2 * H.26L/H.264/AVC/JVT/14496-10/... decoder
3 * Copyright (c) 2003 Michael Niedermayer <michaelni@gmx.at>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * H.264 / AVC / MPEG4 part10 codec.
25 * @author Michael Niedermayer <michaelni@gmx.at>
28 #include "libavutil/avassert.h"
29 #include "libavutil/imgutils.h"
30 #include "libavutil/timer.h"
33 #include "cabac_functions.h"
34 #include "error_resilience.h"
38 #include "h264chroma.h"
39 #include "h264_mvpred.h"
42 #include "mpegutils.h"
43 #include "rectangle.h"
47 static const uint8_t rem6[QP_MAX_NUM + 1] = {
48 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
49 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
50 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2,
51 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5, 0, 1, 2, 3, 4, 5,
55 static const uint8_t div6[QP_MAX_NUM + 1] = {
56 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 3, 3, 3,
57 3, 3, 3, 4, 4, 4, 4, 4, 4, 5, 5, 5, 5, 5, 5, 6, 6, 6, 6, 6, 6,
58 7, 7, 7, 7, 7, 7, 8, 8, 8, 8, 8, 8, 9, 9, 9, 9, 9, 9, 10, 10, 10,
59 10,10,10,11,11,11,11,11,11,12,12,12,12,12,12,13,13,13, 13, 13, 13,
63 static const uint8_t field_scan[16+1] = {
64 0 + 0 * 4, 0 + 1 * 4, 1 + 0 * 4, 0 + 2 * 4,
65 0 + 3 * 4, 1 + 1 * 4, 1 + 2 * 4, 1 + 3 * 4,
66 2 + 0 * 4, 2 + 1 * 4, 2 + 2 * 4, 2 + 3 * 4,
67 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4, 3 + 3 * 4,
70 static const uint8_t field_scan8x8[64+1] = {
71 0 + 0 * 8, 0 + 1 * 8, 0 + 2 * 8, 1 + 0 * 8,
72 1 + 1 * 8, 0 + 3 * 8, 0 + 4 * 8, 1 + 2 * 8,
73 2 + 0 * 8, 1 + 3 * 8, 0 + 5 * 8, 0 + 6 * 8,
74 0 + 7 * 8, 1 + 4 * 8, 2 + 1 * 8, 3 + 0 * 8,
75 2 + 2 * 8, 1 + 5 * 8, 1 + 6 * 8, 1 + 7 * 8,
76 2 + 3 * 8, 3 + 1 * 8, 4 + 0 * 8, 3 + 2 * 8,
77 2 + 4 * 8, 2 + 5 * 8, 2 + 6 * 8, 2 + 7 * 8,
78 3 + 3 * 8, 4 + 1 * 8, 5 + 0 * 8, 4 + 2 * 8,
79 3 + 4 * 8, 3 + 5 * 8, 3 + 6 * 8, 3 + 7 * 8,
80 4 + 3 * 8, 5 + 1 * 8, 6 + 0 * 8, 5 + 2 * 8,
81 4 + 4 * 8, 4 + 5 * 8, 4 + 6 * 8, 4 + 7 * 8,
82 5 + 3 * 8, 6 + 1 * 8, 6 + 2 * 8, 5 + 4 * 8,
83 5 + 5 * 8, 5 + 6 * 8, 5 + 7 * 8, 6 + 3 * 8,
84 7 + 0 * 8, 7 + 1 * 8, 6 + 4 * 8, 6 + 5 * 8,
85 6 + 6 * 8, 6 + 7 * 8, 7 + 2 * 8, 7 + 3 * 8,
86 7 + 4 * 8, 7 + 5 * 8, 7 + 6 * 8, 7 + 7 * 8,
89 static const uint8_t field_scan8x8_cavlc[64+1] = {
90 0 + 0 * 8, 1 + 1 * 8, 2 + 0 * 8, 0 + 7 * 8,
91 2 + 2 * 8, 2 + 3 * 8, 2 + 4 * 8, 3 + 3 * 8,
92 3 + 4 * 8, 4 + 3 * 8, 4 + 4 * 8, 5 + 3 * 8,
93 5 + 5 * 8, 7 + 0 * 8, 6 + 6 * 8, 7 + 4 * 8,
94 0 + 1 * 8, 0 + 3 * 8, 1 + 3 * 8, 1 + 4 * 8,
95 1 + 5 * 8, 3 + 1 * 8, 2 + 5 * 8, 4 + 1 * 8,
96 3 + 5 * 8, 5 + 1 * 8, 4 + 5 * 8, 6 + 1 * 8,
97 5 + 6 * 8, 7 + 1 * 8, 6 + 7 * 8, 7 + 5 * 8,
98 0 + 2 * 8, 0 + 4 * 8, 0 + 5 * 8, 2 + 1 * 8,
99 1 + 6 * 8, 4 + 0 * 8, 2 + 6 * 8, 5 + 0 * 8,
100 3 + 6 * 8, 6 + 0 * 8, 4 + 6 * 8, 6 + 2 * 8,
101 5 + 7 * 8, 6 + 4 * 8, 7 + 2 * 8, 7 + 6 * 8,
102 1 + 0 * 8, 1 + 2 * 8, 0 + 6 * 8, 3 + 0 * 8,
103 1 + 7 * 8, 3 + 2 * 8, 2 + 7 * 8, 4 + 2 * 8,
104 3 + 7 * 8, 5 + 2 * 8, 4 + 7 * 8, 5 + 4 * 8,
105 6 + 3 * 8, 6 + 5 * 8, 7 + 3 * 8, 7 + 7 * 8,
108 // zigzag_scan8x8_cavlc[i] = zigzag_scan8x8[(i/4) + 16*(i%4)]
109 static const uint8_t zigzag_scan8x8_cavlc[64+1] = {
110 0 + 0 * 8, 1 + 1 * 8, 1 + 2 * 8, 2 + 2 * 8,
111 4 + 1 * 8, 0 + 5 * 8, 3 + 3 * 8, 7 + 0 * 8,
112 3 + 4 * 8, 1 + 7 * 8, 5 + 3 * 8, 6 + 3 * 8,
113 2 + 7 * 8, 6 + 4 * 8, 5 + 6 * 8, 7 + 5 * 8,
114 1 + 0 * 8, 2 + 0 * 8, 0 + 3 * 8, 3 + 1 * 8,
115 3 + 2 * 8, 0 + 6 * 8, 4 + 2 * 8, 6 + 1 * 8,
116 2 + 5 * 8, 2 + 6 * 8, 6 + 2 * 8, 5 + 4 * 8,
117 3 + 7 * 8, 7 + 3 * 8, 4 + 7 * 8, 7 + 6 * 8,
118 0 + 1 * 8, 3 + 0 * 8, 0 + 4 * 8, 4 + 0 * 8,
119 2 + 3 * 8, 1 + 5 * 8, 5 + 1 * 8, 5 + 2 * 8,
120 1 + 6 * 8, 3 + 5 * 8, 7 + 1 * 8, 4 + 5 * 8,
121 4 + 6 * 8, 7 + 4 * 8, 5 + 7 * 8, 6 + 7 * 8,
122 0 + 2 * 8, 2 + 1 * 8, 1 + 3 * 8, 5 + 0 * 8,
123 1 + 4 * 8, 2 + 4 * 8, 6 + 0 * 8, 4 + 3 * 8,
124 0 + 7 * 8, 4 + 4 * 8, 7 + 2 * 8, 3 + 6 * 8,
125 5 + 5 * 8, 6 + 5 * 8, 6 + 6 * 8, 7 + 7 * 8,
128 static const uint8_t dequant4_coeff_init[6][3] = {
137 static const uint8_t dequant8_coeff_init_scan[16] = {
138 0, 3, 4, 3, 3, 1, 5, 1, 4, 5, 2, 5, 3, 1, 5, 1
141 static const uint8_t dequant8_coeff_init[6][6] = {
142 { 20, 18, 32, 19, 25, 24 },
143 { 22, 19, 35, 21, 28, 26 },
144 { 26, 23, 42, 24, 33, 31 },
145 { 28, 25, 45, 26, 35, 33 },
146 { 32, 28, 51, 30, 40, 38 },
147 { 36, 32, 58, 34, 46, 43 },
151 static void release_unused_pictures(H264Context *h, int remove_current)
155 /* release non reference frames */
156 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
157 if (h->DPB[i].f.buf[0] && !h->DPB[i].reference &&
158 (remove_current || &h->DPB[i] != h->cur_pic_ptr)) {
159 ff_h264_unref_picture(h, &h->DPB[i]);
164 static int alloc_scratch_buffers(H264Context *h, int linesize)
166 int alloc_size = FFALIGN(FFABS(linesize) + 32, 32);
168 if (h->bipred_scratchpad)
171 h->bipred_scratchpad = av_malloc(16 * 6 * alloc_size);
172 // edge emu needs blocksize + filter length - 1
173 // (= 21x21 for h264)
174 h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 21);
176 if (!h->bipred_scratchpad || !h->edge_emu_buffer) {
177 av_freep(&h->bipred_scratchpad);
178 av_freep(&h->edge_emu_buffer);
179 return AVERROR(ENOMEM);
185 static int init_table_pools(H264Context *h)
187 const int big_mb_num = h->mb_stride * (h->mb_height + 1) + 1;
188 const int mb_array_size = h->mb_stride * h->mb_height;
189 const int b4_stride = h->mb_width * 4 + 1;
190 const int b4_array_size = b4_stride * h->mb_height * 4;
192 h->qscale_table_pool = av_buffer_pool_init(big_mb_num + h->mb_stride,
194 h->mb_type_pool = av_buffer_pool_init((big_mb_num + h->mb_stride) *
195 sizeof(uint32_t), av_buffer_allocz);
196 h->motion_val_pool = av_buffer_pool_init(2 * (b4_array_size + 4) *
197 sizeof(int16_t), av_buffer_allocz);
198 h->ref_index_pool = av_buffer_pool_init(4 * mb_array_size, av_buffer_allocz);
200 if (!h->qscale_table_pool || !h->mb_type_pool || !h->motion_val_pool ||
201 !h->ref_index_pool) {
202 av_buffer_pool_uninit(&h->qscale_table_pool);
203 av_buffer_pool_uninit(&h->mb_type_pool);
204 av_buffer_pool_uninit(&h->motion_val_pool);
205 av_buffer_pool_uninit(&h->ref_index_pool);
206 return AVERROR(ENOMEM);
212 static int alloc_picture(H264Context *h, H264Picture *pic)
216 av_assert0(!pic->f.data[0]);
219 ret = ff_thread_get_buffer(h->avctx, &pic->tf, pic->reference ?
220 AV_GET_BUFFER_FLAG_REF : 0);
224 h->linesize = pic->f.linesize[0];
225 h->uvlinesize = pic->f.linesize[1];
226 pic->crop = h->sps.crop;
227 pic->crop_top = h->sps.crop_top;
228 pic->crop_left= h->sps.crop_left;
230 if (h->avctx->hwaccel) {
231 const AVHWAccel *hwaccel = h->avctx->hwaccel;
232 av_assert0(!pic->hwaccel_picture_private);
233 if (hwaccel->frame_priv_data_size) {
234 pic->hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
235 if (!pic->hwaccel_priv_buf)
236 return AVERROR(ENOMEM);
237 pic->hwaccel_picture_private = pic->hwaccel_priv_buf->data;
240 if (!h->avctx->hwaccel && CONFIG_GRAY && h->flags & CODEC_FLAG_GRAY && pic->f.data[2]) {
241 int h_chroma_shift, v_chroma_shift;
242 av_pix_fmt_get_chroma_sub_sample(pic->f.format,
243 &h_chroma_shift, &v_chroma_shift);
245 for(i=0; i<FF_CEIL_RSHIFT(h->avctx->height, v_chroma_shift); i++) {
246 memset(pic->f.data[1] + pic->f.linesize[1]*i,
247 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
248 memset(pic->f.data[2] + pic->f.linesize[2]*i,
249 0x80, FF_CEIL_RSHIFT(h->avctx->width, h_chroma_shift));
253 if (!h->qscale_table_pool) {
254 ret = init_table_pools(h);
259 pic->qscale_table_buf = av_buffer_pool_get(h->qscale_table_pool);
260 pic->mb_type_buf = av_buffer_pool_get(h->mb_type_pool);
261 if (!pic->qscale_table_buf || !pic->mb_type_buf)
264 pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;
265 pic->qscale_table = pic->qscale_table_buf->data + 2 * h->mb_stride + 1;
267 for (i = 0; i < 2; i++) {
268 pic->motion_val_buf[i] = av_buffer_pool_get(h->motion_val_pool);
269 pic->ref_index_buf[i] = av_buffer_pool_get(h->ref_index_pool);
270 if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i])
273 pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;
274 pic->ref_index[i] = pic->ref_index_buf[i]->data;
279 ff_h264_unref_picture(h, pic);
280 return (ret < 0) ? ret : AVERROR(ENOMEM);
283 static inline int pic_is_unused(H264Context *h, H264Picture *pic)
287 if (pic->needs_realloc && !(pic->reference & DELAYED_PIC_REF))
292 static int find_unused_picture(H264Context *h)
296 for (i = 0; i < H264_MAX_PICTURE_COUNT; i++) {
297 if (pic_is_unused(h, &h->DPB[i]))
300 if (i == H264_MAX_PICTURE_COUNT)
301 return AVERROR_INVALIDDATA;
303 if (h->DPB[i].needs_realloc) {
304 h->DPB[i].needs_realloc = 0;
305 ff_h264_unref_picture(h, &h->DPB[i]);
312 static void init_dequant8_coeff_table(H264Context *h)
315 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
317 for (i = 0; i < 6; i++) {
318 h->dequant8_coeff[i] = h->dequant8_buffer[i];
319 for (j = 0; j < i; j++)
320 if (!memcmp(h->pps.scaling_matrix8[j], h->pps.scaling_matrix8[i],
321 64 * sizeof(uint8_t))) {
322 h->dequant8_coeff[i] = h->dequant8_buffer[j];
328 for (q = 0; q < max_qp + 1; q++) {
331 for (x = 0; x < 64; x++)
332 h->dequant8_coeff[i][q][(x >> 3) | ((x & 7) << 3)] =
333 ((uint32_t)dequant8_coeff_init[idx][dequant8_coeff_init_scan[((x >> 1) & 12) | (x & 3)]] *
334 h->pps.scaling_matrix8[i][x]) << shift;
339 static void init_dequant4_coeff_table(H264Context *h)
342 const int max_qp = 51 + 6 * (h->sps.bit_depth_luma - 8);
343 for (i = 0; i < 6; i++) {
344 h->dequant4_coeff[i] = h->dequant4_buffer[i];
345 for (j = 0; j < i; j++)
346 if (!memcmp(h->pps.scaling_matrix4[j], h->pps.scaling_matrix4[i],
347 16 * sizeof(uint8_t))) {
348 h->dequant4_coeff[i] = h->dequant4_buffer[j];
354 for (q = 0; q < max_qp + 1; q++) {
355 int shift = div6[q] + 2;
357 for (x = 0; x < 16; x++)
358 h->dequant4_coeff[i][q][(x >> 2) | ((x << 2) & 0xF)] =
359 ((uint32_t)dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] *
360 h->pps.scaling_matrix4[i][x]) << shift;
365 void ff_h264_init_dequant_tables(H264Context *h)
368 init_dequant4_coeff_table(h);
369 memset(h->dequant8_coeff, 0, sizeof(h->dequant8_coeff));
371 if (h->pps.transform_8x8_mode)
372 init_dequant8_coeff_table(h);
373 if (h->sps.transform_bypass) {
374 for (i = 0; i < 6; i++)
375 for (x = 0; x < 16; x++)
376 h->dequant4_coeff[i][0][x] = 1 << 6;
377 if (h->pps.transform_8x8_mode)
378 for (i = 0; i < 6; i++)
379 for (x = 0; x < 64; x++)
380 h->dequant8_coeff[i][0][x] = 1 << 6;
385 * Mimic alloc_tables(), but for every context thread.
387 static void clone_tables(H264Context *dst, H264SliceContext *sl,
388 H264Context *src, int i)
390 sl->intra4x4_pred_mode = src->intra4x4_pred_mode + i * 8 * 2 * src->mb_stride;
391 sl->mvd_table[0] = src->mvd_table[0] + i * 8 * 2 * src->mb_stride;
392 sl->mvd_table[1] = src->mvd_table[1] + i * 8 * 2 * src->mb_stride;
394 dst->non_zero_count = src->non_zero_count;
395 dst->slice_table = src->slice_table;
396 dst->cbp_table = src->cbp_table;
397 dst->mb2b_xy = src->mb2b_xy;
398 dst->mb2br_xy = src->mb2br_xy;
399 dst->chroma_pred_mode_table = src->chroma_pred_mode_table;
400 dst->direct_table = src->direct_table;
401 dst->list_counts = src->list_counts;
403 dst->cur_pic_ptr = src->cur_pic_ptr;
404 dst->cur_pic = src->cur_pic;
405 dst->bipred_scratchpad = NULL;
406 dst->edge_emu_buffer = NULL;
407 ff_h264_pred_init(&dst->hpc, src->avctx->codec_id, src->sps.bit_depth_luma,
408 src->sps.chroma_format_idc);
411 #define IN_RANGE(a, b, size) (((a) >= (b)) && ((a) < ((b) + (size))))
413 #define REBASE_PICTURE(pic, new_ctx, old_ctx) \
414 (((pic) && (pic) >= (old_ctx)->DPB && \
415 (pic) < (old_ctx)->DPB + H264_MAX_PICTURE_COUNT) ? \
416 &(new_ctx)->DPB[(pic) - (old_ctx)->DPB] : NULL)
418 static void copy_picture_range(H264Picture **to, H264Picture **from, int count,
419 H264Context *new_base,
420 H264Context *old_base)
424 for (i = 0; i < count; i++) {
425 assert((IN_RANGE(from[i], old_base, sizeof(*old_base)) ||
426 IN_RANGE(from[i], old_base->DPB,
427 sizeof(H264Picture) * H264_MAX_PICTURE_COUNT) ||
429 to[i] = REBASE_PICTURE(from[i], new_base, old_base);
433 static int copy_parameter_set(void **to, void **from, int count, int size)
437 for (i = 0; i < count; i++) {
438 if (to[i] && !from[i]) {
440 } else if (from[i] && !to[i]) {
441 to[i] = av_malloc(size);
443 return AVERROR(ENOMEM);
447 memcpy(to[i], from[i], size);
453 #define copy_fields(to, from, start_field, end_field) \
454 memcpy(&(to)->start_field, &(from)->start_field, \
455 (char *)&(to)->end_field - (char *)&(to)->start_field)
457 static int h264_slice_header_init(H264Context *h, int reinit);
459 int ff_h264_update_thread_context(AVCodecContext *dst,
460 const AVCodecContext *src)
462 H264Context *h = dst->priv_data, *h1 = src->priv_data;
463 int inited = h->context_initialized, err = 0;
464 int context_reinitialized = 0;
471 (h->width != h1->width ||
472 h->height != h1->height ||
473 h->mb_width != h1->mb_width ||
474 h->mb_height != h1->mb_height ||
475 h->sps.bit_depth_luma != h1->sps.bit_depth_luma ||
476 h->sps.chroma_format_idc != h1->sps.chroma_format_idc ||
477 h->sps.colorspace != h1->sps.colorspace)) {
479 /* set bits_per_raw_sample to the previous value. the check for changed
480 * bit depth in h264_set_parameter_from_sps() uses it and sets it to
481 * the current value */
482 h->avctx->bits_per_raw_sample = h->sps.bit_depth_luma;
484 av_freep(&h->bipred_scratchpad);
486 h->width = h1->width;
487 h->height = h1->height;
488 h->mb_height = h1->mb_height;
489 h->mb_width = h1->mb_width;
490 h->mb_num = h1->mb_num;
491 h->mb_stride = h1->mb_stride;
492 h->b_stride = h1->b_stride;
494 if ((ret = copy_parameter_set((void **)h->sps_buffers,
495 (void **)h1->sps_buffers,
496 MAX_SPS_COUNT, sizeof(SPS))) < 0)
499 if ((ret = copy_parameter_set((void **)h->pps_buffers,
500 (void **)h1->pps_buffers,
501 MAX_PPS_COUNT, sizeof(PPS))) < 0)
505 if ((err = h264_slice_header_init(h, 1)) < 0) {
506 av_log(h->avctx, AV_LOG_ERROR, "h264_slice_header_init() failed\n");
509 context_reinitialized = 1;
512 h264_set_parameter_from_sps(h);
513 //Note we set context_reinitialized which will cause h264_set_parameter_from_sps to be reexecuted
514 h->cur_chroma_format_idc = h1->cur_chroma_format_idc;
517 /* update linesize on resize for h264. The h264 decoder doesn't
518 * necessarily call ff_mpv_frame_start in the new thread */
519 h->linesize = h1->linesize;
520 h->uvlinesize = h1->uvlinesize;
522 /* copy block_offset since frame_start may not be called */
523 memcpy(h->block_offset, h1->block_offset, sizeof(h->block_offset));
526 H264SliceContext *orig_slice_ctx = h->slice_ctx;
528 for (i = 0; i < MAX_SPS_COUNT; i++)
529 av_freep(h->sps_buffers + i);
531 for (i = 0; i < MAX_PPS_COUNT; i++)
532 av_freep(h->pps_buffers + i);
534 av_freep(&h->rbsp_buffer[0]);
535 av_freep(&h->rbsp_buffer[1]);
536 ff_h264_unref_picture(h, &h->last_pic_for_ec);
537 memcpy(h, h1, sizeof(H264Context));
539 memset(h->sps_buffers, 0, sizeof(h->sps_buffers));
540 memset(h->pps_buffers, 0, sizeof(h->pps_buffers));
542 memset(&h->er, 0, sizeof(h->er));
543 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
544 memset(&h->last_pic_for_ec, 0, sizeof(h->last_pic_for_ec));
546 h->slice_ctx = orig_slice_ctx;
548 memset(&h->slice_ctx[0].mb, 0, sizeof(h->slice_ctx[0].mb));
549 memset(&h->slice_ctx[0].mb_luma_dc, 0, sizeof(h->slice_ctx[0].mb_luma_dc));
550 memset(&h->slice_ctx[0].mb_padding, 0, sizeof(h->slice_ctx[0].mb_padding));
554 h->qscale_table_pool = NULL;
555 h->mb_type_pool = NULL;
556 h->ref_index_pool = NULL;
557 h->motion_val_pool = NULL;
558 h->intra4x4_pred_mode= NULL;
559 h->non_zero_count = NULL;
560 h->slice_table_base = NULL;
561 h->slice_table = NULL;
563 h->chroma_pred_mode_table = NULL;
564 memset(h->mvd_table, 0, sizeof(h->mvd_table));
565 h->direct_table = NULL;
566 h->list_counts = NULL;
569 for (i = 0; i < 2; i++) {
570 h->rbsp_buffer[i] = NULL;
571 h->rbsp_buffer_size[i] = 0;
574 if (h1->context_initialized) {
575 h->context_initialized = 0;
577 memset(&h->cur_pic, 0, sizeof(h->cur_pic));
578 av_frame_unref(&h->cur_pic.f);
579 h->cur_pic.tf.f = &h->cur_pic.f;
581 ret = ff_h264_alloc_tables(h);
583 av_log(dst, AV_LOG_ERROR, "Could not allocate memory\n");
586 ret = ff_h264_context_init(h);
588 av_log(dst, AV_LOG_ERROR, "context_init() failed.\n");
593 h->bipred_scratchpad = NULL;
594 h->edge_emu_buffer = NULL;
596 h->thread_context[0] = h;
597 h->context_initialized = h1->context_initialized;
600 h->avctx->coded_height = h1->avctx->coded_height;
601 h->avctx->coded_width = h1->avctx->coded_width;
602 h->avctx->width = h1->avctx->width;
603 h->avctx->height = h1->avctx->height;
604 h->coded_picture_number = h1->coded_picture_number;
605 h->first_field = h1->first_field;
606 h->picture_structure = h1->picture_structure;
607 h->droppable = h1->droppable;
608 h->low_delay = h1->low_delay;
610 for (i = 0; h->DPB && i < H264_MAX_PICTURE_COUNT; i++) {
611 ff_h264_unref_picture(h, &h->DPB[i]);
612 if (h1->DPB && h1->DPB[i].f.buf[0] &&
613 (ret = ff_h264_ref_picture(h, &h->DPB[i], &h1->DPB[i])) < 0)
617 h->cur_pic_ptr = REBASE_PICTURE(h1->cur_pic_ptr, h, h1);
618 ff_h264_unref_picture(h, &h->cur_pic);
619 if (h1->cur_pic.f.buf[0]) {
620 ret = ff_h264_ref_picture(h, &h->cur_pic, &h1->cur_pic);
625 h->workaround_bugs = h1->workaround_bugs;
626 h->low_delay = h1->low_delay;
627 h->droppable = h1->droppable;
629 // extradata/NAL handling
630 h->is_avc = h1->is_avc;
633 if ((ret = copy_parameter_set((void **)h->sps_buffers,
634 (void **)h1->sps_buffers,
635 MAX_SPS_COUNT, sizeof(SPS))) < 0)
638 if ((ret = copy_parameter_set((void **)h->pps_buffers,
639 (void **)h1->pps_buffers,
640 MAX_PPS_COUNT, sizeof(PPS))) < 0)
644 // Dequantization matrices
645 // FIXME these are big - can they be only copied when PPS changes?
646 copy_fields(h, h1, dequant4_buffer, dequant4_coeff);
648 for (i = 0; i < 6; i++)
649 h->dequant4_coeff[i] = h->dequant4_buffer[0] +
650 (h1->dequant4_coeff[i] - h1->dequant4_buffer[0]);
652 for (i = 0; i < 6; i++)
653 h->dequant8_coeff[i] = h->dequant8_buffer[0] +
654 (h1->dequant8_coeff[i] - h1->dequant8_buffer[0]);
656 h->dequant_coeff_pps = h1->dequant_coeff_pps;
659 copy_fields(h, h1, poc_lsb, redundant_pic_count);
662 copy_fields(h, h1, short_ref, cabac_init_idc);
664 copy_picture_range(h->short_ref, h1->short_ref, 32, h, h1);
665 copy_picture_range(h->long_ref, h1->long_ref, 32, h, h1);
666 copy_picture_range(h->delayed_pic, h1->delayed_pic,
667 MAX_DELAYED_PIC_COUNT + 2, h, h1);
669 h->frame_recovered = h1->frame_recovered;
671 if (context_reinitialized)
672 ff_h264_set_parameter_from_sps(h);
678 err = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
679 h->prev_poc_msb = h->poc_msb;
680 h->prev_poc_lsb = h->poc_lsb;
682 h->prev_frame_num_offset = h->frame_num_offset;
683 h->prev_frame_num = h->frame_num;
684 h->outputed_poc = h->next_outputed_poc;
686 h->recovery_frame = h1->recovery_frame;
691 static int h264_frame_start(H264Context *h)
695 const int pixel_shift = h->pixel_shift;
697 1<<(h->sps.bit_depth_luma-1),
698 1<<(h->sps.bit_depth_chroma-1),
699 1<<(h->sps.bit_depth_chroma-1),
703 if (!ff_thread_can_start_frame(h->avctx)) {
704 av_log(h->avctx, AV_LOG_ERROR, "Attempt to start a frame outside SETUP state\n");
708 release_unused_pictures(h, 1);
709 h->cur_pic_ptr = NULL;
711 i = find_unused_picture(h);
713 av_log(h->avctx, AV_LOG_ERROR, "no frame buffer available\n");
718 pic->reference = h->droppable ? 0 : h->picture_structure;
719 pic->f.coded_picture_number = h->coded_picture_number++;
720 pic->field_picture = h->picture_structure != PICT_FRAME;
723 * Zero key_frame here; IDR markings per slice in frame or fields are ORed
725 * See decode_nal_units().
727 pic->f.key_frame = 0;
730 pic->invalid_gap = 0;
731 pic->sei_recovery_frame_cnt = h->sei_recovery_frame_cnt;
733 if ((ret = alloc_picture(h, pic)) < 0)
735 if(!h->frame_recovered && !h->avctx->hwaccel &&
736 !(h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU))
737 avpriv_color_frame(&pic->f, c);
739 h->cur_pic_ptr = pic;
740 ff_h264_unref_picture(h, &h->cur_pic);
741 if (CONFIG_ERROR_RESILIENCE) {
742 ff_h264_set_erpic(&h->er.cur_pic, NULL);
745 if ((ret = ff_h264_ref_picture(h, &h->cur_pic, h->cur_pic_ptr)) < 0)
748 if (CONFIG_ERROR_RESILIENCE) {
749 ff_er_frame_start(&h->er);
750 ff_h264_set_erpic(&h->er.last_pic, NULL);
751 ff_h264_set_erpic(&h->er.next_pic, NULL);
754 assert(h->linesize && h->uvlinesize);
756 for (i = 0; i < 16; i++) {
757 h->block_offset[i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
758 h->block_offset[48 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);
760 for (i = 0; i < 16; i++) {
761 h->block_offset[16 + i] =
762 h->block_offset[32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
763 h->block_offset[48 + 16 + i] =
764 h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7) << pixel_shift) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);
767 /* We mark the current picture as non-reference after allocating it, so
768 * that if we break out due to an error it can be released automatically
769 * in the next ff_mpv_frame_start().
771 h->cur_pic_ptr->reference = 0;
773 h->cur_pic_ptr->field_poc[0] = h->cur_pic_ptr->field_poc[1] = INT_MAX;
775 h->next_output_pic = NULL;
777 assert(h->cur_pic_ptr->long_ref == 0);
782 static av_always_inline void backup_mb_border(H264Context *h, uint8_t *src_y,
783 uint8_t *src_cb, uint8_t *src_cr,
784 int linesize, int uvlinesize,
789 const int pixel_shift = h->pixel_shift;
790 int chroma444 = CHROMA444(h);
791 int chroma422 = CHROMA422(h);
794 src_cb -= uvlinesize;
795 src_cr -= uvlinesize;
797 if (!simple && FRAME_MBAFF(h)) {
800 top_border = h->top_borders[0][h->mb_x];
801 AV_COPY128(top_border, src_y + 15 * linesize);
803 AV_COPY128(top_border + 16, src_y + 15 * linesize + 16);
804 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
807 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
808 AV_COPY128(top_border + 48, src_cb + 15 * uvlinesize + 16);
809 AV_COPY128(top_border + 64, src_cr + 15 * uvlinesize);
810 AV_COPY128(top_border + 80, src_cr + 15 * uvlinesize + 16);
812 AV_COPY128(top_border + 16, src_cb + 15 * uvlinesize);
813 AV_COPY128(top_border + 32, src_cr + 15 * uvlinesize);
815 } else if (chroma422) {
817 AV_COPY128(top_border + 32, src_cb + 15 * uvlinesize);
818 AV_COPY128(top_border + 48, src_cr + 15 * uvlinesize);
820 AV_COPY64(top_border + 16, src_cb + 15 * uvlinesize);
821 AV_COPY64(top_border + 24, src_cr + 15 * uvlinesize);
825 AV_COPY128(top_border + 32, src_cb + 7 * uvlinesize);
826 AV_COPY128(top_border + 48, src_cr + 7 * uvlinesize);
828 AV_COPY64(top_border + 16, src_cb + 7 * uvlinesize);
829 AV_COPY64(top_border + 24, src_cr + 7 * uvlinesize);
834 } else if (MB_MBAFF(h)) {
840 top_border = h->top_borders[top_idx][h->mb_x];
841 /* There are two lines saved, the line above the top macroblock
842 * of a pair, and the line above the bottom macroblock. */
843 AV_COPY128(top_border, src_y + 16 * linesize);
845 AV_COPY128(top_border + 16, src_y + 16 * linesize + 16);
847 if (simple || !CONFIG_GRAY || !(h->flags & CODEC_FLAG_GRAY)) {
850 AV_COPY128(top_border + 32, src_cb + 16 * linesize);
851 AV_COPY128(top_border + 48, src_cb + 16 * linesize + 16);
852 AV_COPY128(top_border + 64, src_cr + 16 * linesize);
853 AV_COPY128(top_border + 80, src_cr + 16 * linesize + 16);
855 AV_COPY128(top_border + 16, src_cb + 16 * linesize);
856 AV_COPY128(top_border + 32, src_cr + 16 * linesize);
858 } else if (chroma422) {
860 AV_COPY128(top_border + 32, src_cb + 16 * uvlinesize);
861 AV_COPY128(top_border + 48, src_cr + 16 * uvlinesize);
863 AV_COPY64(top_border + 16, src_cb + 16 * uvlinesize);
864 AV_COPY64(top_border + 24, src_cr + 16 * uvlinesize);
868 AV_COPY128(top_border + 32, src_cb + 8 * uvlinesize);
869 AV_COPY128(top_border + 48, src_cr + 8 * uvlinesize);
871 AV_COPY64(top_border + 16, src_cb + 8 * uvlinesize);
872 AV_COPY64(top_border + 24, src_cr + 8 * uvlinesize);
879 * Initialize implicit_weight table.
880 * @param field 0/1 initialize the weight for interlaced MBAFF
881 * -1 initializes the rest
883 static void implicit_weight_table(H264Context *h, H264SliceContext *sl, int field)
885 int ref0, ref1, i, cur_poc, ref_start, ref_count0, ref_count1;
887 for (i = 0; i < 2; i++) {
888 sl->luma_weight_flag[i] = 0;
889 sl->chroma_weight_flag[i] = 0;
893 if (h->picture_structure == PICT_FRAME) {
894 cur_poc = h->cur_pic_ptr->poc;
896 cur_poc = h->cur_pic_ptr->field_poc[h->picture_structure - 1];
898 if (sl->ref_count[0] == 1 && sl->ref_count[1] == 1 && !FRAME_MBAFF(h) &&
899 sl->ref_list[0][0].poc + sl->ref_list[1][0].poc == 2 * cur_poc) {
901 sl->use_weight_chroma = 0;
905 ref_count0 = sl->ref_count[0];
906 ref_count1 = sl->ref_count[1];
908 cur_poc = h->cur_pic_ptr->field_poc[field];
910 ref_count0 = 16 + 2 * sl->ref_count[0];
911 ref_count1 = 16 + 2 * sl->ref_count[1];
915 sl->use_weight_chroma = 2;
916 sl->luma_log2_weight_denom = 5;
917 sl->chroma_log2_weight_denom = 5;
919 for (ref0 = ref_start; ref0 < ref_count0; ref0++) {
920 int poc0 = sl->ref_list[0][ref0].poc;
921 for (ref1 = ref_start; ref1 < ref_count1; ref1++) {
923 if (!sl->ref_list[0][ref0].long_ref && !sl->ref_list[1][ref1].long_ref) {
924 int poc1 = sl->ref_list[1][ref1].poc;
925 int td = av_clip_int8(poc1 - poc0);
927 int tb = av_clip_int8(cur_poc - poc0);
928 int tx = (16384 + (FFABS(td) >> 1)) / td;
929 int dist_scale_factor = (tb * tx + 32) >> 8;
930 if (dist_scale_factor >= -64 && dist_scale_factor <= 128)
931 w = 64 - dist_scale_factor;
935 sl->implicit_weight[ref0][ref1][0] =
936 sl->implicit_weight[ref0][ref1][1] = w;
938 sl->implicit_weight[ref0][ref1][field] = w;
945 * initialize scan tables
947 static void init_scan_tables(H264Context *h)
950 for (i = 0; i < 16; i++) {
951 #define TRANSPOSE(x) ((x) >> 2) | (((x) << 2) & 0xF)
952 h->zigzag_scan[i] = TRANSPOSE(zigzag_scan[i]);
953 h->field_scan[i] = TRANSPOSE(field_scan[i]);
956 for (i = 0; i < 64; i++) {
957 #define TRANSPOSE(x) ((x) >> 3) | (((x) & 7) << 3)
958 h->zigzag_scan8x8[i] = TRANSPOSE(ff_zigzag_direct[i]);
959 h->zigzag_scan8x8_cavlc[i] = TRANSPOSE(zigzag_scan8x8_cavlc[i]);
960 h->field_scan8x8[i] = TRANSPOSE(field_scan8x8[i]);
961 h->field_scan8x8_cavlc[i] = TRANSPOSE(field_scan8x8_cavlc[i]);
964 if (h->sps.transform_bypass) { // FIXME same ugly
965 memcpy(h->zigzag_scan_q0 , zigzag_scan , sizeof(h->zigzag_scan_q0 ));
966 memcpy(h->zigzag_scan8x8_q0 , ff_zigzag_direct , sizeof(h->zigzag_scan8x8_q0 ));
967 memcpy(h->zigzag_scan8x8_cavlc_q0 , zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
968 memcpy(h->field_scan_q0 , field_scan , sizeof(h->field_scan_q0 ));
969 memcpy(h->field_scan8x8_q0 , field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
970 memcpy(h->field_scan8x8_cavlc_q0 , field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
972 memcpy(h->zigzag_scan_q0 , h->zigzag_scan , sizeof(h->zigzag_scan_q0 ));
973 memcpy(h->zigzag_scan8x8_q0 , h->zigzag_scan8x8 , sizeof(h->zigzag_scan8x8_q0 ));
974 memcpy(h->zigzag_scan8x8_cavlc_q0 , h->zigzag_scan8x8_cavlc , sizeof(h->zigzag_scan8x8_cavlc_q0));
975 memcpy(h->field_scan_q0 , h->field_scan , sizeof(h->field_scan_q0 ));
976 memcpy(h->field_scan8x8_q0 , h->field_scan8x8 , sizeof(h->field_scan8x8_q0 ));
977 memcpy(h->field_scan8x8_cavlc_q0 , h->field_scan8x8_cavlc , sizeof(h->field_scan8x8_cavlc_q0 ));
982 * Replicate H264 "master" context to thread contexts.
984 static int clone_slice(H264Context *dst, H264Context *src)
986 memcpy(dst->block_offset, src->block_offset, sizeof(dst->block_offset));
987 dst->cur_pic_ptr = src->cur_pic_ptr;
988 dst->cur_pic = src->cur_pic;
989 dst->linesize = src->linesize;
990 dst->uvlinesize = src->uvlinesize;
991 dst->first_field = src->first_field;
993 dst->prev_poc_msb = src->prev_poc_msb;
994 dst->prev_poc_lsb = src->prev_poc_lsb;
995 dst->prev_frame_num_offset = src->prev_frame_num_offset;
996 dst->prev_frame_num = src->prev_frame_num;
997 dst->short_ref_count = src->short_ref_count;
999 memcpy(dst->short_ref, src->short_ref, sizeof(dst->short_ref));
1000 memcpy(dst->long_ref, src->long_ref, sizeof(dst->long_ref));
1001 memcpy(dst->default_ref_list, src->default_ref_list, sizeof(dst->default_ref_list));
1003 memcpy(dst->dequant4_coeff, src->dequant4_coeff, sizeof(src->dequant4_coeff));
1004 memcpy(dst->dequant8_coeff, src->dequant8_coeff, sizeof(src->dequant8_coeff));
1009 static enum AVPixelFormat get_pixel_format(H264Context *h, int force_callback)
1011 #define HWACCEL_MAX (CONFIG_H264_DXVA2_HWACCEL + \
1012 CONFIG_H264_VAAPI_HWACCEL + \
1013 (CONFIG_H264_VDA_HWACCEL * 2) + \
1014 CONFIG_H264_VDPAU_HWACCEL)
1015 enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
1016 const enum AVPixelFormat *choices = pix_fmts;
1019 switch (h->sps.bit_depth_luma) {
1022 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1023 *fmt++ = AV_PIX_FMT_GBRP9;
1025 *fmt++ = AV_PIX_FMT_YUV444P9;
1026 } else if (CHROMA422(h))
1027 *fmt++ = AV_PIX_FMT_YUV422P9;
1029 *fmt++ = AV_PIX_FMT_YUV420P9;
1033 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1034 *fmt++ = AV_PIX_FMT_GBRP10;
1036 *fmt++ = AV_PIX_FMT_YUV444P10;
1037 } else if (CHROMA422(h))
1038 *fmt++ = AV_PIX_FMT_YUV422P10;
1040 *fmt++ = AV_PIX_FMT_YUV420P10;
1044 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1045 *fmt++ = AV_PIX_FMT_GBRP12;
1047 *fmt++ = AV_PIX_FMT_YUV444P12;
1048 } else if (CHROMA422(h))
1049 *fmt++ = AV_PIX_FMT_YUV422P12;
1051 *fmt++ = AV_PIX_FMT_YUV420P12;
1055 if (h->avctx->colorspace == AVCOL_SPC_RGB) {
1056 *fmt++ = AV_PIX_FMT_GBRP14;
1058 *fmt++ = AV_PIX_FMT_YUV444P14;
1059 } else if (CHROMA422(h))
1060 *fmt++ = AV_PIX_FMT_YUV422P14;
1062 *fmt++ = AV_PIX_FMT_YUV420P14;
1065 #if CONFIG_H264_VDPAU_HWACCEL
1066 *fmt++ = AV_PIX_FMT_VDPAU;
1069 if (h->avctx->colorspace == AVCOL_SPC_YCGCO)
1070 av_log(h->avctx, AV_LOG_WARNING, "Detected unsupported YCgCo colorspace.\n");
1071 if (h->avctx->colorspace == AVCOL_SPC_RGB)
1072 *fmt++ = AV_PIX_FMT_GBRP;
1073 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1074 *fmt++ = AV_PIX_FMT_YUVJ444P;
1076 *fmt++ = AV_PIX_FMT_YUV444P;
1077 } else if (CHROMA422(h)) {
1078 if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1079 *fmt++ = AV_PIX_FMT_YUVJ422P;
1081 *fmt++ = AV_PIX_FMT_YUV422P;
1083 #if CONFIG_H264_DXVA2_HWACCEL
1084 *fmt++ = AV_PIX_FMT_DXVA2_VLD;
1086 #if CONFIG_H264_VAAPI_HWACCEL
1087 *fmt++ = AV_PIX_FMT_VAAPI_VLD;
1089 #if CONFIG_H264_VDA_HWACCEL
1090 *fmt++ = AV_PIX_FMT_VDA_VLD;
1091 *fmt++ = AV_PIX_FMT_VDA;
1093 if (h->avctx->codec->pix_fmts)
1094 choices = h->avctx->codec->pix_fmts;
1095 else if (h->avctx->color_range == AVCOL_RANGE_JPEG)
1096 *fmt++ = AV_PIX_FMT_YUVJ420P;
1098 *fmt++ = AV_PIX_FMT_YUV420P;
1102 av_log(h->avctx, AV_LOG_ERROR,
1103 "Unsupported bit depth %d\n", h->sps.bit_depth_luma);
1104 return AVERROR_INVALIDDATA;
1107 *fmt = AV_PIX_FMT_NONE;
1109 for (i=0; choices[i] != AV_PIX_FMT_NONE; i++)
1110 if (choices[i] == h->avctx->pix_fmt && !force_callback)
1112 return ff_thread_get_format(h->avctx, choices);
1115 /* export coded and cropped frame dimensions to AVCodecContext */
1116 static int init_dimensions(H264Context *h)
1118 int width = h->width - (h->sps.crop_right + h->sps.crop_left);
1119 int height = h->height - (h->sps.crop_top + h->sps.crop_bottom);
1120 int crop_present = h->sps.crop_left || h->sps.crop_top ||
1121 h->sps.crop_right || h->sps.crop_bottom;
1122 av_assert0(h->sps.crop_right + h->sps.crop_left < (unsigned)h->width);
1123 av_assert0(h->sps.crop_top + h->sps.crop_bottom < (unsigned)h->height);
1125 /* handle container cropping */
1126 if (!crop_present &&
1127 FFALIGN(h->avctx->width, 16) == h->width &&
1128 FFALIGN(h->avctx->height, 16) == h->height) {
1129 width = h->avctx->width;
1130 height = h->avctx->height;
1133 if (width <= 0 || height <= 0) {
1134 av_log(h->avctx, AV_LOG_ERROR, "Invalid cropped dimensions: %dx%d.\n",
1136 if (h->avctx->err_recognition & AV_EF_EXPLODE)
1137 return AVERROR_INVALIDDATA;
1139 av_log(h->avctx, AV_LOG_WARNING, "Ignoring cropping information.\n");
1140 h->sps.crop_bottom =
1150 h->avctx->coded_width = h->width;
1151 h->avctx->coded_height = h->height;
1152 h->avctx->width = width;
1153 h->avctx->height = height;
1158 static int h264_slice_header_init(H264Context *h, int reinit)
1160 int nb_slices = (HAVE_THREADS &&
1161 h->avctx->active_thread_type & FF_THREAD_SLICE) ?
1162 h->avctx->thread_count : 1;
1165 ff_set_sar(h->avctx, h->sps.sar);
1166 av_pix_fmt_get_chroma_sub_sample(h->avctx->pix_fmt,
1167 &h->chroma_x_shift, &h->chroma_y_shift);
1169 if (h->sps.timing_info_present_flag) {
1170 int64_t den = h->sps.time_scale;
1171 if (h->x264_build < 44U)
1173 av_reduce(&h->avctx->framerate.den, &h->avctx->framerate.num,
1174 h->sps.num_units_in_tick * h->avctx->ticks_per_frame, den, 1 << 30);
1178 ff_h264_free_tables(h, 0);
1180 h->prev_interlaced_frame = 1;
1182 init_scan_tables(h);
1183 ret = ff_h264_alloc_tables(h);
1185 av_log(h->avctx, AV_LOG_ERROR, "Could not allocate memory\n");
1189 if (nb_slices > H264_MAX_THREADS || (nb_slices > h->mb_height && h->mb_height)) {
1192 max_slices = FFMIN(H264_MAX_THREADS, h->mb_height);
1194 max_slices = H264_MAX_THREADS;
1195 av_log(h->avctx, AV_LOG_WARNING, "too many threads/slices %d,"
1196 " reducing to %d\n", nb_slices, max_slices);
1197 nb_slices = max_slices;
1199 h->slice_context_count = nb_slices;
1201 if (!HAVE_THREADS || !(h->avctx->active_thread_type & FF_THREAD_SLICE)) {
1202 ret = ff_h264_context_init(h);
1204 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1208 for (i = 1; i < h->slice_context_count; i++) {
1210 c = h->thread_context[i] = av_mallocz(sizeof(H264Context));
1212 ret = AVERROR(ENOMEM);
1215 c->avctx = h->avctx;
1217 c->h264dsp = h->h264dsp;
1218 c->h264qpel = h->h264qpel;
1219 c->h264chroma = h->h264chroma;
1222 c->pixel_shift = h->pixel_shift;
1223 c->cur_chroma_format_idc = h->cur_chroma_format_idc;
1224 c->width = h->width;
1225 c->height = h->height;
1226 c->linesize = h->linesize;
1227 c->uvlinesize = h->uvlinesize;
1228 c->chroma_x_shift = h->chroma_x_shift;
1229 c->chroma_y_shift = h->chroma_y_shift;
1230 c->droppable = h->droppable;
1231 c->low_delay = h->low_delay;
1232 c->mb_width = h->mb_width;
1233 c->mb_height = h->mb_height;
1234 c->mb_stride = h->mb_stride;
1235 c->mb_num = h->mb_num;
1236 c->flags = h->flags;
1237 c->workaround_bugs = h->workaround_bugs;
1238 c->pict_type = h->pict_type;
1240 h->slice_ctx[i].h264 = c;
1242 init_scan_tables(c);
1243 clone_tables(c, &h->slice_ctx[i], h, i);
1244 c->context_initialized = 1;
1247 for (i = 0; i < h->slice_context_count; i++)
1248 if ((ret = ff_h264_context_init(h->thread_context[i])) < 0) {
1249 av_log(h->avctx, AV_LOG_ERROR, "context_init() failed.\n");
1254 h->context_initialized = 1;
1258 ff_h264_free_tables(h, 0);
1259 h->context_initialized = 0;
1263 static enum AVPixelFormat non_j_pixfmt(enum AVPixelFormat a)
1266 case AV_PIX_FMT_YUVJ420P: return AV_PIX_FMT_YUV420P;
1267 case AV_PIX_FMT_YUVJ422P: return AV_PIX_FMT_YUV422P;
1268 case AV_PIX_FMT_YUVJ444P: return AV_PIX_FMT_YUV444P;
1275 * Decode a slice header.
1276 * This will (re)intialize the decoder and call h264_frame_start() as needed.
1278 * @param h h264context
1279 * @param h0 h264 master context (differs from 'h' when doing sliced based
1280 * parallel decoding)
1282 * @return 0 if okay, <0 if an error occurred, 1 if decoding must not be multithreaded
1284 int ff_h264_decode_slice_header(H264Context *h, H264SliceContext *sl, H264Context *h0)
1286 unsigned int first_mb_in_slice;
1287 unsigned int pps_id;
1289 unsigned int slice_type, tmp, i, j;
1290 int last_pic_structure, last_pic_droppable;
1292 int needs_reinit = 0;
1293 int field_pic_flag, bottom_field_flag;
1294 int first_slice = h == h0 && !h0->current_slice;
1295 int frame_num, picture_structure, droppable;
1298 h->qpel_put = h->h264qpel.put_h264_qpel_pixels_tab;
1299 h->qpel_avg = h->h264qpel.avg_h264_qpel_pixels_tab;
1301 first_mb_in_slice = get_ue_golomb_long(&h->gb);
1303 if (first_mb_in_slice == 0) { // FIXME better field boundary detection
1304 if (h0->current_slice && h->cur_pic_ptr && FIELD_PICTURE(h)) {
1305 ff_h264_field_end(h, sl, 1);
1308 h0->current_slice = 0;
1309 if (!h0->first_field) {
1310 if (h->cur_pic_ptr && !h->droppable) {
1311 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX,
1312 h->picture_structure == PICT_BOTTOM_FIELD);
1314 h->cur_pic_ptr = NULL;
1318 slice_type = get_ue_golomb_31(&h->gb);
1319 if (slice_type > 9) {
1320 av_log(h->avctx, AV_LOG_ERROR,
1321 "slice type %d too large at %d %d\n",
1322 slice_type, h->mb_x, h->mb_y);
1323 return AVERROR_INVALIDDATA;
1325 if (slice_type > 4) {
1327 sl->slice_type_fixed = 1;
1329 sl->slice_type_fixed = 0;
1331 slice_type = golomb_to_pict_type[slice_type];
1332 sl->slice_type = slice_type;
1333 sl->slice_type_nos = slice_type & 3;
1335 if (h->nal_unit_type == NAL_IDR_SLICE &&
1336 sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1337 av_log(h->avctx, AV_LOG_ERROR, "A non-intra slice in an IDR NAL unit.\n");
1338 return AVERROR_INVALIDDATA;
1342 (h->avctx->skip_frame >= AVDISCARD_NONREF && !h->nal_ref_idc) ||
1343 (h->avctx->skip_frame >= AVDISCARD_BIDIR && sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1344 (h->avctx->skip_frame >= AVDISCARD_NONINTRA && sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1345 (h->avctx->skip_frame >= AVDISCARD_NONKEY && h->nal_unit_type != NAL_IDR_SLICE) ||
1346 h->avctx->skip_frame >= AVDISCARD_ALL) {
1347 return SLICE_SKIPED;
1350 // to make a few old functions happy, it's wrong though
1351 h->pict_type = sl->slice_type;
1353 pps_id = get_ue_golomb(&h->gb);
1354 if (pps_id >= MAX_PPS_COUNT) {
1355 av_log(h->avctx, AV_LOG_ERROR, "pps_id %u out of range\n", pps_id);
1356 return AVERROR_INVALIDDATA;
1358 if (!h0->pps_buffers[pps_id]) {
1359 av_log(h->avctx, AV_LOG_ERROR,
1360 "non-existing PPS %u referenced\n",
1362 return AVERROR_INVALIDDATA;
1364 if (h0->au_pps_id >= 0 && pps_id != h0->au_pps_id) {
1365 av_log(h->avctx, AV_LOG_ERROR,
1366 "PPS change from %d to %d forbidden\n",
1367 h0->au_pps_id, pps_id);
1368 return AVERROR_INVALIDDATA;
1371 pps = h0->pps_buffers[pps_id];
1373 if (!h0->sps_buffers[pps->sps_id]) {
1374 av_log(h->avctx, AV_LOG_ERROR,
1375 "non-existing SPS %u referenced\n",
1377 return AVERROR_INVALIDDATA;
1380 h->pps = *h0->pps_buffers[pps_id];
1382 if (pps->sps_id != h->sps.sps_id ||
1383 pps->sps_id != h->current_sps_id ||
1384 h0->sps_buffers[pps->sps_id]->new) {
1387 av_log(h->avctx, AV_LOG_ERROR,
1388 "SPS changed in the middle of the frame\n");
1389 return AVERROR_INVALIDDATA;
1392 h->sps = *h0->sps_buffers[h->pps.sps_id];
1394 if (h->mb_width != h->sps.mb_width ||
1395 h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) ||
1396 h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma ||
1397 h->cur_chroma_format_idc != h->sps.chroma_format_idc
1401 if (h->bit_depth_luma != h->sps.bit_depth_luma ||
1402 h->chroma_format_idc != h->sps.chroma_format_idc) {
1403 h->bit_depth_luma = h->sps.bit_depth_luma;
1404 h->chroma_format_idc = h->sps.chroma_format_idc;
1407 if ((ret = ff_h264_set_parameter_from_sps(h)) < 0)
1411 h->avctx->profile = ff_h264_get_profile(&h->sps);
1412 h->avctx->level = h->sps.level_idc;
1413 h->avctx->refs = h->sps.ref_frame_count;
1415 must_reinit = (h->context_initialized &&
1416 ( 16*h->sps.mb_width != h->avctx->coded_width
1417 || 16*h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag) != h->avctx->coded_height
1418 || h->avctx->bits_per_raw_sample != h->sps.bit_depth_luma
1419 || h->cur_chroma_format_idc != h->sps.chroma_format_idc
1420 || h->mb_width != h->sps.mb_width
1421 || h->mb_height != h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag)
1423 if (non_j_pixfmt(h0->avctx->pix_fmt) != non_j_pixfmt(get_pixel_format(h0, 0)))
1426 if (first_slice && av_cmp_q(h->sps.sar, h->avctx->sample_aspect_ratio))
1429 h->mb_width = h->sps.mb_width;
1430 h->mb_height = h->sps.mb_height * (2 - h->sps.frame_mbs_only_flag);
1431 h->mb_num = h->mb_width * h->mb_height;
1432 h->mb_stride = h->mb_width + 1;
1434 h->b_stride = h->mb_width * 4;
1436 h->chroma_y_shift = h->sps.chroma_format_idc <= 1; // 400 uses yuv420p
1438 h->width = 16 * h->mb_width;
1439 h->height = 16 * h->mb_height;
1441 ret = init_dimensions(h);
1445 if (h->sps.video_signal_type_present_flag) {
1446 h->avctx->color_range = h->sps.full_range>0 ? AVCOL_RANGE_JPEG
1448 if (h->sps.colour_description_present_flag) {
1449 if (h->avctx->colorspace != h->sps.colorspace)
1451 h->avctx->color_primaries = h->sps.color_primaries;
1452 h->avctx->color_trc = h->sps.color_trc;
1453 h->avctx->colorspace = h->sps.colorspace;
1457 if (h->context_initialized &&
1458 (must_reinit || needs_reinit)) {
1460 av_log(h->avctx, AV_LOG_ERROR,
1461 "changing width %d -> %d / height %d -> %d on "
1463 h->width, h->avctx->coded_width,
1464 h->height, h->avctx->coded_height,
1465 h0->current_slice + 1);
1466 return AVERROR_INVALIDDATA;
1469 av_assert1(first_slice);
1471 ff_h264_flush_change(h);
1473 if ((ret = get_pixel_format(h, 1)) < 0)
1475 h->avctx->pix_fmt = ret;
1477 av_log(h->avctx, AV_LOG_INFO, "Reinit context to %dx%d, "
1478 "pix_fmt: %s\n", h->width, h->height, av_get_pix_fmt_name(h->avctx->pix_fmt));
1480 if ((ret = h264_slice_header_init(h, 1)) < 0) {
1481 av_log(h->avctx, AV_LOG_ERROR,
1482 "h264_slice_header_init() failed\n");
1486 if (!h->context_initialized) {
1488 av_log(h->avctx, AV_LOG_ERROR,
1489 "Cannot (re-)initialize context during parallel decoding.\n");
1490 return AVERROR_PATCHWELCOME;
1493 if ((ret = get_pixel_format(h, 1)) < 0)
1495 h->avctx->pix_fmt = ret;
1497 if ((ret = h264_slice_header_init(h, 0)) < 0) {
1498 av_log(h->avctx, AV_LOG_ERROR,
1499 "h264_slice_header_init() failed\n");
1504 if (first_slice && h->dequant_coeff_pps != pps_id) {
1505 h->dequant_coeff_pps = pps_id;
1506 ff_h264_init_dequant_tables(h);
1509 frame_num = get_bits(&h->gb, h->sps.log2_max_frame_num);
1511 if (h0->frame_num != frame_num) {
1512 av_log(h->avctx, AV_LOG_ERROR, "Frame num change from %d to %d\n",
1513 h0->frame_num, frame_num);
1514 return AVERROR_INVALIDDATA;
1519 h->mb_aff_frame = 0;
1520 last_pic_structure = h0->picture_structure;
1521 last_pic_droppable = h0->droppable;
1522 droppable = h->nal_ref_idc == 0;
1523 if (h->sps.frame_mbs_only_flag) {
1524 picture_structure = PICT_FRAME;
1526 if (!h->sps.direct_8x8_inference_flag && slice_type == AV_PICTURE_TYPE_B) {
1527 av_log(h->avctx, AV_LOG_ERROR, "This stream was generated by a broken encoder, invalid 8x8 inference\n");
1530 field_pic_flag = get_bits1(&h->gb);
1532 if (field_pic_flag) {
1533 bottom_field_flag = get_bits1(&h->gb);
1534 picture_structure = PICT_TOP_FIELD + bottom_field_flag;
1536 picture_structure = PICT_FRAME;
1537 h->mb_aff_frame = h->sps.mb_aff;
1540 if (h0->current_slice) {
1541 if (last_pic_structure != picture_structure ||
1542 last_pic_droppable != droppable) {
1543 av_log(h->avctx, AV_LOG_ERROR,
1544 "Changing field mode (%d -> %d) between slices is not allowed\n",
1545 last_pic_structure, h->picture_structure);
1546 return AVERROR_INVALIDDATA;
1547 } else if (!h0->cur_pic_ptr) {
1548 av_log(h->avctx, AV_LOG_ERROR,
1549 "unset cur_pic_ptr on slice %d\n",
1550 h0->current_slice + 1);
1551 return AVERROR_INVALIDDATA;
1555 h->picture_structure = picture_structure;
1556 h->droppable = droppable;
1557 h->frame_num = frame_num;
1558 h->mb_field_decoding_flag = picture_structure != PICT_FRAME;
1560 if (h0->current_slice == 0) {
1561 /* Shorten frame num gaps so we don't have to allocate reference
1562 * frames just to throw them away */
1563 if (h->frame_num != h->prev_frame_num) {
1564 int unwrap_prev_frame_num = h->prev_frame_num;
1565 int max_frame_num = 1 << h->sps.log2_max_frame_num;
1567 if (unwrap_prev_frame_num > h->frame_num)
1568 unwrap_prev_frame_num -= max_frame_num;
1570 if ((h->frame_num - unwrap_prev_frame_num) > h->sps.ref_frame_count) {
1571 unwrap_prev_frame_num = (h->frame_num - h->sps.ref_frame_count) - 1;
1572 if (unwrap_prev_frame_num < 0)
1573 unwrap_prev_frame_num += max_frame_num;
1575 h->prev_frame_num = unwrap_prev_frame_num;
1579 /* See if we have a decoded first field looking for a pair...
1580 * Here, we're using that to see if we should mark previously
1581 * decode frames as "finished".
1582 * We have to do that before the "dummy" in-between frame allocation,
1583 * since that can modify h->cur_pic_ptr. */
1584 if (h0->first_field) {
1585 assert(h0->cur_pic_ptr);
1586 assert(h0->cur_pic_ptr->f.buf[0]);
1587 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1589 /* Mark old field/frame as completed */
1590 if (h0->cur_pic_ptr->tf.owner == h0->avctx) {
1591 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1592 last_pic_structure == PICT_BOTTOM_FIELD);
1595 /* figure out if we have a complementary field pair */
1596 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1597 /* Previous field is unmatched. Don't display it, but let it
1598 * remain for reference if marked as such. */
1599 if (last_pic_structure != PICT_FRAME) {
1600 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1601 last_pic_structure == PICT_TOP_FIELD);
1604 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1605 /* This and previous field were reference, but had
1606 * different frame_nums. Consider this field first in
1607 * pair. Throw away previous field except for reference
1609 if (last_pic_structure != PICT_FRAME) {
1610 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1611 last_pic_structure == PICT_TOP_FIELD);
1614 /* Second field in complementary pair */
1615 if (!((last_pic_structure == PICT_TOP_FIELD &&
1616 h->picture_structure == PICT_BOTTOM_FIELD) ||
1617 (last_pic_structure == PICT_BOTTOM_FIELD &&
1618 h->picture_structure == PICT_TOP_FIELD))) {
1619 av_log(h->avctx, AV_LOG_ERROR,
1620 "Invalid field mode combination %d/%d\n",
1621 last_pic_structure, h->picture_structure);
1622 h->picture_structure = last_pic_structure;
1623 h->droppable = last_pic_droppable;
1624 return AVERROR_INVALIDDATA;
1625 } else if (last_pic_droppable != h->droppable) {
1626 avpriv_request_sample(h->avctx,
1627 "Found reference and non-reference fields in the same frame, which");
1628 h->picture_structure = last_pic_structure;
1629 h->droppable = last_pic_droppable;
1630 return AVERROR_PATCHWELCOME;
1636 while (h->frame_num != h->prev_frame_num && !h0->first_field &&
1637 h->frame_num != (h->prev_frame_num + 1) % (1 << h->sps.log2_max_frame_num)) {
1638 H264Picture *prev = h->short_ref_count ? h->short_ref[0] : NULL;
1639 av_log(h->avctx, AV_LOG_DEBUG, "Frame num gap %d %d\n",
1640 h->frame_num, h->prev_frame_num);
1641 if (!h->sps.gaps_in_frame_num_allowed_flag)
1642 for(i=0; i<FF_ARRAY_ELEMS(h->last_pocs); i++)
1643 h->last_pocs[i] = INT_MIN;
1644 ret = h264_frame_start(h);
1646 h0->first_field = 0;
1650 h->prev_frame_num++;
1651 h->prev_frame_num %= 1 << h->sps.log2_max_frame_num;
1652 h->cur_pic_ptr->frame_num = h->prev_frame_num;
1653 h->cur_pic_ptr->invalid_gap = !h->sps.gaps_in_frame_num_allowed_flag;
1654 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 0);
1655 ff_thread_report_progress(&h->cur_pic_ptr->tf, INT_MAX, 1);
1656 ret = ff_generate_sliding_window_mmcos(h, 1);
1657 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1659 ret = ff_h264_execute_ref_pic_marking(h, h->mmco, h->mmco_index);
1660 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1662 /* Error concealment: If a ref is missing, copy the previous ref
1664 * FIXME: Avoiding a memcpy would be nice, but ref handling makes
1665 * many assumptions about there being no actual duplicates.
1666 * FIXME: This does not copy padding for out-of-frame motion
1667 * vectors. Given we are concealing a lost frame, this probably
1668 * is not noticeable by comparison, but it should be fixed. */
1669 if (h->short_ref_count) {
1671 av_image_copy(h->short_ref[0]->f.data,
1672 h->short_ref[0]->f.linesize,
1673 (const uint8_t **)prev->f.data,
1678 h->short_ref[0]->poc = prev->poc + 2;
1680 h->short_ref[0]->frame_num = h->prev_frame_num;
1684 /* See if we have a decoded first field looking for a pair...
1685 * We're using that to see whether to continue decoding in that
1686 * frame, or to allocate a new one. */
1687 if (h0->first_field) {
1688 assert(h0->cur_pic_ptr);
1689 assert(h0->cur_pic_ptr->f.buf[0]);
1690 assert(h0->cur_pic_ptr->reference != DELAYED_PIC_REF);
1692 /* figure out if we have a complementary field pair */
1693 if (!FIELD_PICTURE(h) || h->picture_structure == last_pic_structure) {
1694 /* Previous field is unmatched. Don't display it, but let it
1695 * remain for reference if marked as such. */
1696 h0->missing_fields ++;
1697 h0->cur_pic_ptr = NULL;
1698 h0->first_field = FIELD_PICTURE(h);
1700 h0->missing_fields = 0;
1701 if (h0->cur_pic_ptr->frame_num != h->frame_num) {
1702 ff_thread_report_progress(&h0->cur_pic_ptr->tf, INT_MAX,
1703 h0->picture_structure==PICT_BOTTOM_FIELD);
1704 /* This and the previous field had different frame_nums.
1705 * Consider this field first in pair. Throw away previous
1706 * one except for reference purposes. */
1707 h0->first_field = 1;
1708 h0->cur_pic_ptr = NULL;
1710 /* Second field in complementary pair */
1711 h0->first_field = 0;
1715 /* Frame or first field in a potentially complementary pair */
1716 h0->first_field = FIELD_PICTURE(h);
1719 if (!FIELD_PICTURE(h) || h0->first_field) {
1720 if (h264_frame_start(h) < 0) {
1721 h0->first_field = 0;
1722 return AVERROR_INVALIDDATA;
1725 release_unused_pictures(h, 0);
1727 /* Some macroblocks can be accessed before they're available in case
1728 * of lost slices, MBAFF or threading. */
1729 if (FIELD_PICTURE(h)) {
1730 for(i = (h->picture_structure == PICT_BOTTOM_FIELD); i<h->mb_height; i++)
1731 memset(h->slice_table + i*h->mb_stride, -1, (h->mb_stride - (i+1==h->mb_height)) * sizeof(*h->slice_table));
1733 memset(h->slice_table, -1,
1734 (h->mb_height * h->mb_stride - 1) * sizeof(*h->slice_table));
1736 h0->last_slice_type = -1;
1738 if (h != h0 && (ret = clone_slice(h, h0)) < 0)
1741 /* can't be in alloc_tables because linesize isn't known there.
1742 * FIXME: redo bipred weight to not require extra buffer? */
1743 for (i = 0; i < h->slice_context_count; i++)
1744 if (h->thread_context[i]) {
1745 ret = alloc_scratch_buffers(h->thread_context[i], h->linesize);
1750 h->cur_pic_ptr->frame_num = h->frame_num; // FIXME frame_num cleanup
1752 av_assert1(h->mb_num == h->mb_width * h->mb_height);
1753 if (first_mb_in_slice << FIELD_OR_MBAFF_PICTURE(h) >= h->mb_num ||
1754 first_mb_in_slice >= h->mb_num) {
1755 av_log(h->avctx, AV_LOG_ERROR, "first_mb_in_slice overflow\n");
1756 return AVERROR_INVALIDDATA;
1758 h->resync_mb_x = h->mb_x = first_mb_in_slice % h->mb_width;
1759 h->resync_mb_y = h->mb_y = (first_mb_in_slice / h->mb_width) <<
1760 FIELD_OR_MBAFF_PICTURE(h);
1761 if (h->picture_structure == PICT_BOTTOM_FIELD)
1762 h->resync_mb_y = h->mb_y = h->mb_y + 1;
1763 av_assert1(h->mb_y < h->mb_height);
1765 if (h->picture_structure == PICT_FRAME) {
1766 h->curr_pic_num = h->frame_num;
1767 h->max_pic_num = 1 << h->sps.log2_max_frame_num;
1769 h->curr_pic_num = 2 * h->frame_num + 1;
1770 h->max_pic_num = 1 << (h->sps.log2_max_frame_num + 1);
1773 if (h->nal_unit_type == NAL_IDR_SLICE)
1774 get_ue_golomb(&h->gb); /* idr_pic_id */
1776 if (h->sps.poc_type == 0) {
1777 h->poc_lsb = get_bits(&h->gb, h->sps.log2_max_poc_lsb);
1779 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1780 h->delta_poc_bottom = get_se_golomb(&h->gb);
1783 if (h->sps.poc_type == 1 && !h->sps.delta_pic_order_always_zero_flag) {
1784 h->delta_poc[0] = get_se_golomb(&h->gb);
1786 if (h->pps.pic_order_present == 1 && h->picture_structure == PICT_FRAME)
1787 h->delta_poc[1] = get_se_golomb(&h->gb);
1790 ff_init_poc(h, h->cur_pic_ptr->field_poc, &h->cur_pic_ptr->poc);
1792 if (h->pps.redundant_pic_cnt_present)
1793 h->redundant_pic_count = get_ue_golomb(&h->gb);
1795 ret = ff_set_ref_count(h, sl);
1799 if (slice_type != AV_PICTURE_TYPE_I &&
1800 (h0->current_slice == 0 ||
1801 slice_type != h0->last_slice_type ||
1802 memcmp(h0->last_ref_count, sl->ref_count, sizeof(sl->ref_count)))) {
1804 ff_h264_fill_default_ref_list(h, sl);
1807 if (sl->slice_type_nos != AV_PICTURE_TYPE_I) {
1808 ret = ff_h264_decode_ref_pic_list_reordering(h, sl);
1810 sl->ref_count[1] = sl->ref_count[0] = 0;
1815 if ((h->pps.weighted_pred && sl->slice_type_nos == AV_PICTURE_TYPE_P) ||
1816 (h->pps.weighted_bipred_idc == 1 &&
1817 sl->slice_type_nos == AV_PICTURE_TYPE_B))
1818 ff_pred_weight_table(h, sl);
1819 else if (h->pps.weighted_bipred_idc == 2 &&
1820 sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1821 implicit_weight_table(h, sl, -1);
1824 for (i = 0; i < 2; i++) {
1825 sl->luma_weight_flag[i] = 0;
1826 sl->chroma_weight_flag[i] = 0;
1830 // If frame-mt is enabled, only update mmco tables for the first slice
1831 // in a field. Subsequent slices can temporarily clobber h->mmco_index
1832 // or h->mmco, which will cause ref list mix-ups and decoding errors
1833 // further down the line. This may break decoding if the first slice is
1834 // corrupt, thus we only do this if frame-mt is enabled.
1835 if (h->nal_ref_idc) {
1836 ret = ff_h264_decode_ref_pic_marking(h0, &h->gb,
1837 !(h->avctx->active_thread_type & FF_THREAD_FRAME) ||
1838 h0->current_slice == 0);
1839 if (ret < 0 && (h->avctx->err_recognition & AV_EF_EXPLODE))
1840 return AVERROR_INVALIDDATA;
1843 if (FRAME_MBAFF(h)) {
1844 ff_h264_fill_mbaff_ref_list(h, sl);
1846 if (h->pps.weighted_bipred_idc == 2 && sl->slice_type_nos == AV_PICTURE_TYPE_B) {
1847 implicit_weight_table(h, sl, 0);
1848 implicit_weight_table(h, sl, 1);
1852 if (sl->slice_type_nos == AV_PICTURE_TYPE_B && !sl->direct_spatial_mv_pred)
1853 ff_h264_direct_dist_scale_factor(h, sl);
1854 ff_h264_direct_ref_list_init(h, sl);
1856 if (sl->slice_type_nos != AV_PICTURE_TYPE_I && h->pps.cabac) {
1857 tmp = get_ue_golomb_31(&h->gb);
1859 av_log(h->avctx, AV_LOG_ERROR, "cabac_init_idc %u overflow\n", tmp);
1860 return AVERROR_INVALIDDATA;
1862 h->cabac_init_idc = tmp;
1865 sl->last_qscale_diff = 0;
1866 tmp = h->pps.init_qp + get_se_golomb(&h->gb);
1867 if (tmp > 51 + 6 * (h->sps.bit_depth_luma - 8)) {
1868 av_log(h->avctx, AV_LOG_ERROR, "QP %u out of range\n", tmp);
1869 return AVERROR_INVALIDDATA;
1872 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
1873 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
1874 // FIXME qscale / qp ... stuff
1875 if (sl->slice_type == AV_PICTURE_TYPE_SP)
1876 get_bits1(&h->gb); /* sp_for_switch_flag */
1877 if (sl->slice_type == AV_PICTURE_TYPE_SP ||
1878 sl->slice_type == AV_PICTURE_TYPE_SI)
1879 get_se_golomb(&h->gb); /* slice_qs_delta */
1881 h->deblocking_filter = 1;
1882 h->slice_alpha_c0_offset = 0;
1883 h->slice_beta_offset = 0;
1884 if (h->pps.deblocking_filter_parameters_present) {
1885 tmp = get_ue_golomb_31(&h->gb);
1887 av_log(h->avctx, AV_LOG_ERROR,
1888 "deblocking_filter_idc %u out of range\n", tmp);
1889 return AVERROR_INVALIDDATA;
1891 h->deblocking_filter = tmp;
1892 if (h->deblocking_filter < 2)
1893 h->deblocking_filter ^= 1; // 1<->0
1895 if (h->deblocking_filter) {
1896 h->slice_alpha_c0_offset = get_se_golomb(&h->gb) * 2;
1897 h->slice_beta_offset = get_se_golomb(&h->gb) * 2;
1898 if (h->slice_alpha_c0_offset > 12 ||
1899 h->slice_alpha_c0_offset < -12 ||
1900 h->slice_beta_offset > 12 ||
1901 h->slice_beta_offset < -12) {
1902 av_log(h->avctx, AV_LOG_ERROR,
1903 "deblocking filter parameters %d %d out of range\n",
1904 h->slice_alpha_c0_offset, h->slice_beta_offset);
1905 return AVERROR_INVALIDDATA;
1910 if (h->avctx->skip_loop_filter >= AVDISCARD_ALL ||
1911 (h->avctx->skip_loop_filter >= AVDISCARD_NONKEY &&
1912 h->nal_unit_type != NAL_IDR_SLICE) ||
1913 (h->avctx->skip_loop_filter >= AVDISCARD_NONINTRA &&
1914 sl->slice_type_nos != AV_PICTURE_TYPE_I) ||
1915 (h->avctx->skip_loop_filter >= AVDISCARD_BIDIR &&
1916 sl->slice_type_nos == AV_PICTURE_TYPE_B) ||
1917 (h->avctx->skip_loop_filter >= AVDISCARD_NONREF &&
1918 h->nal_ref_idc == 0))
1919 h->deblocking_filter = 0;
1921 if (h->deblocking_filter == 1 && h0->max_contexts > 1) {
1922 if (h->avctx->flags2 & CODEC_FLAG2_FAST) {
1923 /* Cheat slightly for speed:
1924 * Do not bother to deblock across slices. */
1925 h->deblocking_filter = 2;
1927 h0->max_contexts = 1;
1928 if (!h0->single_decode_warning) {
1929 av_log(h->avctx, AV_LOG_INFO,
1930 "Cannot parallelize slice decoding with deblocking filter type 1, decoding such frames in sequential order\n"
1931 "To parallelize slice decoding you need video encoded with disable_deblocking_filter_idc set to 2 (deblock only edges that do not cross slices).\n"
1932 "Setting the flags2 libavcodec option to +fast (-flags2 +fast) will disable deblocking across slices and enable parallel slice decoding "
1933 "but will generate non-standard-compliant output.\n");
1934 h0->single_decode_warning = 1;
1937 av_log(h->avctx, AV_LOG_ERROR,
1938 "Deblocking switched inside frame.\n");
1939 return SLICE_SINGLETHREAD;
1943 sl->qp_thresh = 15 -
1944 FFMIN(h->slice_alpha_c0_offset, h->slice_beta_offset) -
1946 h->pps.chroma_qp_index_offset[0],
1947 h->pps.chroma_qp_index_offset[1]) +
1948 6 * (h->sps.bit_depth_luma - 8);
1950 h0->last_slice_type = slice_type;
1951 memcpy(h0->last_ref_count, sl->ref_count, sizeof(h0->last_ref_count));
1952 sl->slice_num = ++h0->current_slice;
1955 h0->slice_row[(sl->slice_num-1)&(MAX_SLICES-1)]= h->resync_mb_y;
1956 if ( h0->slice_row[sl->slice_num&(MAX_SLICES-1)] + 3 >= h->resync_mb_y
1957 && h0->slice_row[sl->slice_num&(MAX_SLICES-1)] <= h->resync_mb_y
1958 && sl->slice_num >= MAX_SLICES) {
1959 //in case of ASO this check needs to be updated depending on how we decide to assign slice numbers in this case
1960 av_log(h->avctx, AV_LOG_WARNING, "Possibly too many slices (%d >= %d), increase MAX_SLICES and recompile if there are artifacts\n", sl->slice_num, MAX_SLICES);
1963 for (j = 0; j < 2; j++) {
1965 int *ref2frm = sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][j];
1966 for (i = 0; i < 16; i++) {
1968 if (j < sl->list_count && i < sl->ref_count[j] &&
1969 sl->ref_list[j][i].f.buf[0]) {
1971 AVBuffer *buf = sl->ref_list[j][i].f.buf[0]->buffer;
1972 for (k = 0; k < h->short_ref_count; k++)
1973 if (h->short_ref[k]->f.buf[0]->buffer == buf) {
1977 for (k = 0; k < h->long_ref_count; k++)
1978 if (h->long_ref[k] && h->long_ref[k]->f.buf[0]->buffer == buf) {
1979 id_list[i] = h->short_ref_count + k;
1987 for (i = 0; i < 16; i++)
1988 ref2frm[i + 2] = 4 * id_list[i] + (sl->ref_list[j][i].reference & 3);
1990 ref2frm[18 + 1] = -1;
1991 for (i = 16; i < 48; i++)
1992 ref2frm[i + 4] = 4 * id_list[(i - 16) >> 1] +
1993 (sl->ref_list[j][i].reference & 3);
1996 h0->au_pps_id = pps_id;
1998 h0->sps_buffers[h->pps.sps_id]->new = 0;
1999 h->current_sps_id = h->pps.sps_id;
2001 if (h->avctx->debug & FF_DEBUG_PICT_INFO) {
2002 av_log(h->avctx, AV_LOG_DEBUG,
2003 "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",
2005 (h->picture_structure == PICT_FRAME ? "F" : h->picture_structure == PICT_TOP_FIELD ? "T" : "B"),
2007 av_get_picture_type_char(sl->slice_type),
2008 sl->slice_type_fixed ? " fix" : "",
2009 h->nal_unit_type == NAL_IDR_SLICE ? " IDR" : "",
2010 pps_id, h->frame_num,
2011 h->cur_pic_ptr->field_poc[0],
2012 h->cur_pic_ptr->field_poc[1],
2013 sl->ref_count[0], sl->ref_count[1],
2015 h->deblocking_filter,
2016 h->slice_alpha_c0_offset, h->slice_beta_offset,
2018 sl->use_weight == 1 && sl->use_weight_chroma ? "c" : "",
2019 sl->slice_type == AV_PICTURE_TYPE_B ? (sl->direct_spatial_mv_pred ? "SPAT" : "TEMP") : "");
2025 int ff_h264_get_slice_type(const H264SliceContext *sl)
2027 switch (sl->slice_type) {
2028 case AV_PICTURE_TYPE_P:
2030 case AV_PICTURE_TYPE_B:
2032 case AV_PICTURE_TYPE_I:
2034 case AV_PICTURE_TYPE_SP:
2036 case AV_PICTURE_TYPE_SI:
2039 return AVERROR_INVALIDDATA;
2043 static av_always_inline void fill_filter_caches_inter(H264Context *h,
2044 H264SliceContext *sl,
2045 int mb_type, int top_xy,
2046 int left_xy[LEFT_MBS],
2048 int left_type[LEFT_MBS],
2049 int mb_xy, int list)
2051 int b_stride = h->b_stride;
2052 int16_t(*mv_dst)[2] = &sl->mv_cache[list][scan8[0]];
2053 int8_t *ref_cache = &sl->ref_cache[list][scan8[0]];
2054 if (IS_INTER(mb_type) || IS_DIRECT(mb_type)) {
2055 if (USES_LIST(top_type, list)) {
2056 const int b_xy = h->mb2b_xy[top_xy] + 3 * b_stride;
2057 const int b8_xy = 4 * top_xy + 2;
2058 int (*ref2frm)[64] = (void*)(sl->ref2frm[h->slice_table[top_xy] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2059 AV_COPY128(mv_dst - 1 * 8, h->cur_pic.motion_val[list][b_xy + 0]);
2060 ref_cache[0 - 1 * 8] =
2061 ref_cache[1 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 0]];
2062 ref_cache[2 - 1 * 8] =
2063 ref_cache[3 - 1 * 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 1]];
2065 AV_ZERO128(mv_dst - 1 * 8);
2066 AV_WN32A(&ref_cache[0 - 1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2069 if (!IS_INTERLACED(mb_type ^ left_type[LTOP])) {
2070 if (USES_LIST(left_type[LTOP], list)) {
2071 const int b_xy = h->mb2b_xy[left_xy[LTOP]] + 3;
2072 const int b8_xy = 4 * left_xy[LTOP] + 1;
2073 int (*ref2frm)[64] =(void*)( sl->ref2frm[h->slice_table[left_xy[LTOP]] & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2074 AV_COPY32(mv_dst - 1 + 0, h->cur_pic.motion_val[list][b_xy + b_stride * 0]);
2075 AV_COPY32(mv_dst - 1 + 8, h->cur_pic.motion_val[list][b_xy + b_stride * 1]);
2076 AV_COPY32(mv_dst - 1 + 16, h->cur_pic.motion_val[list][b_xy + b_stride * 2]);
2077 AV_COPY32(mv_dst - 1 + 24, h->cur_pic.motion_val[list][b_xy + b_stride * 3]);
2079 ref_cache[-1 + 8] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 0]];
2080 ref_cache[-1 + 16] =
2081 ref_cache[-1 + 24] = ref2frm[list][h->cur_pic.ref_index[list][b8_xy + 2 * 1]];
2083 AV_ZERO32(mv_dst - 1 + 0);
2084 AV_ZERO32(mv_dst - 1 + 8);
2085 AV_ZERO32(mv_dst - 1 + 16);
2086 AV_ZERO32(mv_dst - 1 + 24);
2089 ref_cache[-1 + 16] =
2090 ref_cache[-1 + 24] = LIST_NOT_USED;
2095 if (!USES_LIST(mb_type, list)) {
2096 fill_rectangle(mv_dst, 4, 4, 8, pack16to32(0, 0), 4);
2097 AV_WN32A(&ref_cache[0 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2098 AV_WN32A(&ref_cache[1 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2099 AV_WN32A(&ref_cache[2 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2100 AV_WN32A(&ref_cache[3 * 8], ((LIST_NOT_USED) & 0xFF) * 0x01010101u);
2105 int8_t *ref = &h->cur_pic.ref_index[list][4 * mb_xy];
2106 int (*ref2frm)[64] = (void*)(sl->ref2frm[sl->slice_num & (MAX_SLICES - 1)][0] + (MB_MBAFF(h) ? 20 : 2));
2107 uint32_t ref01 = (pack16to32(ref2frm[list][ref[0]], ref2frm[list][ref[1]]) & 0x00FF00FF) * 0x0101;
2108 uint32_t ref23 = (pack16to32(ref2frm[list][ref[2]], ref2frm[list][ref[3]]) & 0x00FF00FF) * 0x0101;
2109 AV_WN32A(&ref_cache[0 * 8], ref01);
2110 AV_WN32A(&ref_cache[1 * 8], ref01);
2111 AV_WN32A(&ref_cache[2 * 8], ref23);
2112 AV_WN32A(&ref_cache[3 * 8], ref23);
2116 int16_t(*mv_src)[2] = &h->cur_pic.motion_val[list][4 * h->mb_x + 4 * h->mb_y * b_stride];
2117 AV_COPY128(mv_dst + 8 * 0, mv_src + 0 * b_stride);
2118 AV_COPY128(mv_dst + 8 * 1, mv_src + 1 * b_stride);
2119 AV_COPY128(mv_dst + 8 * 2, mv_src + 2 * b_stride);
2120 AV_COPY128(mv_dst + 8 * 3, mv_src + 3 * b_stride);
2126 * @return non zero if the loop filter can be skipped
2128 static int fill_filter_caches(H264Context *h, H264SliceContext *sl, int mb_type)
2130 const int mb_xy = h->mb_xy;
2131 int top_xy, left_xy[LEFT_MBS];
2132 int top_type, left_type[LEFT_MBS];
2136 top_xy = mb_xy - (h->mb_stride << MB_FIELD(h));
2138 /* Wow, what a mess, why didn't they simplify the interlacing & intra
2139 * stuff, I can't imagine that these complex rules are worth it. */
2141 left_xy[LBOT] = left_xy[LTOP] = mb_xy - 1;
2142 if (FRAME_MBAFF(h)) {
2143 const int left_mb_field_flag = IS_INTERLACED(h->cur_pic.mb_type[mb_xy - 1]);
2144 const int curr_mb_field_flag = IS_INTERLACED(mb_type);
2146 if (left_mb_field_flag != curr_mb_field_flag)
2147 left_xy[LTOP] -= h->mb_stride;
2149 if (curr_mb_field_flag)
2150 top_xy += h->mb_stride &
2151 (((h->cur_pic.mb_type[top_xy] >> 7) & 1) - 1);
2152 if (left_mb_field_flag != curr_mb_field_flag)
2153 left_xy[LBOT] += h->mb_stride;
2157 sl->top_mb_xy = top_xy;
2158 sl->left_mb_xy[LTOP] = left_xy[LTOP];
2159 sl->left_mb_xy[LBOT] = left_xy[LBOT];
2161 /* For sufficiently low qp, filtering wouldn't do anything.
2162 * This is a conservative estimate: could also check beta_offset
2163 * and more accurate chroma_qp. */
2164 int qp_thresh = sl->qp_thresh; // FIXME strictly we should store qp_thresh for each mb of a slice
2165 int qp = h->cur_pic.qscale_table[mb_xy];
2166 if (qp <= qp_thresh &&
2167 (left_xy[LTOP] < 0 ||
2168 ((qp + h->cur_pic.qscale_table[left_xy[LTOP]] + 1) >> 1) <= qp_thresh) &&
2170 ((qp + h->cur_pic.qscale_table[top_xy] + 1) >> 1) <= qp_thresh)) {
2171 if (!FRAME_MBAFF(h))
2173 if ((left_xy[LTOP] < 0 ||
2174 ((qp + h->cur_pic.qscale_table[left_xy[LBOT]] + 1) >> 1) <= qp_thresh) &&
2175 (top_xy < h->mb_stride ||
2176 ((qp + h->cur_pic.qscale_table[top_xy - h->mb_stride] + 1) >> 1) <= qp_thresh))
2181 top_type = h->cur_pic.mb_type[top_xy];
2182 left_type[LTOP] = h->cur_pic.mb_type[left_xy[LTOP]];
2183 left_type[LBOT] = h->cur_pic.mb_type[left_xy[LBOT]];
2184 if (h->deblocking_filter == 2) {
2185 if (h->slice_table[top_xy] != sl->slice_num)
2187 if (h->slice_table[left_xy[LBOT]] != sl->slice_num)
2188 left_type[LTOP] = left_type[LBOT] = 0;
2190 if (h->slice_table[top_xy] == 0xFFFF)
2192 if (h->slice_table[left_xy[LBOT]] == 0xFFFF)
2193 left_type[LTOP] = left_type[LBOT] = 0;
2195 sl->top_type = top_type;
2196 sl->left_type[LTOP] = left_type[LTOP];
2197 sl->left_type[LBOT] = left_type[LBOT];
2199 if (IS_INTRA(mb_type))
2202 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2203 top_type, left_type, mb_xy, 0);
2204 if (sl->list_count == 2)
2205 fill_filter_caches_inter(h, sl, mb_type, top_xy, left_xy,
2206 top_type, left_type, mb_xy, 1);
2208 nnz = h->non_zero_count[mb_xy];
2209 nnz_cache = sl->non_zero_count_cache;
2210 AV_COPY32(&nnz_cache[4 + 8 * 1], &nnz[0]);
2211 AV_COPY32(&nnz_cache[4 + 8 * 2], &nnz[4]);
2212 AV_COPY32(&nnz_cache[4 + 8 * 3], &nnz[8]);
2213 AV_COPY32(&nnz_cache[4 + 8 * 4], &nnz[12]);
2214 sl->cbp = h->cbp_table[mb_xy];
2217 nnz = h->non_zero_count[top_xy];
2218 AV_COPY32(&nnz_cache[4 + 8 * 0], &nnz[3 * 4]);
2221 if (left_type[LTOP]) {
2222 nnz = h->non_zero_count[left_xy[LTOP]];
2223 nnz_cache[3 + 8 * 1] = nnz[3 + 0 * 4];
2224 nnz_cache[3 + 8 * 2] = nnz[3 + 1 * 4];
2225 nnz_cache[3 + 8 * 3] = nnz[3 + 2 * 4];
2226 nnz_cache[3 + 8 * 4] = nnz[3 + 3 * 4];
2229 /* CAVLC 8x8dct requires NNZ values for residual decoding that differ
2230 * from what the loop filter needs */
2231 if (!CABAC(h) && h->pps.transform_8x8_mode) {
2232 if (IS_8x8DCT(top_type)) {
2233 nnz_cache[4 + 8 * 0] =
2234 nnz_cache[5 + 8 * 0] = (h->cbp_table[top_xy] & 0x4000) >> 12;
2235 nnz_cache[6 + 8 * 0] =
2236 nnz_cache[7 + 8 * 0] = (h->cbp_table[top_xy] & 0x8000) >> 12;
2238 if (IS_8x8DCT(left_type[LTOP])) {
2239 nnz_cache[3 + 8 * 1] =
2240 nnz_cache[3 + 8 * 2] = (h->cbp_table[left_xy[LTOP]] & 0x2000) >> 12; // FIXME check MBAFF
2242 if (IS_8x8DCT(left_type[LBOT])) {
2243 nnz_cache[3 + 8 * 3] =
2244 nnz_cache[3 + 8 * 4] = (h->cbp_table[left_xy[LBOT]] & 0x8000) >> 12; // FIXME check MBAFF
2247 if (IS_8x8DCT(mb_type)) {
2248 nnz_cache[scan8[0]] =
2249 nnz_cache[scan8[1]] =
2250 nnz_cache[scan8[2]] =
2251 nnz_cache[scan8[3]] = (sl->cbp & 0x1000) >> 12;
2253 nnz_cache[scan8[0 + 4]] =
2254 nnz_cache[scan8[1 + 4]] =
2255 nnz_cache[scan8[2 + 4]] =
2256 nnz_cache[scan8[3 + 4]] = (sl->cbp & 0x2000) >> 12;
2258 nnz_cache[scan8[0 + 8]] =
2259 nnz_cache[scan8[1 + 8]] =
2260 nnz_cache[scan8[2 + 8]] =
2261 nnz_cache[scan8[3 + 8]] = (sl->cbp & 0x4000) >> 12;
2263 nnz_cache[scan8[0 + 12]] =
2264 nnz_cache[scan8[1 + 12]] =
2265 nnz_cache[scan8[2 + 12]] =
2266 nnz_cache[scan8[3 + 12]] = (sl->cbp & 0x8000) >> 12;
2273 static void loop_filter(H264Context *h, H264SliceContext *sl, int start_x, int end_x)
2275 uint8_t *dest_y, *dest_cb, *dest_cr;
2276 int linesize, uvlinesize, mb_x, mb_y;
2277 const int end_mb_y = h->mb_y + FRAME_MBAFF(h);
2278 const int old_slice_type = sl->slice_type;
2279 const int pixel_shift = h->pixel_shift;
2280 const int block_h = 16 >> h->chroma_y_shift;
2282 if (h->deblocking_filter) {
2283 for (mb_x = start_x; mb_x < end_x; mb_x++)
2284 for (mb_y = end_mb_y - FRAME_MBAFF(h); mb_y <= end_mb_y; mb_y++) {
2286 mb_xy = h->mb_xy = mb_x + mb_y * h->mb_stride;
2287 sl->slice_num = h->slice_table[mb_xy];
2288 mb_type = h->cur_pic.mb_type[mb_xy];
2289 sl->list_count = h->list_counts[mb_xy];
2293 h->mb_field_decoding_flag = !!IS_INTERLACED(mb_type);
2297 dest_y = h->cur_pic.f.data[0] +
2298 ((mb_x << pixel_shift) + mb_y * h->linesize) * 16;
2299 dest_cb = h->cur_pic.f.data[1] +
2300 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2301 mb_y * h->uvlinesize * block_h;
2302 dest_cr = h->cur_pic.f.data[2] +
2303 (mb_x << pixel_shift) * (8 << CHROMA444(h)) +
2304 mb_y * h->uvlinesize * block_h;
2305 // FIXME simplify above
2308 linesize = sl->mb_linesize = h->linesize * 2;
2309 uvlinesize = sl->mb_uvlinesize = h->uvlinesize * 2;
2310 if (mb_y & 1) { // FIXME move out of this function?
2311 dest_y -= h->linesize * 15;
2312 dest_cb -= h->uvlinesize * (block_h - 1);
2313 dest_cr -= h->uvlinesize * (block_h - 1);
2316 linesize = sl->mb_linesize = h->linesize;
2317 uvlinesize = sl->mb_uvlinesize = h->uvlinesize;
2319 backup_mb_border(h, dest_y, dest_cb, dest_cr, linesize,
2321 if (fill_filter_caches(h, sl, mb_type))
2323 sl->chroma_qp[0] = get_chroma_qp(h, 0, h->cur_pic.qscale_table[mb_xy]);
2324 sl->chroma_qp[1] = get_chroma_qp(h, 1, h->cur_pic.qscale_table[mb_xy]);
2326 if (FRAME_MBAFF(h)) {
2327 ff_h264_filter_mb(h, sl, mb_x, mb_y, dest_y, dest_cb, dest_cr,
2328 linesize, uvlinesize);
2330 ff_h264_filter_mb_fast(h, sl, mb_x, mb_y, dest_y, dest_cb,
2331 dest_cr, linesize, uvlinesize);
2335 sl->slice_type = old_slice_type;
2337 h->mb_y = end_mb_y - FRAME_MBAFF(h);
2338 sl->chroma_qp[0] = get_chroma_qp(h, 0, sl->qscale);
2339 sl->chroma_qp[1] = get_chroma_qp(h, 1, sl->qscale);
2342 static void predict_field_decoding_flag(H264Context *h, H264SliceContext *sl)
2344 const int mb_xy = h->mb_x + h->mb_y * h->mb_stride;
2345 int mb_type = (h->slice_table[mb_xy - 1] == sl->slice_num) ?
2346 h->cur_pic.mb_type[mb_xy - 1] :
2347 (h->slice_table[mb_xy - h->mb_stride] == sl->slice_num) ?
2348 h->cur_pic.mb_type[mb_xy - h->mb_stride] : 0;
2349 h->mb_mbaff = h->mb_field_decoding_flag = IS_INTERLACED(mb_type) ? 1 : 0;
2353 * Draw edges and report progress for the last MB row.
2355 static void decode_finish_row(H264Context *h, H264SliceContext *sl)
2357 int top = 16 * (h->mb_y >> FIELD_PICTURE(h));
2358 int pic_height = 16 * h->mb_height >> FIELD_PICTURE(h);
2359 int height = 16 << FRAME_MBAFF(h);
2360 int deblock_border = (16 + 4) << FRAME_MBAFF(h);
2362 if (h->deblocking_filter) {
2363 if ((top + height) >= pic_height)
2364 height += deblock_border;
2365 top -= deblock_border;
2368 if (top >= pic_height || (top + height) < 0)
2371 height = FFMIN(height, pic_height - top);
2373 height = top + height;
2377 ff_h264_draw_horiz_band(h, sl, top, height);
2379 if (h->droppable || h->er.error_occurred)
2382 ff_thread_report_progress(&h->cur_pic_ptr->tf, top + height - 1,
2383 h->picture_structure == PICT_BOTTOM_FIELD);
2386 static void er_add_slice(H264Context *h, H264SliceContext *sl,
2387 int startx, int starty,
2388 int endx, int endy, int status)
2390 if (CONFIG_ERROR_RESILIENCE) {
2391 ERContext *er = &h->er;
2393 ff_er_add_slice(er, startx, starty, endx, endy, status);
2397 static int decode_slice(struct AVCodecContext *avctx, void *arg)
2399 H264SliceContext *sl = arg;
2400 H264Context *h = sl->h264;
2401 int lf_x_start = h->mb_x;
2403 h->mb_skip_run = -1;
2405 av_assert0(h->block_offset[15] == (4 * ((scan8[15] - scan8[0]) & 7) << h->pixel_shift) + 4 * h->linesize * ((scan8[15] - scan8[0]) >> 3));
2407 h->is_complex = FRAME_MBAFF(h) || h->picture_structure != PICT_FRAME ||
2408 avctx->codec_id != AV_CODEC_ID_H264 ||
2409 (CONFIG_GRAY && (h->flags & CODEC_FLAG_GRAY));
2411 if (!(h->avctx->active_thread_type & FF_THREAD_SLICE) && h->picture_structure == PICT_FRAME && h->er.error_status_table) {
2412 const int start_i = av_clip(h->resync_mb_x + h->resync_mb_y * h->mb_width, 0, h->mb_num - 1);
2414 int prev_status = h->er.error_status_table[h->er.mb_index2xy[start_i - 1]];
2415 prev_status &= ~ VP_START;
2416 if (prev_status != (ER_MV_END | ER_DC_END | ER_AC_END))
2417 h->er.error_occurred = 1;
2423 align_get_bits(&h->gb);
2426 ff_init_cabac_decoder(&sl->cabac,
2427 h->gb.buffer + get_bits_count(&h->gb) / 8,
2428 (get_bits_left(&h->gb) + 7) / 8);
2430 ff_h264_init_cabac_states(h, sl);
2434 int ret = ff_h264_decode_mb_cabac(h, sl);
2436 // STOP_TIMER("decode_mb_cabac")
2439 ff_h264_hl_decode_mb(h, sl);
2441 // FIXME optimal? or let mb_decode decode 16x32 ?
2442 if (ret >= 0 && FRAME_MBAFF(h)) {
2445 ret = ff_h264_decode_mb_cabac(h, sl);
2448 ff_h264_hl_decode_mb(h, sl);
2451 eos = get_cabac_terminate(&sl->cabac);
2453 if ((h->workaround_bugs & FF_BUG_TRUNCATED) &&
2454 sl->cabac.bytestream > sl->cabac.bytestream_end + 2) {
2455 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2456 h->mb_y, ER_MB_END);
2457 if (h->mb_x >= lf_x_start)
2458 loop_filter(h, sl, lf_x_start, h->mb_x + 1);
2461 if (sl->cabac.bytestream > sl->cabac.bytestream_end + 2 )
2462 av_log(h->avctx, AV_LOG_DEBUG, "bytestream overread %"PTRDIFF_SPECIFIER"\n", sl->cabac.bytestream_end - sl->cabac.bytestream);
2463 if (ret < 0 || sl->cabac.bytestream > sl->cabac.bytestream_end + 4) {
2464 av_log(h->avctx, AV_LOG_ERROR,
2465 "error while decoding MB %d %d, bytestream %"PTRDIFF_SPECIFIER"\n",
2467 sl->cabac.bytestream_end - sl->cabac.bytestream);
2468 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2469 h->mb_y, ER_MB_ERROR);
2470 return AVERROR_INVALIDDATA;
2473 if (++h->mb_x >= h->mb_width) {
2474 loop_filter(h, sl, lf_x_start, h->mb_x);
2475 h->mb_x = lf_x_start = 0;
2476 decode_finish_row(h, sl);
2478 if (FIELD_OR_MBAFF_PICTURE(h)) {
2480 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2481 predict_field_decoding_flag(h, sl);
2485 if (eos || h->mb_y >= h->mb_height) {
2486 tprintf(h->avctx, "slice end %d %d\n",
2487 get_bits_count(&h->gb), h->gb.size_in_bits);
2488 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x - 1,
2489 h->mb_y, ER_MB_END);
2490 if (h->mb_x > lf_x_start)
2491 loop_filter(h, sl, lf_x_start, h->mb_x);
2497 int ret = ff_h264_decode_mb_cavlc(h, sl);
2500 ff_h264_hl_decode_mb(h, sl);
2502 // FIXME optimal? or let mb_decode decode 16x32 ?
2503 if (ret >= 0 && FRAME_MBAFF(h)) {
2505 ret = ff_h264_decode_mb_cavlc(h, sl);
2508 ff_h264_hl_decode_mb(h, sl);
2513 av_log(h->avctx, AV_LOG_ERROR,
2514 "error while decoding MB %d %d\n", h->mb_x, h->mb_y);
2515 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2516 h->mb_y, ER_MB_ERROR);
2520 if (++h->mb_x >= h->mb_width) {
2521 loop_filter(h, sl, lf_x_start, h->mb_x);
2522 h->mb_x = lf_x_start = 0;
2523 decode_finish_row(h, sl);
2525 if (FIELD_OR_MBAFF_PICTURE(h)) {
2527 if (FRAME_MBAFF(h) && h->mb_y < h->mb_height)
2528 predict_field_decoding_flag(h, sl);
2530 if (h->mb_y >= h->mb_height) {
2531 tprintf(h->avctx, "slice end %d %d\n",
2532 get_bits_count(&h->gb), h->gb.size_in_bits);
2534 if ( get_bits_left(&h->gb) == 0
2535 || get_bits_left(&h->gb) > 0 && !(h->avctx->err_recognition & AV_EF_AGGRESSIVE)) {
2536 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y,
2537 h->mb_x - 1, h->mb_y, ER_MB_END);
2541 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y,
2542 h->mb_x, h->mb_y, ER_MB_END);
2544 return AVERROR_INVALIDDATA;
2549 if (get_bits_left(&h->gb) <= 0 && h->mb_skip_run <= 0) {
2550 tprintf(h->avctx, "slice end %d %d\n",
2551 get_bits_count(&h->gb), h->gb.size_in_bits);
2553 if (get_bits_left(&h->gb) == 0) {
2554 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y,
2555 h->mb_x - 1, h->mb_y, ER_MB_END);
2556 if (h->mb_x > lf_x_start)
2557 loop_filter(h, sl, lf_x_start, h->mb_x);
2561 er_add_slice(h, sl, h->resync_mb_x, h->resync_mb_y, h->mb_x,
2562 h->mb_y, ER_MB_ERROR);
2564 return AVERROR_INVALIDDATA;
2572 * Call decode_slice() for each context.
2574 * @param h h264 master context
2575 * @param context_count number of contexts to execute
2577 int ff_h264_execute_decode_slices(H264Context *h, unsigned context_count)
2579 AVCodecContext *const avctx = h->avctx;
2583 av_assert0(h->mb_y < h->mb_height);
2585 if (h->avctx->hwaccel ||
2586 h->avctx->codec->capabilities & CODEC_CAP_HWACCEL_VDPAU)
2588 if (context_count == 1) {
2589 return decode_slice(avctx, &h->slice_ctx[0]);
2591 av_assert0(context_count > 0);
2592 for (i = 1; i < context_count; i++) {
2593 hx = h->thread_context[i];
2594 if (CONFIG_ERROR_RESILIENCE) {
2595 hx->er.error_count = 0;
2597 hx->x264_build = h->x264_build;
2600 avctx->execute(avctx, decode_slice, h->slice_ctx,
2601 NULL, context_count, sizeof(h->slice_ctx[0]));
2603 /* pull back stuff from slices to master context */
2604 hx = h->thread_context[context_count - 1];
2607 h->droppable = hx->droppable;
2608 h->picture_structure = hx->picture_structure;
2609 if (CONFIG_ERROR_RESILIENCE) {
2610 for (i = 1; i < context_count; i++)
2611 h->er.error_count += h->thread_context[i]->er.error_count;