFALIGN(height, 16)) {
877 width = h->avctx->width;
878 height = h->avctx->height;
881 if (width <= 0 || height <= 0) {
882 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
884 if (h->avctx->err_recognition & AV_EF_EXPLODE)
885 return AVERROR_INVALIDDATA;
887 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
898 h->avctx->coded_width = h->width;
899 h->avctx->coded_height = h->height;
900 h->avctx->width = width;
901 h->avctx->height = height;
906 static int h264_slice_header_init(H264Context *h)
908 int nb_slices = (HAVE_THREADS &&
909 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
910 h->avctx->thread_count : 1;
913 ff_set_sar(h->avctx, h->sps.sar);
914 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
915 &h->chroma_x_shift, &h->chroma_y_shift);
917 if (h->sps.timing_info_present_flag) {
918 int64_t den = h->sps.time_scale;
919 if (h->x264_build < 44U)
921 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
922 h->sps.num_units_in_tick, den, 1 << 30);
925 ff_h264_free_tables(h);
928 h->prev_interlaced_frame = 1;
931 ret = ff_h264_alloc_tables(h);
933 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
937 if (h->sps.bit_depth_luma < 8 || h->sps.bit_depth_luma > 10) {
938 av_log(h->avctx, AV_LOG_ERROR, "Unsupported bit depth %d\n",
939 h->sps.bit_depth_luma);
940 return AVERROR_INVALIDDATA;
943 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
944 h->pixel_shift = h->sps.bit_depth_luma > 8;
945 h->chroma_format_idc = h->sps.chroma_format_idc;
946 h->bit_depth_luma = h->sps.bit_depth_luma;
948 ff_h264dsp_init(&h->h264dsp, h->sps.bit_depth_luma,
949 h->sps.chroma_format_idc);
950 ff_h264chroma_init(&h->h264chroma, h->sps.bit_depth_chroma);
951 ff_h264qpel_init(&h->h264qpel, h->sps.bit_depth_luma);
952 ff_h264_pred_init(&h->hpc, h->avctx->codec_id, h->sps.bit_depth_luma,
953 h->sps.chroma_format_idc);
954 ff_videodsp_init(&h->vdsp, h->sps.bit_depth_luma);
956 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
959 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
961 max_slices = H264_MAX_THREADS;
962 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
963 " reducing to %d\n", nb_slices, max_slices);
964 nb_slices = max_slices;
966 h->slice_context_count = nb_slices;
968 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
969 ret = ff_h264_slice_context_init(h, &h->slice_ctx[0]);
971 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
975 for (i = 0; i < h->slice_context_count; i++) {
976 H264SliceContext *sl = &h->slice_ctx[i];
979 sl->intra4x4_pred_mode = h->intra4x4_pred_mode + i * 8 * 2 * h->mb_stride;
980 sl->mvd_table[0] = h->mvd_table[0] + i * 8 * 2 * h->mb_stride;
981 sl->mvd_table[1] = h->mvd_table[1] + i * 8 * 2 * h->mb_stride;
983 if ((ret = ff_h264_slice_context_init(h, sl)) < 0) {
984 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
990 h->context_initialized = 1;
996 * Decode a slice header.
997 * This will (re)intialize the decoder and call h264_frame_start() as needed.
999 * @param h h264context
1001 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1003 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl)
1005 unsigned int first_mb_in_slice;
1006 unsigned int pps_id;
1008 unsigned int slice_type, tmp, i, j;
1009 int last_pic_structure, last_pic_droppable;
1010 int needs_reinit = 0;
1011 int field_pic_flag, bottom_field_flag;
1012 int frame_num, droppable, picture_structure;
1013 int mb_aff_frame = 0;
1015 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1016 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1018 first_mb_in_slice = get_ue_golomb(&sl->gb);
1020 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1021 if (h->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1022 ff_h264_field_end(h, sl, 1);
1025 h->current_slice = 0;
1026 if (!h->first_field) {
1027 if (h->cur_pic_ptr && !h->droppable) {
1028 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1029 h->picture_structure == PICT_BOTTOM_FIELD);
1031 h->cur_pic_ptr = NULL;
1035 slice_type = get_ue_golomb_31(&sl->gb);
1036 if (slice_type > 9) {
1037 av_log(h->avctx, AV_LOG_ERROR,
1038 "slice type %d too large at %d\n",
1039 slice_type, first_mb_in_slice);
1040 return AVERROR_INVALIDDATA;
1042 if (slice_type > 4) {
1044 sl->slice_type_fixed = 1;
1046 sl->slice_type_fixed = 0;
1048 slice_type = ff_h264_golomb_to_pict_type[slice_type];
1049 sl->slice_type = slice_type;
1050 sl->slice_type_nos = slice_type & 3;
1052 if (h->nal_unit_type == NAL_IDR_SLICE &&
1053 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1054 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1055 return AVERROR_INVALIDDATA;
1058 // to make a few old functions happy, it's wrong though
1059 if (!h->setup_finished)
1060 h->pict_type = sl->slice_type;
1062 pps_id = get_ue_golomb(&sl->gb);
1063 if (pps_id >= MAX_PPS_COUNT) {
1064 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1065 return AVERROR_INVALIDDATA;
1067 if (!h->pps_buffers[pps_id]) {
1068 av_log(h->avctx, AV_LOG_ERROR,
1069 "non-existing PPS %u referenced\n",
1071 return AVERROR_INVALIDDATA;
1073 if (!h->setup_finished) {
1074 h->pps = *h->pps_buffers[pps_id];
1075 } else if (h->dequant_coeff_pps != pps_id) {
1076 av_log(h->avctx, AV_LOG_ERROR, "PPS changed between slices\n");
1077 return AVERROR_INVALIDDATA;
1080 if (!h->sps_buffers[h->pps.sps_id]) {
1081 av_log(h->avctx, AV_LOG_ERROR,
1082 "non-existing SPS %u referenced\n",
1084 return AVERROR_INVALIDDATA;
1087 if (h->pps.sps_id != h->sps.sps_id ||
1088 h->sps_buffers[h->pps.sps_id]->new) {
1089 h->sps_buffers[h->pps.sps_id]->new = 0;
1091 h->sps = *h->sps_buffers[h->pps.sps_id];
1093 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1094 h->chroma_format_idc != h->sps.chroma_format_idc)
1097 if (h->flags & AV_CODEC_FLAG_LOW_DELAY ||
1098 (h->sps.bitstream_restriction_flag &&
1099 !h->sps.num_reorder_frames)) {
1100 if (h->avctx->has_b_frames > 1 || h->delayed_pic[0])
1101 av_log(h->avctx, AV_LOG_WARNING, "Delayed frames seen. "
1102 "Reenabling low delay requires a codec flush.\n");
1107 if (h->avctx->has_b_frames < 2)
1108 h->avctx->has_b_frames = !h->low_delay;
1112 if (!h->setup_finished) {
1113 h->avctx->profile = ff_h264_get_profile(&h->sps);
1114 h->avctx->level = h->sps.level_idc;
1115 h->avctx->refs = h->sps.ref_frame_count;
1117 if (h->mb_width != h->sps.mb_width ||
1118 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag))
1121 h->mb_width = h->sps.mb_width;
1122 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1123 h->mb_num = h->mb_width * h->mb_height;
1124 h->mb_stride = h->mb_width + 1;
1126 h->b_stride = h->mb_width * 4;
1128 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1130 h->width = 16 * h->mb_width;
1131 h->height = 16 * h->mb_height;
1133 ret = init_dimensions(h);
1137 if (h->sps.video_signal_type_present_flag) {
1138 h->avctx->color_range = h->sps.full_range ? AVCOL_RANGE_JPEG
1140 if (h->sps.colour_description_present_flag) {
1141 if (h->avctx->colorspace != h->sps.colorspace)
1143 h->avctx->color_primaries = h->sps.color_primaries;
1144 h->avctx->color_trc = h->sps.color_trc;
1145 h->avctx->colorspace = h->sps.colorspace;
1150 if (h->context_initialized && needs_reinit) {
1151 h->context_initialized = 0;
1152 if (sl != h->slice_ctx) {
1153 av_log(h->avctx, AV_LOG_ERROR,
1154 "changing width %d -> %d / height %d -> %d on "
1156 h->width, h->avctx->coded_width,
1157 h->height, h->avctx->coded_height,
1158 h->current_slice + 1);
1159 return AVERROR_INVALIDDATA;
1162 ff_h264_flush_change(h);
1164 if ((ret = get_pixel_format(h)) < 0)
1166 h->avctx->pix_fmt = ret;
1168 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1169 "pix_fmt: %d\n", h->width, h->height, h->avctx->pix_fmt);
1171 if ((ret = h264_slice_header_init(h)) < 0) {
1172 av_log(h->avctx, AV_LOG_ERROR,
1173 "h264_slice_header_init() failed\n");
1177 if (!h->context_initialized) {
1178 if (sl != h->slice_ctx) {
1179 av_log(h->avctx, AV_LOG_ERROR,
1180 "Cannot (re-)initialize context during parallel decoding.\n");
1181 return AVERROR_PATCHWELCOME;
1184 if ((ret = get_pixel_format(h)) < 0)
1186 h->avctx->pix_fmt = ret;
1188 if ((ret = h264_slice_header_init(h)) < 0) {
1189 av_log(h->avctx, AV_LOG_ERROR,
1190 "h264_slice_header_init() failed\n");
1195 if (sl == h->slice_ctx && h->dequant_coeff_pps != pps_id) {
1196 h->dequant_coeff_pps = pps_id;
1197 h264_init_dequant_tables(h);
1200 frame_num = get_bits(&sl->gb, h->sps.log2_max_frame_num);
1201 if (!h->setup_finished)
1202 h->frame_num = frame_num;
1206 last_pic_structure = h->picture_structure;
1207 last_pic_droppable = h->droppable;
1209 droppable = h->nal_ref_idc == 0;
1210 if (h->sps.frame_mbs_only_flag) {
1211 picture_structure = PICT_FRAME;
1213 field_pic_flag = get_bits1(&sl->gb);
1214 if (field_pic_flag) {
1215 bottom_field_flag = get_bits1(&sl->gb);
1216 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1218 picture_structure = PICT_FRAME;
1219 mb_aff_frame = h->sps.mb_aff;
1222 if (!h->setup_finished) {
1223 h->droppable = droppable;
1224 h->picture_structure = picture_structure;
1225 h->mb_aff_frame = mb_aff_frame;
1227 sl->mb_field_decoding_flag = h->picture_structure != PICT_FRAME;
1229 if (h->current_slice != 0) {
1230 if (last_pic_structure != picture_structure ||
1231 last_pic_droppable != droppable) {
1232 av_log(h->avctx, AV_LOG_ERROR,
1233 "Changing field mode (%d -> %d) between slices is not allowed\n",
1234 last_pic_structure, h->picture_structure);
1235 return AVERROR_INVALIDDATA;
1236 } else if (!h->cur_pic_ptr) {
1237 av_log(h->avctx, AV_LOG_ERROR,
1238 "unset cur_pic_ptr on slice %d\n",
1239 h->current_slice + 1);
1240 return AVERROR_INVALIDDATA;
1243 /* Shorten frame num gaps so we don't have to allocate reference
1244 * frames just to throw them away */
1245 if (h->frame_num != h->prev_frame_num) {
1246 int unwrap_prev_frame_num = h->prev_frame_num;
1247 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1249 if (unwrap_prev_frame_num > h->frame_num)
1250 unwrap_prev_frame_num -= max_frame_num;
1252 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1253 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1254 if (unwrap_prev_frame_num < 0)
1255 unwrap_prev_frame_num += max_frame_num;
1257 h->prev_frame_num = unwrap_prev_frame_num;
1261 /* See if we have a decoded first field looking for a pair...
1262 * Here, we're using that to see if we should mark previously
1263 * decode frames as "finished".
1264 * We have to do that before the "dummy" in-between frame allocation,
1265 * since that can modify s->current_picture_ptr. */
1266 if (h->first_field) {
1267 assert(h->cur_pic_ptr);
1268 assert(h->cur_pic_ptr->f->buf[0]);
1269 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1271 /* figure out if we have a complementary field pair */
1272 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1273 /* Previous field is unmatched. Don't display it, but let it
1274 * remain for reference if marked as such. */
1275 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1276 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1277 last_pic_structure == PICT_TOP_FIELD);
1280 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1281 /* This and previous field were reference, but had
1282 * different frame_nums. Consider this field first in
1283 * pair. Throw away previous field except for reference
1285 if (!last_pic_droppable && last_pic_structure != PICT_FRAME) {
1286 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1287 last_pic_structure == PICT_TOP_FIELD);
1290 /* Second field in complementary pair */
1291 if (!((last_pic_structure == PICT_TOP_FIELD &&
1292 h->picture_structure == PICT_BOTTOM_FIELD) ||
1293 (last_pic_structure == PICT_BOTTOM_FIELD &&
1294 h->picture_structure == PICT_TOP_FIELD))) {
1295 av_log(h->avctx, AV_LOG_ERROR,
1296 "Invalid field mode combination %d/%d\n",
1297 last_pic_structure, h->picture_structure);
1298 h->picture_structure = last_pic_structure;
1299 h->droppable = last_pic_droppable;
1300 return AVERROR_INVALIDDATA;
1301 } else if (last_pic_droppable != h->droppable) {
1302 avpriv_request_sample(h->avctx,
1303 "Found reference and non-reference fields in the same frame, which");
1304 h->picture_structure = last_pic_structure;
1305 h->droppable = last_pic_droppable;
1306 return AVERROR_PATCHWELCOME;
1312 while (h->frame_num != h->prev_frame_num &&
1313 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1314 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1315 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1316 h->frame_num, h->prev_frame_num);
1317 ret = initialize_cur_frame(h);
1323 h->prev_frame_num++;
1324 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1325 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1326 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1327 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1328 ret = ff_generate_sliding_window_mmcos(h, 1);
1329 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1331 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1332 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1334 /* Error concealment: If a ref is missing, copy the previous ref
1336 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1337 * many assumptions about there being no actual duplicates.
1338 * FIXME: This does not copy padding for out-of-frame motion
1339 * vectors. Given we are concealing a lost frame, this probably
1340 * is not noticeable by comparison, but it should be fixed. */
1341 if (h->short_ref_count) {
1343 h->short_ref[0]->f->width == prev->f->width &&
1344 h->short_ref[0]->f->height == prev->f->height &&
1345 h->short_ref[0]->f->format == prev->f->format) {
1346 av_image_copy(h->short_ref[0]->f->data,
1347 h->short_ref[0]->f->linesize,
1348 (const uint8_t **)prev->f->data,
1353 h->short_ref[0]->poc = prev->poc + 2;
1355 h->short_ref[0]->frame_num = h->prev_frame_num;
1359 /* See if we have a decoded first field looking for a pair...
1360 * We're using that to see whether to continue decoding in that
1361 * frame, or to allocate a new one. */
1362 if (h->first_field) {
1363 assert(h->cur_pic_ptr);
1364 assert(h->cur_pic_ptr->f->buf[0]);
1365 assert(h->cur_pic_ptr->reference != DELAYED_PIC_REF);
1367 /* figure out if we have a complementary field pair */
1368 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1369 /* Previous field is unmatched. Don't display it, but let it
1370 * remain for reference if marked as such. */
1371 h->cur_pic_ptr = NULL;
1372 h->first_field = FIELD_PICTURE(h);
1374 if (h->cur_pic_ptr->frame_num != h->frame_num) {
1375 /* This and the previous field had different frame_nums.
1376 * Consider this field first in pair. Throw away previous
1377 * one except for reference purposes. */
1379 h->cur_pic_ptr = NULL;
1381 /* Second field in complementary pair */
1386 /* Frame or first field in a potentially complementary pair */
1387 h->first_field = FIELD_PICTURE(h);
1390 if (!FIELD_PICTURE(h) || h->first_field) {
1391 if (h264_frame_start(h) < 0) {
1393 return AVERROR_INVALIDDATA;
1396 release_unused_pictures(h, 0);
1400 assert(h->mb_num == h->mb_width * h->mb_height);
1401 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1402 first_mb_in_slice >= h->mb_num) {
1403 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1404 return AVERROR_INVALIDDATA;
1406 sl->resync_mb_x = sl->mb_x = first_mb_in_slice % h->mb_width;
1407 sl->resync_mb_y = sl->mb_y = (first_mb_in_slice / h->mb_width) <<
1408 FIELD_OR_MBAFF_PICTURE(h);
1409 if (h->picture_structure == PICT_BOTTOM_FIELD)
1410 sl->resync_mb_y = sl->mb_y = sl->mb_y + 1;
1411 assert(sl->mb_y < h->mb_height);
1413 if (h->picture_structure == PICT_FRAME) {
1414 h->curr_pic_num = h->frame_num;
1415 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1417 h->curr_pic_num = 2 * h->frame_num + 1;
1418 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1421 if (h->nal_unit_type == NAL_IDR_SLICE)
1422 get_ue_golomb(&sl->gb); /* idr_pic_id */
1424 if (h->sps.poc_type == 0) {
1425 int poc_lsb = get_bits(&sl->gb, h->sps.log2_max_poc_lsb);
1427 if (!h->setup_finished)
1428 h->poc_lsb = poc_lsb;
1430 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1431 int delta_poc_bottom = get_se_golomb(&sl->gb);
1432 if (!h->setup_finished)
1433 h->delta_poc_bottom = delta_poc_bottom;
1437 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1438 int delta_poc = get_se_golomb(&sl->gb);
1440 if (!h->setup_finished)
1441 h->delta_poc[0] = delta_poc;
1443 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME) {
1444 delta_poc = get_se_golomb(&sl->gb);
1446 if (!h->setup_finished)
1447 h->delta_poc[1] = delta_poc;
1451 if (!h->setup_finished)
1452 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1454 if (h->pps.redundant_pic_cnt_present)
1455 sl->redundant_pic_count = get_ue_golomb(&sl->gb);
1457 ret = ff_set_ref_count(h, sl);
1461 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1462 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1464 sl->ref_count[1] = sl->ref_count[0] = 0;
1469 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1470 (h->pps.weighted_bipred_idc == 1 &&
1471 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1472 ff_h264_pred_weight_table(&sl->gb, &h->sps, sl->ref_count,
1473 sl->slice_type_nos, &sl->pwt);
1474 else if (h->pps.weighted_bipred_idc == 2 &&
1475 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1476 implicit_weight_table(h, sl, -1);
1478 sl->pwt.use_weight = 0;
1479 for (i = 0; i < 2; i++) {
1480 sl->pwt.luma_weight_flag[i] = 0;
1481 sl->pwt.chroma_weight_flag[i] = 0;
1485 // If frame-mt is enabled, only update mmco tables for the first slice
1486 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1487 // or h->mmco, which will cause ref list mix-ups and decoding errors
1488 // further down the line. This may break decoding if the first slice is
1489 // corrupt, thus we only do this if frame-mt is enabled.
1490 if (h->nal_ref_idc) {
1491 ret = ff_h264_decode_ref_pic_marking(h, &sl->gb,
1492 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1493 h->current_slice == 0);
1494 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1495 return AVERROR_INVALIDDATA;
1498 if (FRAME_MBAFF(h)) {
1499 ff_h264_fill_mbaff_ref_list(h, sl);
1501 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1502 implicit_weight_table(h, sl, 0);
1503 implicit_weight_table(h, sl, 1);
1507 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1508 ff_h264_direct_dist_scale_factor(h, sl);
1509 ff_h264_direct_ref_list_init(h, sl);
1511 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1512 tmp = get_ue_golomb_31(&sl->gb);
1514 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1515 return AVERROR_INVALIDDATA;
1517 sl->cabac_init_idc = tmp;
1520 sl->last_qscale_diff = 0;
1521 tmp = h->pps.init_qp + get_se_golomb(&sl->gb);
1522 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1523 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1524 return AVERROR_INVALIDDATA;
1527 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1528 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1529 // FIXME qscale / qp ... stuff
1530 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1531 get_bits1(&sl->gb); /* sp_for_switch_flag */
1532 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1533 sl->slice_type == AV_PICTURE_TYPE_SI)
1534 get_se_golomb(&sl->gb); /* slice_qs_delta */
1536 sl->deblocking_filter = 1;
1537 sl->slice_alpha_c0_offset = 0;
1538 sl->slice_beta_offset = 0;
1539 if (h->pps.deblocking_filter_parameters_present) {
1540 tmp = get_ue_golomb_31(&sl->gb);
1542 av_log(h->avctx, AV_LOG_ERROR,
1543 "deblocking_filter_idc %u out of range\n", tmp);
1544 return AVERROR_INVALIDDATA;
1546 sl->deblocking_filter = tmp;
1547 if (sl->deblocking_filter < 2)
1548 sl->deblocking_filter ^= 1; // 1<->0
1550 if (sl->deblocking_filter) {
1551 sl->slice_alpha_c0_offset = get_se_golomb(&sl->gb) * 2;
1552 sl->slice_beta_offset = get_se_golomb(&sl->gb) * 2;
1553 if (sl->slice_alpha_c0_offset > 12 ||
1554 sl->slice_alpha_c0_offset < -12 ||
1555 sl->slice_beta_offset > 12 ||
1556 sl->slice_beta_offset < -12) {
1557 av_log(h->avctx, AV_LOG_ERROR,
1558 "deblocking filter parameters %d %d out of range\n",
1559 sl->slice_alpha_c0_offset, sl->slice_beta_offset);
1560 return AVERROR_INVALIDDATA;
1565 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1566 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1567 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1568 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1569 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1570 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1571 h->nal_ref_idc == 0))
1572 sl->deblocking_filter = 0;
1574 if (sl->deblocking_filter == 1 && h->max_contexts > 1) {
1575 if (h->avctx->flags2 & AV_CODEC_FLAG2_FAST) {
1576 /* Cheat slightly for speed:
1577 * Do not bother to deblock across slices. */
1578 sl->deblocking_filter = 2;
1580 h->max_contexts = 1;
1581 if (!h->single_decode_warning) {
1582 av_log(h->avctx, AV_LOG_INFO,
1583 "Cannot parallelize deblocking type 1, decoding such frames in sequential order\n");
1584 h->single_decode_warning = 1;
1586 if (sl != h->slice_ctx) {
1587 av_log(h->avctx, AV_LOG_ERROR,
1588 "Deblocking switched inside frame.\n");
1593 sl->qp_thresh = 15 -
1594 FFMIN(sl->slice_alpha_c0_offset, sl->slice_beta_offset) -
1596 h->pps.chroma_qp_index_offset[0],
1597 h->pps.chroma_qp_index_offset[1]) +
1598 6 * (h->sps.bit_depth_luma - 8);
1600 sl->slice_num = ++h->current_slice;
1601 if (sl->slice_num >= MAX_SLICES) {
1602 av_log(h->avctx, AV_LOG_ERROR,
1603 "Too many slices, increase MAX_SLICES and recompile\n");
1606 for (j = 0; j < 2; j++) {
1608 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1609 for (i = 0; i < 16; i++) {
1611 if (j < sl->list_count && i < sl->ref_count[j] &&
1612 sl->ref_list[j][i].parent->f->buf[0]) {
1614 AVBuffer *buf = sl->ref_list[j][i].parent->f->buf[0]->buffer;
1615 for (k = 0; k < h->short_ref_count; k++)
1616 if (h->short_ref[k]->f->buf[0]->buffer == buf) {
1620 for (k = 0; k < h->long_ref_count; k++)
1621 if (h->long_ref[k] && h->long_ref[k]->f->buf[0]->buffer == buf) {
1622 id_list[i] = h->short_ref_count + k;
1630 for (i = 0; i < 16; i++)
1631 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1633 ref2frm[18 + 1] = -1;
1634 for (i = 16; i < 48; i++)
1635 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1636 (sl->ref_list[j][i].reference & 3);
1639 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
1640 av_log(h->avctx, AV_LOG_DEBUG,
1641 "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",
1643 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
1645 av_get_picture_type_char(sl->slice_type),
1646 sl->slice_type_fixed ? " fix" : "",
1647 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
1648 pps_id, h->frame_num,
1649 h->cur_pic_ptr->field_poc[0],
1650 h->cur_pic_ptr->field_poc[1],
1651 sl->ref_count[0], sl->ref_count[1],
1653 sl->deblocking_filter,
1654 sl->slice_alpha_c0_offset, sl->slice_beta_offset,
1656 sl->pwt.use_weight == 1 && sl->pwt.use_weight_chroma ? "c" : "",
1657 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
1663 int ff_h264_get_slice_type(const H264SliceContext *sl)
1665 switch (sl->slice_type) {
1666 case AV_PICTURE_TYPE_P:
1668 case AV_PICTURE_TYPE_B:
1670 case AV_PICTURE_TYPE_I:
1672 case AV_PICTURE_TYPE_SP:
1674 case AV_PICTURE_TYPE_SI:
1677 return AVERROR_INVALIDDATA;
1681 static av_always_inline void fill_filter_caches_inter(const H264Context *h,
1682 H264SliceContext *sl,
1683 int mb_type, int top_xy,
1684 int left_xy[LEFT_MBS],
1686 int left_type[LEFT_MBS],
1687 int mb_xy, int list)
1689 int b_stride = h->b_stride;
1690 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
1691 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
1692 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
1693 if (USES_LIST(top_type, list)) {
1694 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
1695 const int b8_xy = 4 * top_xy + 2;
1696 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1697 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
1698 ref_cache[0 - 1 * 8] =
1699 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
1700 ref_cache[2 - 1 * 8] =
1701 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
1703 AV_ZERO128(mv_dst - 1 * 8);
1704 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1707 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
1708 if (USES_LIST(left_type[LTOP], list)) {
1709 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
1710 const int b8_xy = 4 * left_xy[LTOP] + 1;
1711 int (*ref2frm)[64] = sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1712 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
1713 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
1714 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
1715 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
1717 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
1718 ref_cache[-1 + 16] =
1719 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
1721 AV_ZERO32(mv_dst - 1 + 0);
1722 AV_ZERO32(mv_dst - 1 + 8);
1723 AV_ZERO32(mv_dst - 1 + 16);
1724 AV_ZERO32(mv_dst - 1 + 24);
1727 ref_cache[-1 + 16] =
1728 ref_cache[-1 + 24] = LIST_NOT_USED;
1733 if (!USES_LIST(mb_type, list)) {
1734 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
1735 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1736 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1737 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1738 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
1743 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
1744 int (*ref2frm)[64] = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(sl) ? 20 : 2);
1745 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
1746 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
1747 AV_WN32A(&ref_cache[0 * 8], ref01);
1748 AV_WN32A(&ref_cache[1 * 8], ref01);
1749 AV_WN32A(&ref_cache[2 * 8], ref23);
1750 AV_WN32A(&ref_cache[3 * 8], ref23);
1754 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * sl->mb_x + 4 * sl->mb_y * b_stride];
1755 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
1756 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
1757 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
1758 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
1763 * @return non zero if the loop filter can be skipped
1765 static int fill_filter_caches(const H264Context *h, H264SliceContext *sl, int mb_type)
1767 const int mb_xy = sl->mb_xy;
1768 int top_xy, left_xy[LEFT_MBS];
1769 int top_type, left_type[LEFT_MBS];
1773 top_xy = mb_xy - (h->mb_stride << MB_FIELD(sl));
1775 /* Wow, what a mess, why didn't they simplify the interlacing & intra
1776 * stuff, I can't imagine that these complex rules are worth it. */
1778 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
1779 if (FRAME_MBAFF(h)) {
1780 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
1781 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
1783 if (left_mb_field_flag != curr_mb_field_flag)
1784 left_xy[LTOP] -= h->mb_stride;
1786 if (curr_mb_field_flag)
1787 top_xy += h->mb_stride &
1788 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
1789 if (left_mb_field_flag != curr_mb_field_flag)
1790 left_xy[LBOT] += h->mb_stride;
1794 sl->top_mb_xy = top_xy;
1795 sl->left_mb_xy[LTOP] = left_xy[LTOP];
1796 sl->left_mb_xy[LBOT] = left_xy[LBOT];
1798 /* For sufficiently low qp, filtering wouldn't do anything.
1799 * This is a conservative estimate: could also check beta_offset
1800 * and more accurate chroma_qp. */
1801 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
1802 int qp = h->cur_pic.qscale_table[mb_xy];
1803 if (qp <= qp_thresh &&
1804 (left_xy[LTOP] < 0 ||
1805 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
1807 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
1808 if (!FRAME_MBAFF(h))
1810 if ((left_xy[LTOP] < 0 ||
1811 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
1812 (top_xy < h->mb_stride ||
1813 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
1818 top_type = h->cur_pic.mb_type[top_xy];
1819 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
1820 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
1821 if (sl->deblocking_filter == 2) {
1822 if (h->slice_table[top_xy] != sl->slice_num)
1824 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
1825 left_type[LTOP] = left_type[LBOT] = 0;
1827 if (h->slice_table[top_xy] == 0xFFFF)
1829 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
1830 left_type[LTOP] = left_type[LBOT] = 0;
1832 sl->top_type = top_type;
1833 sl->left_type[LTOP] = left_type[LTOP];
1834 sl->left_type[LBOT] = left_type[LBOT];
1836 if (IS_INTRA(mb_type))
1839 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1840 top_type, left_type, mb_xy, 0);
1841 if (sl->list_count == 2)
1842 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
1843 top_type, left_type, mb_xy, 1);
1845 nnz = h->non_zero_count[mb_xy];
1846 nnz_cache = sl->non_zero_count_cache;
1847 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
1848 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
1849 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
1850 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
1851 sl->cbp = h->cbp_table[mb_xy];
1854 nnz = h->non_zero_count[top_xy];
1855 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
1858 if (left_type[LTOP]) {
1859 nnz = h->non_zero_count[left_xy[LTOP]];
1860 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
1861 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
1862 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
1863 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
1866 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
1867 * from what the loop filter needs */
1868 if (!CABAC(h) && h->pps.transform_8x8_mode) {
1869 if (IS_8x8DCT(top_type)) {
1870 nnz_cache[4 + 8 * 0] =
1871 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
1872 nnz_cache[6 + 8 * 0] =
1873 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
1875 if (IS_8x8DCT(left_type[LTOP])) {
1876 nnz_cache[3 + 8 * 1] =
1877 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
1879 if (IS_8x8DCT(left_type[LBOT])) {
1880 nnz_cache[3 + 8 * 3] =
1881 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
1884 if (IS_8x8DCT(mb_type)) {
1885 nnz_cache[scan8[0]] =
1886 nnz_cache[scan8[1]] =
1887 nnz_cache[scan8[2]] =
1888 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
1890 nnz_cache[scan8[0 + 4]] =
1891 nnz_cache[scan8[1 + 4]] =
1892 nnz_cache[scan8[2 + 4]] =
1893 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
1895 nnz_cache[scan8[0 + 8]] =
1896 nnz_cache[scan8[1 + 8]] =
1897 nnz_cache[scan8[2 + 8]] =
1898 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
1900 nnz_cache[scan8[0 + 12]] =
1901 nnz_cache[scan8[1 + 12]] =
1902 nnz_cache[scan8[2 + 12]] =
1903 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
1910 static void loop_filter(const H264Context *h, H264SliceContext *sl, int start_x, int end_x)
1912 uint8_t *dest_y, *dest_cb, *dest_cr;
1913 int linesize, uvlinesize, mb_x, mb_y;
1914 const int end_mb_y = sl->mb_y + FRAME_MBAFF(h);
1915 const int old_slice_type = sl->slice_type;
1916 const int pixel_shift = h->pixel_shift;
1917 const int block_h = 16 >> h->chroma_y_shift;
1919 if (sl->deblocking_filter) {
1920 for (mb_x = start_x; mb_x < end_x; mb_x++)
1921 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
1923 mb_xy = sl->mb_xy = mb_x + mb_y * h->mb_stride;
1924 sl->slice_num = h->slice_table[mb_xy];
1925 mb_type = h->cur_pic.mb_type[mb_xy];
1926 sl->list_count = h->list_counts[mb_xy];
1930 sl->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
1934 dest_y = h->cur_pic.f->data[0] +
1935 ((mb_x << pixel_shift) + mb_y * sl->linesize) * 16;
1936 dest_cb = h->cur_pic.f->data[1] +
1937 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1938 mb_y * sl->uvlinesize * block_h;
1939 dest_cr = h->cur_pic.f->data[2] +
1940 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
1941 mb_y * sl->uvlinesize * block_h;
1942 // FIXME simplify above
1945 linesize = sl->mb_linesize = sl->linesize * 2;
1946 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize * 2;
1947 if (mb_y & 1) { // FIXME move out of this function?
1948 dest_y -= sl->linesize * 15;
1949 dest_cb -= sl->uvlinesize * (block_h - 1);
1950 dest_cr -= sl->uvlinesize * (block_h - 1);
1953 linesize = sl->mb_linesize = sl->linesize;
1954 uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;
1956 backup_mb_border(h, sl, dest_y, dest_cb, dest_cr, linesize,
1958 if (fill_filter_caches(h, sl, mb_type))
1960 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
1961 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
1963 if (FRAME_MBAFF(h)) {
1964 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
1965 linesize, uvlinesize);
1967 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
1968 dest_cr, linesize, uvlinesize);
1972 sl->slice_type = old_slice_type;
1974 sl->mb_y = end_mb_y - FRAME_MBAFF(h);
1975 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1976 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1979 static void predict_field_decoding_flag(const H264Context *h, H264SliceContext *sl)
1981 const int mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;
1982 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
1983 h->cur_pic.mb_type[mb_xy - 1] :
1984 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
1985 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
1986 sl->mb_mbaff = sl->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
1990 * Draw edges and report progress for the last MB row.
1992 static void decode_finish_row(const H264Context *h, H264SliceContext *sl)
1994 int top = 16 * (sl->mb_y >> FIELD_PICTURE(h));
1995 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
1996 int height = 16 << FRAME_MBAFF(h);
1997 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
1999 if (sl->deblocking_filter) {
2000 if ((top + height) >= pic_height)
2001 height += deblock_border;
2002 top -= deblock_border;
2005 if (top >= pic_height || (top + height) < 0)
2008 height = FFMIN(height, pic_height - top);
2010 height = top + height;
2014 ff_h264_draw_horiz_band(h, sl, top, height);
2019 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2020 h->picture_structure == PICT_BOTTOM_FIELD);
2023 static void er_add_slice(H264SliceContext *sl,
2024 int startx, int starty,
2025 int endx, int endy, int status)
2027 #if CONFIG_ERROR_RESILIENCE
2028 ERContext *er = &sl->er;
2030 if (!sl->h264->enable_er)
2033 er->ref_count = sl->ref_count[0];
2034 ff_er_add_slice(er, startx, starty, endx, endy, status);
2038 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2040 H264SliceContext *sl = arg;
2041 const H264Context *h = sl->h264;
2042 int lf_x_start = sl->mb_x;
2045 sl->linesize = h->cur_pic_ptr->f->linesize[0];
2046 sl->uvlinesize = h->cur_pic_ptr->f->linesize[1];
2048 ret = alloc_scratch_buffers(sl, sl->linesize);
2052 sl->mb_skip_run = -1;
2054 sl->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2055 avctx->codec_id != AV_CODEC_ID_H264 ||
2056 (CONFIG_GRAY && (h->flags & AV_CODEC_FLAG_GRAY));
2060 align_get_bits(&sl->gb);
2063 ff_init_cabac_decoder(&sl->cabac,
2064 sl->gb.buffer + get_bits_count(&sl->gb) / 8,
2065 (get_bits_left(&sl->gb) + 7) / 8);
2067 ff_h264_init_cabac_states(h, sl);
2073 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2074 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2075 sl->next_slice_idx);
2076 return AVERROR_INVALIDDATA;
2079 ret = ff_h264_decode_mb_cabac(h, sl);
2080 // STOP_TIMER("decode_mb_cabac")
2083 ff_h264_hl_decode_mb(h, sl);
2085 // FIXME optimal? or let mb_decode decode 16x32 ?
2086 if (ret >= 0 && FRAME_MBAFF(h)) {
2089 ret = ff_h264_decode_mb_cabac(h, sl);
2092 ff_h264_hl_decode_mb(h, sl);
2095 eos = get_cabac_terminate(&sl->cabac);
2097 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2098 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2099 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2100 sl->mb_y, ER_MB_END);
2101 if (sl->mb_x >= lf_x_start)
2102 loop_filter(h, sl, lf_x_start, sl->mb_x + 1);
2105 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2106 av_log(h->avctx, AV_LOG_ERROR,
2107 "error while decoding MB %d %d, bytestream %td\n",
2109 sl->cabac.bytestream_end - sl->cabac.bytestream);
2110 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2111 sl->mb_y, ER_MB_ERROR);
2112 return AVERROR_INVALIDDATA;
2115 if (++sl->mb_x >= h->mb_width) {
2116 loop_filter(h, sl, lf_x_start, sl->mb_x);
2117 sl->mb_x = lf_x_start = 0;
2118 decode_finish_row(h, sl);
2120 if (FIELD_OR_MBAFF_PICTURE(h)) {
2122 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2123 predict_field_decoding_flag(h, sl);
2127 if (eos || sl->mb_y >= h->mb_height) {
2128 ff_tlog(h->avctx, "slice end %d %d\n",
2129 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2130 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x - 1,
2131 sl->mb_y, ER_MB_END);
2132 if (sl->mb_x > lf_x_start)
2133 loop_filter(h, sl, lf_x_start, sl->mb_x);
2141 if (sl->mb_x + sl->mb_y * h->mb_width >= sl->next_slice_idx) {
2142 av_log(h->avctx, AV_LOG_ERROR, "Slice overlaps with next at %d\n",
2143 sl->next_slice_idx);
2144 return AVERROR_INVALIDDATA;
2147 ret = ff_h264_decode_mb_cavlc(h, sl);
2150 ff_h264_hl_decode_mb(h, sl);
2152 // FIXME optimal? or let mb_decode decode 16x32 ?
2153 if (ret >= 0 && FRAME_MBAFF(h)) {
2155 ret = ff_h264_decode_mb_cavlc(h, sl);
2158 ff_h264_hl_decode_mb(h, sl);
2163 av_log(h->avctx, AV_LOG_ERROR,
2164 "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);
2165 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2166 sl->mb_y, ER_MB_ERROR);
2170 if (++sl->mb_x >= h->mb_width) {
2171 loop_filter(h, sl, lf_x_start, sl->mb_x);
2172 sl->mb_x = lf_x_start = 0;
2173 decode_finish_row(h, sl);
2175 if (FIELD_OR_MBAFF_PICTURE(h)) {
2177 if (FRAME_MBAFF(h) && sl->mb_y < h->mb_height)
2178 predict_field_decoding_flag(h, sl);
2180 if (sl->mb_y >= h->mb_height) {
2181 ff_tlog(h->avctx, "slice end %d %d\n",
2182 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2184 if (get_bits_left(&sl->gb) == 0) {
2185 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2186 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2190 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2191 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2193 return AVERROR_INVALIDDATA;
2198 if (get_bits_left(&sl->gb) <= 0 && sl->mb_skip_run <= 0) {
2199 ff_tlog(h->avctx, "slice end %d %d\n",
2200 get_bits_count(&sl->gb), sl->gb.size_in_bits);
2202 if (get_bits_left(&sl->gb) == 0) {
2203 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y,
2204 sl->mb_x - 1, sl->mb_y, ER_MB_END);
2205 if (sl->mb_x > lf_x_start)
2206 loop_filter(h, sl, lf_x_start, sl->mb_x);
2210 er_add_slice(sl, sl->resync_mb_x, sl->resync_mb_y, sl->mb_x,
2211 sl->mb_y, ER_MB_ERROR);
2213 return AVERROR_INVALIDDATA;
2221 * Call decode_slice() for each context.
2223 * @param h h264 master context
2224 * @param context_count number of contexts to execute
2226 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2228 AVCodecContext *const avctx = h->avctx;
2229 H264SliceContext *sl;
2232 if (h->avctx->hwaccel)
2234 if (context_count == 1) {
2237 h->slice_ctx[0].next_slice_idx = h->mb_width * h->mb_height;
2239 ret = decode_slice(avctx, &h->slice_ctx[0]);
2240 h->mb_y = h->slice_ctx[0].mb_y;
2243 for (i = 0; i < context_count; i++) {
2244 int next_slice_idx = h->mb_width * h->mb_height;
2247 sl = &h->slice_ctx[i];
2248 sl->er.error_count = 0;
2250 /* make sure none of those slices overlap */
2251 slice_idx = sl->mb_y * h->mb_width + sl->mb_x;
2252 for (j = 0; j < context_count; j++) {
2253 H264SliceContext *sl2 = &h->slice_ctx[j];
2254 int slice_idx2 = sl2->mb_y * h->mb_width + sl2->mb_x;
2256 if (i == j || slice_idx2 < slice_idx)
2258 next_slice_idx = FFMIN(next_slice_idx, slice_idx2);
2260 sl->next_slice_idx = next_slice_idx;
2263 avctx->execute(avctx, decode_slice, h->slice_ctx,
2264 NULL, context_count, sizeof(h->slice_ctx[0]));
2266 /* pull back stuff from slices to master context */
2267 sl = &h->slice_ctx[context_count - 1];
2269 for (i = 1; i < context_count; i++)
2270 h->slice_ctx[0].er.error_count += h->slice_ctx[i].er.error_count;